The way I see it, artificial intelligence (or AI), really leapt into the zeitgeist in late-2022 or early-2023 with the public introduction of DALL-E2 and ChatGPT. Both are provided by OpenAI and are software products that use AI to generate art and writing, respectively (and often at astounding quality). Since then, developments in AI have progressed at a breathtaking pace.

We’re thick in the action of the latest earnings season for the US stock market – for the first quarter of 2025 – and I thought it would be useful to collate some of the interesting commentary I’ve come across in earnings conference calls, from the leaders of technology companies that I follow or have a vested interest in, on the topic of AI and how the technology could impact their industry and the business world writ large. This is an ongoing series. For the older commentary:

• 2023 Q1 – here and here
• 2023 Q2 – here and here
• 2023 Q3 – here and here
• 2023 Q4 – here and here
• 2024 Q1 – here and here
• 2024 Q2 – here and here
• 2024 Q3 – here and here
• 2024 Q4 – here, here, and here

With that, here are the latest commentary, in no particular order:

Airbnb (NASDAQ: ABNB)

Airbnb’s management thinks that designing end-to-end travel is very difficult and travelers often find planning travel to be very complicated, so travelers do it very infrequently; management thinks that a great user interface is the key to designing a great end-to-end travel experience for Airbnb users, and AI will be an important way to do it

I think a lot of companies have tried to like design an end-to-end travel. I think designing end-to-end travel is very, very hard. It’s funny — there’s this funny thing. One of the most common start-up ideas for entrepreneurs is to do a travel planning app. And yet travel planning apps almost always fail. So it’s almost like a riddle, why do travel planning apps fail, and everyone really tries to do it? And the reason why is because to plan travel is very complicated. In fact, it’s so complicated many people have assistants and a big part of their job is to plan travel for them. And yet you use it infrequently. So it’s a very difficult thing to do and you do it infrequently. And so therefore, a lot of companies have failed to design like a so-called connected trip. So I think to do this, a lot of it is to design a really good user experience. And I think that’s one of the things that we’re going to try to do to really design a great end-to-end experience, to able to book your entire trip, and much more. I think the user interface will be important. I think AI will be an important way to do this as well…

…We’re focused on making everything instant book and easy to use. We’re trying to make sure that the end-to-end travel experience is really, really wonderful with great Airbnb design, and we’re going to bring more AI into the application so that Airbnb, you can really solve your own problems with great self-solve through AI customer service agents.

Airbnb’s management recently rolled out an AI customer service agent; 50% of Airbnb’s US users are already using the customer service agent and it will soon be rolled out to 100% of Airbnb’s US users; management thinks Airbnb’s AI customer service agent is the best of its kind in travel, having already led to a 15% reduction in users needing to contact human agents; the AI customer service agent will be more personalised and agentic in the years ahead

We just rolled out our AI customer service agent this past month. 50% U.S. users are now using the agent, and we’ll roll it out to 100% of U.S. users this month. We believe this is the best AI-supported customers travel agent in travel. It’s already led to a 15% reduction in people needing to contact live human agents and it’s going to get significantly more personalized and agentic over the years to come.

Alphabet (NASDAQ: GOOG)

AI Overviews in Search now has more than 1.5 billion monthly users; AI Mode has received early positive reaction; usage growth of AI Overviews continues to increase nearly a year after its launch; management is leaning heavily into AI Overviews; management released the AI Mode in March as an experiment; AI Mode searches are twice as long as traditional search queries; AI Mode is getting really positive feedback from early users; the volume of commercial queries on Google Search has increased with the launch of AI Overviews; AI Overviews is now available in 15 languages and 140 countries; AI Overviews continues to monetise at a similar rate to traditional Search; reminder that ads within AI Overviews was launched in mobile in the USA in late-2024; an example of longer search queries in AI Mode is product comparisons; management thinks AI Overviews in Search and Gemini as 2 distinct consumer experiences; management thinks of AI Mode as a way to discover how the most advanced users are using AI-powered search

AI Overviews is going very well with over 1.5 billion users per month, and we are excited by the early positive reaction to AI Mode…

…Nearly a year after we launched AI Overviews in the U.S., we continue to see that usage growth is increasing as people learn that Search is more useful for more of their queries. So we are leaning in heavily here, continuing to roll the feature out in new countries to more users and to more queries. Building on the positive feedback for AI Overviews, in March, we released AI Mode, an experiment in labs. It expands what AI Overviews can do with more advanced reasoning, thinking and multimodal capabilities to help with questions that need further exploration and comparisons. On average, AI Mode queries are twice as long as traditional search queries. We’re getting really positive feedback from early users about its design, fast response time and ability to understand complex, nuanced questions…

…As we’ve mentioned before, with the launch of AI Overviews, the volume of commercial queries has increased. Q1 marked our largest expansion to date for AI Overviews, both in terms of launching to new users and providing responses for more questions. The feature is now available in more than 15 languages across 140 countries. For AI Overviews, overall, we continue to see monetization at approximately the same rate, which gives us a strong base in which we can innovate even more…

…On the ads of — in AI Overviews, last — late last year, actually, we launched them within the AI Overviews on mobile in the U.S. And this builds on our previous rollout of ads above and below. So this was a change that we have…

…I mentioned people typing in longer queries. There’s a lot more complex, nuanced questions. People are following through more. People are appreciating the clean design, the fast response time and the fact that they can kind of be much more open-ended, can undertake more complicated tasks. Product comparisons, for example, has been a positive one, exploring how tos, planning a trip…

…On AI-powered search and how do we see our consumer experience. Look, I do think Search and Gemini, obviously, will be 2 distinct efforts, right? I think there are obviously some areas of overlap, but they’re also — like expose very, very different use cases. And so for example, in Gemini, we see people iteratively coding and going much deeper on a coding workflow, as an example. So I think both will be around…

…AI Mode is the tip of the tree for us pushing forward on an AI-forward experience. There will be things which we discover there, which will make sense in the context of AI Overviews, so I think will flow through to our user base. But you almost want to think of what are the most advanced 1 million people using Search for, the most advanced 10 million people and then how do 1.5 billion people use Search for.

Alphabet’s management rolled out Alphabet’s latest foundation model, Gemini 2.5, in 2025 Q1; Gemini 2.5 is widely recognised as the best model in the industry; Gemini 2.5 Pro debuted at No.1 on the Chatbot Arena in 2025 Q1 by a significant margin; activer users in AI Studio and Gemini API is up 200% since the start of 2025; Alphabet introduced Gemini 2.5 Flash in April 2025; Gemini models are now found in all of Alphabet’s 15 products with at least 0.5 billion users each; Alphabet is upgrading Google Assistant on mobile devices to Gemini, and will also upgrade tablets, cars, and devices that connect to phones later this year; the Pixel 9a phone with Gemini integration was launched to strong reviews; the Gemini Live camera feature, among others, will soon be rolled out to all Android devices

This quarter was super exciting as we rolled out Gemini 2.5, our most intelligent AI model, which is achieving breakthroughs in performance, and it’s widely recognized as the best model in the industry…

…We released Gemini 2.5 Pro last month, receiving extremely positive feedback from both developers and consumers. 2.5 Pro is state-of-the-art on a wide range of benchmarks and debuted at #1 on the Chatbot Arena by a significant margin. 2.5 Pro achieved big leaps in reasoning, coding, science and math capabilities, opening up new possibilities for developers and customers. Active users in AI Studio and Gemini API have grown over 200% since the beginning of the year…

…Last week, we introduced 2.5 Flash, which enables developers to optimize quality and cost…

…All 15 of our products with 0.5 billion users now use Gemini models…

…We are upgrading Google Assistant on mobile devices to Gemini. And later this year, we’ll upgrade tablets, cars and devices that connect to your phones such as headphones and watches. The Pixel 9a launched very strong reviews, providing the best of Google’s AI offerings like Gemini Live and AI-powered camera features. And Gemini Live camera and screen sharing is now rolling out to all Android devices, including Pixel and Samsung S25.

Google Cloud is offering the industry’s widest range of TPUs and GPUs; Alphabet’s 7th generation TPU, Ironwood, has 10x better compute power and 2x better power efficiency than the previous generation TPU; Google Cloud is the first cloud provider to offer NVIDIA’s Blackwell family of GPUs; Google Cloud will be offering NVIDIA’s upcoming Rubin family of GPUs

Complementing this, we offer the industry’s widest range of TPUs and GPUs and continue to invest in next-generation capabilities. Ironwood, our seventh-generation TPU and most powerful to date, is the first designed specifically for inference at scale. It delivers more than 10x improvement in compute power or a recent high-performance TPU while being nearly twice as power efficient. Our strong relationship with NVIDIA continues to be a key advantage for us and our customers. We were the first cloud provider to offer NVIDIA’s groundbreaking B200 and GB200 Blackwell GPUs, and we’ll be offering their next-generation Vera Rubin GPUs.

Alphabet’s management is rolling out the company’s latest image and video generation models; Alphabet has launched its open-sourced Gemma 3 model in March 2025; Gemma models have been downloaded more than 140 million times; Alphabet is developing robotics AI models; Alphabet has launched a multi-agent AI research system called AI Co-Scientist; the AlphaFold model has been used by more than 2.5 million researchers

Our latest image and video generation models, Imagen 3 and Veo 2, are rolling out broadly and are powering incredible creativity. Turning to open models. We launched Gemma 3 last month, delivering state-of-the-art performance for its size. Gemma models have been downloaded more than 140 million times. Lastly, we are developing AI models in new areas where there’s enormous opportunity, for example, our new Gemini Robotics models. And in health, we launched AI Co-Scientist, a multi-agent AI research system, while AlphaFold has now been used by over 2.5 million researchers.

Google Cloud’s AI developer platform, Vertex AI, now has more than 200 foundation models available, including Alphabet’s in-house models and third-party models

Our Vertex AI platform makes over 200 foundation models available, helping customers like Lowe’s integrate AI. We offer industry-leading models, including Gemini 2.5 Pro, 2.5 Flash, Imagen 3, Veo 2, Chirp and Lyria, plus open-source and third-party models like Llama 4 and Anthropic.

Google Cloud is the leading cloud platform for building AI agents; Google Cloud has an open source framework for building AI agents and multi-agent systems called Agent Development Kit; Google Cloud has a low-code agent-building tool called Agent Designer; KPMG is using Google Cloud to deploy AI agents to employees; Google Cloud has the Google Agentspace product that helps employees in organisations use AI agents widely; Google Cloud offers pre-packaged AI agents across various functions including coding and customer engagement; Alphabet is working on agentic experiences internally and deploying it across the company; Alphabet’s customer service teams have deployed AI agents to dramatically enhance the user experience and is teaching Google Cloud customers how to do so

We are the leading cloud solution for companies looking to the new era of AI agents, a big opportunity. Our Agent Development Kit is a new open-source framework to simplify the process of building sophisticated AI agents and multi-agent systems. And Agent Designer is a low-code tool to build AI agents and automate tasks in over 100 enterprise applications and systems.

We are putting AI agents in the hands of employees at major global companies like KPMG. With Google Agentspace, employees can find and synthesize information from within their organization, converse with AI agents and take action with their enterprise applications. It combines enterprise search, conversational AI or chat and access to Gemini and third-party agents. We also offer pre-packaged agents across customer engagement, coding, creativity and more that are helping to provide conversational customer experiences, accelerate software development, and improve decision-making…

…Particularly with the newer models, I think we are working on early agentic workflows and how we can get those coding experiences to be much deeper. We are deploying it across all parts of the company. Our customer service teams are deeply leading the way there. We’ve both dramatically enhanced our user experience as well as made it much more efficient to do so. And we are actually bringing all our learnings and expertise in our solutions through cloud to our other customers. But beyond that, all the way from the finance team preparing for this earnings call to everything, it’s deeply embedded in everything we do.

Waymo is now serving 250,000 trips per week (was 150,000 in 2024 Q4), up 5x from a year ago; Waymo launched its paid service in Silicon Valley in 2025 Q1; Waymo has expanded in Austin, Texas, and will launch in Atlanta later this year; Waymo will launch in Washington DC and Miami in 2026; Waymo continues to make progress in airport access and freeway driving; management thinks Alphabet will not be able to scale Waymo by themselves, so partners are needed

Waymo is now safely serving over 0.25 million paid passenger trips each week. That’s up 5x from a year ago. This past quarter, Waymo opened up paid service in Silicon Valley. Through our partnership with Uber, we expanded in Austin and are preparing for our public launch in Atlanta later this summer. We recently announced Washington, D.C. as a future ride-hailing city, going live in 2026 alongside Miami. Waymo continues progressing on 2 important capabilities for riders, airport access and freeway driving…

More businesses are adopting Alphabet’s AI-powered campaigns; Alphabet’s recent work with AI is helping advertisers reach customers and searches where advertising would previously not be showed; Alphabet is infusing AI at every step of the marketing process for advertisers, for example, (1) advertisers can now generate a broader variety of lifestyle imagery customized to their business, (2) in PMax, advertisers can automatically source images from their landing pages and crop them, (3) on media buying, AI-powered campaigns continue to help advertisers find new customers, (4) in Demand Gen, advertisers can more precisely manage ad placements and understand which assets work best at a channel level; users of Demand Gen now see an average 26% year-on-year increase in conversions per dollar spend; when Demand Gen is paired with Product Feed, advertisers see double the conversion per dollar spend year-over-year on average; Royal Canin used Demand Gen and PMax campaigns and achieved a 2.7x higher conversion rate, a 70% lower cost per acquisition for purchases, a 8% higher value per user

More businesses, big and small, are adopting AI-powered campaigns, and the deployment of AI across our Ads business is driving results for our customers and for our business. Throughout 2024, we launched several features that leverage LLMs to enhance advertiser value, and we’re seeing this work pay off. The combination of these launches now allows us to match ads to more relevant search queries. And this helps advertisers reach customers and searches where we would not previously have shown their ads.

Focusing on our customers, we continue to solve advertisers’ pain points and find opportunities to help them create, distribute and measure more performant ads, infusing AI at every step of the marketing process. On Audience Insights, we released new asset audience recommendations, which tell businesses the themes that resonate most with their top audiences. On creatives, advertisers can now generate a broader variety of lifestyle imagery customized to their business to better engage their customers and use them across PMax, demand gen, display and app campaigns. Additionally, in PMax, advertisers can automatically source images from their landing pages and crop them, increasing the variety of their assets. On media buying, advertisers continue to see how AI-powered campaigns help them find new customers. In Demand Gen, advertisers can more precisely manage ad placements across YouTube, Gmail, Discover and Google Display Network globally and understand which assets work best at a channel level. Thanks to dozens of AI-powered improvements launched in 2024, businesses using Demand Gen now see an average 26% year-on-year increase in conversions per dollar spend for goals like purchases and leads. And when using Demand Gen with Product Feed, on average, they see more than double the conversion per dollar spend year-over-year…

…Royal Canin combined Demand Gen and PMax campaigns to find more customers for its cat and dog food products. The integration resulted in a 2.7x higher conversion rate, a 70% lower cost per acquisition for purchases and increased the value per user by 8%.

Google Cloud still has more AI demand than capacity in 2025 Q1 (as it did in 2024 Q4) 

Recall I’ve stated on the Q4 call that we exited the year in Cloud specifically with more customer demand than we had capacity. And that was the case this quarter as well.

30% of new code at Alphabet is now generated by AI (it was 25% in 2024 Q3)

We’re continuing to make a lot of progress there in terms of people using coding suggestions. I think the last time I had said, the number was like 25% of code that’s checked in. It involves people accepting AI-suggested solutions. That number is well over 30% now. But more importantly, we have deployed more deeper flows.

Amazon (NASDAQ: AMZN)

AWS grew 17% year-on-year in 2025 Q1, and is now at a US$117 billion annualised revenue run rate (was US$115 billion in 2024 Q4); management used to think AWS could be a multi-hundred billion dollar revenue run rate business without AI and now that there’s AI, they think AWS could be even bigger; AWS’s AI business is now at a multi-billion annual revenue run rate and is growing triple-digits year-on-year; the shifting from on-premise to the cloud is still a huge tailwind for AWS, and now even more so as companies that want realize the full potential of AI will need to shift to the cloud; AWS is currently still supply constrained and it will be on a lot more new chips in the coming months; management thinks that the supply chain issues with chips will get better as the year progresses

AWS grew 17% year-over-year in Q1 and now sits at a $117 billion annualized revenue run rate…

…Before this generation of AI, we thought AWS had the chance to ultimately be a multi-hundred billion dollar revenue run rate business. We now think it could be even larger…

…Our AI business has a multibillion-dollar annual revenue run rate, continues to grow triple-digit year-over-year percentages and is still in its very early days…

…Infrastructure modernization is much less sexy to talk about than AI, but fundamental to any company’s technology and invention capabilities, developer productivity, speed and cost structure. And for companies to realize the full potential of AI, they’re going to need their infrastructure and data in the cloud…

…During the first quarter, we continued to see growth in both generative AI business and non-generative AI offerings as companies turn their attention to newer initiatives, bring more workloads to the cloud, restart or accelerate existing migrations from on-premises to the cloud and tap into the power of Generative AI…

…We — as fast as we actually put the capacity in, it’s being consumed. So I think we could be driving — we could be helping more customers driving more revenue for the business if we had more capacity. We have a lot more Trainium2 instances and the next generation of NVIDIA’s instances landing in the coming months…

…I do believe that the supply chain issues and the capacity issues will continue to get better as the year proceeds.

Management is directing Amazon to invest aggressively in AI; Amazon is building 1000-plus AI applications across the company; the next generation of Alexa is Alexa+; Amazon is using AI in its fulfilment network, robotics, shopping, and more

If you believe your mission is to make customers’ lives easier and better every day, and you believe that every customer experience will be reinvented with AI, you’re going to invest very aggressively in AI, and that’s what we’re doing. You can see that in the 1,000-plus AI applications we’re building across Amazon. You can see that with our next generation of Alexa, named Alexa+. You can see that in how we’re using AI in our fulfillment network, robotics, shopping, Prime Video and advertising experiences. And you can see that in the building blocks AWS is constructing for external and internal builders to build their own AI solutions.

AWS’s in-house AI chip, Trainium 2, is starting to lay in capacity in larger quantities with significant appeal and demand; AWS will always be offering AI chips from multiple providers, but Trainium 2 offers a compelling option with 30%-40% better price performance; management believes that the price of inference needs to be much lower for AI to be successful, and they think the price of inference will go down; Anthropic is still building its next few models with Trainium 2

Our new custom AI chip Trainium2 is starting to lay in capacity in larger quantities with significant appeal and demand. While we offer customers the ability to do AI in multiple chip providers and will for as long as I can foresee, customers doing AI at any significant scale realize that it can get expensive quickly. So the 30% to 40% better price performance that Trainium2 offers versus other GPU-based instances is compelling. For AI to be as successful as we believe it can be, the price of inference needs to come down significantly…

…I would say that we’ve been bringing on a lot of P5, which is a form of NVIDIA chip instances, as well as landing more and more Trainium2 instances as fast as we can…

…Anthropic is running — building the next few training models on top of our Trainium2 chip on AWS…

…As they’re waiting to see the cost of inference continue to go down, which it will.

The latest premier Amazon Nova model was launched yesterday and it delivers frontier intelligence and industry-leading price performance; thousands of customers are already using Amazon Nova models; Amazon Nova Sonic, a speech-to-speech foundation model, was recently released and it enables developers to build voice-based AI applications; Amazon Nova Sonic has lower word error rates and higher win rates over other comparable models; AWS recently released a research preview of Amazon Nova Act, a new AI model that can perform actions within a web browser; Amazon Nova Act aims to move the current state-of-the-art accuracy of multi-step agentic actions from 30%-60% to 90%-plus

We offer our own Amazon Nova state-of-the-art foundation models in Bedrock with the latest premier model launching yesterday. They deliver frontier intelligence and industry-leading price performance, and we have thousands of customers already using them, including Slack, Siemens, Sumo Logic, Coinbase, FanDuel, Glean and Blue Origin. A few weeks ago, we released Amazon Nova Sonic, a new speech-to-speech foundation model that enables developers to build voice-based AI applications that are highly accurate, expressive and human-like. Nova Sonic has lower word error rates and higher win rates over other comparable models for speech interactions…

…We’ve just released a research preview of Amazon Nova Act, a new AI model trained to perform actions within a web browser. It enables developers to break down complex workflows into reliable atomic commands like search or checkout or answer questions about the screen. It also enables them to add more detailed instructions to these commands where needed, like don’t accept the insurance upsell. Nova Act aims to move the current state-of-the-art accuracy of multistep agentic actions from 30% to 60% to 90-plus percent with the right set of building blocks to build these action-oriented agents.

Amazon’s management sees question-and-answer being the only current use-case for AI agents, but they want AI agents to be be capable of performing a wide variety of complex tasks and they have built Alexa+ to be such an agent; management launched a new lightning fast AI agent coding experience in Amazon Q in 2025 Q1 and customers are loving it; management has made generally available GitLab Duo with Amazon Q, which enables AI agents to assist multi-step tasks; Alexa+ is meaningfully smarter and more capable than the previous Alexa; Alexa+ is free with Prime and available for non-Prime customers at $19.99 per month; Alexa+ is just starting to be rolled out in the USA and will be introduced to other countries later in 2025; users really like Alexa+ thus far; Alexa+ is now with more than 100,000 users; Amazon already has 0.5 billion devices in people’s homes and cars that can easily distribute Alexa+; management thinks users will have to relearn a little on how to communicate with Alexa+, but the communication experience is now much better; management asked Alexa+  about good Italian restaurants in New York and Alexa+ helped to make a reservation

To date, virtually all of the agentic use cases have been of the question-answer variety. Our intention is for agents to perform wide-ranging complex multistep tasks by organizing a trip or setting the lighting, temperature and music ambience in your house for dinner guests or handling complex IT tasks to increase business productivity. There haven’t been action-oriented agents like this until Alexa+…

…This past quarter, Amazon Q, the most capable generative AI-powered assistant for accelerating software development and leveraging your own data, launched a lightning fast new agent coating experience within the command line interface that can execute complex workflows autonomously. Customers are loving this. We also made generally available GitLab Duo with Amazon Q, enabling AI agents to assist multi-step tasks such as new feature development, code-based upgrades for Java 8 and 11, while also offering code review and unit testing, all within the same familiar GitLab platform…

…We introduced Alexa+, our next-generation Alexa personal assistant, who is meaningfully smarter and more capable than our prior self can both answer virtually any question and take actions and is free with Prime or available to non-Prime customers for $19.99 a month. We’re just starting to roll this out in the U.S., and we’ll be expanding to additional countries later this year. People are really liking Alexa+ this far…

…So we’ve worked hard on that in Alexa+. We’ve been — we started rolling out over the last several weeks. It’s with now over 100,000 users with more rolling out in the coming months. And so far, the response from our customers has been very, very positive…

…We’re very fortunate in that we have over 0.5 billion devices out there in people’s homes and offices and cars. So we have a lot of distribution already…

…To some degree, there will be a little bit of rewiring for people on what they can do because you get used to patterns. I mean even the simple thing of not having to speak, Alexa speak anymore, we we’re all used to saying, Alexa, before we want every action to happen. And what you find is you really don’t have to do it the first time, and then really the conversation is ongoing where you don’t have to say Alexa anymore. And I’ve been lucky enough to have the alpha and the beta that I’ve been playing with for several months, and it took me a little bit of time to realize they didn’t have to keep saying Alexa, it’s very freeing when you don’t have to do that…

…When I was in New York, when we were announcing, I asked her, what were the — we did the event way downtown. I asked her what was great Italian restaurants or pizza restaurants, she gave me a list and she asked me if she wanted me to make a reservation. I said yes. And she made the reservation and confirmed the time, like that. When you get into those types of routines and you have those types of experience, they’re very, very useful.

The majority of Amazon’s capital expenditure (capex) in 2025 Q1 was for AWS’s technology infrastructure, including the Trainium chips

Turning to our cash CapEx, which was $24.3 billion in Q1. The majority of this spend is to support the growing need for technology infrastructure. It primarily relates to AWS as we invest to support demand for our AI services and increasingly in custom silicon like Trainium as well as tech infrastructure to support our North America and International segments. We’re also investing in our fulfillment and transportation network to support future growth and improve delivery speeds and our cost structure. This investment will support growth for many years to come.

The vast majority of successful startups are built on AWS; high-profile startups building AI coding agents are on AWS

If you look at the start-up space, the vast majority of successful start-ups over the last 10 to 15 years have run on top of AWS…

…If you just look at the growth of these coding agents in the last few months, these are companies like Cursor or Vercel, both of them run significantly on AWS.

Amazon’s management thinks that current AI apps have yet to really tackle customer experiences that are going to be reinvented and many other agents that are going to be built

What’s interesting in AI is that we still haven’t gotten to all the other customer experiences that are going to be reinvented and all the other agents that are going to be built. They’re going to take the role of a lot of different functions today. And those are — they’re — even though we have a lot of combined inference in those areas, I would say we’re not even at the second strike of the first batter in the first inning. It is so early right now.

AWS operating margin improved from 37.6% in 2024 Q1 to 39.5% in 2025 Q1, but margins will fluctuate from time to time; AWS’s margin strength is from the business’s strong growth, the impact of some continued investments, and AWS’s custom chips; the investments include software optimisations for server capacity, low-cost custom networking equipment, and power usage in data centers

AWS operating income was $11.5 billion and reflects our continued growth coupled with our focus on driving efficiencies across the business. As we said before, we expect AWS operating margins to fluctuate over time, driven in part by the level of investments we’re making at any point in time…

…We had a strong quarter in AWS, as you mentioned, the margin performance. I would attribute it to the strong growth that we’re seeing, coupled with the impact of some continued investment we’re making in innovation and technology. I’ll give you some examples. So we invest in software and process improvements and ends up optimizing our server capacity, which helps our infrastructure cost. We’ve been developing more efficient network using our low-cost custom networking gear. We’re working to maximize the power usage in our existing data centers, which both lowers our costs and also reclaims power for other newer workloads. And we’re also seeing the impact of advancing custom silicon like Graviton. It provides lower cost not only for us, but also for our customers, better price performance for them.

Apple (NASDAQ: AAPL)

Apple is currently shipping an LLM (large language model) on the iPhone 16 where some of the queries are being handled on the device itself

As you know, we’re shipping an LLM on the iPhone 16 today. And there are — some of the queries that are being used by our customers are on-device, and then others go to the private cloud where we’ve essentially mimicked the security and privacy of the device into the cloud. nd then others, for world knowledge, are with the integration with ChatGPT.

The new Mac Studio has Apple’s M4 Max and M3 Ultra chips, and it can run large language models with over 600 billion parameters entirely in memory

The new Mac Studio is the most powerful Mac we’ve ever shipped, equipped with M4 Max and our new M3 Ultra chip. It’s a true AI powerhouse capable of running large language models with over 600 billion parameters entirely in memory.

Apple has released VisionOS 2.4 which unlocks the first set of Apple Intelligence features for Vision Pro users

VisionOS 2.4 unlocks the first set of Apple Intelligence features for Vision Pro users while inviting them to explore a curated and regularly updated collection of spatial experiences with the Spatial Gallery app.

Apple’s management has released iOS 18.4, which brings Apple Intelligence to more languages (including Singlish); Apple has built its own foundation models for everyday tasks; new Apple Intelligence features in iOS 18 include, Writing Tools, Genmoji, Image Playground, Image Wand, Clean Up, Visual Intelligence, and a seamless connection to ChatGPT

Turning to software. We just released iOS 18.4, which brought Apple Intelligence to more languages, including French, German, Italian, Portuguese, Spanish, Japanese, Korean and simplified Chinese as well as localized English to Singapore and India…

…At WWDC24, we announced Apple Intelligence and shared our vision for integrating generative AI across our ecosystem into the apps and features our users rely on every day. To achieve this goal, we built our own highly capable foundation models that are specialized for everyday tasks. We designed helpful features that are right where our users need them and are easy to use. And we went to great lengths to build a system that protects user privacy whether requests are processed on-device or in the cloud with Private Cloud Compute, an extraordinary step forward for privacy and AI.

Since we launched iOS 18, we’ve released a number of Apple Intelligence features from helpful Writing Tools to Genmoji, Image Playground, Image Wand, Clean Up, visual intelligence and a seamless connection to ChatGPT. We made it possible for users to create movies of their memories with a simple prompt and added AI-powered photo search, smart replies, priority notifications, summaries for mail, messages and more. We’ve also expanded these capabilities to more languages and regions.

Apple’s in-house chips are designed with a neural engine that powers AI features across Apple’s products and 3rd-party apps; management thinks the neural engine makes Apple products the best devices for generative AI

AI and machine learning are core to so many profound features we’ve rolled out over the years to help our users live a better day. It’s why we designed Apple silicon with a neural engine that powers so many AI features across our products and third-party apps. It’s also what makes Apple products the best devices for generative AI.

Apple still needs more time to work on the more personalised Siri that was unveiled by management recently

With regard to the more personal Siri features we announced, we need more time to complete our work on these features so they meet our high-quality bar. We are making progress and we look forward to getting these features into customers’ hands.

Apple has low capital expenditures for AI relative to other US technology giants because it uses 3rd-party data centers so they are mostly operating expenses; Apple’s new $500 billion investment in the USA could signal more capital expenditures and data center investments

On the data center side, we have a hybrid strategy. And so we utilize third parties in addition to the data center investments that we’re making. And as I’ve mentioned in the $500 billion, there’s a number of states that we’re expanding in. Some of those are data center investments. And so we do plan on making investments in that area

Arista Networks (NYSE: ANET)

Arista Networks’ management remains confident of reaching $750 million in back-end AI revenue in 2025 even with the uncertainty surrounding US tariffs; the 1:1 ratio between front-end and back-end AI spending for Arista Networks’ products still remains, but management thinks it’s increasingly hard to parse between front-end and back-end

Our cloud and AI momentum continues as we remain confident of our $750 million front-end AI goal in 2025…

…Just a quick clarification before we go into Q&A. Jayshree meant we were reiterating our back-end goal of $750 million, not front-end AI…

…[Question] Is that 1:1 ratio for the front-end back is still intact in your perspective?

[Answer] On the front-end ratio, yes, we’ve said it’s generally 1:1. It’s getting harder and harder to measure front end and back end. Maybe we’ll look at the full AI cluster differently next year. But I think 1:1 is still a good ratio. It varies. Some of them just build a cluster and don’t worry about the front end and others worry about it entirely holistically. So it does vary, but I think the 1:1 is still a good ratio…

…[Question] You reiterated the $750 million back-end target, but you’ve kind of had this $1.5 billion kind of AI target for 2025. And just wondering, is the capability of that more dependent on kind of the tariffs given kind of some of the front-end spend?

[Answer] Regarding tariffs, I don’t think it will have a material difference on the $750 million number or the $1.5 billion. We got the demand. So unless we have some real trouble shipping it or customers change their mind, I think we’re good with both those targets for the year.

Arista Networks is progressing well with its 4 major AI customers; 1 of the 4 customers have been in NVIDIA’s Infiniband solution for a long time, so they’ll be small for Arista Networks; 2 of the 4 are heading towards 50,000 GPU deployments by end-2025, maybe even 100,000 GPUs; 3 of the 4 customers are already in production with the 4th progressing well towards production; management has a lot of visibility from the 4 major AI customers for 2025 and 2026 and it’s looking good; the 4 major AI customers are mostly deploying Arista Networks’ 800-gig switches

We are progressing well in all 4 customers and continue to add smaller ones as well…

…Let me start with the 4 customers. All of them are progressing well. One of them is still new to us. They’ve been in Infiniband for a long time, so they’ll be small. I would say 2 of them are heading towards 50,000 GPU deployments by end of the year, maybe they’ll be at 100 but I can be most certainly sure of 50,000, heading to 100,000. And then the other one is also in production. So I had talked about all 4 going into production. Three are already in production, the fourth one is well underway…

…[Question] If I can go back to the 4 Tier 1s that you’re working with on the AI back end and the progress that you updated on that front. Are these customers now giving you more visibility just given the tariff landscape and that you would need to sort of build inventory for some of the finished codes? And can you just update us how they’re handling the situation on that front? And particularly then, as you think about — I think the investor focus is a lot about sort of 2026 and potential sort of changes in the CapEx landscape from these customers at that point. Are you getting any forward visibility from them? Any sort of early signs for 2026 on these customers?

[Answer] We definitely have all the visibility in the world for this year, and we’re feeling good. We’re getting unofficial visibility because they all know our lead times are tied to some fairly long lead times from our partners and suppliers. So I would say 2026 is looking good. And based on our execution of 2025 and plans we’re putting together, we should have a great year in 2026 as well for AI sector specifically…

…[Question] Do you see the general cadence of hyperscalers deploying 800-gig switch ports this year? I ask because I believe your Etherlink family of switches became generally available in late 2024.

[Answer] I alluded to this earlier in 2024, the majority of our AI trials were on 400 gig at that time. So you’re right to observe that with our Etherlink portfolio really getting introduced in the second half of ’24 that a lot of our 800-gig activity has picked up in 2025, some of which will be reflected in shipments and some of it which will be part of our deferred. So it’s a good observation and an accurate one that this is the year of 800, like last year it was the year of 400.

Arista Networks’ management plans for the company to be the premier and preferred network for NVIDIA’s next-generation GPUs; Arista Networks’ Etherlink portfolio makes it easy to identify and localise performance issues in accelerated compute AI clusters

At the GTC event in March of 2025, we heard all about NVIDIA’s planned GPU road map every 12 to 18 months, and Arista intends to be the premier and preferred scale-out network for all of those GPUs and AI accelerators. Traditional GPUs have a collective communication libraries or CCL, as they’re known, that try to discover the underlying network topology using localization techniques. With this accelerated compute approach, the discrepancies between the discovered topology and the one that actually happens can impact AI job completion times. Arista’s ethylene portfolio highlights the accelerated networking approach, bringing that single point of network control and visibility as a differentiation. This makes it extremely crisp to identify and localize performance issues especially as the size of the AI cluster grows to 50,000 and 100,000 XPUs with the Arista AI Spine and leaf network designs.

Arista Networks’ campus portfolio provides cost-effective access points for agentic AI applications

Arista’s cognitive campus portfolio features our advanced spine with power or Ethernet-wired lease capabilities, along with a wide range of cost-effective wireless or 7 indoor and outdoor access points for the newer IoT and agentic applications.

The data center ecosystem is still somewhat new to AI and the suppliers are figuring things out together

But everybody is new to AI, they’ve never really put together a network design for 4-rail or 8-rail or how does it connect into the GPUs and what is the NIC [network interface card] attachment? What is the accessories in terms of cables or optics that connect? So this movement from trials to production causes us to bring a whole ecosystem together for the first time.

Arista Networks’ management thinks that when it comes to AI use-cases, Arista Networks’ products will play a far bigger role than whitebox networking manufacturers, even though whiteboxes will always be around and management is even happy to help customers build networking solutions that encompass both Arista Networks’ products and whiteboxes; Arista Networks was able to help a small AI customer build a network for a cluster of a few hundred GPUs very quickly after the customer struggled to do so with whiteboxes

I’ve always said, that white box is not new. It’s been with us since the beginning of time. In fact, when Arista got started, a couple of our customers had already implemented internally various implementations of white box. So there is a class of customers who will make the investments in engineering and operations to build their own network and manage it. And it’s a very different business model. It operates typically at 10% gross margins. I don’t think you want Arista to go there. And it’s very hardware-centric and doesn’t require the rich software foundation and investments that we’ve made. So first, I’ll start by saying we will always and will continue to coexist with white box. There are times that you’ve noticed this, too, that because Arista builds some very superior hardware, that even if they don’t use our EOS, they like to have our blue box, as I often call it, the Arista hardware that’s engineered much better than any others with a more open OS like Sonic or FBOSS or at least the attributes of running both EOS and an open-source networking system. So I think we view this as a natural part of selection in a customer base where if it’s a simple use case, they’re going to use something cost effective. But if it’s a really complex use case, like the AI spine or roles that require and demand more mission-critical features, Arista always plays a far bigger role in premium, highly scalable, highly valued software and hardware combinations than we do in a stand-alone white box. So we’ll remain coexistent peacefully, and we’re not in any way threatened by it. In fact, I would say we work with our customers to make sure as they’re building permutations and combinations of the white box, that we can work with that and build the right complement to that with our Etherlink portfolio…

…We had a customer, again, not material. We said, “I can’t get these boxes. I can’t make them run. I cannot get an AI network.” And one of my most technical sales leaders said, hey, we got a chance to build an AI cluster here for a few hundred GPUs. We jumped on it. Obviously, that customer is small and have been largely using white boxes and is now about to install an AI leaf and an AI spine, and we had to get it to him before the tariff deadline. So as an example of not material, but how quickly these decisions get made when you have the right product, right performance, right quality, right mission-critical nature and you can deal with that traffic pattern better than anyone else can. So it happens. It’s not big because we’ve got so much commitment in a given quarter from a customer, but when it is, we ask with great deal of nimbleness and agility to do that.

Arista Networks’ management is happy to support any kind of advanced packaging technologies – such as co-packaged optics or co-packaged copper – for back-end AI networks in the company’s products; management has yet to see any major adoption of co-packaged optics for back-end AI networks

[Question] I’d love to get your latest views around co-packaged optics. NVIDIA introduced its first CPO switches, GCC, for scale-out. And I was wondering whether that had any impact on your views regarding CPO adoption in back-end AI networks in coming years.

[Answer] It’s had no impact. It’s very early days. I think you’ve seen — Arista doesn’t build optics, but Arista enables optics and we’ve always been at the forefront, especially with Andy Bechtolsheim and his team of talented tech individuals that whether it is pluggable optics with LPO or how we define the OSFP connector for MSAs or 100 gig, 400 gig, it’s something we take seriously. And our views on CPOs, it’s not a new idea. It’s been demonstrated in prototype for, I don’t know, 10 to 20 years. The fundamental lack of adoption to date on CPO, it’s relatively high failure rates and it’s mostly been in the labs. So what are some of the advantages of CPO? Well, it has a linear interface. It has lower power than DSP for long-haul optics. It has a higher channel count. And I think if pluggable optics can achieve some of that in the best of both worlds, then you can overcome that with pluggable optics or even co-packaged copper. So Arista has no religion. We will do co-package copper. We’ll do co-package optics. We will do pluggable optics, but it’s too early to call this a real production-ready product that’s still in very early experiments and trials.

Arista Networks’ management is not seeing any material pull-forward in demand for its products because of US tariffs

[Question] We know tariffs are coming later in the year. Whether the strength you’re seeing is the result of early purchases of customers ahead of tariffs in order to save some dollars?

[Answer] Even if our customers try to pull it in and get it all by July, we would be unable to supply it. So that would be the first thing. So I’m not seeing the pull-ins that are really material in any fashion. I am seeing a few customers trying to save $1 here, $1 there to try and ship it before the tariff date but nothing material. Regarding pull-ins for 4 to 6 quarters, again, our best visibility is near term. And if we saw that kind of behavior, we would see a lot of inventory sitting in our customers, which we don’t. In fact, that’s long enough to ship faster and ship more.

2 years ago, Arista Networks’ management saw all its Cloud Titan customers pivot to AI and slow down their cloud spending; management is seeing more balanced spending now, with a more surgical focus on AI

2 years ago, I was very nervous because the entire cloud titans pivoted to AI and slowed down their cloud. Now we see a more balanced spend. And while we can’t measure how much of this cloud and how much of it is AI, if they’re kind of cobbled together, we are seeing less of a pivot, more of a surgical focus on AI and then a continued upgrade of the cloud networks as well. So compared to ’23, I would say the environment is much more balanced between AI and cloud.

Arista Networks’ management sees competitive advantages in the company’s hardware design, development, and operation that are hard to replicate even for its Cloud Titan customers

[Question] What functionality about the blue box actually makes it defensible versus what hyperscalers can kind of self-develop?

[Answer] Let me give you a few attributes of what I call the blue box, and I’m not saying others don’t have it, but Arista has built this as a mission, although we’re known for our software. We’re just as well known for our hardware. When you look at everything from a form factor of a one RU that we build to a chassis, we’ve got a tremendous focus on signal integrity, for example, all the way from layer 1, multilayer PCB boards, a focus on quality, a focus on driving distances, a focus on integrating optics for longer distances, a focus on driving MACsec, et cetera. So that’s a big focus. The second is hardware diagnostics. Internal to the company, we call it Arista boot. We’ve got a dedicated team focused on not just the hardware but the firmware to make it all possible in terms of troubleshooting because when these boards get super complex, you know where the failure is and you’re running at high-speed 200 [indiscernible] 30s. So things are very complex. So the ability to pinpoint and troubleshoot is a big part of what we do. And then there’s additional focus on the mechanical, the power supplies, the cooling, all of which translate to better power characteristics. Along with our partners and chip vendors, there’s a maniacal focus on not just high performance but low power. So some of the best attributes come from our blue boxes, not only for 48 ports, but all the way up to 576 ports of an AI spine or double that if you’re looking for dual capabilities. So well-designed, high-quality hardware is a thing of beauty, but also think of complexity that not everyone can do.

With neo AI cloud customers, Arista Networks’ management is observing that they are very willing to forsake NVIDIA’s GPUs and networking solutions and try other AI accelerators and Ethernet; management thinks that the establishment of the Ultra Ethernet Consortium in 2024 has a role to play in the increasing adoption of Ethernet for AI networking; with the Cloud Titans, management is also observing that they are shifting towards Ethernet; management thinks that the shift from Infiniband to Ethernet is faster than the the shift from NVIDIA’s GPUs to other companies’ GPUs

[Question] There’s a general perception that most of them are buying NVIDIA-defined clusters and networking. So I wonder if you could comment on those trends, their interest in moving past InfiniBand? And also are there opportunities developing with some of these folks to kind of multi-source their AI connectivity to different providers?

[Answer] We’re seeing more adventurous spirit in the neo-cloud customers because they want to try alternatives. So some of them are absolutely trying other AI accelerators like Lisa and AMD and my friends there. Some of them are absolutely looking at Ethernet, not InfiniBand as a scale-out. And that momentum has really shifted in the last year with the Ultra Ethernet Consortium and the spec coming out in May. I just want to give a shout-out to that team and what we have done. So I think Ethernet is a given that there’s an awful lot of legacy of InfiniBand that will obviously sort itself out. And a new class of AI accelerators we are seeing more niche players, more internal developments from the cloud titans, all of which is mandating more Ethernet. So I think between your 2 questions, I would say the progress from InfiniBand to Ethernet is faster, the progress from the ones they know and the high-performance GPU from NVIDIA versus the others is still taking time.

ASML (NASDAQ: ASML)

ASML’s management still sees AI (artificial intelligence) as the key growth driver; ASML will hit upper range of guidance for 2025 if AI demand continues to be strong, while ASML will hit the lower range of guidance if there is uncertainty among its customers

Consistent with our view from last quarter, the growth in artificial intelligence remains the key driver for growth in our industry. If AI demand continues to be strong and customers are successful in bringing on additional capacity to support the demand, there is a potential opportunity towards the upper end of our range. On the other hand, there is still quite some uncertainty for a number of our customers that can lead to the lower end of our range. 

ASML’s management is still positive on the long-term outlook for ASML, with AI being a driver for growth

Looking longer term, the semiconductor market remains strong with artificial intelligence, creating growth in recent quarters, and we see some of the future demand for AI solidifying, which is encouraging. 

ASML’s management thinks inference will become a larger part of AI demand going forward

I think there has been a lot of emphasis in the past quarters on the training side of life. I think more and more, which I think is logical, that you also see more and more emphasis being put on the inferencing side of the equation. So I think you will see the inferencing part becoming a larger component of AI demand on a go-forward basis.

ASML’s management is unable to tell what 2027 will look like for AI demand, but the commitment to AI chips in the next 2 years is very strong

You are looking at major investment, investment has been committed, investment that a lot of company believe they have to make in order to basically enter this AI race, I think the threshold to change this behavior is pretty high. And this is why — this is what our customers are telling us. And that’s also why we mentioned that, based on those conversations, we still see ’25, ’26 as growth years. That’s largely driven by AI and by that dynamic. Now ’27 start to be a bit further away, so you’re asking us too much, I think, to be able to answer basically what AI may look like in ’27. But if you look at the next couple of year, so far, the commitment to the AI investment and, therefore, the commitment also to deliver the chips for AI has been very solid.

Coupang (NYSE: CPNG)

Coupang’s management is investing in automation (such as automated picking, packing and sorting) and machine learning to deploy inventory more precisely to improve the customer experience and reduce costs

This quarter, we saw benefits from advances in our automated picking, packing and sorting systems and machine learning utilization that deploys inventory with more precise prediction of demand. This, coupled with our focus on operational excellence, enables us to continually improve the customer experience while also lowering their cost of service.

Datadog (NASDAQ: DDOG)

Existing customer usage growth in 2025 Q1 was in line with management’s expectations; management is seeing high growth in Datadog’s AI cohort, and stable growth in the other cohorts

Overall, we saw trends for usage growth from existing customers in Q1 that were in line with our expectations. We are seeing high growth in our AI cohort as well as consistent and stable growth in the rest of the business.

Datadog’s ,anagement continues to see increase in interest in next-gen AI capabilities and analysis; 4,000 Datadog customers at the end of 2025 Q1 used 1 or more Datadog AI integrations (was 3,500 in 2024 Q4), up 100% year-on-year; companies using end-to-end data observability to manage model performance, security, and quality, has more than doubled in the past 6 months; management has observed that data observability has become a big enabler of building AI workloads; the acquisition of Metaplane helps Datadog build towards a comprehensive data observability suite; management thinks data observability will be a big opportunity for Datadog

We continue to see rising customer for next-gen AI capabilities and analysis. At the end of Q1, more than 4,000 customers used one or more Datadog AI integrations, and this number has doubled year-over-year. With end-to-end data observability, we are seeing continued growth in customers and usage as they seek to manage end-to-end model performance, security and quality. I’ll call out the fact that the number of companies using end-to-end data observability has more than doubled in the past 6 months…

…[Question] What the vision is about moving into data observability and how consequential an opportunity it could be for Datadog?

[Answer] The field is evolving into a big enabler or it can be positive enabler, if you don’t do it right, for building enterprise workloads — for AI workloads, sorry. So in other words, making sure the data is being extracted from the the right place, transformed the right way and is being fed into the right AI models on the other hand…

…We only had some building blocks for data observability. We built data streams monitoring product for streaming data that comes out of few, such as Kafka, for example. We built their job monitoring product that monitors back jobs and large transformation jobs. We have a database monitoring product that looks at the way you optimize queries and optimize base performance and cost. And by adding data quality and data pipelines, with Metaplane, we have a full suite basically that allows our customers to manage everything from getting the data from their core data storage into all of the products and AI workloads and reports they need to go populate that data. And so we think it’s a big opportunity for us.

Datadog’s management has improved Bits AI, and is using next-gen AI to help solve customer issues quickly and move towards auto remediation

We are adding to Bits AI, with capabilities for customers to take action with workflow automation and App Builder, using next GenAI to help our customers immediate issues more quickly and move towards auto remediation in the future.

Datadog has made 2 recent acquisitions; Eppo is a feature management and experimentation platform; management sees automated experimentation as an important part of modern application development because of the use of AI in coding; Metaplane is a data observability platform that works well for new enterprise AI workloads; management is seeing more AI-written code in both its customers and the company itself; management thinks that as AI writes more code, more value will come from being able to observe and understand the AI-written code in production environments, which is Datadog’s expertise; the acquisitions of Eppo and Metaplane are to position Datadog for the transition towards a world of AI-written code

We recently announced a couple of acquisitions.

First, we acquired Eppo, a next-generation feature management and experimentation platform. The Eppo platform helps increase the velocity of releases, while also lowering risk by helping customers to release and validate features in a controlled manner. Eppo augments our efforts in product analytics, helping customers improve the variance and tie feature performance to business outcomes. More broadly, we see automated experimentation as a key part of modern application development, with the rapid adoption of the agent generative code, as well as more and more of the application logic itself being implemented with nondeterministic AI models. 

Second, we also acquired Metaplane, the data observability platform built for modern data teams. Metaplane helps prevent, detect and resolve their availability and quality issues across the company’s data warehouses and data pipelines. We’ve seen for several years now that better freshness and quality were critical for applications and business analytic. And we believe that they are becoming key enablers of the creation of new enterprise AI workloads, which is why we intend to integrate the Metaplane capabilities into our end-to-end dataset offerings…

…There is definitely a big transition that is happening right now, like we see the rise of AI written code. We see it across our customers. We also see it inside of Datadog, where we’ve had very rapid adoption of this technology as well…

…The way we see it is that it means that there’s a lot less value in writing the code itself, like everybody can do it pretty quickly, can do a lot of it. You can have the machine to do a lot of it, and you complement it with a little bit of your own work. But the real difficulty is in validating that code, making sure that it’s safe, making sure it runs well, that it’s performing and that it does what it’s supposed to do for the business. Also making sure that when 15 different people are changing the code at the same time, all of these different changes come together and work the right way, and you understand the way these different pieces interact in the way. So the way we see it is this move out a lot of their value from writing the code to observing it and understanding it in production environments, which is what we do. So a lot of the investments we’re making right now, including some of the acquisitions we’ve announced, build towards that, and making sure that we’re in the right spot.

Datadog signed a 7-figure expansion deal with a leading generative AI company; the generative AI company needs to reduce tool fragmentation; the generative AI company is replacing commercial tools for APM (application performance monitoring) and log management with Datadog, and is expanding to 5 Datadog products

We signed a 7-figure expansion as an annualized contract with a leading next GenAI company. This customer needs to reduce tool fragmentation to keep on top of its hyper growth in usage and employee headcount. With this expansion, the customer will use 5 Datadog products and will replace commercial tool for APM and log management.

AI-native customers accounted for 8.5% of Datadog’s ARR in 2024 Q4 (was 6% in 2024 Q4); AI-native customers contributed 6 percentage points to Datadog’s year-on-year growth in 2025 Q1, compared to 2 percentage points in 2024 Q1; management thinks AI-native customers will continue to optimise cloud and observability usage in the future; AI-native contracts that come up for renewal are healthy; Datadog has huge customer concentration with the AI-native cohort; Datadog has more than 10 AI-native customers that are spending $1 million or more with Datadog; the strong performance of the AI-native cohort in 2025 Q1 is fairly broad-based; Datadog is helping the AI-native customers mostly with inference, and not training; when Datadog sees growth among AI-native customers, that’s growth of AI adoption because the AI-native customers’ workloads are mostly customer-facing

We saw a continued rise in contribution from AI-native customers who represented about 8.5% of Q1 ARR, up from about 6% of ARR last quarter and up from about 3.5% of ARR in the year ago quarter. AI-native customers contributed about 6 points of year-over-year revenue growth in Q1 versus about 5 points last quarter and about 2 points in the year ago quarter. We continue to believe that adoption of AI will benefit Datadog in the long term, but we remain mindful that we may see volatility in our revenue growth on the backdrop of long-term volume growth from this cohort as customers renew with us on different terms and as they may choose to optimize cloud and observability usage…

…[Question] Could you talk about what you’re seeing from some of those AI-native contracts that have already come up for renewal and just how those conversations have been trending?

[Answer] All the contracts that come up for renewal, they are healthy. The trick with the cohort is that it’s growing fast. There’s also a revenue concentration there. We now have our largest customer in the cohort, and they’re growing very fast. And on the flip side of that, we also have a larger number of large customers that are also growing. So we — I think we mentioned more than 10 customers now that are spending $1 million or more with us in that AI-native cohort and that are also growing fast…

…On the AI side, we do have, as I mentioned, one customer large and the others there, they’re contributing more of the new revenue than the others. But we see growth in the rest of the cohort as well. So again, it’s fairly typical…

…For the AI natives, actually, what we help them with mostly is not training. It’s running their applications and their inference workloads as customer-facing. Because what’s training for the AI natives tends to be largely homegrown one-off and different from — between each and every one of them. We expect that as and if most other companies and enterprises do significant training, that this will not be the case. This will not be one-off and homegrown. But right now, it is still the AI natives that do most of the training, and they still do it in a way that’s largely homegrown. So when we see growth on the AI-native cohorts, that’s growth of AI adoption because that’s growth of customer-facing workloads by and large.

Datadog’s management sees the trend of cloud migration as being steady; management sees cloud migration being partly driven by customers’ desires to adopt AI, because migrating to the cloud is a prerequisite for AI

[Question] What are the trend lines on the cloud migration side?

[Answer] It’s consistent with what we’ve seen before. It’s also consistent with what you’ve heard from the hyperscalers over the past couple of weeks. So I would say it’s steady, unremarkable. It’s not really trending up nor trending down right now. But we see the same desire from customers to move more into the cloud and to lay the groundwork so they can also add up AI, because digital transformation and cloud migrations are prerequisites for that.

Datadog’s management thinks there will be more products for Datadog to build as AI workloads shift towards inferencing; management is seeing its LLM Observability product getting increasing usage as customers move AI workloads into production; management wants to build more products across the stack, from closer to the GPU to AI agents; 

On the workloads turning more towards inference, so there’s definitely more product to build there. So we have a — so we built an LLM Observability product that is being — that is getting increasing usage from customers as they move into production. And we think there’s more that we need to build both down the stack closer to the GPUs and up the stack closer to the agents that are being built on top of these models.

Datadog’s management is already seeing returns on Datadog’s internal investments in AI in terms of employee productivity; in the long-term, there’s the possibility that Datadog may need lesser headcount because of AI

[Question] Internally, how do you think about AI from an efficiency perspective?

[Answer] For right now, I think we’re seeing the returns in productivity, whether that be salespeople getting more information or R&D. We’re essentially trying to create an environment where we’re encouraging the various departments to use it and learning from it. Long term, there might well be efficiency gains — there may be efficiency gains that can be manifested in headcount.

Mastercard (NYSE: MA)

Mastercard’s management sees contactless payments and tokenised transactions as important parts of agentic AI digital commerce; Mastercard has announced Mastercard Agent Pay, which will facilitate safe, frictionless and programmable transactions across AI platforms; Mastercard is working with important AI companies such as Microsoft and OpenAI to deliver agentic payments

Today, 73% of all in-person switched transactions are contactless and approximately 35% of all our switch transactions are tokenized. These technologies will continue to play an important role as we move forward into the next phase of digital commerce, such as Agentic AI. We announced Mastercard Agent Pay to leverage our Agentic tokens as well as franchise rules, fraud and cybersecurity solutions. Combined, these will help partners like Microsoft to facilitate safe, frictionless and programmable transactions across AI platforms. We will also work with companies like OpenAI to deliver smarter, more secure and more personalized agentic payments. The launch of Agent Pay is an important step in redefining commerce in the AI era.

Mastercard closed the Recorded Future acquisition in 2024 Q4 (Recorded Future provides AI-powered solutions for real-time visibility into potential threats related to fraud); Recorded Future just unveiled the AI-powered Malware Intelligence; Malware Intelligence enables proactive threat prevention

On the cybersecurity front, Recorded Future just unveiled malware intelligence. It’s a new capability enabling proactive threat prevention for any business using real-time AI-powered intelligence insights.

Mastercard’s management sees AI as being deeply ingrained in Mastercard’s business; Mastercard’s access to an enormous amount of data is an advantage for Mastercard in deploying AI; in 2024, a third of Mastercard’s products in its value-added services and solutions segment was powered by AI

AI is deeply ingrained in our business. We have access to an enormous amount of data, and this uniquely positions us to enhance our AI’s performance, resulting in greater accuracy and reliability. And we’re deploying AI to enable many solutions in market today. In fact, in 2024, AI enabled approximately 1 in 3 of our products within value-added services and solutions.

Meta Platforms (NASDAQ: META)

Meta’s management is focused on 5 opportunities within AI namely, improved advertising, more engaging experiences, business messaging, Meta AI and AI devices; the 5 opportunities are downstream of management’s attempt to build artificial general intelligence and leading AI models and infrastructure in an efficient manner; management thinks the ROI of Meta’s investment in AI will be good even if Meta does not succeed in all the 5 opportunities;

As we continue to increase our investments and focus more of our resources on AI, I thought it would be useful today to lay out the 5 major opportunities that we are focused on. Those are improved advertising, more engaging experiences, business messaging, Meta AI and AI devices. And these are each long-term investments that are downstream from us building general intelligence and leading AI models and infrastructure. Even with our significant investments, we don’t need to succeed in all of these areas to have a good ROI. But if we do, then I think that we will be wildly happy with the investments that we are making…

…We are focused on building full general intelligence. All of the opportunities that I’ve discussed today are downstream of delivering general intelligence and doing so efficiently.

Meta’s management’s goal with the company’s advertising business is for businesses to simply tell Meta their objectives and budget, and for Meta to do all the rest with AI; management thinks that Meta can redefine advertising into an AI agent that delivers measurable business results at scale

Our goal is to make it so that any business can basically tell us what objective they’re trying to achieve like selling something or getting a new customer and how much they’re willing to pay for each result and then we just do the rest. Businesses used to have to generate their own ad creative and define what audiences they wanted to reach, but AI has already made us better at targeting and finding the audiences that will be interested in their products than many businesses are themselves, and that keeps improving. And now AI is generating better creative options for many businesses as well. I think that this is really redefining what advertising is into an AI agent that delivers measurable business results at scale.

Meta tested a new advertising recommendation model for Reels in 2025 Q1 called Generative Ads Recommendation Model, or GEM, that has improved conversion rates by 5%; 30% more advertisers are using Meta’s AI creative tools in 2025 Q1; GEM is twice as efficient at improving ad performance for a given amount of data and compute; GEM’s better efficiency helped Meta significantly scale up the amount of compute used for model training; GEM is now being rolled out to additional surfaces across Meta’s apps; the initial test of Advantage+’s streamlined campaign creation flow for sales, app and lead campaigns is encouraging and will be rolled out globally later in 2025; Advantage+ Creative is seeing strong adoption; all eligible advertisers can now automatically adjust the aspect ratio of their existing videos and generate images; management is testing a feature that uses gen AI to place clothing on virtual models; management has seen a 46% lift in incremental conversions in the testing of the incremental attribution feature and will roll out the feature to all advertisers in the coming weeks; improvements in Meta’s advertising ranking and modeling drove conversion growth that outpaced advertising impressions growth in 2025 Q1

In just the last quarter, we are testing a new ads recommendation model for Reels, which has already increased conversion rates by 5%. We’re seeing 30% more advertisers are using AI creative tools in the last quarter as well…

…In Q1, we introduced our new Generative Ads Recommendation Model, or GEM, for ads ranking. This model uses a new architecture we developed that is twice as efficient at improving ad performance for a given amount of data and compute. This efficiency gain enabled us to significantly scale up the amount of compute we use for model training with GEM trained on thousands of GPUs, our largest cluster for ads training to date. We began testing the new model for ads recommendations on Facebook Reels earlier this year and have seen up to a 5% increase in ad conversions. We’re now rolling it out to additional surfaces across our apps…

…We’re seeing continued momentum with our Advantage+ suite of AI-powered solutions. We’ve been encouraged by the initial test of our streamlined campaign creation flow for sales, app and lead campaigns, which starts with Advantage+ turned on from the beginning for advertisers. In April, we rolled this out to more advertisers and expect to complete the global rollout later this year. We’re also seeing strong adoption of Advantage+ Creative. This week, we are broadening access of video expansion to Facebook Reels for all eligible advertisers, enabling them to automatically adjust the aspect ratio of their existing videos by generating new pixels in each frame to optimize their ads for full screen surfaces. We also rolled out image generation to all eligible advertisers. And this quarter, we plan to continue testing a new virtual try-on feature that uses gen AI to place clothing on virtual models, helping customers visualize how an item may look and fit…

…We continue to evolve our ads platform to drive results that are optimized for each business’ objectives and the way they measure value. One example of this is our incremental attribution feature, which enables advertisers to optimize for driving incremental conversions or conversions we believe would not have occurred without an ad being shown. We’re seeing strong results in testing so far with advertisers using incremental attribution in tests seeing an average 46% lift in incremental conversions compared to their business-as-usual approach. We expect to make this available to all advertisers in the coming weeks…

…Year-over-year conversion growth remains strong. And in fact, we continue to see conversions grow at a faster rate than ad impressions in Q1, so reflecting increased conversion rates. And ads ranking and modeling improvements are a big driver of overall performance gains.

Improvements in the past 6 months to Meta’s content recommendation systems have driven increases of 7% in time spent on Facebook, 6% on Instagram, and 35% on Threads; video consumption in Facebook and Instagram grew strongly in 2025 Q1 because of improvements to Meta’s content recommendation systems; management sees opportunities for further gains in improving the content recommendation systems in 2025; Meta is making progress on longer-term efforts to improve its content recommendation systems in two areas, (1) develop increasingly efficient recommendation systems by incorporating innovations from LLM model architectures, and (2) integrating LLMs into content recommendation systems to better identify what is interesting to a user; management’s testing of Llama in Threads’ recommendation systems has led to a 4% increase in time spent from launch; management is exploring how Llama can be deployed in recommendation systems for photo and video content, which management expects can improve Meta AI’s personalisation by better understanding users’ interests and preferences through their use of Meta’s apps; management launched a new feed in Instagram in the US in 2025 Q1 of content a user’s friends have left a note on or liked and the new feed is producing good results; management has launched the Blend experience that blends a user’s Reels algorithm in direct messages with friends; the increases of 7% in time spent on Facebook and 6% on Instagram seen in the last 6 months is on top of uplift in time spent on Facebook and Instagram that management had already produced in the first 9 months of 2024

In the last 6 months, improvements to our recommendation systems have led to a 7% increase in time spent on Facebook, 6% increase on Instagram and 35% on Threads…

…In the first quarter, we saw strong growth in video consumption across both Facebook and Instagram, particularly in the U.S., where video time spent grew double digits year-over-year. This growth continues to be driven primarily by ongoing enhancements to our recommendation systems, and we see opportunities to deliver further gains this year.

We’re also progressing on longer-term efforts to develop innovative new approaches to recommendations. A big focus of this work will be on developing increasingly efficient recommendation systems so that we can continue scaling up the complexity and compute used to train our models while avoiding diminishing returns. There are promising techniques we’re working on that will incorporate the innovations from LLM model architectures to achieve this. Another area that is showing early promise is integrating LLM technology into our content recommendation systems. For example, we’re finding that LLM’s ability to understand a piece of content more deeply than traditional recommendation systems can help better identify what is interesting to someone about a piece of content, leading to better recommendations.

We began testing using Llama in Threads recommendation systems at the end of last year given the app’s text-based content and have already seen a 4% lift in time spent from the first launch. It remains early here, but a big focus this year will be on exploring how we can deploy this for other content types, including photos and videos. We also expect this to be complementary to Meta AI as it can provide more relevant responses to people’s queries by better understanding their interests and preferences through their interactions across Facebook, Instagram and Threads…

…In Q1, we launched a new experience on Instagram in the U.S. that consists of a feed of content your friends have left a note on or liked, and we’re seeing good results. We also just launched Blend, which is an opt-in experience in direct messages that enables you to blend your Reels algorithm with your friends to spark conversations over each other’s interest…

…We shared on the Q3 2024 call that improvements to our AI-driven feed and video recommendations drove a roughly 8% lift in time spent on Facebook and a 6% lift on Instagram over the first 9 months of last year. Since then, we’ve been able to deliver similar gains in just 6 months’ time with improvements to our AI recommendations delivering 7% and 6% time spent gains on Facebook and Instagram, respectively.

AI is enabling the creation of better content on Meta’s apps; the better content includes AI generating content directly for users and AI helping users produce better content; management thinks that the content created on Meta’s apps will be increasingly interactive over time; management recently launched the stand-alone Edits app that contains an ultra-high resolution, short-form video camera, and generative AI tools to remove backgrounds of video or animate still images; more features on Edits are coming soon; 

AI is also enabling the creation of better content as well. Some of this will be helping people produce better content to share themselves. Some of this will be AI generating content directly for people that is personalized for them. Some of this will be in existing formats like photos and videos, and some of it will be increasingly interactive…

…Our feeds started mostly with text and then became mostly photos when we all got mobile phones with cameras and then became mostly video when mobile networks became fast enough to handle that well. We are now in the video era, but I don’t think that this is the end of the line. In the near future, I think that we’re going to have content in our feeds that you can interact with and that it will interact back with you rather than you just watching it…

…Last week, we launched our stand-alone Edits app, which supports the full creative process for video creators from inspiration and creation to performance insights. Edits has an ultra-high resolution, short-form video camera and includes generative AI tools that enable people to remove the background of any video or animate still images with more features coming soon.

Countries like Thailand and Vietnam with low-cost labour actually conduct a lot of business through Meta’s messaging apps but management thinks this phenomena is absent in developed economies because of the high cost of labour; management thinks that AI will allow businesses in developed economies to conduct business through Meta’s messaging apps; management thinks that every business in the future will have AI business agents that are easy to set up and can perform customer support and sales; Meta is currently testing AI business agents with small businesses in the USA and a few countries across Meta’s apps; management has launched a new agent management experience to make it easier for businesses to train their AI; management’s vision is for that to be one agent that’s interacting with a consumer regardless of where he/she is engaging with the business AI; feedback from the tests are that the AI business agents are saving businesses a lot of time and helping them determine which conversations to spend more time on

In countries like Thailand and Vietnam, where there is a low cost of labor, we see many businesses conduct commerce through our messaging apps. There’s actually so much business through messaging that those countries are both in our top 10 or 11 by revenue even though they’re ranked in the 30s in global GDP. This phenomenon hasn’t yet spread to developed countries because the cost of labor is too high to make this a profitable model before AI, but AI should solve this. So in the next few years, I expect that just like every business today has an e-mail address, social media account and website, they’ll also have an AI business agent that can do customer support and sales. And they should be able to set that up very easily given all the context that they’ve already put into our business platforms…

…We are currently testing business AIs with a limited set of businesses in the U.S. and a few additional countries on WhatsApp, Messenger and on ads on Facebook and Instagram. We’ve been starting with small business and focusing first on helping them sell their goods and services with business AIs…

…We’ve launched a new agent management experience and dashboard that makes it easier for businesses to train their AI based on existing information on their website or WhatsApp profile or their Instagram and Facebook pages. And we’re starting with the ability for businesses to activate AI in their chats with customers. We are also testing business AIs on Facebook and Instagram ads that you can ask about product and return policies or assist you in making a purchase within our in-app browser…

…No matter where you engage with the business AI, it should be one agent that recalls your history and your preferences. And we’re hearing encouraging feedback, particularly that adopting these AIs are saving the business that we’re testing with a lot of time and helping to determine which conversations make sense for them to spend more time on.

Meta AI now has nearly 1 billion monthly actives; management’s focus for Meta AI in 2025 is to establish Meta AI as the leading personal AI for personalization, voice conversations, and entertainment; management thinks people will eventually have an AI to talk to throughout the day on smart-glasses and this AI will be one of the most important and valuable services that has ever been created; management recently released the first Meta AI stand-alone app; the Meta AI stand-alone app is personalised to the user’s behaviour on other Meta apps, and it also has a social feed for discovery on how others are using Meta AI; initial feedback on the Meta AI stand-alone app is good; management expects to focus on scaling and deepening engagement on Meta AI for at least the next year before attempting to monetise; management saw engagement on Meta AI improve when testing Meta AI’s ability to personalize responses by remembering people’s prior queries and their usage of Meta’s apps; management has built personalisation into Meta AI across all of Meta’s apps; the top use cases for Meta AI currently include information gathering, writing assistance, interacting with visual content, and seeking help; WhatsApp has the strongest usage of Meta AI, followed by Facebook; a standalone Meta AI app is important for Meta AI to become the leading personal AI assistant because WhatsApp is currently not the primary messaging app used in the USA; management thinks that people are going to use different AI agents for different things; management thinks having memory of a user will be a differentiator for AI agents

Across our apps, there are now almost 1 billion monthly actives using Meta AI. Our focus for this year is deepening the experience and making Meta AI the leading personal AI with an emphasis on personalization, voice conversations and entertainment. I think that we’re all going to have an AI that we talk to throughout the day, while we’re browsing content on our phones, and eventually, as we’re going through our days with glasses. And I think that this is going to be one of the most important and valuable services that has ever been created.

In addition to building Meta AI into our apps, we just released our first Meta AI stand-alone app. It is personalized. So you can talk to it about interests that you’ve shown while browsing Reels or different content across our apps. And we built a social feed into it. So you can discover entertaining ways that others are using Meta AI. And initial feedback on the app has been good so far.

Over time, I expect the business opportunity for Meta AI to follow our normal product development playbook. First, we build and scale the product. And then once it is at scale, then we focus on revenue. In this case, I think that there will be a large opportunity to show product recommendations or ads as well as a premium service for people who want to unlock more compute for additional functionality or intelligence. But I expect that we’re going to be largely focused on scaling and deepening engagement for at least the next year before we’ll really be ready to start building out the business here…

…Earlier this year, we began testing the ability for Meta AI to better personalize its responses by remembering certain details from people’s prior queries and considering what that person engages with on our apps. We are already seeing this lead to deeper engagement with people we’ve rolled it out to, and it is now built into Meta AI across Facebook, Instagram, Messenger and our new stand-alone Meta AI app in the U.S. and Canada…

…The top use case right now for Meta AI from a query perspective is really around information gathering as people are using it to search for and understand and analyze information followed by social interactions from — ranging from casual chatting to more in-depth discussion or debate. We also see people use it for writing assistance, interacting with visual content, seeking help…

…WhatsApp continues to see the strongest Meta AI usage across our Family of Apps. Most of that WhatsApp engagement is in one-on-one Threads, followed by Facebook, which is the second largest driver of Meta AI engagement, where we’re seeing strong engagement from our feed deep dives integration that lets people ask Meta AI questions about the content that’s recommended to them…

…I also think that the stand-alone app is going to be particularly important in the United States because WhatsApp, as Susan said, is the largest surface that people use Meta AI and which makes sense. If you want to text an AI, having that be closely integrated and a good experience in the messaging app that you use makes a lot of sense. But we’re — while we have more than 100 million people use WhatsApp in the United States, we’re clearly not the primary messaging app in the United States at this point. iMessage is. We hope to become the leader over time. But we’re in a different position there than we are in most of the rest of the world on WhatsApp. So I think that the Meta AI app as a stand-alone is going to be particularly important in the United States to establishing leadership in — as the main personal AI that people use…

…I think that there are going to be a number of different agents that people use, just like people use different apps for different things. I’m not sure that people are going to use multiple agents for the same exact things, but I’d imagine that something that is more focused on kind of enterprise productivity might be different from something that is somewhat more optimized for personal productivity. And that might be somewhat different from something that is optimized for entertainment and social connectivity. So I think there will be different experiences…

…Once an AI starts getting to know you and what you care about in context and can build up memory from the conversations that you’ve had with it over time, I think that will start to become somewhat more of a differentiator.

Meta’s management continues to think of glasses as the ideal form factor for an AI device; management thinks that the 1 billion people in the world today who wear glasses will likely all be wearing smart glasses in the next 5-10 years; management thinks that building the devices people use for Meta’s apps lets the company deliver the best AI and social experiences; sales of the Ray-Ban Meta AI glasses have tripled in the last year and usage of the glasses is high; Meta has new launches of smart glasses lined up for later this year; monthly actives of Ray-Ban Meta AI glasses is up 4x from a year ago, with the number of people using voice commands growing even faster; management has rolled out live translations on Ray-Ban Meta AI glasses to all markets for English, French, Italian and Spanish; management continues to want to scale the Ray-Ban Meta AI glasses to 10 million units or more for its 3rd generation; management intends to run the same monetisation playbook with the Ray-Ban Meta AI glasses as Meta’s other products

Glasses are the ideal form factor for both AI and the metaverse. They enable you to let an AI see what you see, hear what you hear and talk to you throughout the day. And they let you blend the physical and digital worlds together with holograms. More than 1 billion people worldwide wear glasses today, and it seems highly likely that these will become AI glasses over the next 5 to 10 years. Building the devices that people use to experience our services lets us deliver the highest-quality AI and social experiences…

…Ray-Ban Meta AI glasses have tripled in sales in the last year. The people who have them are using them a lot. We’ve got some exciting new launches with our partner, EssilorLuxottica, later this year as well that should expand that category and add some new technological capabilities to the glasses…

…We’re seeing very strong traction with Ray-Ban Meta AI glasses with over 4x as many monthly actives as a year ago. And the number of people using voice commands is growing even faster as people use it to answer questions and control their glasses. This month, we fully rolled out live translations on Ray-Ban Meta AI glasses to all markets for English, French, Italian and Spanish. Now when you are speaking to someone in one of these languages, you’ll hear what they say in your preferred language through the glasses in real time…

…If you look at some of the leading consumer electronics products of other categories, by the time they get to their third generation, they’re often selling 10 million units and scaling from there. And I’m not sure if we’re going to do exactly that, but I think that that’s like the ballpark of the opportunity that we have…

…As a bunch of the products start to hit and start to grow even bigger than the number that I just said is just sort of like the sort of a near-term milestone, then I think we’ll continue scaling in terms of distribution. And then at some point, just like the other products that we build out, we will feel like we’re at a sufficient scale that we’re going to primarily focus on making sure that we’re monetizing and building an efficient business around it.

Meta released the first few Llama 4 models in April 2025 and more Llama 4 models are on the way, including the massive Llama 4 Behemoth model; management thinks leading-edge AI models are critical for Meta’s business, so they want the company to control its own destiny; by developing its own models, Meta is also able to optimise the model to its infrastructure and use-cases; an example of the optimisation is the Llama 4 17-billion model that comes with low latency to suit voice interactions; another example of the optimisation is the models’ industry-leading context window length which helps Meta AI’s personalisation efforts; Llama 4 Behemoth is important for Meta because all the models the company is using internally, and some of the models the company will develop in the future, are distilled from Behemoth

We released the first Llama 4 models earlier this month. They are some of the most intelligent, best multimodal, lowest latency and most efficient models that anyone has built. We have more models on the way, including the massive Llama 4 Behemoth model…

…On the LLM, yes, there’s a lot of progress being made in a lot of different dimensions. And the reason why we want to build this out is — one is that we think it’s important that for kind of how critical this is for our business that we sort of have control of our own destiny and are not depending on another company for something so critical. But two, we want to make sure that we can shape the development to be optimized for our infrastructure and the use cases that we want.

So to that end, Llama 4, the shape of the model with 17 billion parameters per expert was designed specifically for the infrastructure that we have in order to provide the low latency experience to be voice optimized. One of the key things, if you’re having a voice conversation with AI, is it needs to be low latency. So that way, when you’re having a conversation with it, there’s isn’t a large gap between when you stop speaking and it starts. So everything from the shape of the model to the research that we’re doing to techniques that go into it are kind of fit into that.

Similarly, another thing that we focused on was context window length. And in some of our models, we have really — we’re industry-leading on context window length. And part of the reason why we think that that’s important is because we’re very focused on providing a personalized experience. And there are different ways that you can put personalization context into an LLM, but one of the ways to do it is to include some of that context in the context window. And having a long context window that can incorporate a lot of the background that the person has shared across our apps is one way to do that…

…I think it’s also very important to deliver big models like Behemoth, not because we’re going to end up serving them in production, but because of the technique of distilling from larger models, right? The Llama 4 models that we’ve published so far and the ones that we’re using internally and some of the ones that we’ll build in the future are basically distilled from the Behemoth model in order to get the 90%, 95% of the intelligence of the large model in a form factor that is much lower latency and much more efficient.

Meta’s management is accelerating the buildout of Meta’s AI capacity, leading to higher planned investment for 2025; Meta’s capex growth in 2025 is for both generative AI and core business needs with the majority of overall capex supporting Meta’s core business; management continues to build infrastructure in a flexible way where the company can react to how the AI ecosystem develops in the coming years; management is increasing the efficiency of Meta’s workloads and this has helped the company to achieve strong returns from its core AI initiatives

We are accelerating some of our efforts to bring capacity online more quickly this year as well as some longer-term projects that will give us the flexibility to add capacity in the coming years as well. And that has increased our planned investment for this year…

…Our primary focus remains investing capital back into the business with infrastructure and talent being our top priorities…

…Our CapEx growth this year is going toward both generative AI and core business needs with the majority of overall CapEx supporting the core. We expect the significant infrastructure footprint we are building will not only help us meet the demands of our business in the near term but also provide us an advantage in the quality and scale of AI services we can deliver. We continue to build this capacity in a way that grants us maximum flexibility in how and when we deploy it to ensure we have the agility to react to how the technology and industry develop in the coming years…

…The second way we’re meeting our compute needs is by increasing the efficiency of our workloads. In fact, many of the innovations coming out of our ranking work are focused on increasing the efficiency of our systems. This emphasis on efficiency is helping us deliver consistently strong returns from our core AI initiatives.

Meta’s management sees a number of long-term tailwinds that AI can provide for Meta’s business, including making advertising a larger share of global GDP, and freeing up more time for people to engage in entertainment

Over the coming years, I think that the increased productivity from AI will make advertising a meaningfully larger share of global GDP than it is today…

…Over the long term, as AI unlocks more productivity in the economy, I also expect that people will spend more of their time on entertainment and culture, which will create an even larger opportunity to create more engaging experiences across all of these apps.

Meta’s management still expects to develop an AI coding agent sometime in 2025 that can operate as a mid-level engineer; management expects this AI coding agent to be do a substantial part of Meta’s AI research and development in 2026 H2; management is focused on building AI that can run experiments to improve Meta’s recommendation systems

I’d say it’s basically still on track for something around a mid-level engineer kind of starting to become possible sometime this year, scaling into next year. So I’d expect that by the middle to end of next year, AI coding agents are going to be doing a substantial part of AI research and development. So we’re focused on that. Internally, we’re also very focused on building AI agents or systems that can help run different experiments to increase recommendations across our other AI products like the ones that do recommendations across our feeds and things like that.

Microsoft (NASDAQ: MSFT)

Microsoft’s management is seeing accelerating demand across industries for cloud migrations; there are 4 things happening to drive cloud migrations, (1) classic migration, (2) data growth, (3) growth in cloud-native companies’ consumption, and (4) growth in AI consumption, which also requires non-AI consumption 

When it comes to cloud migrations, we saw accelerating demand with customers in every industry, from Abercrombie in French, to Coca-Cola and ServiceNow expanding their footprints on Azure…

…[Question] On your comment about accelerating demand for cloud migrations. I’m curious if you could dig in and extrapolate a little more what you’re seeing there.

[Answer] One is, I’ll just say, the classic migration of whether it’s SQL, Windows Server. And so that sort of again got good steady-state progress because the reality is, I think everyone is now, perhaps there’s another sort of kick in the data center migrations just because of the efficiency the cloud provides. So that’s sort of one part.

The second piece is good data growth. You saw some — like Postgres on Azure — I mean, forgetting even SQL server, Postgres on Azure is growing. Cosmos is growing. The analytics stuff I talked about with Fabric. It’s even the others, whether it is Databricks or even Snowflake on Azure are growing. So we feel very good about Fabric growth and our data growth.

Then the cloud-native growth. So this is again before we even get to AI, some of the core compute consumption of cloud-native players is also pretty very healthy. It was healthy throughout the quarter. We projected to go moving forward as well.

Then the thing to notice is the ratio, and I think we mentioned this multiple times before, if you look underneath even ChatGPT, in fact, that team does a fantastic job of thinking about not only their growth in terms of AI accelerators they need, they use Cosmos DB, they use Postgres. They use core compute and storage. And so there’s even a ratio between any AI workload in terms of AI accelerator to others.

So those are the 4 pockets, I’d say, or 4 different trend lines, which all have a relationship with each other.

Foundry is now used by developers in over 70,000 companies, from enterprises to startups, to design, customize and manage their AI apps and agents; Foundry processed  more than 100 trillion tokens in 2025 Q1, up 5x from a year ago; Foundry now has industry-leading model fine tuning tools; the latest models from AI heavyweights including OpenAI and Meta are available on Foundry;  Microsoft’s Phi family of SLMs (small language model) now has over 38 million downloads (20 million downloads in 2024 Q4); Foundry will soon introduce an LLM (large language model) with 1 billion parameters that can run on just CPUs

Foundry is the agent in AI app factory. It’s now used by developers at over 70,000 enterprises and digital natives from Atomicwork to Epic, Fujitsu and Gainsight to H&R Block and LG Electronics to design, customize and manage their AI apps and agents. We processed over 100 trillion tokens this quarter, up 5x year-over-year, including a record 50 trillion tokens last month alone. And 4 months in, over 10,000 organizations have used our new agent service to build, deploy and scale their agents.

This quarter, we also made a new suite of fine-tuning tools available to customers with industry-leading reliability, and we brought the latest models from OpenAI along with new models from Cohere, DeepSeek, Meta, Mistral, Stability to Foundry. And we’ve expanded our Phi family of SLMs with new multimodal and mini models. All-up, Phi has been downloaded 38 million times. And our research teams are taking it one step further with BitNet b1.58, a billion parameter, large language model that can run on just CPUs coming to the Foundry.

With agent mode in VS Code, Github Copilot can now iterate on code, recognize errors, and fix them automatically; there are other Github agent modes that provide coding support to developers; Microsoft is previewing a first-of-its-kind SWE (software engineering) agent that can execute developer tasks; GitHub Copilot now has 15 million users, up 4x from a year ago; GitHub Copilot is used by a wide range of companies; VS Code has more than 50 million monthly active users

We’re evolving GitHub Copilot from paired to peer programmer with agent mode in VS Code, Copilot can now iterate on code, recognize errors and fix them automatically. This adds to other Copilot agents like Autofix, which helps developers remediate vulnerabilities as well as code review agent, which has already reviewed over 8 million pull requests. And we are previewing a first-of-its-kind SWE-agent capable of asynchronously executing developer tasks. All-up, we now have over 15 million GitHub Copilot users, up over 4x year-over-year. And both digital natives like Twilio and enterprises like Cisco, HPE, Skyscanner and Target continue to choose GitHub Copilot to their developers with AI throughout the entire dev life cycle. With Visual Studio and VS Code, we have the world’s most popular editor with over 50 million monthly active users.

Microsoft 365 Copilot is now used hundreds of thousands of customers, up 3x from a year ago; deal sizes for Microsoft 365 Copilot continue to grow; a record number of customers in 2025 Q1 returned to buy more seats for Microsoft 365 Copilot; new researcher and analyst deep reasoning agents can analyze vast amounts of web and enterprise data on-demand directly within Microsoft 365 Copilot; Microsoft is introducing agents for every role and business process; customers can build their own AI agents with no/low code with Copilot Studio and these agents can handle complex tasks, including taking action across desktop and web apps; 230,000 organisations, including 90% of the Fortune 500, have already used Copilot Studio; customers created more than 1 million custom agents across SharePoint and Copilot Studio, up 130% sequentially

Microsoft 365 Copilot is built to facilitate human agent collaboration, hundreds of thousands of customers across geographies and industries now use Copilot, up 3x year-over-year. Our overall deal size continues to grow. In this quarter, we saw a record number of customers returning to buy more seats. And we’re going further. Just last week, we announced a major update, bringing together agents, notebooks, search and create into a new scaffolding for work. Our new researcher and analyst deep reasoning agents analyze vast amounts of web and enterprise data to deliver highly skilled expertise on demand directly within Copilot…

…We are introducing agents for every role and business process. Our sales agent turns contacts into qualified leads and with sales chat reps can quickly get up to speed on new accounts. And our customer service agent is deflecting customer inquiries and helping service reps resolve issues faster.

With Copilot Studio, customers can extend Copilot and build their own agents with no code, low code. More than 230,000 organizations, including 90% of the Fortune 500 have already used Copilot Studio. With deep reasoning and agent flows in Copilot Studio, customers can build agents that perform more complex tasks and also handle deterministic scenarios like document processing and financial approvals. And they can now build Computer Use Agents that take action on the UI across desktop and web apps. And with just a click, they can turn any SharePoint site into an agent, too. This quarter alone, customers created over 1 million custom agents across SharePoint and Copilot Studio, up 130% quarter-over-quarter.

Azure grew revenue by 33% in 2025 Q1 (was 31% in 2024 Q4), with 16 points of growth from AI services (was 13 points in 2024 Q4); management brought capacity online for Azure AI services faster than expected;  Azure’s non-AI business saw accelerated growth in its Enterprise customer segment as well as some improvement in its scale motions; management thinks the real outperfomer within Azure in 2025 Q1 is the non-AI business; the strength in the AI business in 2025 Q1 came because Microsoft was able to match supply and demand somewhat, and also deliver supply early to some customers; management thinks it’s getting harder to separate an AI workload from a non-AI workload

In Azure and other cloud services, revenue grew 33% and 35% in constant currency, including 16 points from AI services. Focused execution drove non-AI services results, where we saw accelerated growth in our Enterprise customer segment as well as some improvement in our scale motions. And in Azure AI services, we brought capacity online faster than expected…

…The real outperformance in Azure this quarter was in our non-AI business. So then to talk about the AI business, really, what was better was precisely what we said. We talked about this. We knew Q3 that we had and hadn’t really match supply and demand pretty carefully and so didn’t expect to do much better than we had guided to on the AI side. We’ve been quite consistent on that. So the only real upside we saw on the AI side of the business was that we were able to deliver supply early to a number of customers…

…[Question] You mentioned that the upside on Azure came from the non-AI services this time around. I was wondering if you could just talk a little bit more about that.

[Answer] In general, we saw better-than-expected performance across our segments, but we saw acceleration in our largest customers. We call that the Enterprise segment in general. And then in what we talked about of our scale motions, where we had some challenges in Q2, things were a little better. And we still have some work to do in our scale motions, and we’re encouraged by our progress. We’re excited to stay focused on that as, of course, we work through the final quarter of our fiscal year…

…It’s getting harder and harder to separate what an AI workload is from a non-AI workload.

Around half of Microsoft’s cloud and AI-related capex in 2025 Q1 (FY2025 Q3) are for long-lived assets that will support monetisation over the next 15 years and more, while the other half are for CPUs and GPUs; management expects Microsoft’s capex in 2025 Q2 (FY2025 Q4) to increase sequentially, but the guidance for total capex for FY2025 H2 is unchanged from previous guidance (previously, expectation was for capex for 2025 Q1 and 2025 Q2 to be at similar levels as 2024 Q4 (FY2025 Q2); FY2026’s capex is still expected to grow at a lower rate than in FY2025; the mix of spend in FY2026 will shift to short-lived assets in FY2026; demand for Azure’s AI services is growing faster than capacity is being brought online and management expects to have some AI capacity constraints beyond June 2025 (or FY2025 Q4); management’s goal with Microsoft’s data center investments is to be positioned for the workload growth of the future; management thinks pretraining plus test-time compute is a big change in terms of model-training workloads; Microsoft is short of power in fulfilling its data center growth plans; Microsoft’s data center builds have very long lead-times; in Microsoft’s 2024 Q4 (FY 2025 Q1) earnings call, management expected Azure to no longer be capacity-constrained by the end of 2025 Q2 (FY2025 Q4) but demand was stronger than expected in 2025 Q1 (FY2025 Q3); management still thinks they can get better and better capital efficiency from the cloud and AI capex; Azure’s margin on the AI business now is far better than what the margin was when the cloud transition was at a similar stage

Roughly half of our cloud and AI-related spend was on long-lived assets that will support monetization over the next 15 years and beyond. The remaining cloud and AI spend was primarily for servers, both CPUs and GPUs, to serve customers based on demand signals, including our customer contracted backlog of $315 billion…

…We expect Q4 capital expenditures to increase on a sequential basis. H2 CapEx in total remains unchanged from our January H2 guidance. As a reminder, there can be quarterly spend variability from cloud infrastructure build-outs and the timing of delivery of finance leases…

…Our earlier comments on FY ’26 capital expenditures remain unchanged. We expect CapEx to grow. It will grow at a lower rate than FY ’25 and will include a greater mix of short-lived assets, which are more directly correlated to revenue than long-lived assets…

… In our AI services, while we continue to bring data center capacity online as planned, demand is growing a bit faster. Therefore, we now expect to have some AI capacity constraints beyond June…

…the key thing for us is to have our builds and lease be positioned for what is the workload growth of the future, right? So that’s what you have to [ goal ] seek to. So there’s a demand part to it, there is the shape of the workload part to it, and there is a location part to it. So you don’t want to be upside down on having one big data center in one region when you have a global demand footprint. You don’t want to be upside down when the shape of demand changes because, after all, with essentially pretraining plus test-time compute, that’s a big change in terms of how you think about even what is training, right, forget inferencing…

…We will be short power. And so therefore — but it’s not a blanket statement. I need power in specific places so that we can either lease or build at the pace at which we want…

…From land to build to build-outs can be lead times of 5 to 7 years, 2 to 3 years. So we’re constantly in a balancing position as we watch demand curves…

…I did talk about in my comments, we had hoped to be in balance by the end of Q4. We did see some increased demand as you saw through the quarter. So we are going to be a little short still, say, a little tight as we exit the year…

…[Question] You’ve said in the past that you can attain better and better capital efficiency with the cloud business and probably cloud and AI business. Where do you stand today?

[Answer] The way, of course, you’ve seen that historically is right when we went through the prior cloud transitions, you see CapEx accelerate, you build out data center footprint.,, You slowly filled GPU capacity. And over time, you see software efficiencies and hardware efficiencies build on themselves. And you saw that process for us for goodness now quite a long time. And what Satya’s talking about is how quickly that’s happening on the AI side of the business and you add to that model diversity. So think about the same levers plus model efficiency, those compounds. Now the one thing that’s a little different this time is just the pace. And so when you’re seeing that happen, pace in terms of efficiency side, but also pace in terms of the build-out. So it can mask some of the progress… Our margins on the AI side of the business are better than they were at this point by far than when we went through the same transition in the server to cloud transition…

…I think the way to think about this is you can ask the question, what’s the difference between a hosting business and a hyperscale business? It’s software. That’s, I think, the gist of it. Yes, for sure, it’s a capital-intensive business, but capital efficiency comes from that system-wide software optimization. And that’s what makes the hyperscale business attractive and that’s what we want to just keep executing super well on.

Microsoft’s management sees Azure as Microsoft’s largest business; management thinks that the next platform shift in technology, which is AI, is built on the last major platform, which was for cloud computing, so this benefits Microsoft

There’s nothing certain for sure in the future, except for one thing, which is our largest business is our infrastructure business. And the good news here is the next big platform shift builds on that. So it’s not a complete rebuild, having gone through all these platform shifts where you have to come out on the other side with a full rebuild. If there is good news here is that we have a good business in Azure that continues to grow and the new platform depends on that.

It’s possible that software optimizations with AI model development and deployment could lead to even longer useful lives for GPUs, but management wants to observe this for longer

[Question] Could we start to consider the possibility that software enhancements might extend the useful life assumption that you’re using for GPUs?

[Answer] In terms of thinking about the depreciable life of an asset, we like to have a long history before we make any of those changes. So we’re focused on getting every bit of useful life we can, of course, out of assets. But to Satya’s point, that tends to be a software question more than a hardware one.

Netflix (NASDAQ: NFLX)

Netflix’s content talent are already using AI tools to improve the content production process; management thinks AI tools can enable lower-budget projects to access top-grade VFX; Rodrigo Prieto is directing his first feature film with Netflix in 2025, Pedro Paramo, and he’s able to use AI tools for de-aging VFX at a much lower cost than The Irishman film that Prieto worked on 5 years ago; the entire budget for Pedro Paramo is similar to the cost of VFX alone for The Irishman; management’s focus with AI is to find ways for AI to improve the member and creator experience

So our talent today is using AI tools to do set references or previs, VFX sequence prep, shop planning, all kinds of things today that kind of make the process better. Traditionally, only big budget projects would have access to things like advanced visual effects such as de-aging. So today, you can use these AI-powered tools so to enable smaller budget projects to have access to big VFX on screen.

A recent example, I think, is really exciting. Rodrigo Prieto was the DP on The Irishman just 5 years ago. And if you remember that movie, we were using very cutting edge, very expensive de-aging technology that still had massive limitations, still creating a bunch of complexity on set for the actors. It was a giant leap forward for sure, but nowhere near what we needed for that film. So this year, just 5 years later, Rodrigo is directing his first feature film for us, Pedro Páramo in Mexico. Using AI-powered tools he was able to deliver this de-aging VFX to the screen for a fraction of what it cost on The Irishman. In fact, the entire budget of the film was about the VFX cost on The Irishman…

…So our focus is simple, find ways for AI to improve the member and the creator experience.

Netflix’s management is building interactive search into Netflix which is based on generative AI

We’re also building out like new capabilities, an example would be interactive search. That’s based on generative technologies. We expect that will improve that aspect of discovery for members.

Paycom Software (NYSE: PAYC)

Paycom’s GONE is the industry’s first fully automated time-off solution, utilising AI, that automates all time off requests; prior to GONE, 10% of an organisation’s labour cost was unmanaged; GONE can generate ROI of up to 800%, according to Forrester; GONE helped Paycom be named by Fast Company as one of the world’s most innovative companies

Our award-winning solution, GONE, is a perfect example of how Paycom simplifies tests through automation and AI. GONE is the industry’s first fully automated time-off solution that decisions all time-off requests based on customizable guidelines set by the company’s time-off rules. Before GONE, 10% of an organization’s labor cost went substantially unmanaged, creating scheduling errors, increased cost from overpayments, staffing shortages and employee uncertainty over pending time-off requests. According to a Forrester study, GONE’s automation delivers an ROI of up to 800% for clients. GONE continues to receive recognition. Most recently, Fast Company magazine named Paycom, one of the world’s most innovative companies for a second time. This honor specifically recognized GONE and is a testament to how Paycom is shaping our industry by setting new standards for automation across the globe.

PayPal (NASDAQ: PYPL)

PayPal’s management is leaning into agentic commerce; PayPal recently launched the payments industry’s first remote MCP (Model Context Protocol) server to enable AI agent frameworks to integrate with PayPal APIs; the introduction of the MCP allows any business to create an agentic commerce experience; all major AI players are involved with PayPal’s annual Developer Days to engage PayPal’s developer community

At Investor Day, I told you we were leaning into agentic commerce…

…Just a few weeks ago, we launched the industry’s first remote MCP server and enabled the leading AI agent frameworks to seamlessly integrate with PayPal APIs. Now any business can create agentic experience that allow customers to pay, track shipments, manage invoices and more, all powered by PayPal and all within an AI client. As we speak, developers are gathering in our San Jose headquarters for our annual Developer Days. Every major player in AI is represented, providing demos and engaging with our developer community.

Shopify (NASDAQ: SHOP)

Shopify’s management recently launched TariffGuide.ai, an AI-powered tool that provides duty rates based on just a product description and the country of origin, helping merchants source the right products in minutes

And just this past week, we launched TariffGuide.ai. This AI driven tool provides duty rates based on just a product description and the country of origin. Sourcing the right products from the right country can mean the difference between a 0% and a 15% duty rate or higher, And TariffGuide.ai allows merchants to do this in minutes, not days.

Shopify CEO Tobi Lutke penned a memo recently on his vision on how Shopify should be workin with AI; AI is becoming 2nd nature to how Shopify’s employees work, where employees use AI reflexively; before any team requests for additional headcount, they need to first assess if AI can meet their goals; Shopify has built a dozen MCP (model context protocol) servers in the last few weeks to enable anyone in Shopify to ask questions and find resources more easily; management sees AI being a cornerstone of how Shopify delivers value; management is investing more in AI, but the increased investment is not a driver for the lower gross margin in Shopify’s Subscription Solutions segment in 2025 Q1; management does not expect the Subscription Solutions segment’s gross margin to change much in the near term; Shopify has shown strong operating leverage partly because of its growing internal use of AI

AI is at the core of how we operate and is transforming our work processes. For those who have not seen it, I encourage you to check out Toby’s recent company wide email on AI that has now been shared publicly. At Shopify, we take AI seriously. In fact, it’s becoming second nature to how we work. By fostering a culture of reflexive AI usage, our teams default to using AI first, reflexive being the key term here. This also means that before requesting additional headcount or resources, teams are required to start with assessing how they can meet their goals using AI first. This approach is sparking some really fascinating explorations and discussions around the company, challenging the way we think, the way we operate, and pushing us to look ahead as we redefine our decision making processes. In the past couple of weeks, we built a dozen MCP servers that make Shopify’s work legible and accessible. And now anyone within Shopify can ask questions, find resources, and leverage those tools for greater efficiency. This reflexive use of AI goes well beyond internal improvements. It supercharges our team’s capabilities and drives operational efficiencies, keeping us agile. And as we continue to innovate, AI will remain a cornerstone of how we deliver value across the board…

…Gross profit for Subscription Solutions grew 19%, slightly less than the 21% revenue growth for Subscription Solutions. The lower rate was driven primarily by higher cloud and infrastructure hosting costs needed to support higher volumes and geographic expansion. Although we are investing more in AI, it is not a significant factor in this increase. Over the past 5 years, the gross margin for Subscription Solutions has centered around 80%, plus or minus a couple of hundred basis points in any given quarter, and we do not anticipate that trend changing in the near term…

…Our continued discipline on head count across all 3 of R&D, sales and marketing and G&A continued to yield strong operating leverage, all while helping us move even faster on product development aided by our increasing use of AI.

Shopify’s management rearchitected the AI engine of Sidekick, Shopify’s AI merchant assistant, in 2025 Q1; monthly average users of Sidekick has more than doubled since the start of 2025; early results of Sidekick are really strong for both large and small merchants

In Q1, key developments for Sidekick included a complete rearchitecture of the AI engine for deeper reasoning capabilities, enhancing processing of larger business datasets and accessibility in all supported languages, allowing every Shopify merchant to use Sidekick in their preferred language. And these changes, well, they’re working. In fact, our monthly average users of Sidekick continue to climb more than doubling since the start of 2025. Now this is still really early days, but the progress we are making is already yielding some really strong results for merchants, both large and small. 

Shopify acquired Vantage Discovery in 2025 Q1; Vantage Discovery works on AI-powered, multi-vector search; management thinks the acquisition will improve the overall consumer search experience delivered by Shopify’s merchants

In March, we closed the acquisition of Vantage Discovery, which helps accelerate the development of AI-powered, multi-vector search across our search, APIs, shop and storefront search offerings. This acquisition is one piece of a broader strategy to ensure that our merchants are able to continue meeting buyers regardless of where they’re shopping or discovering great products…

…The Vantage team coming in who are rock stars in AI are going to help take our search abilities to the next level.

Shopify’s management is seeing more and more commerce searches starting away from a search engine; Shopify is already working with AI chatbot providers on AI shopping; management thinks that AI shopping is a huge opportunity; management thinks AI agents will be a great opportunity for Shopify too

One of the things we think about is that wherever commerce is taking place, Shopify will be there. And obviously, one of the things we are seeing is that more and more searches are starting on places beyond just somebody’s search engine. That’s a huge opportunity whereby more consumers are going to be searching for great products…

…We’ve talked about some of the partnerships in the past. You’ve seen what we’ve done with Perplexity and OpenAI. We will continue doing that. We’re not going to front run our product road map when it comes to anything, frankly. But we do think though that AI shopping, in particular, is a huge opportunity…

…[Question] How does Shopify view the emergence of AI agents in terms of do you guys see this as an opportunity or more of a threat because, on one hand, they could facilitate direct checkout with their own platforms. On the other hand, this may also unlock some new sales channel for Shopify merchants, very similar to sort of what happened with social media commerce

[Answer] We think it’s a great opportunity. Look, the more channels that exist in the world, the more complexity it is for merchants and brands, that’s where the value of Shopify really shines. So if there’s a new surface area, whether it’s through AI agents or through just simply LLMs and AI wrappers, that consumer goes to, to look for a new pair of sneakers or a new cosmetic or a piece of furniture, they want to have access to the most interesting products for the most important brands, and those are all on Shopify. So for us, we think that all of these new areas where commerce is happening is a great thing. It allows Shopify to increase its value.

Taiwan Semiconductor Manufacturing Company (NYSE: TSM)

TSMC’s management continues to expect AI accelerators revenue to double in 2025; management has factored China-bans on US chips into TSMC’s 2025 outlook; AI-related demand outside of China appears to have become even stronger over the last 3 months

We reaffirm our revenue from AI accelerated to double in 2025. The AI accelerators we define as AI GPU, AI ASIC and HPM controllers for AI training and inference in the data center. Based on our customers’ strong demand, we are also working hard to double our CoWoS capacity in 2025 to support their needs…

…[Question] The geopolitical risk, micro concerns is one of the major uncertainty nowadays. Last 2 days, we have like H20 being banned in China, blah, blah, blah. So how does that impact to TSMC’s focus and production planning, right? Do we have enough other customers and demand to keep our advanced node capacity fully utilized? Or how does that change our long-term production planning moving forward?

[Answer] Of course, we do not comment on specific customers or product, but let me assure you that we have taken this into consideration when providing our full year’s growth outlook. Did I answer the question?…

…[Question] AI is still expected to double this year despite the U.S. ban on AI GPUs into China. And I guess, China was a meaningful portion of accelerated shipments well over 10% of volumes. So factoring this in, it would imply your AI outlook this year, still doubling would mean that the AI orders have improved meaningfully outside of China in the last sort of 3 months. Is that how we should interpret your comment about you still expect the business to double?

[Answer] 3 months ago, we are — we just cannot supply enough wafer to our customer. And now it’s a little bit balanced, but still, the demand is very strong. And you are right, other than China, the demand is still very strong, especially in U.S.

TSMC’s management has a disciplined approach when building capacity and management recognises how important the discipline is given the high forecasted demand for AI-related chips

At TSMC, higher level of capital expenditures is always correlated with higher growth opportunities in the following years. We reiterate our 2025 capital budget is expected to be between USD 38 billion and USD 42 billion as we continue to invest to support customers’ growth. About 70% of the capital budget will be allocated for advanced process technologies. About 10% to 20% will be spent for specialty technologies and about 10% to 20% will be spent for advanced packaging, testing, mass-making and others. Our 2025 CapEx also includes a small amount related to our recently announced additional $100 billion investment plan to expand our capacity in Arizona…

…To address the structural increase in the long-term market demand profile, TSMC employed a disciplined and robust capacity planning system. This is especially important when we have such high forecasted demand from AI-related business. Externally, we work closely with our customers and our customers’ customers to plan our capacity. Internally, our planning system involves multiple teams across several functions to assess and evaluate the market demand from both a top-down and bottom-up approach to determine the appropriate capacity build.

TSMC’s management expects the Foundry 2.0 industry to grow 10% year-on-year in 2025, driven by AI-related demand and mild recovery in other end markets; management expects TSMC to outperform the Foundry 2.0 industry in 2025

Looking at the full year of 2025, we expect Foundry 2.0 industry growth to be supported by robust AI-related demand and a mild recovery in other end market segment. In January, we had forecasted a Foundry 2.0 industry to grow 10 points year-over-year in 2025, which is consistent with IDC’s forecast of 11% year-over-year growth for Foundry 2.0…

…We are confident TSMC can continue to outperform the Foundry 2.0 industry growth in 2025.

TSMC’s management thinks impact from recent AI models, including DeepSeek, will lower the barrier to future long-term AI development; TSMC’s management continues to expect mid-40% revenue CAGR from AI accelerators in the 5-years starting from 2024

Recent developments are also positive to AI’s long-term demand outlook. In our assessment, the impact from AI recent models, including DeepSeek, will drive greater efficiency and help lower the barrier to future AI development. This will lead to wider usage and greater adoption of AI models, which all require use of leading-edge silicon. These developments only serve to strengthen our conviction in the long-term growth opportunities from the industry megatrend of 5G, AI and HPC…

…Based on our planning framework, we are confident that our revenue growth from AI accelerators will approach a mid-40s percentage CAGR for the next 5 years period starting from 2024.

TSMC’s 2nd fab in Arizona will utilise N3 process technology and is already complete and management wants to speed up volume production schedule to meet AI-related demand

Our first fab in Arizona has already successfully entered high-volume production in 4Q ’24, utilizing N4 process technology with a yield comparable to our fab in Taiwan. The construction of our second fab, which will utilize the 3-nanometer process technology, is already complete and we are working on speeding up the volume production schedule based on the strong AI-related demand from our customers. Our third and fourth fab will utilize N2 and A16 process technologies and with the expectation of receiving all the necessary permits are scheduled to begin construction later this year. Our fifth and sixth fab will use even more advanced technologies. The construction and ramp schedule for this fab will be based on our customers’ demand.

TSMC’s management believes its A16 technology has a best-in-class backside power delivery solution that is also the first in the industry; A16 is best suited for specific HPC (high-performance computing) products, which means it is best suited for AI-related workloads; A16 is scheduled for volume production in 2026 H2

We also introduced a 16 feature in super power rail or SPR as a separate offering. Compared with the N2P, A16 provides a further 8% to 10% speed improvement at the same power or 15% to 20% power improvement at the same speed and additional 7% to 10% chip density gain. A16 is best suited for specific HPC products with complex signal route and dense power delivery network. Volume production is scheduled for second half 2026.

Tesla (NASDAQ: TSLA)

Tesla’s management continues to expect fully autonomous Tesla rides in Austin, Texas in June 2025; management will sell full autonomy software for Model Y in Austin; management now demarcates CyberCab as a separate product, and all of the other models (S, 3, X, Y) that is compatible with autonomous software as being robotaxis; management reiterates that once Tesla can solve for autonomy in 1 city, it can very quickly scale because Tesla’s autonomous solution is a general solution, not a city-specific solution; Tesla’s autonomous solution involves AI and a specific Tesla-designed AI chip, as opposed to expensive sensors and high-precision maps; the fully autonomous Teslas in June 2025 in Austin will be Model Ys; management expects full autonomy in Tesla’s fleet to ramp up very quickly; management is confident that Tesla will have large-scale autonomy by 2026 H2, meaning, millions of fully autonomous Tesla vehicles by 2026 H2; even with the introduction of full autonomy, management thinks there will be some localised parameters – effectively a mixture of experts model – set for safety; management thinks Tesla’s autonomous solution can scale well because when the FSD (Full Self Driving) software was deployed in China, it used very minimal China-specific data and yet could work well in China; validation of Tesla’s autonomous solution will be important in determining its rate of acceptance; there are now convoys of Teslas in Austin running autonomously in testing in order to compress Tesla’s AI’s learning curve; a consumer in China used FSD on a narrow mountain dirt road; management expects FSD unsupervised to be available for personal use by end of 2025; Musk thinks the first Model Y ro drive itself from factory to customer will happen later in 2025; newly-manufactured Model Ys are already driving themselves around in Tesla factories

We expect to have — be selling fully autonomous rides in June in Austin as we’ve been saying for now several months. So that’s continued…

…Unsupervised autonomy will first be sold for the Model Y in Austin, and then actually, should parse out the term for robotic taxi or robotaxi and just generally like what’s the Cybercab because we’ve got a product called the Cybercab. And then any Tesla, which could be an S, 3, X or Y that is autonomous is a robotic taxi or a robotaxi. It’s very confusing. So the vast majority of the Tesla fleet that we’ve made is capable of being a robotaxi or a robotic taxi…

…Once we can make the system work where you can have paid rides, fully autonomously with no one in the car in 1 city, that is a very scalable thing for us to go broadly within whatever jurisdiction allows us to operate. So because what we’re solving for is a general solution to autonomy, not a city-specific solution for autonomy, once we make it work in a few cities, we can basically make it work in all cities in that legal jurisdiction. So if it’s — once we can make the pace to work in a few cities in America, we can make it work anywhere in America. Once we can make it work in a few cities in China, we can make it work anywhere in China, likewise in Europe, limited only by regulatory approvals. So this is the advantage of having a generalized solution using artificial intelligence and an AI chip that Tesla designed specifically for this purpose, as opposed to very expensive sensors and high-precision maps on a particular neighborhood where that neighborhood may change or often changes and then the car stops working. So we have a general solution instead of a specific solution…

…The Teslas that will be fully autonomous in June in Austin are fully Model Ys. So that is — it’s currently on track to be able to do paid rides fully autonomously in Austin in June, and then to be in many other cities in the U.S. by the end of this year.

It’s difficult to predict the exact ramp sort of week by week and month by month, except that it will ramp up very quickly. So it’s going to be like some — basically an S-curve where it’s very difficult to predict the intermediate slope of the S-curve, but you kind of know where the S-curve is going to end up, which is the vast majority of the Tesla fleet being autonomous. So that’s why I feel confident in predicting large-scale autonomy around the middle of next year, certainly the second half next year, meaning I bet that there will be millions of Teslas operating autonomously, fully autonomously in the second half of next year, yes…

…It does seem increasingly likely that there will be a localized parameter set sort of — especially for places that have, say, a very snowy weather, like I say, if you’re in the Northeast or something like this — you can think of — it’s kind of like a human. Like you can be a very good driver in California but are you going to be also a good driver in a blizzard in Manhattan? You’re not going to be as good. So there is actually some value in — you can still drive but your probability of an accident is higher. So the — it’s increasingly obvious that there’s some value to having a localized set of parameters for different regions and localities…

…You can see that from our deployment of FSD supervised in China with this very minimal data that’s China-specific, the model is generalized quite well to completely different driving styles. That just like shows that the AI-based solution that we have is the right one because if you had gone down the previous rule-based solutions, sort of like more hard-coded HD map-based solutions, it would have taken like many, many years to get China to work. You can see those in the videos that people post online themselves. So the generalized solution that we are pursuing is the right one that’s going to scale well…

…You can think of this like location-specific parameters that Elon alluded to as a mixture of experts. And if you are sort of familiar with the AI models, Grok and others, they all use this mixture of experts to sort of specialize the parameters to specific tasks while still being general…

…What are the critical things that need to get right, one thing I would like to note is validation. Self-driving is a long-tail problem where there can be a lot of edge cases that only happen very, very rarely. Currently, we are driving around in Austin using our QA fleet, but then super [ rare ] to get interventions that are critical for robotaxi operation. And so you can go many days without getting a single intervention. So you can’t easily know whether you are improving or regressing in your capacity. And we need to build out sophisticated simulations, including neural network-based video generation…

…There’s just always a convoy of Teslas going — just going all over to Austin in circles. But yes, I just can’t emphasize this enough. In order to get a figure on the long-tail things, it’s 1 in 10,000, that says 1 in 20,000 miles or 1 in 30,000. The average person drives 10,000 miles in a year. So not trying to compress that test cycle into a matter of a few months. It means you need a lot of cars doing a lot of driving in order to compress that to do in a matter of a month what would normally take someone a year…

…I saw one guy take a Tesla on — autonomously on a narrow dirt road across like a mountain. And I’m like, still a very brave person. And I said this driving along the road with no barriers where he makes a mistake, he’s going to plunge to his doom. But it worked…

…[Question] when will FSD unsupervised be available for personal use on personally-owned cars?

[Answer] Before the end of this year… the acid test being you should — can you go to sleep in your car and wait until your destination? And I’m confident that will be available in many cities in the U.S. by the end of this year…

…I’m confident also that later this year, the first Model Y will drive itself all the way to the customer. So from our — probably from a factory in Austin and our one in here in Fremont, California, I’m confident that from both factories, we’ll be able to drive directly to a customer from the factory…

…We have — it has been put to use — it’s doing useful work fully autonomously at the factories, as Ashok was mentioning, the cars drive themselves from end of line to where they supposed to be picked up by a truck to be taken to the customer… It’s important to note in the factories, we don’t have dedicated lengths or anything. People are coming out every day, trucks delivering supplies, parts, construction.

Tesla’s management expects thousands of Optimus robots to be working in Tesla factories by end-2025; management expects Optimus to be the fastest product to get to millions of units per year; management thinks Tesla can get to 1 million units annually in 4-5 years; management expects to make thousands of Optimus robots at the end of this year; there’s no existing supply chain for all of Optimus’s components, so Tesla has to build a supply chain from scratch; the speed of manufacturing of a product is governed by the speed of the slowest item in the supply chain, but in Optimus’s case, there are many, many such items since it’s so new; Optimus production is currently rate-limited by restrictions on rare-earth magnets from China but management is working on it; management still has no idea how Optimus’s supply chain will look like at maturity

Making good progress in Optimus. We expect to have thousands of Optimus robots working in Tesla factories by the end of this year beginning this fall. And we expect to see Optimus faster than any product, I think, in history to get to millions of units per year as soon as possible. I think we feel confident in getting to 1 million units per year in less than 5 years, maybe 4 years. So by 2030, I feel confident in predicting 1 million Optimus units per year. It might be 2029…

…This year, we’ll make a few — we do expect to make thousands of Optimus robots, but most of that production is going to be at the end of the year…

…Almost everything in Optimus is new. There’s not like an existing supply chain for the motors, gearboxes, electronics, actuators, really anything in the Optimus apart from the AI for Tesla, the Tesla AI computer, which is the same as the one in the car. So when you have a new complex manufactured product, it will move as fast as the slowest and the least lucky component in the entire thing. And as a first order approximation, there’s like 10,000 unique things. So that’s why anyone who tells you they can predict with precision, the production ramp of the truly new product is — doesn’t know what they’re talking about. It is literally impossible…

…Now Optimus was affected by the magnet issue from China because the Optimus actuators in the arm to use permanent magnet. Now Tesla, as a whole, does not need to use permanent magnets. But when something is volume constrained like an arm of the robot, then you want to try to make the motor as small as possible. And then — so we did design in permanent magnets for those motors and those were affected by the supply chain by basically China requiring an export license to send out any rare earth magnets. So we’re working through that with China. Hopefully, we’ll get a license to use the rare earth magnets. China wants some assurances that these are not used for military purposes, which obviously they’re not. They’re just going into a humanoid robot. So — and it’s a nonweapon system…

…[Question] Wanted to ask about the Optimus supply chain going forward. You mentioned a very fast ramp-up. What do you envision that supply chain looking like? Is it going to require many more suppliers to be in the U.S. now because of the tariffs?

[Answer] We’ll have to see how things settle out. I don’t know yet. I mean some things we’re doing, as we’ve already talked about, which is that we’ve already taken tremendous steps to localize our supply chain. We’re more localized than any other manufacturer. And we have a lot of things kind of underway that to increase the localization to reduce supply chain risk associated with geopolitical uncertainty.

Tesla’s supervised FSD (full-self driving) software is safer than a human driver; management has been using social media (X, or Twitter) to encourage people to try out Tesla’s FSD software; management did not directly answer a question on FSD pricing once the vehicle can be totally unsupervised

Not only is FSD supervised safer than a human driver, but it is also improving the lifestyle of individuals who experience it. And again, this is something you have to experience and anybody who has experienced just knows it. And we’ve been doing a lot lately to try and get those stories out, at least on X, so that people can see how other people have benefited from this…

…[Question] Can we envision when you launch unsupervised FSD that there could be sort of a multitiered pricing approach to unsupervised versus supervised similar to what you did with autopilot versus FSD in the past?

[Answer] I mean this is something which we’ve been thinking about. I mean just so now for people who have been trying FSD and who’ve been using FSD, they think given the current pricing is too cheap because for $99, basically getting a personal shop… I mean we do need to give people more time to — if they want to look at — like a key breakpoint is, can you read your text messages or not? Can you write a text message or not? Because obviously, people are doing this, by the way, with unautonomous cars all the time. And if you just go over and drive down the highway and you’ll see people texting while driving doing 80-mile an hour… So that value — it will really be profound when you can basically do whatever you want, including sleep. And then that $99 is going to seem like the best $99 you ever spent in your life.

Tesla’s management thinks Waymo vehicles are too expensive compared to Teslas; Waymo has expensive sensor suites; management thinks Tesla will have lion’s share of the robotaxi market; a big difference between Tesla and Waymo is that Tesla is also manufacturing the cars whereas Waymo is retrofitting cars from other parties; management thinks Tesla’s vision-only approach will not have issues with cameras becoming blinded by glare and stuff because the system uses direct photon counting and bypasses image signal processing

The issue with Waymo’s cars is it costs way more money, but that is the issue. The car is very expensive, made in low-volume. Teslas are probably cost 1/4, 20% of what a Waymo costs and made in very high volume. Ironically, like we’re the ones who made the bet that a pure AI solution with cameras and [ already ] what the car actually will listen for sirens and that kind of thing. It’s the right move. And Waymo decided that an expensive sensor suite is the way to go, even though Google is very good at AI. So I’m wondering…

….As far as I’m aware, Tesla will have, I don’t know, 99% market share or something ridiculous…

…The other thing which people forget is that we’re not just developing the software solution, we are also manufacturing the cars. And like you know what like Waymo has, they’re taking cars and then trying to…

…[Question] You’re still sticking with the vision-only approach. A lot of autonomous people still have a lot of concerns about sun glare, fog and dust. Any color on how you anticipate on getting around those issues? Because my understanding, it kind of blinds the camera when you get glare and stuff.

[Answer] Actually, it does not blind the camera. We use an approach which is a direct photon count. So when you see a processed image, so the image that goes from the — with sort of photon counter, the silicon photon counter, that they get — goes through a digital signal processor or image signal processor. That’s normally what happens. And then the image that you see looks all washed out because if it’s — you pointed a camera at the sun, the post-processing of the photon counting washes things out. It actually adds noise. So quite a big breakthrough that we made some time ago was to go with direct photon counting and bypass the image signal processor. And then you can drive pretty much straight at the sun, and you can also see in what appears to be the blackest of night. And then here in fog, we can see as well as people can, probably better, but in fact probably slightly better than people than the average person anyway.

Tesla’s AI software team and chip-design team was built from scratch with no acquisitions; management thinks Tesla’s team is the best

It is worth noting that Tesla has built an incredible AI software team and AI hardware chip design team from scratch, didn’t acquire anyone. We just built it. So yes, it’s really — I mean I don’t see anyone being able to compete with Tesla at present.

Tesla’s management thinks China is ahead of the USA in physical AI with respect to autonomous drones because China has the ability to manufacture autonomous drones, but the USA does not;  management thinks Tesla is ahead of any company in the world, even Chinese companies, in terms of humanoid autonomous robots 

[Question] Between China and United States, who, in your opinion, is further ahead on the development of physical AI, specifically on humanoid and also drones?

[Answer] A friend of mine posted on X, I reposted it. I think of a prophetic statement, which is any country that cannot manufacture its own drones is going to be the vassal state of any country that can. And we can’t — America cannot currently manufacture its own drones. Let that sink in, unfortunately. So China, I believe manufactures about 70% of all drones. And if you look at the total supply chain, China is almost 100% of drones are — have a supply chain dependency on China. So China is in a very strong position. And here in America, we need to tip more of our people and resources to manufacturing because this is — and I have a lot of respect for China because I think China is amazing, actually. But the United States does have such a severe dependency on China for drones and be unable to make them unless China gives us the parts, which is currently the situation.

With respect to humanoid robots, I don’t think there’s any company and any country that can match as well. Tesla and SpaceX are #1. And then I’m a little concerned that on the leaderboard, ranks 2 through 10 will be Chinese companies. I’m confident that rank 1 will be Tesla.

The Trade Desk (NASDAQ: TTD)

Trade Desk’s industry-leading Koa AI tools are embedded across Kokai; adoption of Kokai is now ahead of schedule, with 2/3 of clients using it; the bulk of spending on Trade Desk now takes place on Kokai; management continues to expect all Trade Desk clients to be using Kokai by end-2025; management is confident that Kokai will be seen as the most powerful buying platform by the industry by end-2025

The injection of our industry-leading Koa AI tools across every aspect of our platform has been a game changer, and we are just getting started…

…The core of Kokai has been delivered and adoption is now ahead of schedule. Around 2/3 of our clients are now using it and the bulk of the spend in our platform is now running through Kokai. We expect all clients to be using it by the end of the year…

…I’m confident that by the end of this year, we will reflect on Kokai as the most powerful buying platform the industry has ever seen, precisely because it combines client needs with the strong point of view on where value is shifting and how to deliver the most efficient return on ad spend.

…Kokai adoption now represents the majority of our spend, almost 2/3, a significant acceleration from where we ended 2024.

Deutsche Telekom used Kokai’s AI tools and saw an 11x improvement in post-click conversions and an 18x improvement in the cost of conversions; Deutsche Telekom is now planning to use Kokai across more campaigns and transition from Trade Desk’s previous platform, Solimar, like many other Trade Desk clients

Deutsche Telekom. They’re running the streaming TV service called Magenta TV, and they use our platform to try to grow their subscriber base…

…Using seed data from their existing customers, Deutsche Telekom was able to use the advanced AI tools in our Kokai platform to find new customers and define the right ad impressions across display and CTV, most relevant to retain those new customers successfully, and the results were very impressive. They saw an 11x improvement in post-click conversions attributed to advertising and an 18x improvement in the cost of those conversions. Deutsche Telekom is now planning to use Kokai across more campaigns, a transition that is fairly typical as clients move from our previous platform, Solimar to our newer, more advanced AI fuel platform, Kokai.

Visa (NASDAQ: V)

Visa recently announced the Authorize.net product that features AI capabilities, including an AI agent; Authorize.net enables all different types of payments;

In Acceptance Solutions, we recently announced 2 new product offerings. The first is a completely new version of Authorize.net, launching in the U.S. next quarter and additional countries next year. It features a streamlined user in base; AI capabilities with an AI agent, Anet; improved dashboards for day-to-day management and support for in-person card readers and Tap to Phone. It will help businesses analyze data, summarize insights and adapt to rapidly changing customer trends…

……I talked about the Authorize.net platform that we’ve relaunched and we’re relaunching. That’s a great example of enabling all different types of payments. And that’s going to be, we think, a really positive impact in the market specifically focused on growing our share in small business checkout.

Visa has an enhanced holistic fraud protection solution known as Adaptive Real-time Individual Change identification (ARIC) Risk Hub; ARIC Risk Hub uses AI to build more accurate risk profiles;

We also now provide an enhanced holistic fraud protection solution from Featurespace called the Adaptive Real-time Individual Change identification, or ARIC, Risk Hub. This solution utilizes machine learning and AI solutions to enable clients to build more accurate risk profiles and more confidently detect and block fraudulent transactions, ultimately helping to increase approvals and stop bad actors in real time. 

---

Disclaimer: The Good Investors is the personal investing blog of two simple guys who are passionate about educating Singaporeans about stock market investing. By using this Site, you specifically agree that none of the information provided constitutes financial, investment, or other professional advice. It is only intended to provide education. Speak with a professional before making important decisions about your money, your professional life, or even your personal life. I have a vested interest in Alphabet, Amazon, Apple, ASML, Coupang, Datadog, Mastercard, Meta Platforms, Microsoft, Netflix, Paycom Software, PayPal, Shopify, TSMC, Tesla, The Trade Desk and Visa. Holdings are subject to change at any time.

Warren Buffett is one of my investment heroes. On 3 May 2025, he held court at the 2025 Berkshire Hathaway AGM (annual general meeting).

For many years, I’ve anticipated the AGM to hear his latest thoughts. This year’s session holds special significance because it may well be his last – during the AGM, he announced that he would be stepping down as CEO of Berkshire Hathaway by the end of this year, ending an amazing 60-year run since becoming the company’s leader in 1965. Greg Abel is slated to be Berkshire Hathaway’s next CEO.

The most recent Berkshire meeting contained great insights from Buffett and other senior Berkshire executives that I wish to share and document. Before I get to them, I would like to thank my friend Thomas Chua for performing a great act of public service. Shortly after the AGM ended, Thomas posted a transcript of the session at his excellent investing website Steady Compounding. 

Without further ado, the italicised passages between the two horizontal lines below are my favourite takeaways after I went through Thomas’ transcript.

---

Buffett thinks his idea on import certificates is different from tariffs and that it’s important to have more balanced trade between countries; he also thinks that trade should not be wielded as a weapon, and that the more prosperous the world becomes, the better the USA would be

Becky Quick: Thanks Warren. This first question comes from Bill Mitchell. I received more questions about this than any other question. He writes, “Warren, in a 2003 Fortune article, you argued for import certificates to limit trade deficits and said these import certificates basically amounted to a tariff, but recently you called tariffs an act of economic war. Has your view on trade barriers changed or do you see import certificates as somehow distinct from tariffs?”

Warren Buffett: Well, the import certificates were distinct, but their goal was to balance imports against exports so that the trade deficit would not grow in an enormous way. It had various provisions to help third world countries catch up a little bit. They were designed to balance trade, and I think you can make very good arguments that balanced trade is good for the world. It makes sense for cocoa to be raised in Ghana and coffee in Colombia and a few other things…

…There’s no question that trade can be an act of war, and I think it’s led to bad things like the attitudes it’s brought out in the United States. We should be looking to trade with the rest of the world. We should do what we do best, and they should do what they do best…

…The main thing is that trade should not be a weapon. The United States has become an incredibly important country starting from nothing 250 years ago – there’s never been anything like it. And it’s a big mistake when you have 7.5 billion people who don’t like you very well and you have 300 million people crowing about how well they’ve done. I don’t think it’s right and I don’t think it’s wise. The more prosperous the rest of the world becomes, it won’t be at our expense – the more prosperous we’ll become and the safer we’ll feel and your children will feel someday.

Buffett did not look at macroeconomic factors in Japan when making the huge investments he did in five Japanese trading houses; Berkshire won’t be selling the Japanese investments for a long, long time, if at all; Berkshire would be happy to invest a lot more in Japan if there was capacity to do so; the fact that Berkshire could borrow in Japanese Yen to hedge the Japanese investments’ currency risk is merely a lucky coincidence

Question: Mr. Buffett and Mr. Munger did a very good and successful investment in Japan in the past five or six years. The recent CPI in Japan is currently above 3%, not far away from its 2% target. Bank of Japan seems very determined in raising rates while Fed, ECB, and other central banks are considering cutting them. Do you think BOJ makes sense to proceed with the rate hike? Will its planned rate hike deter you from further investing in the Japanese stock market or even considering realizing your current profits?

Warren Buffett: Well, I’m going to extend the same goodwill to Japan that you’ve just extended to me. I’ll let the people of Japan determine their best course of action in terms of economics. It’s an incredible story. It’s been about six years now since our Japanese investments. I was just going through a little handbook that probably had two or three thousand Japanese companies in it. One problem I have is that I can’t read that handbook anymore – the print’s too small. But there were these five trading companies selling at ridiculously low prices. So I spent about a year acquiring them. And then we got to know the people better, and everything that Greg and I saw, we liked better as we went along…

…Greg Abel: When you think of the five companies, there’s definitely a couple meetings a year, Warren. The thing we’re building with the five companies is, one, it’s been a very good investment, but we really envision holding the investment for 50 years or forever…

…Warren Bufett: We will not be selling any stock. That will not happen in decades, if then…

…It’s too bad that Berkshire has gotten as big as it is because we love that position and I’d like it to be a lot larger. Even with the five companies being very large in Japan, we’ve got at market in the range of $20 billion invested, but I’d rather have $100 billion than $20 billion…

…The Japanese situation is different because we intend to stay so long with that position and the funding situation is so cheap that we’ve attempted to some degree to match purchases against yen-denominated funding. But that’s not a policy of ours…

…Greg Abel: There’s no question we were fundamentally very comfortable with investing in the five Japanese companies and recognizing we’re investing in yen. The fact we could then borrow in yen was almost just a nice incremental opportunity. But we were very comfortable both with the Japanese companies and with the currency we would ultimately realize in yen.

Just the simple act of reading about companies can lead to great investment opportunities

Warren Buffett: It’s been about six years now since our Japanese investments. I was just going through a little handbook that probably had two or three thousand Japanese companies in it…

…I never dreamt of that when I picked up that handbook. It’s amazing what you can find when you just turn the page. We showed a movie last year about “turn every page,” and I would say that turning every page is one important ingredient to bring to the investment field. Very few people do turn every page, and the ones who turn every page aren’t going to tell you what they’re finding. So you’ve got to do a little of it yourself.

Berkshire’s current huge cash position is the result of Buffett not being able to find sufficiently attractive investment opportunities; Buffett thinks that great investment opportunities appear infrequently

Becky Quick: This next question comes from Advate Prasad in New York. He writes, “Today, Berkshire holds over $300 billion in cash and short-term investments, representing about 27% of total assets, a historically high figure compared to the 13% average over the last 25 years. This has also led Berkshire to effectively own nearly 5% of the entire US Treasury market. Beyond the need for liquidity to meet insurance obligations, is the decision to raise cash primarily a de-risking strategy in response to high market valuations?…

…Warren Buffett: Well, I wouldn’t do anything nearly so noble as to withhold investing myself just so that Greg could look good later on. If he gets any edge of what I leave behind, I’ll resent it. The amount of cash we have – we would spend $100 billion if something is offered that makes sense to us, that we understand, offers good value, and where we don’t worry about losing money. The problem with the investment business is that things don’t come along in an orderly fashion, and they never will. I’ve had about 16,000 trading days in my career. It would be nice if every day you got four opportunities or something like that with equal attractiveness. If I was running a numbers racket, every day would have the same expectancy that I would keep 40% of whatever the handle was, and the only question would be how much we transacted. But we’re not running that kind of business. We’re running a business which is very opportunistic.

Investing in stocks is a much better investment-bet than investing in real estate

Warren Buffett: Well, in respect to real estate, it’s so much harder than stocks in terms of negotiation of deals, time spent, and the involvement of multiple parties in the ownership. Usually when real estate gets in trouble, you find out you’re dealing with more than just the equity holder. There have been times when large amounts of real estate have changed hands at bargain prices, but usually stocks were cheaper and they were a lot easier to do.

Charlie did more real estate. Charlie enjoyed real estate transactions, and he actually did a fair number of them in the last 5 years of his life. But he was playing a game that was interesting to him. I think if you’d asked him to make a choice when he was 21 – either be in stocks exclusively for the rest of his life or real estate for the rest of his life – he would have chosen stocks. There’s just so much more opportunity, at least in the United States, that presents itself in the security market than in real estate…

…When you walk down to the New York Stock Exchange, you can do billions of dollars worth of business, totally anonymous, and you can do it in 5 minutes. The trades are complete when they’re complete. In real estate, when you make a deal with a distressed lender, when you sign the deal, that’s just the beginning. Then people start negotiating more things, and it’s a whole different game with a different type of person who enjoys the game.

Berkshire’s leaders think AI will have a massive impact on the insurance business, but they are not in a hurry to pour money into AI as they think there’s plenty of faddish capital in the space

Ajit Jain: There is no question in my mind that AI is going to be a real game-changer. It’s going to change the way we assess risk, we price risk, we sell risk, and then the way we end up paying claims. Having said that, I certainly also feel that people end up spending enormous amounts of money trying to chase the next fashionable thing…

…Right now the individual insurance operations do dabble in AI and try to figure out the best way to exploit it. But we have not yet made a conscious big-time effort in terms of pouring a lot of money into this opportunity.

Buffett prefers Ajit Jain to any kind of sophisticated AI systems when pricing insurance risks

Warren Buffett: I wouldn’t trade everything that’s developed in AI in the next 10 years for Ajit. If you gave me a choice of having a hundred billion dollars available to participate in the property casualty insurance business for the next 10 years and a choice of getting the top AI product from whoever’s developing it or having Ajit making the decisions, I would take Ajit anytime – and I’m not kidding about that.

Despite the political upheaval happening in the USA right now, Buffett still thinks the long-term future of the country is incredibly bright; in Buffett’s eyes, the USA has been through plenty of tumultuous periods and emerged stronger

Warren Buffett: America has been undergoing significant and revolutionary change ever since it was developed. I mentioned that we started out as an agricultural society with high promises that we didn’t deliver on very well. We said all men were created equal, and then we wrote a constitution that counted blacks as three-fifths of a person. In Article 2, you’ll find male pronouns used 20 times and no female pronouns. So it took until 1920, with the 19th amendment, to finally give women the vote that we had promised back in 1776.

We’re always in the process of change, and we’ll always find all kinds of things to criticize in the country. But the luckiest day in my life is the day I was born, because I was born in the United States. At that time, about 3% of all births in the world were taking place in the United States. I was just lucky, and I was lucky to be born white, among other things…

…We’ve gone through all kinds of things – great recessions, world wars, the development of the atomic bomb that we never dreamt of when I was born. So I would not get discouraged about the fact that we haven’t solved every problem that’s come along. If I were being born today, I would just keep negotiating in the womb until they said I could be in the United States.

It’s important to be patient while waiting for opportunities, but equally important to pounce when the opportunity appears

Warren Buffett: The trick when you get in business with somebody who wants to sell you something for $6 million that’s got $2 million of cash, a couple million of real estate, and is making $2 million a year, is you don’t want to be patient at that moment. You want to be patient in waiting to get the occasional call. My phone will ring sometime with something that wakes me up. You just never know when it’ll happen. That’s what makes it fun. So patience is a combination of patience and a willingness to do something that afternoon if it comes to you.

It does not pay to invest in a way that depends on the appearance of a greater fool

Warren Buffett: If people are making more money because they’re borrowing money or participating in securities that are pieces of junk but they hope to find a bigger sucker later on, you have to forget that.

Buffett does not think it’s important to manage currency risk with Berkshire’s international investments, but he avoids investments denominated in currencies that are at risk of depreciating wildly

Warren Buffett: We’ve owned lots of securities in foreign currencies. We do nothing in terms of its impact on quarterly and annual earnings. We don’t do anything based on its impact on quarterly and annual earnings. There’s never been a board meeting I can remember where I’ve said, “If we do this, our annual earnings will be this, therefore we ought to do it.” The number will turn out to be what it’ll be. What counts is where we are five or 10 or 20 years from now…

…Obviously, we wouldn’t want to own anything in a currency that we thought was really going to hell.

Buffett is worried about the tendency for governments to want to devalue their currencies, the USA included, but there’s nothing much that can be done about it; Buffett thinks the USA is running a fiscal deficit that is unsustainable over a long period of time; Buffett thinks a 3% fiscal deficit appears sustainable

Warren Buffett: That’s the big thing we worry about with the United States currency. The tendency of a government to want to debase its currency over time – there’s no system that beats that. You can pick dictators, you can pick representatives, you can do anything, but there will be a push toward weaker currencies. I mentioned very briefly in the annual report that fiscal policy is what scares me in the United States because of the way it’s made, and all the motivations are toward doing things that can cause trouble with money. But that’s not limited to the United States – it’s all over the world, and in some places, it gets out of control regularly. They devalue at rates that are breathtaking, and that’s continued…

…So currency value is a scary thing, and we don’t have any great system for beating that…

…We’re operating at a fiscal deficit now that is unsustainable over a very long period of time. We don’t know whether that means two years or 20 years because there’s never been a country like the United States. But as Herbert Stein, the famous economist, said, “If something can’t go on forever, it will end.” We are doing something that is unsustainable, and it has the aspect to it that it gets uncontrollable to a certain point….

…I wouldn’t want the job of trying to correct what’s going on in revenue and expenditures of the United States with roughly a 7% gap when probably a 3% gap is sustainable…

…We’ve got a lot of problems always as a country, but this is one we bring on ourselves. We have a revenue stream, a capital-producing stream, a brains-producing machine like the world has never seen. And if you picked a way to screw it up, it would involve the currency. That’s happened a lot of places.

Buffett thinks the key factors for a developing economy to attract investors are having a solid currency, and being business-friendly

Audience member: What advice would you give to government and business leaders of emerging markets like Mongolia to attract institutional investors like yourself?

Warren Buffett: If you’re looking for advice to give the government over there, it’s to develop a reputation for having a solid currency over time. We don’t really want to go into any country where we think there’s a significant probability of runaway inflation. That’s too hard to figure…

…If the country develops a reputation for being business-friendly and currency-conscious, that bodes very well for the residents of that country, particularly if it has some natural assets that it can build around.

Private equity firms are flooding the life insurance market, but they are doing so by taking on lots of leverage and credit risk

Ajit Jain: There’s no question the private equity firms have come into the space, and we are no longer competitive in the space. We used to do a fair amount in this space, but in the last 3-4 years, I don’t think we’ve done a single deal.

You should separate this whole segment into two parts: the property casualty end of the business and the life end of the business. The private equity firms you mentioned are all very active in the life end of the business, not the property casualty end.

You are right in identifying the risks these private equity firms are taking on both in terms of leverage and credit risk. While the economy is doing great and credit spreads are low, these firms have taken the assets from very conservative investments to ones where they get a lot more return. As long as the economy is good and credit spreads are low, they will make money – they’ll make a lot of money because of leverage.

However, there is always the danger that at some point the regulators might get cranky and say they’re taking too much risk on behalf of their policyholders, and that could end in tears. We do not like the risk-reward that these situations offer, and therefore we put up the white flag and said we can’t compete in this segment right now.

Buffett thinks Berkshire’s insurance operation is effectively unreplicable

Warren Buffett: I think there are people that want to copy Berkshire’s model, but usually they don’t want to copy it by also copying the model of the CEO having all of his money in the company forever. They have a different equation – they’re interested in something else. That’s capitalism, but they have a whole different situation and probably a somewhat different fiduciary feeling about what they’re doing. Sometimes it works and sometimes it doesn’t work. If it doesn’t work, they go on to other things. If what we do at Berkshire doesn’t work, I spend the end of my life regretting what I’ve created. So it’s just a whole different personal equation.

There is no property casualty company that can basically replicate Berkshire. That wasn’t the case at the start – at the start we just had National Indemnity a few miles from here, and anybody could have duplicated what we had. But that was before Ajit came with us in 1986, and at that point the other fellows should have given up.

Buffett thinks recent market volatility is not noteworthy at all; it’s nearly certain that significant downward moves in stocks will happen sometime in the next 20 years

Warren Buffett: What has happened in the last 30-45 days, 100 days, whatever this period has been, is really nothing. There have been three times since we acquired Berkshire that Berkshire has gone down 50% in a fairly short period of time – three different times. Nothing was fundamentally wrong with the company at any time. This is not a huge move. The Dow Jones average was at 381 in September of 1929 and got down to 42. That’s going from 100 to 11. This has not been a dramatic bear market or anything of the sort. I’ve had about 17,000 or 18,000 trading days. There have been plenty of periods that are dramatically different than this…

…You will see a period in the next 20 years that will be a “hair curler” compared to anything you’ve seen before. That just happens periodically. The world makes big mistakes, and surprises happen in dramatic ways. The more sophisticated the system gets, the more the surprises can come out of left field. That’s part of the stock market, and that’s what makes it a good place to focus your efforts if you’ve got the proper temperament for it and a terrible place to get involved if you get frightened by markets that decline and get excited when stock markets go up.

Berkshire’s leaders think the biggest change autonomous vehicles will bring to the automotive insurance industry is substitution of operator error policies by product liability policies; Berkshire’s leaders also think that the cost per repair in the event of an accident will rise significantly; the total cost of providing insurance for autonomous vehicles is still unclear; from the 1950s to today, cars have gotten 6x safer but auto insurance has become 50x pricier

Ajit Jain: There’s no question that insurance for automobiles is going to change dramatically once self-driving cars become a reality. The big change will be what you identified. Most of the insurance that is sold and bought revolves around operator errors – how often they happen, how severe they are, and therefore what premium we ought to charge. To the extent these new self-driving cars are safer and involved in fewer accidents, that insurance will be less required. Instead, it’ll be substituted by product liability. So we at GEICO and elsewhere are certainly trying to get ready for that switch, where we move from providing insurance for operator errors to being more ready to provide protection for product errors and errors and omissions in the construction of these automobiles…

…We talked about the shift to product liability and protection for accidents that take place because of an error in product design or supply. In addition to that shift, I think what we’ll see is a major shift where the number of accidents will drop dramatically because of automatic driving. But on the other hand, the cost per repair every time there’s an accident will go up very significantly because of the amount of technology in the car. How those two variables interact with each other in terms of the total cost of providing insurance, I think, is still an open issue…

…Warren Buffett: When I walked into GEICO’s office in 1951, the average price of a policy was around $40 a year. Now it’s easy to get up to $2,000 depending on location and other factors. During that same time, the number of people killed in auto accidents has fallen from roughly six per 100 million miles driven to a little over one. So the car has become incredibly safer, and it costs 50 times as much to buy an insurance policy.

There’s a tax now when American companies conduct share buybacks

Warren Buffett: I don’t think people generally know that, but there is a tax that was introduced a year or so ago where we pay 1%. That not only hurts us because we pay more for it than you do – it’s a better deal for you than for us – but it actually hurts some of our investee companies quite substantially. Tim Cook has done a wonderful job running Apple, but he spent about $100 billion in a year repurchasing shares, and there’s a 1% charge attached to that now. So that’s a billion dollars a year that he pays when he buys Apple stock compared to what you pay.

Buffett is very careful with the risks that come with derivative contracts on a company’s balance sheet

Greg Abel: I’ll maybe go back to the very first meeting with Warren because it still stands out in my mind. Warren was thinking about acquiring Mid-America Energy Holdings Company at that time, and we had the opportunity with my partners to go over there on a Saturday morning. We were discussing the business and Warren had the financial statements in front of him. Like anybody, I was sort of expecting a few questions on how the business was performing, but Warren locked in immediately to what was on the balance sheet and the fact we had some derivative contracts, the “weapons of mass destruction.”

In the utility business, we do have derivatives because they’re used to match certain positions. They’re never matched perfectly, but we have them and they’re required in the regulated business. I remember Warren going to it immediately and asking about the composition and what was the underlying risk, wanting to thoroughly understand. It wasn’t that big of a position, but it was absolutely one of the risks he was concerned about as he was acquiring Mid-America, especially in light of Enron and everything that had gone on.

The followup to that was a year or 18 months later. There was an energy crisis in the US around electricity and natural gas, and various companies were making significant sums of money. Warren’s follow-up question to me was, “How much money are we making during this energy crisis? Are we making a lot? Do we have speculative positions in place?” The answer was we weren’t making any more than we would have been six months ago because all those derivatives were truly to support our business and weren’t speculative. That focus on understanding the business and the risks around it still stands out in my mind.

Buffett spends more time analysing a company’s balance sheet than other financial statements

Warren Buffett: I spend more time looking at balance sheets than I do income statements. Wall Street doesn’t pay much attention to balance sheets, but I like to look at balance sheets over an 8 or 10 year period before I even look at the income account because there are certain things it’s harder to hide or play games with on the balance sheet than with the income statement.

Buffett thinks America’s electric grid needs a massive overhaul and it can only be done in via a partnership between the private sector and the government – unfortunately, nobody has figured out the partnership model yet

Warren Buffett: t’s very obvious that the country needs an incredible improvement, rethinking, redirection to some extent in the electric grid. We’ve outgrown what would be the model that America should have. In a sense, it’s a problem something akin to the interstate highway system where you needed the power of the government really to get things done because it doesn’t work so well when you get 48 or 50 jurisdictions that each has their own way of thinking about things…

…There are certain really major investment situations where we have capital like nobody else has in the private system. We have particular knowhow in the whole generation and transmission arena. The country is going to need it. But we have to figure out a way that makes sense from the standpoint of the government, from the standpoint of the public, and from the standpoint of Berkshire, and we haven’t figured that out yet. It’s a clear and present use of hundreds of billions of dollars. You have people that set up funds and they’re getting paid for just assembling stuff, but that’s not the way to handle it. The way to handle it is to have some kind of government-private industry cooperation similar to what you do in a war.

The risk of wildfires to electric utilities is not going to go away, and in fact, will increase over time

Greg Abel: The reality is the risk around wildfires – do the wildfires occur – they’re not going away, and we know that. The risk probably goes up each year.

Berkshire’s leaders think it’s important for utilities to de-energise when wildfires occur to minimise societal damage; Berkshire is the only utility operator so far that’s willing to de-energise; but de-energising also has its drawbacks; Berkshire may not be able to solve the conundrum of de-energising

Greg Abel: the one thing we hadn’t tackled – this is very relevant to the significant event we had back in 2020 in PacifiCorp – is we didn’t de-energize the system as the fire was approaching. Our employees and the whole management team have been trained all their lives to keep the lights on, and the last thing they want to do is turn those lights off and have a system de-energized. After those events and as we looked at how we’re going to move forward in managing the assets and reducing risk, we recognized as a team that we have to de-energize those assets. Now as we get fires encroaching at a certain number of miles, we de-energize because we do not want to contribute to the fire nor harm any of our consumers or contribute to a death. We had to take our team to managing a different risk now. It’s not around keeping the lights on, it’s around protecting the general public and ensuring the fire does not spread further. We’re probably the one utility or across our utilities that does that today, and we strongly believe in that approach.

Becky Quick: Doesn’t that open you up to other risk if you shut down your system, a hospital gets shut down, somebody dies?

Greg Abel: That’s something we do deal with a lot because we have power outages that occur by accident. When we look at critical infrastructure, that’s an excellent point and we’re constantly re-evaluating it. We do receive a lot of feedback from our customer groups as to how to manage that…

…Warren Buffett: There’s some problems that can’t be solved, and we shouldn’t be in the business of taking investors’ money and tackling things that we don’t know the solution for. You can present the arguments, but it’s a political decision when you are dealing with states or the federal government. If you’re in something where you’re going to lose, the big thing to do is quit.

Buffett thinks the value of electric utility companies have fallen a lot over the past two years because of societal trends and his enthusiasm for investing in electric utilities has waned considerably

Becky Quick: Ricardo Bri, a longtime shareholder based in Panama, says that he was very happy to see Berkshire acquire 100% of BHE. It was done in two steps: one in late 2022 – 1% was purchased from Greg Abel for $870 million implying a valuation of BHE of $87 billion, and then in 2024 the remaining 8% was purchased from the family of Walter Scott Jr. for $3.9 billion implying a valuation of $48.8 billion for the enterprise. That second larger transaction represented a 44% reduction in valuation in just two years. Ricardo writes that PacifiCorp liabilities seem too small to explain this. Therefore, what factors contributed to the difference in value for BHE between those two moments in time?

Warren Buffett: Well, we don’t know how much we’ll lose out of PacifiCorp and decisions that are made, but we also know that certain of the attitudes demonstrated by that particular example have analogues throughout the utility system. There are a lot of states that so far have been very good to operate in, and there are some now that are rat poison, as Charlie would say, to operate in. That knowledge was accentuated when we saw what happened in the Pacific Northwest, and it’s eventuated by what we’ve seen as to how utilities have been treated in certain other situations. So it wasn’t just a direct question of what was involved at PacifiCorp. It was an extrapolation of a societal trend…

…We’re not in the mood to sell any business. But Berkshire Hathaway Energy is worth considerably less money than it was two years ago based on societal factors. And that happens in some of our businesses. It certainly happened to our textile business. The public utility business is not as good a business as it was a couple of years ago. If anybody doesn’t believe that, they can look at Hawaiian Electric and look at Edison in the current wildfires situation in California. There are societal trends that are changing things…

…I would say that our enthusiasm for buying public utility companies is different now than it would have been a couple years ago. That happens in other industries, too, but it’s pretty dramatic in public utilities. And it’s particularly dramatic in public utilities because they are going to need lots of money. So, if you’re going to need lots of money, you probably ought to behave in a way that encourages people to give you lots of money.

Buffett thinks the future capital intensity of the USA’s large technology companies remains to be seen

Warren Buffett: It’ll be interesting to see how much capital intensity there is now with the Magnificent 7 compared to a few years ago. Basically, Apple has not really needed any capital over the years and it’s repurchased shares with a dramatic reduction. Whether that world is the same in the future or not is something yet to be seen.

Buffett thinks there’s no better system than capitalism that has been discovered so far

Warren Buffett: Capitalism in the United States has succeeded like nothing you’ve ever seen. But what it is is a combination of this magnificent cathedral which has produced an economy like nothing the world’s ever seen, and then it’s got this massive casino attached…

…In the cathedral, they’re designing things that will be producing goods and services for 300 and some million people like it’s never been done before in history. It’s an interesting system we developed, but it’s worked. It dispenses rewards in what seems like a terribly capricious manner. The idea that people get what they deserve in life – it’s hard to make that argument. But if you argue with it that any other system works better, the answer is we haven’t found one.

---

Disclaimer: The Good Investors is the personal investing blog of two simple guys who are passionate about educating Singaporeans about stock market investing. By using this Site, you specifically agree that none of the information provided constitutes financial, investment, or other professional advice. It is only intended to provide education. Speak with a professional before making important decisions about your money, your professional life, or even your personal life. I currently have a vested interest in Alphabet, Amazon, Apple, Meta Platforms, Microsoft, and Tesla (they are all part of the Magnificent 7). Holdings are subject to change at any time.

We’ve constantly been sharing a list of our recent reads in our weekly emails for The Good Investors.

Do subscribe for our weekly updates through the orange box in the blog (it’s on the side if you’re using a computer, and all the way at the bottom if you’re using mobile) – it’s free!

But since our readership-audience for The Good Investors is wider than our subscriber base, we think sharing the reading list regularly on the blog itself can benefit even more people. The articles we share touch on a wide range of topics, including investing, business, and the world in general. 

Here are the articles for the week ending 04 May 2025:

1. Everyone Says They’ll Pay More for “Made in the USA.” So We Ran an A/B Test – Ramon van Meer

We make filtered showerheads. Clean, sleek design. But more importantly, with the best shower filters on the market. 

Our bestselling model—manufactured in Asia (China and Vietnam)—sells for $129. But this year, as tariffs jumped from 25% to 170%, we wondered: Could we reshore manufacturing to the U.S. while maintaining margins to keep our lights on?…

…We found a U.S.-based supplier. The new unit cost us nearly 3x more to produce. To maintain our margins, we’d have to sell it for $239.

So we ran an experiment.

We created a secret landing page. The product and design were identical. The only difference? One was labeled “Made in Asia” and priced at $129. The other, “Made in the USA,” at $239…

…Add-to-carts for the U.S. version were only 24! Conversion? 0.0% (zero).

Not a single customer purchased the Made-in-USA version…

…We wanted to believe customers would back American labor with their dollars. But when faced with a real decision—not a survey or a comment section—they didn’t…

…Small brands like ours want to manufacture here. We’re willing to invest. But without serious shifts—in consumer incentives, automation, and trade policy—the math doesn’t work. Not for us. Not for our customers.

We’re still committed to exploring local manufacturing. But for now, it’s not viable.

We’re sharing this because the numbers surprised even us. And we think they’re worth talking about.

2. Perspectives from 30 business leaders on the trade war frontlines: preparing for the best, bracing for the worst – Amber Zhang

For companies with strong brand equity and pricing power, the strategy is clear: raise prices.

Anker Innovations was among the first movers, increasing prices on Amazon by roughly 20%. A person familiar with the company said its U.S. market share remains stable despite the hikes. Pop Mart hasn’t adjusted pricing yet, but told Waves that it’s “not ruling out the possibility.”

One cross-border commerce insider noted that for companies like DJI, whose supply chain advantages are hard to replicate, rising costs may squeeze margins, but competitors will struggle to gain market share in the near term.“Tariffs are a toll, not a blockade. Only great products hold long-term passports,” Anker commented…

…According to Waves, most ODM firms that rely heavily on the U.S. market are currently in a holding pattern, halting shipments and refraining from taking new orders for the coming months…

…Wei believes a resolution – or at least a temporary workaround – is likely within the next two to three months. Why? ODM manufacturers need to finalize production plans for Christmas orders by July or August. If the standoff drags on any longer, “America may be headed for a very empty Christmas.”…

…On April 2, Donald Trump signed an executive order eliminating the de minimis exemption for packages from the Chinese mainland and Hong Kong. The White House followed up on April 8, raising tariffs on these goods from 30% to 90%, effective May 2.

This presents a major setback for small Chinese sellers who rely on platforms like TikTok, Temu, and Shein to ship low-value parcels without stocking local warehouses…

…That said, some insiders point out that more than 90% of shipments to the U.S. currently rely on “Gray Customs Clearance” — unofficial customs channels where third-party brokers help exporters bypass formal declarations at minimal cost. From their perspective, the new tariffs are more likely to disrupt clearance efficiency rather than kill off the model entirely…

…Amid persistent U.S.-China trade friction, a viable strategy is to first export raw materials or semi-finished goods from China for overseas assembly and packaging. Over time, upstream production can gradually shift to local markets, supported by regional suppliers…

…In the wake of the de minimis exemption repeal, TikTok Shop recently issued a notice to U.S. sellers: starting May 2, all incoming shipments will face a 30% ad valorem tax, with an additional flat tariff of $25 per item before June 1, rising to $50 afterward. Carriers will also be required to post international bonds as part of the compliance framework. So far, Temu and Shein have yet to publicly announce specific countermeasures.

Ray Bu predicts that platform-based giants will be forced to accelerate full-scale internationalization, separating their domestic and overseas supply chains and localizing operations in major markets…

…Both Temu and Shein are ramping up local presence in multiple countries, aggressively recruiting local merchants. Shein, for example, has been building production capacity in Turkey and Brazil — jurisdictions seen as tariff-safe zones amid the current climate.

Temu, for its part, plans to launch six “native fulfillment hubs” and nine “semi-managed hubs” between April and June. The former are geared toward local legal entities, while the latter target Chinese sellers with the ability to fulfill orders domestically in target markets…

…According to Xue Yi, professor at the University of International Business and Economics, if access to the U.S. market becomes more restricted, the European Union may be China’s best substitute…

…According to Xue Feng, a partner at Fangda Partners, recent tariff hikes have dealt significant blows to sectors including furniture, toys, textiles, auto parts, chemicals, steel, and aluminum. These industries, heavily reliant on low-end supply chains, are structurally tied to the U.S. and face steep challenges when seeking alternative markets…

…At its core, America simply isn’t ready to rebuild industrial capacity at scale.

To start with, U.S. manufacturing wages are 6–8 times higher than those in emerging markets, and the country faces a chronic shortage of skilled labor. But more critical than labor costs is the disintegration of domestic supply chains. After decades of offshoring, the U.S. industrial base is severely fragmented — core components like semiconductor materials and electronic parts still depend heavily on Asian suppliers.

Even if reshoring happens — as envisioned by the Trump administration — these factories would function as isolated islands, unable to form self-sustaining industrial clusters.

3. Did ancient Rome have a stock market? – Swen Lorenz

“The New Deal in Old Rome” was published by H. J. Haskell in 1940. The original book is rare and expensive today, but reprints can easily be found. As it reports on page 11:

“Indications are there was a curiously high degree of commercial organisation in the ancient world. In the time of Cicero, in the last century before Christ, wealthy Romans were busily exploiting the eastern provinces. Companies of contractors were organised to construct public works and to collect government revenue, from which the contractors were took a large cut. They sold shares in offices on the Via Sacra, the Wall Street of Rome. Everybody, says the Greek historian Polybius, meaning all the country club crowd, bought them. … We may imagine how the bottom dropped out of Asiatic stocks on the Roman market when the news came of the concerted massacre of eighty thousand Italians at the instigation of the native ruler of an adjoining kingdom.”…

…How much of Haskell’s claims were true?

A quick Google search of “did the Romans have a stock market” produces contradicting results.

The first search result is an abstract from Prof. Pellegrino Manfra, a professor at City University New York: “Ancient Rome Economy and Investment: The Origins of the Stock Market

The origins of the stock market can be observed as far back as ancient Rome. The earliest example of organized market for equities can be found in the Roman Republic in second century B.C…. Back in Roman times, organizations called ‘Societates Publicanorum’ were formed that offered investments referred to as ‘partes’ or what we now know them as – shares. … The shares were tradable and had fluctuating prices based on the underlying project’s success. …. The place where trading occurred was the forum, near the temple of Castor.”

Prof. Manfra’s conclusion couldn’t be clearer.

However, three search results further down, you find a complete repudiation of his claims.

In 2016, Bocconi University in Milan put out a press release summarising a scientific article that its Prof. Manuela Geranio had published: “Ancient Rome Stock Exchange Is a Myth

Manuela Geranio, in a paper with Geoffrey Poitras, shows that modern claims of the existence of a market for shares of the societates publicanorum in the late Roman Republic are not supported by primary sources … Recent claims of trading in shares (partes) of tax-farming corporations (societates publicanorum) in the late Roman Republic can thus raise some skepticism. ‘Upon closer inspection there is only brief discussion of possible share trading in a few sources that, in turn, depend fundamentally on a debatable interpretation of the commercial and legal context’. The location of the proto-stock-exchange near the temple of Castor results, for example, the fruit of ‘romanticized descriptions’. … The paper … highlights the need for more careful historical and legal analyses before concluding about the existence of its peculiar institutions in ancient times.”…

…The most common source cited as evidence that modern-day share trading took place in ancient Rome is “Publicans and Sinners”, a 1972 book by Ernst Badian, an Austrian-born classical scholar who served as a professor at Harvard University from 1971-1998.

It describes how the Romans used “partes (shares) in public companies” and traded “over the counter” based on a “register” of shareholders that the companies kept. It all sounds like the ancient Romans did have an early version of a stock market!…

…Badian was elected a fellow of the American Academy of Arts and Sciences in 1974, and in 1999 his native Austria awarded him the Cross of Honor for Science and Art. His work may be dated, but it’s impossible to dismiss his writing as that of a crank.

Badian was an early member of a group that critics of this field of study today decry as the “maximalists” who allegedly were too aggressive in interpreting incomplete historical evidence in a favourable way…

…Still, another group of equally serious scientists delivered a broadside to the whole idea of the Romans operating something worthy of comparing it to modern-day stock markets.

In 2015, Bocconi University’s Manuela Geranio teamed up with Geoffrey Poitras to publish “Trading of shares in the Societates Publicanorum?

The often repeated modern claim of significant trading in ‘shares of the societates publicanorum’ (partes) during the late Roman Republic cannot be supported using the available ‘primary sources’.”

As the authors go on to argue, previous analysis of this field failed to take into consideration differences in language, nuances in interpreting complex terms, the historical context, and the lack of sufficient amounts of clear evidence.

“Even where elements related to possible share trading can be identified in the primary sources, evidence is often vague or questionable. … To avoid semantic confusions, understanding the commercial and legal context for claims of share trading and other activities involving the societates publicanorum requires definition of important terms – shares, share trading, company, joint-stock company, corporation. Appropriate definition is essential to clarify various claims made in modern sources.”

Geranio and Poitras argue that previous interpretations in this field relied too much on an “artful interpretation” of key terms.

Their conclusion is supported by the widely-known notion that the day and age a scientist lives in (as well as their language and culture of origin) can play heavily into the conclusions of research.

4. China is trying to create a national network of cloud computing centers – Andrew Stokols

I’ve written before about the Eastern Data Western Compute 东数西算 project, China’s effort to boost its’ “computing power” by constructing new data centers in the country’s West…

…The vision of the EDWC is not only about building new “nodes” of data centers, it is also about creating a “national network of computing”, 全国算力一张网 quanguo suanli yizhang wang so that computing power can be shifted between data centers depending on fluctuations in computing demand and supply in different regions, not unlike the way interconnected or smart power grids can move electricity around the grid depending on where demand is greatest. One of the premises of the EDWC is that data centers, which consume large amounts of energy, should be located in areas with ample renewable energy supply and cheaper land and energy than in the populated east coast. Networking computing resources across the country can make better use of energy in the West without having to transmit electricity from West to East…

…In a televised interview in 2022, Yu Xiaohui 余晓晖 the President of the influential think tank CAICT1, which has played a leading role in developing the EDWC plan, notes that “other countries are doing similar things [to the EDWC] but only within one cloud company, but this is an advantage of China, we are doing a systematic layout.”2 In other words, what he alludes to is the fact that the EDWC project is a state-led effort to coordinate the data center infrastructure of the entire country, which requires going beyond simply encouraging cloud/telecom providers to invest in their own new cloud computing centers. It also means creating a unified national computing network that would allow for a more dynamic allocation of computing demand across the country.

But to do this requires creating a system that can allocate demand from computing centers of different providers. From a business and data privacy standpoint this seems difficult to do. However, China’s three state-owned telecom operators are the ones playing a significant role in building out the EDWC project…

…In various documents and news announcements there have been reference to 调度中心 or “adjustment centers” in each of the data center nodes, which are supposed to function as traffic centers for nationwide computing resources, balancing supply (western data centers) and demand (eastern applications). Such “adjustment centers” allow computing tasks to be dynamically allocated to different data centers, such as allocating workloads between 8 national computing hubs and 10 data center clusters, optimizing latency, and improving energy efficiency by redirecting “non-urgent” data tasks to western renewable-powered hubs during off-peak hours…

…There are technical challenges to developing the national network. But there are also obvious financial/proprietary hurdles as well, namely how to interconnect cloud networks of separate companies who have their own proprietary businesses and systems. This may be easier to do with the three national operators (China Mobile, China Telecom, and China Unicom) than it is with private cloud operators, which are still the leading cloud platform companies (Alibaba, Tencent, Huawei, Baidu)…

…In the U.S., there are some examples of so-called “multi-cloud” interconnections, such as agreements between cloud operators Microsoft and Oracle that allow clients to connect certain cloud databases stored on different cloud platforms.7 But China’s ambition to create a national computing network would require a much greater interconnection and coordination to carry out…

…The degree to which China is able to build out the national network of cloud computing will have implications for its digital innovation, particularly in AI. While the U.S. leads China in the number of data centers by a wide margin, China’s system could develop in ways that diverge from the proprietary open-cloud model in the U.S. in which large enterprise cloud platforms dominate the market (AWS, Microsoft Azure, Google Cloud). Whether and to what degree China’s existing cloud providers continue to dominate the market will depend on their ability to innovate and maintain their edge in the face of increasing entry by state-owned cloud providers into cloud markets.

5. AI Horseless Carriages – Pete Koomen

I noticed something interesting the other day: I enjoy using AI to build software more than I enjoy using most AI applications–software built with AI.

When I use AI to build software I feel like I can create almost anything I can imagine very quickly. AI feels like a power tool. It’s a lot of fun.

Many AI apps don’t feel like that. Their AI features feel tacked-on and useless, even counter-productive.

I am beginning to suspect that these apps are the “horseless carriages” of the AI era. They’re bad because they mimic old ways of building software that unnecessarily constrain the AI models they’re built with…

…Up until very recently, if you wanted a computer to do something you had two options for making that happen:

1. Write a program
2. Use a program written by someone else

Programming is hard, so most of us choose option 2 most of the time. It’s why I’d rather pay a few dollars for an off-the-shelf app than build it myself, and why big companies would rather pay millions of dollars to Salesforce than build their own CRM.

The modern software industry is built on the assumption that we need developers to act as middlemen between us and computers. They translate our desires into code and abstract it away from us behind simple, one-size-fits-all interfaces we can understand.

The division of labor is clear: developers decide how software behaves in the general case, and users provide input that determines how it behaves in the specific case.

By splitting the prompt into System and User components, we’ve created analogs that map cleanly onto these old world domains. The System Prompt governs how the LLM behaves in the general case and the User Prompt is the input that determines how the LLM behaves in the specific case.

With this framing, it’s only natural to assume that it’s the developer’s job to write the System Prompt and the user’s job to write the User Prompt. That’s how we’ve always built software.

But in Gmail’s case, this AI assistant is supposed to represent me. These are my emails and I want them written in my voice, not the one-size-fits-all voice designed by a committee of Google product managers and lawyers.

In the old world I’d have to accept the one-size-fits-all version because the only alternative was to write my own program, and writing programs is hard.

In the new world I don’t need a middleman tell a computer what to do anymore. I just need to be able to write my own System Prompt, and writing System Prompts is easy!…

…In most AI apps, System Prompts should be written and maintained by users, not software developers or even domain experts hired by developers.

Most AI apps should be agent builders, not agents…

…AI-native software should maximize a user’s leverage in a specific domain. An AI-native email client should minimize the time I have to spend on email. AI-native accounting software should minimize the time an accountant spends keeping the books.

---

Disclaimer: The Good Investors is the personal investing blog of two simple guys who are passionate about educating Singaporeans about stock market investing. By using this Site, you specifically agree that none of the information provided constitutes financial, investment, or other professional advice. It is only intended to provide education. Speak with a professional before making important decisions about your money, your professional life, or even your personal life. We currently have a vested interest in Alphabet (parent of Google Cloud), Amazon (parent of AWS), Microsoft (parent of Azure), Salesforce, and Tencent. Holdings are subject to change at any time.

The book mentioned in the title of this article is The Other Half of Macroeconomics and the Fate of Globalization written by economist Richard C. Koo (Gu Chao Ming) and published in 2018.

I first came across Koo’s book in March 2020 when I chanced upon a review of it in Mandarin, written by investor Li Lu. I can read Mandarin and I found myself agreeing to the ideas from the book that Li shared, so much so that I made a self-directed attempt at translating the review into English. But I only began reading the actual book near the start of this year and finished it about a month ago. There was even more richness in the book’s ideas about how economies should operate than what was shared in Li’s already-wonderful review.

Earlier this month, the US government, under the Trump administration, made sweeping changes to the global trading system by introducing the Reciprocal Tariff Policy, which raised tariffs, sometimes significantly so, for many of the US’s trading partners. Major driving forces behind the Reciprocal Tariff Policy ostensibly include the US’s sustained trade deficits (particularly with China) and a desire by the Trump administration to bring manufacturing jobs back to the country.

As I contemplated the Trump administration’s actions, and the Chinese government’s reactions, I realised Koo’s book explained why all these issues happened. So I’m writing my own notes and takeaways from the book for easy reference in the future and I would like to share them in this article in the hopes that they could be useful for you. I will be borrowing from my translation of Li’s Mandarin review in this article. Below the horizontal line, all content in grey font are excerpts from my translation while all italicised content are excerpts from the book. 

---

The three stages of economic development a country goes through

There are six important ideas that Koo discussed in his book. One of them is the concept that a country goes through three distinct stages of economic development over time. 

The first stage of development would be a country that is industrialising and has yet to reach the Lewis Turning Point (LTP). The LTP is the “point at which urban factories have finally absorbed all the surplus rural labour.” When a country starts industrialising, people are mostly living in rural areas and there are only a very few educated elite who have the knowhow to kickstart industrialisation. There is also a surplus of labour. As a result, the educated elite – the industrialists – hold the power and “most of the gains during the initial stage of industrialisation therefore go to the educated few.”  The first stage of economic development is also when income inequality widens – the gains from industrialisation continue to accumulate in the hands of the elite as they reinvest profits into their businesses because there continues to be a surplus of labour.

The second stage of development happens when an industrialising economy reaches the LTP. At this point, labour “gains the bargaining power to demand higher wages for the first time in history, which reduces the share of output accruing to business owners.” But business owners are happy to continue reinvesting their profits as they are still “achieving good returns, leading to further tightness in the labour market.” This dynamic leads to an economy’s “golden era”:

“As labor’s share increases, consumption’s share of GDP will increase at the expense of investment. At the same time, the explosive increase in the purchasing power of ordinary citizens means that most businesses are able to increase profits simply by expanding existing productive capacity. Consequently, both consumption and investment will increase rapidly…

…Inequality also diminishes as workers’ share of output increases relative to that of capital… 

…With incomes rising and inequality falling, this post-LTP maturing phase may be called the golden era of economic growth…

…Higher wages force businesses to look harder for profitable investment opportunities. On the other hand, the explosive increase in the purchasing power of ordinary workers who are paid ever-higher wages creates major investment opportunities. This prompts businesses to invest for two reasons. First, they seek to increase worker productivity so that they can pay ever-higher wages. Second, they want to expand capacity to address workers’ increasing purchasing power. Both productivity- and capacity-enhancing investments increase demand for labor and capital that add to economic growth. In this phase, business investment increases workers’ productivity even if their skill level remains unchanged…

…With rapid improvements in the living standards of most workers, the post-LTP maturing phase is characterised by broadly distributed benefits from economic growth.”

The golden era has its problems too. This is because this period is when “workers begin to utilise their newfound bargaining power” such as by organising strikes. But business owners and labour tend to be able to work out their differences.

The third stage of development is what Koo calls a “post-LTP pursued economy.” When a country is in the golden era, at some point ever-growing wages creates inroads for foreign competitors – and the country starts being chased by the foreign competitors that have lower wages. This is when businesses in the country find it very challenging to “find attractive investment opportunities at home because it often makes more sense for them to buy directly from the “chaser” or to invest in that country themselves.” This is also when “the return on capital is higher abroad than at home.” During the pursued stage, “real wage growth will be minimal” and “economic growth also slows.” Although a pursued country can continue to grow economically, a major problem is that inequality once again rears its ugly head:

“Japan’s emergence in the 1970s shook the U.S. and European industrial establishments. As manufacturing workers lost their jobs, ugly trade frictions ensued between Japan and the West. This marked the first time that Western countries that had already passed their LTPs had been chased by a country with much lower wages…

…While Western companies at the forefront of technology continued to do well, the disappearance of many well-paying manufacturing jobs led to worsening income inequality in these countries…

…Some of the pain Western workers felt was naturally offset by the fact that, as consumers, they benefited from cheaper imports from Asia, which is one characteristic of import-led globalisation. Businesses with advanced technology continued to do well, but it was no longer the case that everyone in society was benefiting from economic growth. Those whose jobs could be transferred to lower-cost locations abroad saw their living standards stagnate or even fall.”

Koo wrote that Western economies – the USA and Europe – entered their golden eras around the 1950s and became pursued starting in the 1970s by Japan. During the golden era of the West, “it was in an export-led globalisation phase as it exported consumer and capital goods to the world.” But as the West started getting pursued, they entered “an import-led globalisation phase as capital seeks higher returns abroad and imports flood the domestic market.”

The four states of an economy

Another important idea from Koo’s book is the concept that an economy has four distinct states, which are summarised in Table 1 below:

“An economy is always in one of four possible states depending on the presence or absence of lenders (savers) and borrowers (investors). They are as follows: (1) both lenders and borrowers are present in sufficient numbers, (2) there are borrowers but not enough lenders even at high interest rates, (3) there are lenders but not enough borrowers even at low interest rates, and (4) both lenders and borrowers are absent.”

Koo’s idea that an economy has four distinct states is important because mainstream economic-thought does not cater for the disappearance of borrowers:

“Of the four, only Cases 1 and 2 are discussed in traditional economics, which implicitly assumes there are always enough borrowers as long as real interest rates are low enough.”

There are two key reasons why an economy would be in Cases 3 and 4, i.e. when borrowers disappear. The first is when private-sector businesses are unable to find attractive investment opportunities (this is related to economies that are in the third-stage of development discussed earlier in this article, when attractive domestic investment opportunities become scarce):

“The first is one in which private‐sector businesses cannot find investment opportunities that will pay for themselves. The private sector will only borrow money if it believes it can pay back the debt with interest. And there is no guarantee that such opportunities will always be available. Indeed, the emergence of such opportunities depends very much on scientific discoveries and technological innovations, both of which are highly irregular and difficult to predict.

In open economies, businesses may also find that overseas investment opportunities are more attractive than those available at home. If the return on capital is higher in emerging markets, for example, pressure from shareholders will force businesses to invest more abroad while reducing borrowings and investments at home. In modern globalized economies, this pressure from shareholders to invest where the return on capital is highest may play a greater role than any technological breakthroughs, or lack thereof, in the decision as to whether to borrow and invest at home.”

The second reason for the disappearance of borrowers is named by Koo as a “balance sheet recession” which is described as such:

“In the second set of circumstances, private‐sector borrowers have sustained huge losses and are forced to rebuild savings or pay down debt to restore their financial health. Such a situation may arise following the collapse of a nationwide asset price bubble in which a substantial part of the private sector participated with borrowed money. The collapse of the bubble leaves borrowers with huge liabilities but no assets to show for the debt. Facing a huge debt overhang, these borrowers have no choice but to pay down debt or increase savings in order to restore their balance sheets, regardless of the level of interest rates.

Even when the economy is doing well, there will always be businesses that experience financial difficulties or go bankrupt because of poor business decisions. But the number of such businesses explodes after a nationwide asset bubble bursts.

For businesses, negative equity or insolvency implies the potential loss of access to all forms of financing, including trade credit. In the worst case, all transactions must be settled in cash, since no supplier or creditor wants to extend credit to an entity that may seek bankruptcy protection at any time. Many banks and other depository institutions are also prohibited by government regulations from extending or rolling over loans to insolvent borrowers in order to safeguard depositors’ money. For households, negative equity means savings they thought they had for retirement or a rainy day are no longer there. Both businesses and households will respond to these life‐threatening conditions by focusing on restoring their financial health—regardless of the level of interest rates—until their survival is no longer at stake.”

A balance sheet recession can be a huge problem for a country’s economy if it is unresolved as it can lead to a rapidly shrinking economy as a manifestation of the “fallacy of composition” problem:

“One person’s expenditure is another person’s income…

…The interaction between thinking and reacting households and businesses create a situation where one plus one does not necessarily equal two. For example, if A decides to buy less from B in order to set aside more savings for an uncertain future, B will have less income to buy things from A. That will lower A’s income, which in turn will reduce the amount A can save.

This interaction between expenditure and income also means that, at the national level, if one group is saving money, another group must be doing the opposite – “dis-saving” – to keep the economy running. In most cases, this dis-saving takes the form of borrowing by businesses that seek to expand their operations. If everyone is saving and no one is dis-saving on borrowing, all of those savings will leak out of the economy’s income stream, resulting in less income for all.

For example, if a person with an income of $1,000 decides to spend $900 and save $100, the $900 that is spent becomes someone else’s income and continues circulating in the economy. The $100 that is saved is typically deposited with a financial institution such as a bank, which then lends it to someone else who can make use of it. When that person borrows and spends the $100, total expenditures in the economy amount to $900 plus $100, which is equal to the original $1,000, and the economy moves forward…

…If there are no borrowers for $100 in savings in the above example, even at zero interest rates, total expenditures in the economy will drop to $900, while the saved $100 remains unborrowed in financial institutions or under mattresses. The economy has effectively shrunk by 10 percent, from $1,000 to $900. That $900 now becomes someone else’s income. If that person decides to save 10 percent, and there are still no borrowers, only $810 will be spent, causing the economy to contract to $810. This cycle will repeat, and the economy will shrink to $730, if borrowers remain on the sidelines. This process of contraction is called a “deflationary spiral.”…

…Keynes had a name for this state of affairs, in which everyone wants to save but is unable to do so because no one is borrowing. He called it the paradox of thrift. It is a paradox because if everyone tries to save, the net result is that no one can save.

The phenomenon of right behaviour at the individual level leading to a bad result collectively is known as the “fallacy of composition.””

Japan was the “first advanced country to experience a private-sector shift to debt minimization for balance sheet reasons since the Great Depression.” Japan’s real estate bubble burst in 1990, where real estate prices fell by 87%, “devastating the balance sheets of businesses and financial institutions across the country” and leading to a disappearance of borrowers:

“Demand for funds shrank rapidly when the bubble finally burst in 1990. Noting that the economy was also slowing sharply, the BOJ took interest rates down from 8 percent at the height of the bubble to almost zero by 1995. But demand for funds not only failed to recover but actually turned negative that year. Negative demand for funds means that Japan’s entire corporate sector was paying down debt at a time of zero interest rates, a world that no economics department in university or business school had ever envisioned. The borrowers not only stopped borrowing but began moving in the opposite direction by paying down debt and continued doing so for a full ten years, until around 2005…

…While in a textbook economy the household sector saves and the corporate sector borrows, both sectors became net-savers in post-1999 Japan, with the corporate sector becoming the largest saver in the country from 2002 onward in spite of zero interest rates.”

The Western economies experienced their own balance sheet recessions starting in 2008 with the bursting of housing bubbles in that year. When the “bubbles collapsed on both sides of the Atlantic in 2008, the balance sheets of millions of households and many financial institutions were devastated.” Borrowers also disappeared; the “private sectors in virtually all major advanced nations have been increasing savings or paying down debt since 2008 in spite of record low interest rates.” For example, “the U.S. private sector saved 4.1 percent of GDP at near-zero interest rates in the four quarters through Q1 2017” and the “Eurozone’s overall private sector is saving 4.6 percent of GDP in spite of negative interest rates.”

Prior to the Japanese episode, the most recent example was the Great Depression:

“Until 2008, the economics profession considered a contractionary equilibrium (the $500 economy) brought about by a lack of borrowers to be an exceptionally rare occurrence – the only recent example was the Great Depression, which was triggered by the stock market crash in October 1929 and during which the U.S. lost 46 percent of nominal GNP. Although Japan fell into a similar predicament when its asset price bubble burst in 1990, its lessons were almost completely ignored by the economics professional until the Lehman shock of 2008.”

The appropriate macroeconomic policies

The third important idea from Koo’s book is that depending on which stage (pre-LTP, golden era, or pursued) and which state (Case 1, Case 2, Case 3, or Case 4) a country’s economy is in, there are different macroeconomic policies that would be appropriate.

First, it’s important to differentiate the two policies governments can wield, namely, monetary policy and fiscal policy:

“The government also has two types of policy, known as monetary and fiscal policy, that it can use to help stabilise the economy by matching private-sector savings and borrowings. Themore frequently used is monetary policy, which involves raising or lowering interest rates to assist the matching process. Since an excess of borrowers is usually associated with a strong economy, a higher policy rate might be appropriate to prevent overheating and inflation. Similarly, a shortage of borrowers is usually associated with a weak economy, in which case a lower policy rate might be needed to avert a recession or deflation.

With fiscal policy, the government itself borrows and spends money on such projects as highways, airports, and other social infrastructure. While monetary policy decisions can be made very quickly by the central bank governor and his or her associates, fiscal policy tends to be very cumbersome in a peacetime democracy because elected representatives must come to an agreement on how much to borrow and where to spend the money. Because of the political nature of these decisions and the time it takes to implement them, most recent economic fluctuations were dealt with by central banks using monetary policy.”

Fiscal policy is more important than monetary policy when a country’s economy is in the pre-LTP stage, but the relative importance of the two types of policies switches once the economy enters the golden era; fiscal policy once again becomes the more important type of policy when the economy is in the pursued stage: 

“In the early phases of industrialisation, economic growth will rely heavily on manufacturing, exports, and the formation of capital etc. At this juncture, the government’s fiscal policies can play a huge role. Through fiscal policies, the government can gather scarce resources and invest them into basic infrastructure, resources, and export-related services etc. These help emerging countries to industrialise rapidly. Nearly every country that was in this stage of development saw their governments implement policies that promote active governmental support.

In the second stage of development, the twin engines of economic growth are rising wages and consumer spending. The economy is already in a state of full employment, so an increase in wages in any sector or field will inevitably lead to higher wages in other areas. Rising wages lead to higher spending and savings, and companies will use these savings to invest in productivity to improve output. In turn, profits will grow, leading to companies having an even stronger ability to raise wages to attract labour. All these combine to create a positive feedback loop of economic growth. Such growth comes mainly from internal sources in the domestic economy. Entrepreneurs, personal and household investing behaviour, and consumer spending patterns are the decisive players in promoting economic growth, since they are able to nimbly grasp business opportunities in the shifting economic landscape. Monetary policies are the most effective tool in this phase, compared to fiscal policies, for a few reasons. First, fiscal policies and private-sector investing both tap on a finite pool of savings. Second, conflicts could arise between the private sector’s investing activities and the government’s if poorly thought-out fiscal policies are implemented, leading to unnecessary competition for resources and opportunities. 

When an economy reaches the third stage of development (the stage where it’s being chased), fiscal policy regains its importance. At this stage, domestic savings are high, but the private sector is unwilling to invest domestically because the investing environment has deteriorated – domestic opportunities have dwindled, and investors can get better returns from investing overseas. The government should step in at this juncture, like what Japan did, and invest heavily in infrastructure, education, basic research and more. The returns are not high. But the government-led investments can make up for the lack of private-sector investments and the lack of consumer-spending because of excessive savings. In this way, the government can protect employment in society and prevent the formation of a vicious cycle of a decline in GDP. In contrast, monetary policy is largely ineffective in the third stage.”

It’s worth noting that an economy that is in the pre-LTP stage is likely to be in Case 4, where borrowers and lenders are both absent. Meanwhile, an economy that is in the Golden Era is likely to be in Case 1 (where both borrowers and lenders are present in abundance) and in Case 2 (where borrowers are present, but lenders are absent) during a run-of-the-mill recession, although a Case 1 Golden Era economy can also quickly be in Case 3 or Case 4. Once an economy is in the pursued stage, it is likely to be in Case 3 (where borrowers are absent but lenders are present) because of a lack of domestic investment opportunities or a balance sheet recession, or Case 4 (where borrowers and lenders are both absent) because of a balance sheet recession.

When a country’s economy is in Case 1 or Case 2, monetary policy is more important:

“Case 1 requires a minimum of policy intervention – such as slight adjustments to interest rates – to match savers and borrowers and keep the economy going. Case 1, therefore, is associated with ordinary interest rates and can be considered the ideal textbook case.

The causes of Case 2 (insufficient lenders) can be traced to both macro and financial factors. The most common macro factor is when the central bank tightens monetary policy to rein in inflation. The tighter credit conditions that result certainly leave lenders less willing to lend. Once inflation is under control, however, the central bank typically eases monetary policy, and the economy returns to Case 1. A country may also be too poor or underdeveloped to save. If the paradox of thrift leaves a country too poor to save, the situation would be classified as Case 3 or 4 because it is actually attributable to a lack of borrowers.

Financial factors weighing on lenders may also push the economy into Case 2. One such factor is an excess of non-performing loans (NPLs) in the banking system, which depresses banks’ capital ratios and prevents them from lending. This is what is typically called a “credit crunch.” Over-regulation of financial institutions by the authorities can also lead to a credit crunch. When many banks encounter NPL problems at the same time, mutual distrust may lead not only to a credit crunch, but also to a dysfunctional interbank market, a state of affairs typically referred to as a “financial crisis.”…

…Non-developmental causes of a shortage of lenders all have well-known remedies… For example, the government can inject capital into the banks to restore their ability to lend, or it can relax regulations preventing financial institutions from serving as financial intermediaries.

In the case of a dysfunctional interbank market, the central bank can act as lender of last resort to ensure the clearing system continues to operate. It can also relax monetary policy. The conventional emphasis on monetary policy and concerns over the crowding-out effect of fiscal policy are justified in Cases 1 and 2 where there are borrowers but (for a variety of reasons in Case 2) not enough lenders.””

When a country’s economy is in Case 3 or Case 4, fiscal policy is more important because monetary policy does not work when borrowers disappear, although the appropriate type of fiscal policy can also differ:

“It should be noted that in the immediate aftermath of a bubble collapse, the economy is usually in Case 4, characterized by a disappearance of both lenders and borrowers. The lenders stop lending because they provided money to borrowers who participated in the bubble and are now facing technical or real insolvency. Banks themselves may be facing severe solvency problems when many of their borrowers are unable to service their debts…

…In a financial crisis, therefore, the central bank must act as lender of last resort to ensure that the settlement system continues to function…

…Once the bubble bursts and households and businesses are left facing debt overhangs, no amount of monetary easing by the central bank will persuade them to resume borrowing until their balance sheets are fully repaired…

…When private-sector borrowers disappear and monetary policy stops working, the correct way to prevent a deflationary spiral is for the government to borrow and spend the excess savings of the private sector… 

…In other words, the government should mobilize fiscal policy and serve as borrower of last resort when the economy is in Case 3 or 4. 

If the government borrows and spends the $100 left unborrowed by the private sector, total expenditures will amount to $900 plus $100, or $1,000, and the economy will move on. This way, the private sector will have the income it needs to pay down debt or rebuild savings…

…It has been argued that the fiscal stimulus is essential when the economy is in Case 3 or 4. But there are two kinds of fiscal stimulus: government spending and tax cuts. If the economy is in a balance sheet recession, the correct form of fiscal stimulus is government spending. If the economy is suffering from a lack of domestic investment opportunities, the proper response would be a combination of tax cuts and deregulation to encourage innovation and risk taking… augmented by government spending…

…The close relationship observed prior to 2008 between central-bank-supplied liquidity, known as the monetary base, and growth in money supply and private-sector credit broke down completely after the bubbles burst and the private sector began minimizing debt. Here money supply refers to the sum of all bank accounts plus bills and coins circulating in the economy, and credit means the amount of money lent to the private sector by financial institutions…

…In this textbook world, a 10 percent increase in central bank liquidity would increase both the money supply and credit by 10 percent. This means there were enough borrowers in the private sector to borrow all the funds supplied by the central bank, and the economies were tin Case 1…

…But after the bubble burst, which forced the private sector to minimize debt in order to repair its balance sheet, no amount of central bank accommodation was able to increase private-sector borrowings. The U.S. Federal Reserve, for example, expanded the monetary base by 349 percent after Lehman Brothers went under. But the money supply grew by only 76 percent and credit by only 27 percent. A 27 percent increase in private-sector credit over a period of nearly nine years represents an average annual increase of only 2.75 percent, which is next to nothing.”

Fiscal stimulus equates to government-spending, which increases public debt. Koo suggests that (1) when an economy is in Case 3 or Case 4, rising and/or high public debt is not necessarily a problem, and (2) the limits of public debt should be determined by the bond market:

“Debt is simply the flip side of savings. Somebody has to be saving for debt to grow, and it is bound to increase as long as someone in the economy continues to save. Moreover, if someone is saving but debt levels fail to grow (i.e., if no one borrows and spends the saved funds), the economy will fall into the $1000 – $900 – $810 – $730 deflationary spiral….

…Growth in debt (excluding debt financed by the central bank) is merely a reflection of the fact that the private sector has continued to save. 

If debt is growing faster than actual savings, it simply means there is double counting somewhere, i.e., somebody has borrowed the money but instead of using it himself, he lent it to someone else, possibly with a different maturity structure (maturity transfer) or interest rates (fixed to floating or vice versa). With the prevalence of carry trades and structured financial products involving multiple counterparties, debt numbers may grow rapidly on the surface, but the actual debt can never be greater than the actual savings. 

Furthermore, the level of debt anyone can carry also depends on the level of interest rates and the quality of projects financed with the debt. If the projects earn enough to pay back both borrowing costs and principal, then no one should care about the debt load, no matter how large, because it does not represent a future burden on anyone. Similarly, no matter how great the national debt, if the funds are invested in public works projects capable of generating returns high enough to pay back both interest and principal, the projects will be self-financing and will not increase the burden on future taxpayers…

…Whether or not fiscal policy has reached its limits should be decided by the bond market, not by some economist using arbitrarily chosen criteria. 

During the golden era, when the private sector has strong demand for funds to finance productivity- and capacity-enhancing investments, fiscal stimulus will have a minimal if not negative impact on the economy because of the crowding-out effect. The bond market during this era correctly assigns very low prices (high yields) to government bonds, indicating that such stimulus is not welcome.

During the pursued era or during balance sheet recessions, however, private-sector demand for funds is minimal if not negative. At such times, fiscal stimulus is not only essential, but it has maximum positive impact on the economy because there is no danger of crowding out. During this period, the bond market correctly sets very high prices (low yields) for government bonds, indicating they are welcome…

…Ultra-low bond yields in economies in Cases 3 and 4 are also a signal to the government to look for public works projects capable of producing a social rate of return in excess of those rates. If such projects can be found, fiscal stimulus centered on them will ultimately place no added burden on future taxpayers.” 

The experience of the economies of the US, the UK, Japan, and Europe in the aftermath of the housing bubble bursting in 2008 which thrust them into balance sheet recessions is instructive on the importance of fiscal policy in combating balance sheet recessions:

“In November 2008, just two months after Lehman Brothers went under, the G20 countries agreed at an emergency meeting in Washington to implement fiscal stimulus. That decision kept the world economy from falling into a deflationary spiral. But in 2010, the fiscal orthodoxy of those who did not understand balance sheet recessions reasserted itself at the Toronto G20 meeting, where members agreed to cut deficits in half even though private-sector balance sheets were nowhere near a healthy state. The result was a sudden loss of forward momentum for the global economy that prolonged the recession unnecessarily in many parts of the world. After 2010, those countries that understood the danger of balance sheet recessions did well, while those that did not fell by the wayside…

…Bernanke and Yellen both understood this, and they used the expression “fiscal cliff” to warn Congress about the danger posed by fiscal consolidation, which the Republicans and many orthodox economists supported. The extent of Bernanke’s concerns about fiscal consolidation can be gleaned from a press conference on April 25, 2012, when he was asked what the Fed would do if Congress pushed the U.S. economy off the fiscal cliff. He responded, “There is . . . absolutely no chance that the Federal Reserve could or would have any ability whatsoever to offset that effect on the economy.”10 Bernanke clearly understood that the Fed’s monetary policy not only cannot offset the negative impact of fiscal consolidation, but would also lose its effectiveness if the government refused to act as borrower of last resort.

Even though the U.S. came frighteningly close to falling off the fiscal cliff on a number of occasions, including government shutdowns, sequesters, and debt‐ceiling debates, it ultimately managed to avoid that outcome thanks to the efforts of officials at the Fed and the Obama administration. And that is why the U.S. economy is doing so much better than Europe, where virtually every country did fall off the fiscal cliff…

…The warnings about the fiscal cliff set the Fed apart from its counterparts in Japan, the UK, and Europe. In the UK, then-BOE Governor Mervyn King publicly supported David Cameron’s rather draconian austerity measures, arguing that his bank’s QE policy would provide necessary support for the British economy. At the time, the UK private sector was saving a full 9 percent of GDP when interest rates were at their lowest levels in 300 years. That judgement led to the disastrous performance of the UK economy during the first two years of the Cameron administration…

…BOJ Governor Haruhiko Kuroda also argued strongly in favor of hiking the consumption tax rate, believing a Japanese economy supported by his quantitative easing regime would be strong enough to withstand the shock of fiscal consolidation. This was in spite of the fact that Japanese private sector was saving 6.2 percent of GDP at a time of zero interest rates. The tax hike, which was carried out in April 2014, threw the Japanese economy back into recession…

…ECB President Mario Draghi has admonished member governments to meet the austerity target imposed by the Stability and Growth Pact at every press conference, even though his own inflation forecasts have been revised downwards almost every time they are updated. He seems to be completely oblivious to the danger posed by fiscal austerity when the Eurozone private sector has been saving an average of 5 percent of GDP since 2008 despite zero or even negative interest rates.” 

Koo also noted that when Japan’s real estate bubble burst in 1990, the government was “quick to administer fiscal stimulus to stop the implosion” and that “the economy responded positively each time fiscal stimulus was implemented, but lost momentum each time the stimulus was removed.” The Japanese government was under enormous pressure to cut fiscal stimulus in the aftermath of the bubble, but the government did not completely cave, and the Japanese economy managed to fare better than it would otherwise:

“The orthodox fiscal hawks who dominated the press and academia also tried to stop fiscal stimulus at every step of the way, arguing that large deficits would soon lead to skyrocketing interest rates and a fiscal crisis. These hawks forced politicians to cut stimulus as soon as the economy showed signs of life, prompting another downturn. The resulting on-again, off-again fiscal stimulus did not imbue the public with confidence in the government’s handling of the economy. Fortunately, the LDP [Liberal Democratic Party] had enough pork-barrel politicians to keep a minimum level of stimulus needed in place, and as a result, Japanese GDP never once fell below its bubble peak. Nor did the Japanese unemployment rate ever exceed 5.5 percent.

That was a fantastic achievement in view of the fact that the Japanese private sector was saving an average of 8 percent of GDP from 1995 to 2005, and the Japanese lost three times as much wealth (as a share of GDP) as Americans did during the Great Depression, when nominal GNP fell 46 percent.”

The reason for US backlash against globalisation & the conflict between free-trade and free-capital

The fourth and fifth important ideas from Koo’s book are connected and they are respectively, (a) the possible reasons behind the backlash against globalisation that is seen from the current US government under the Trump administration, and (b) the possible conflict between free trade and free-movement of capital. Again, Koo’s book was published in 2018, so it was discussing Donald Trump’s first term as President. But the ideas appear to me to be very applicable to today’s context.

Koo advanced that the Western economies’ entrance into the third stage of economic development – pursued stage – is a reason for the backlash against globalisation:

“One reason for the frustration and social backlash witnessed in the advanced countries is that these countries are experiencing the post-Lewis Turning Point (LTP) pursued phase for the first time in history… 

…Many were caught offguard, having assumed that the golden era that they enjoyed into the 1970s would last forever. It comes as no surprise that those who have seen no improvement in their living standards for many years but still remember the golden age, when everyone was hopeful and living standards were steadily improving, would long for the “good old days.”…

…In the U.S. too, the Trump phenomenon, which has depended largely on the support of blue-collar white males, suggests that people are longing for the life they enjoyed during the golden era, when U.S. manufacturing was the undisputed leader of the world.

Participants in this social backlash in many of the pursued economies view globalization as the source of all evil and are trying to slow down the free movement of both goods and people. Donald Trump and others like him are openly hostile toward immigration while arguing in favour of protectionism and the scuttling of agreements such as the TPP that seek even freer trade.”

Koo described the mainstream view that free trade creates overall gains for trading partners, but cautioned that the view has a flawed assumption, in that imports and exports will be largely balanced as free trade grows, and it is that wrong assumption that also contributed to the backlash against globalisation:

“Economists have traditionally argued that while free trade creates both winners and losers within the same country, it offers significant overall welfare gains for both trading partners because the gains of the winners are greater than the losses of the losers. In other words, there should be more winners than losers from free trade…

…This conclusion, however, is based on one key assumption: that imports and exports will be largely balanced as free trade expands. When – as in the U.S. during the past 30 years – that assumption does not hold and a nation continues to run massive trade deficits, free trade may produce far more losers than theory would suggest. With the U.S. running a trade deficit of almost [US]$740bn a year, or about four percent of GDP, there were apparently enough losers from free trade to put the protectionist Donald Trump into the White House. The fact that Hillary Clinton was also nominated to be the Democratic Party’s candidate for president in the arena full of banners saying “No to TPP” indicates that the social backlash has grown very large indeed.”

Koo clarified that free trade is important and has its benefits, but the way free trade has taken place since World War II is hugely problematic because of (1) the way free trade is structured, and (2) the free movement of capital that is happening in parallel:

“Outright protectionism is likely to benefit the working class in the short term only. In the long run, history has repeatedly shown that protected industries always fall behind on competitiveness and technological advances, which means the economy will stagnate and be overtaken by more dynamic competitors…

…This does not mean that free trade as practiced since 1945 and globalism in general have no problems. They both have major issues, but these can be addressed if properly understood. A correct understanding is important here because even though increasing imports is the most visible feature of an economy in a pursued phase, trade deficits and the plight of workers displaced by imports have been made far worse by the free movement of capital since 1980…

…Once the U.S. opened up its massive markets to the world after 1945 and the GATT-based [General Agreement on Tariffs and Trade] system of free trade was adopted, nations belonging to this system found that it was possible to achieve economic growth without territorial expansion as long as they could produce competitive products. The first countries to recognize this were the vanquished nations of Japan and West Germany, which then decided to devote their best people to developing globally competitive products…

…By the end of the 1970s, however, the West began losing its ability to compete with Japanese firms as the latter overtook the U.S. and European rivals in many sectors, including home appliances, shipbuilding, steel, and automobiles. This led to stagnant income growth and disappearing job opportunities for Western workers.

When Japan joined the GATT in 1963, it still had many tariff and non-tariff trade barriers. In other words, while Western nations had been steadily reducing their own trade barriers, they were suddenly confronted with an upstart from Asia that still had many barriers in place. But as long as Japan’s maximum tariff rates were falling as negotiated and the remaining barriers applied to all GATT members equally, GATT members who had opened their markets earlier could do little under the agreement’s framework to force Japan to open its market (the same problem resurfaced when China joined the WTO 38 years later)…

…When U.S.-Japan trade frictions began to flare up in the 1970s, however, exchange rates still responded correctly to trade imbalances. In other words, when Japanese exports to the U.S. outstripped U.S. exports to Japan, there were more Japanese exporters selling dollars and buying yen to pay employees and suppliers in Japan than there were U.S. exporters selling yen and buying dollars to pay employees and suppliers in the U.S.

Since foreign exchange market participants in those days consisted mostly of exporters and importers, excess demand for yen versus the dollar caused the yen to strengthen against the dollar. That, in turn, made Japanese products less competitive in the U.S. As a result, trade frictions between the U.S. and Japan were prevented from growing any worse than they did because the dollar fell from ¥360 in mid-1971 to less than ¥200 in 1978 in response to widening Japanese trade surpluses with the U.S..

But this arrangement, in which the foreign exchange market acted as a trade equalizer, broke down with financial liberalization, which began in the U.S. with the Monetary Control Act of 1980…

…These changes prompted huge capital outflows from Japan as local investors sought higher-yielding U.S. Treasury securities. Since Japanese investors needed dollars to buy Treasuries, their demand for dollars in the currency market outstripped the supply of dollars from Japanese exporters and pushed the yen back to ¥280 against the dollar. This rekindled the two countries’ trade problems, because few U.S. manufacturers were competitive vis-a-vis the Japanese at that exchange rate.

When calls for protectionism engulfed Washington, President Ronald Reagan, a strong supporter of free trade, responded with the September 1985 Plaza Accord, which took the dollar from ¥240 in 1985 down to ¥120 just two years later. The dollar then rose to ¥160 in 1990 but subsequently fell as low as ¥79.75 in April 1995, largely ending the trade-related hostilities that had plagued the two nations’ relationship for nearly two decades…

…Capital transactions made possible by the liberalization of cross-border capital flows also began to dominate the currency market. Consequently, capital inflows to the U.S. have led to continued strength of the dollar – and stagnant or declining incomes for U.S. workers – even as U.S. trade deficits continue to mount. In other words, the foreign exchange market lost its traditional function as an automatic stabilizer for trade balances, and the resulting demands for protectionism in deficit countries are now at least as great as they were before the Plaza Accord in 1985.”

Specifically with regards to the belligerent relationship the US has with China today, Koo suggested that it was because of flaws in the free trade framework of the World Trade Organisation (WTO):

“…a key contradiction in the WTO framework: the fact that China levies high tariffs on imports from all WTO nations is no reason why the U.S.—which runs a huge trade deficit with China—should have to settle for lower tariffs on imports from China.

This problem arose because the developed‐world members of the WTO had already lowered tariffs among themselves before developing countries such as China, with their significantly lower wages and higher tariffs, were allowed to join. When they joined, developing countries could argue that they were still underdeveloped and needed higher tariffs to allow infant domestic industries to grow and to keep their trade deficits under control. Although that was a valid argument for developing countries at the time and their maximum tariff rates have come down as negotiated, the effective rates remained higher than those of advanced countries long after those countries became competitive enough to run trade surpluses with the developed world…

…Because the WTO system is based on the principle of multilateralism, with rules applied equally to all member nations, this framework provides no way of addressing bilateral imbalances between the U.S. and China. It is therefore not surprising that the Trump administration has decided to pursue bilateral, not multilateral, trade negotiations.

In retrospect, what the WTO should have done is to impose a macroeconomic condition stating that new members must lower their tariff and non‐tariff barriers to advanced‐country norms after they start to run significant trade surpluses with the latter. Here the term “significant” might be defined to mean running a trade surplus averaging more than, say, two percent of GDP for three years. If a country fails to reduce its tariffs to the advanced‐nation norm within say five years after reaching that threshold, the rest of the WTO community should then be allowed to raise tariffs on products from that country to the same level that country charges on its imports. The point is that if the country is competitive enough to run trade surpluses vis‐à‐vis advanced countries, then it should be treated as one.

If this requirement had existed when Japan joined the GATT in 1963 or when China joined the WTO in 2001, subsequent trade frictions would have been far more manageable. Under the above rules, Japan would have had to lower its tariffs starting in 1976, and China would have had to lower its tariffs from the day it joined the WTO in 2000! Such a requirement would also have enhanced the WTO’s reputation as an organization that supports not only free trade but also fair trade.”

Koo also noted that the term globalisation actually has two components, namely, free trade and free movement of capital, and that the former is important for countries to continue maintaining because of the benefits it brings, while the latter system needs improvement:

“The term “globalization” as used today actually has two components: free trade and the free movement of capital. 

Of the two, it was argued in previous chapters that the system of free trade introduced by the U.S. after 1947 led to unprecedented global peace and prosperity. Although free trade produces winners and losers and providing a helping hand to the losers is a major issue in the pursued economies, the degree of improvement in real living standards since 1945 has been nothing short of spectacular in both pursued and pursuing countries…

…The same cannot be said for the free movement of capital, the second component of globalization. Manufacturing workers and executives in the pursued economies feel so insecure not only because imports are surging but also because exchange rates driven by portfolio capital flows of questionable value are no longer acting to equilibrate trade.

To better understand this problem, let us take a step back and consider a world in which only two countries – the U.S. and Japan – are engaged in trade, and each country buys $100 in goods from the other. The next year, both countries will have the $100 earned from exporting to its trading partner, enabling it to buy another $100 in goods from that country. The two nations’ trade accounts are in balance, and the trade relationship is sustainable. 

But if the U.S. buys $100 from Japan, and Japan only buys $50 from the U.S., Japan will have $100 to use the next year, but the U.S. will have only $50, and Japanese exports to the U.S. will fall to $50 as a result. Earning only $50 from the U.S., the Japanese may have to reduce their purchases from the U.S. the following year. This sort of negative feedback loop may push trade into a “contractionary equilibrium.”

When exchange rates are added to the equation, the Japanese manufacturer that exported $100 in goods to the U.S. must sell those dollars on the currency market to buy the yen it needs to pay domestic suppliers and employees. However, the only entity that will sell it those yen is the U.S. manufacturer that exported $50 in goods to Japan.

With $100 of dollar selling and only $50 worth of yen selling, the dollar’s value versus the yen will be cut in half. This is how a surplus country’s exchange rate is pushed higher to equilibrate trade…

…If Japanese life insurers, pension funds, or other investors who need dollars to invest in the U.S. Treasury bonds sold yen and bought the remaining $50 the Japanese exporters wanted to sell, there would then be a total of $100 in dollar-buying demand for the $100 the Japanese exporter seeks to sell, and exchange rates would not change. If Japanese investors continued buying $50-worth of dollar investments each year, exchange rates would not change, in spite of the sustained $50 trade imbalances. 

Although the above arrangement may continue for a long time, the Japanese investors would effectively be lending money to the U.S. This means that at some point the money would have to be paid back. 

Unless the U.S. sells goods to Japan, there will be no U.S. exporters to provide the Japanese investors with the yen they need when they sell their U.S. Treasury bonds to pay yen obligations to Japanese pensioners and life insurance policyholders. Unless Japan is willing to continue lending to the U.S. in perpetuity, therefore, the underlying 100:50 trade imbalance will manifest itself when the lending stops.

At that point, the value of the yen will increase, resulting in large foreign exchange losses for Japanese pensioners and life insurance policyholders. Hence this scenario is also unsustainable in the long run. The U.S., too, would prefer a healthy relationship in which it sells goods to Japan and uses the proceeds to purchase goods from Japan to an unhealthy one in which it funds its purchases via constant borrowings…

…When financial markets are liberalized, capital moves to equalize the expected return in all markets. To the extent that countries with strong domestic demand tend to have higher interest rates than those with weak demand, money will flow from the latter to the former. Such flows will strengthen the currency of the former and weaken the currency of the latter. They may also add to already strong investment activity in the former by keeping interest rates lower than they would be otherwise, while depressing already weak investment activity in the latter by pushing interest rates higher than they would be otherwise.

To the extent that countries with strong domestic demand tend to run trade deficits and those with weak domestic demand run trade surpluses, these capital flows will exacerbate trade imbalances between the two by pushing the deficit country’s currency higher and pushing the surplus country’s currency lower. In other words, these flows are not only not in the best interests of individual countries, but are also detrimental to the attainment of balanced trade between countries. The widening imbalances then increase calls for protectionism in deficit countries.”

Prior to the liberalization of capital flows in the 1980s, “trade was free, but capital flows were regulated, so the foreign exchange market was driven largely by trade-related transactions.” This also meant that currency transactions could play the role they were meant to in terms of driving balanced trade:

“The currencies of trade surplus nations therefore tended to strengthen, and those of trade deficit nations to weaken. That encouraged surplus countries to import more and deficit countries to export more. In other words, the currency market acted as a natural stabilizer of trade between nations.”

As a sign of how free movement of capital has distorted the currency market, Koo noted that when the book was published “only about five percent of foreign exchange transactions involve trade, while the remaining 95 percent are attributable to capital flows.”

The problems with China’s economy

The sixth and last important idea from Koo’s book is a discussion of the factors that affect China’s economic growth, and why the country’s growth rate has slowed in recent years from the scorching pace seen in the 1990s and 200s. One issue, described by Koo, is that China no longer has a demographic tailwind to drive rapid economic growth, and is now facing a “middle-income trap” after passing the LTP around 2012:

“China actually passed the LTP around 2012 and is now experiencing sharp increases in wages. This means the country is now in its golden era, or post‐LTP maturing phase. However, because the Chinese government is wary of strikes, labor disputes, or other public disturbances of any kind, it is trying to pre‐empt such conflict by administering significant wage increases each year, with businesses required to raise wages under directives issued by local governments. In some regions, wages had risen at double‐ digit rates in a bid to prevent labor disputes. It remains to be seen whether such top‐down actions can substitute for a process in which employers and employees learn through confrontation what can reasonably be expected from the other party.

Just as China was passing the LTP, its working‐age population—defined as those aged 15 to 594—started shrinking in 2012. From a demographic perspective, it is highly unusual for the entire labor supply curve to begin shifting to the left just as a country reaches the LTP. Japan, Taiwan, and South Korea all enjoyed about 30 years of workforce growth after reaching their LTPs. The huge demographic bonus China enjoyed until 2012 is not only exhausted, but has now reversed… That means China that will not be able to maintain the rapid pace of economic growth seen in the past, and in fact growth has already slowed sharply. 

Higher wages in China are now leading both Chinese and foreign businesses to move factories to lower-wage countries such as Vietnam and Bangladesh, prompting fears that China will become stuck in the so-called “middle-income trap”. This trap arises from the fact that once a country loses its distinction as the lowest-cost producer, many factories may leave for lower-cost destinations, resulting in less investment and less growth. In effect, the laws of globalization and free trade that benefited China when it was the lowest-cost producer are now posing real challenges for the country.”

Koo proposed ideas for China to reinvigorate China’s growth, such as investing in productivity-enhancing measures for domestic workers. 

Another important factor affecting China’s economic growth involves the appropriate type of policy the government should implement to govern the economy. Since the country had passed the LTP more than a decade ago, and is in its golden era, fiscal policy – the act of the government directing the economy – is no longer the effective way to govern the economy. But is the government relinquishing control? To complicate matters, there are early signs now that China may already be in the pursued stage, in which case, fiscal policy will be important again. It remains to be seen what would be the most appropriate way for the government to lead China’s economy. 

---

Disclaimer: The Good Investors is the personal investing blog of two simple guys who are passionate about educating Singaporeans about stock market investing. By using this Site, you specifically agree that none of the information provided constitutes financial, investment, or other professional advice. It is only intended to provide education. Speak with a professional before making important decisions about your money, your professional life, or even your personal life. I don’t have a vested interest in any company mentioned. Holdings are subject to change at any time.

The way I see it, artificial intelligence (or AI), really leapt into the zeitgeist in late-2022 or early-2023 with the public introduction of DALL-E2 and ChatGPT. Both are provided by OpenAI and are software products that use AI to generate art and writing, respectively (and often at astounding quality). Since then, developments in AI have progressed at a breathtaking pace.

With the latest earnings season for the US stock market – for the fourth quarter of 2024 – coming to its tail-end, I thought it would be useful to collate some of the interesting commentary I’ve come across in earnings conference calls, from the leaders of technology companies that I follow or have a vested interest in, on the topic of AI and how the technology could impact their industry and the business world writ large. This is an ongoing series. For the older commentary:

• 2023 Q1 – here and here
• 2023 Q2 – here and here
• 2023 Q3 – here and here
• 2023 Q4 – here and here
• 2024 Q1 – here and here
• 2024 Q2 – here and here
• 2024 Q3 – here and here

I’ve split the latest commentary into two parts for the sake of brevity. This is Part 1, and you can Part 2 here. With that, I’ll let the management teams take the stand… 

Airbnb (NASDAQ: ABNB)

Airbnb’s management thinks AI is early and has yet to fundamentally change the travel market for any of the large travel platforms; most travel companies are starting with AI on trip planning but management thinks AI is still too early for trip planning

Here’s what I think about AI. I think it’s still really early. It’s probably similar to like the mid- to late ’90s for the Internet. So I think it’s going to have a profound impact on travel, but I don’t think it’s yet fundamentally changed for any of the large travel platforms…

…So most companies, what they’re actually doing is they’re doing integrations of these other platforms on trip planning. But the trip planning, it’s still early. I don’t think it’s quite bit ready for prime time.

Airbnb’s management is starting with AI in customer service; Airbnb will roll out AI-powered customer support later in 2025; management thinks AI can provide great customer support partly because it can speak all languages 24/7 and read thousands of pages of documents; management will eventually graduate the customer-support AI into a travel and living concierge; management thinks AI can help improve efficiency at Airbnb in customer service

We’re actually starting with customer service. So later this year, we’re going to be rolling out, as part of our Summer Release, AI-powered customer support. As you imagine, we get millions of contacts every year. AI can do an incredible job of customer service. It can speak every language 24/7. It can read a corpus of thousands of pages of documents. And so we’re starting with customer support. And over the coming years, what we’re going to do is we’re going to take the AI-powered customer service agent, and we’re going to bring it into essentially Airbnb search to eventually graduate to be a travel and living concierge…

…[Question] With respect to the AI, I appreciate your answer with respect to outward-looking and how it might change the landscape. What do you think the potential is internally to apply AI for efficiencies inside the company and create an additional layer of potential margin efficiency and/or free cash flow conversion in the years ahead?

[Answer] There’s like a couple like efficiencies that you could imagine at Airbnb. One is obviously customer service. I think that’s like one of the biggest ones. I’ve kind of already covered that, but I think that’s like a massive change for Airbnb.

Airbnb’s management thinks that AI models are getting cheaper and are starting to be commoditised

I think it’s a really exciting time in the space because you’ve seen like with DeepSeek and more competition with models is models are getting cheaper or nearly free. They’re getting faster and they’re getting more intelligent. And they are, for all intent and purpose, starting to get commoditized.

Airbnb’s management thinks that a lot of the value from AI is going to accrue to platforms, and they want Airbnb to be the platform for travel and living that will reap most of the value from AI

What I think that means is a lot of value is going to accrue to the platform. And ultimately, I think the best platform, the best applications are going to be the ones that like most accrue the value from AI. And I think we’re going to be the one to do that with traveling and living.

Airbnb’s management thinks AI can help improve efficiency at Airbnb in engineering productivity; in the short-term, the improvement in engineering productivity has not been material; over the next few years, management thinks AI can drive a 30% increase in engineering productivity at Airbnb; over the long-term, management thinks there can an order of magnitude more productivity; management think younger, more innovative companies, could benefit from AI more than incumbent enterprises

[Question] With respect to the AI, I appreciate your answer with respect to outward-looking and how it might change the landscape. What do you think the potential is internally to apply AI for efficiencies inside the company and create an additional layer of potential margin efficiency and/or free cash flow conversion in the years ahead?

[Answer] The other, I assume, you refer to is essentially engineering productivity. We are seeing some productivity gains. I’ve talked to a lot of other tech CEOs, and here’s what I’ve heard talking to other like tech CEOs. Most of them haven’t seen a material like change in engineering productivity. Most of the engineers are using AI tools. They’re seeing some productivity. I don’t think it’s flowing to like a fundamental step-change in productivity yet. I think a lot of us believe in some kind of medium term of a few years, you could easily see like a 30% increase in technology and engineering productivity. And then, of course, beyond that, I mean, I think it could be like an order of magnitude more productivity because — but that’s going to be like down the road. And I think that’s going to be something that almost all companies benefit from. I think the kind of younger, more innovative, startup-like companies might benefit a little bit more because they’ll have engineers who are more likely to adopt the tools.

Alphabet (NASDAQ: GOOG)

AI Overviews in Search is now available in more than 100 countries; AI Overviews drive higher user satisfaction and search usage; Google’s Gemini model is being used in AI Overviews; with AI Overviews, usage growth of Search is growing over time, especially with younger users; management recently launched ads in AI Overviews and AI Overviews is currently monetising at nearly the same rate as Google Search; Google Search has continued to perform well in this AI age, as overall usage has continued to grow, with stronger growth seen in AI Overviews across all segments

In Search, AI overviews are now available in more than 100 countries. They continue to drive higher satisfaction and search usage…

…That includes Search where Gemini is pairing our AI overviews. People use search more with AI overviews and usage growth increases over-time as people learn that they can ask new types of questions. This behavior is even more pronounced with younger users who really appreciate the speed and efficiency of this new format…

…We’ve already started testing Gemini 2.0 in AI overviews and plan to roll it out more broadly later in the year…

…We recently launched the ads within AI Overviews on mobile in the U.S., which builds on our previous rollout of ads above and below. And as I talked about before, for the AI Overviews, overall, we actually see monetization at approximately the same rate, which I think really gives us a strong base on which we can innovate even more…

…On Search usage, overall, our metrics are healthy. We are continuing to see growth in Search on a year-on-year basis in terms of overall usage. Of course, within that, AI Overviews has seen stronger growth, particularly across all segments of users, including younger users, so it’s being well received. But overall, I think through this AI moment, I think Search is continuing to perform well.

Circle to Search is now available on more than 200 million Android devices; Circle to Search is opening new Search use cases; Circle to Search is popular with younger users; Circle to Search is used to start more than 10% of searches among users who have tried it before

Circle to Search is now available on over 200 million Android devices…

…Circle to Search is driving additional Search use and opening up even more types of questions. This feature is also popular among younger users. Those who have tried Circle to Search before now use it to start more than 10% of their searches…

…In Search, we’re seeing people increasingly ask entirely new questions using their voice, camera or in ways that were not possible before, like with Circle to search.

Alphabet’s management believes Google has a unique infrastructure advantage in AI because the company has developed each component of its technology stack; Alphabet broke ground on 11 new cloud regions and data center campuses in 2024, and announced plans for 7 new subsea cable projects; Google data centers now deliver 4x more computing power per unit of electricity compared to 5 years ago

We have a unique advantage because we develop every component of our technology stack, including hardware, compilers, models and products. This approach allows us to drive efficiencies at every level from training and serving to develop our productivity. In 2024, we broke ground on 11 new cloud regions and data center campuses in places like South Carolina, Indiana, Missouri and around the world.

We also announced plans for seven new subsea cable projects, strengthening global connectivity. Our leading infrastructure is also among the world’s most efficient. Google data centers deliver nearly four times more computing power per unit of electricity compared to just five years ago.

Google Cloud customers consume 8x more compute capacity for training and inference compared to 18 months ago; first-time commitments to Google Cloud more than doubled in 2024; Google Cloud closed a few deals in 2024 worth more than $1 billion each; Google Cloud’s AI hypercomputer utilises both GPUs (graphics processing units) and TPUs (tensor processing units), and has helped Wayfair improve performance and scalability by 25%; Google saw strong uptake of Trillium, its 6th generation TPU, in 2024 Q4; Trillium is 4x better in training and has 3x higher inference throughput than the 5th generation TPU; Google Cloud is offering NVIDIA’s H200 GPUs to customers; Google Cloud is the first cloud provider to provide NVIDIA’s Blackwell GPUs; the capex for Google Cloud is mostly for Google’s own self-designed data centers and TPUs (tensor processing units)

Today, cloud customers consume more than eight times the compute capacity for training and inferencing compared to 18 months ago…

…In 2024, the number of first-time commitments more than double compared to 2023…

…Last year, we closed several strategic deals over $1 billion and the number of deals over $250 million doubled from the prior year…

…We continue to see strong growth across our broad portfolio of AI-powered cloud solutions. It begins with our AI hypercomputer, which delivers leading performance and cost across both GPUs and TPUs. These advantages help Citadel with modeling markets and training and enabled Wayfair to modernize its platform, improving performance and scalability by nearly 25%. 

In Q4, we saw strong uptake of Trillium, our sixth-generation TPU, which delivers four times better training performance and three times greater inference throughput compared to the previous generation. We also continue our strong relationship with NVIDIA. We recently delivered their H200 based platforms to customers. And just last week, we were the first to announce a customer running on the highly-anticipated Blackwell platform…

……Our strategy is mostly to rely our own self-design and build data centers. So, they’re industry-leading in terms of both cost and power efficiency at scale. We have our own customized TPUs. They’re customized for our own workload, so they do deliver outstanding the superior performance and capex efficiency. So, we’re going to be looking at all that when we make decisions as to how we’re going to progress capital investments throughout the coming years.

Google launched an experimental version of its Gemini 2.0 Flash model in December 2024, but the model will be generally available for developers and customers; Google debuted its experimental Gemini 2.0 Flash Thinking model in late-2024 and it has gathered extremely positive reviews; Google is working on even better thinking models; Gemini 2.0’s advances in multimodality and native tool use helps Google build a universal AI assistant; an example of this universal assistant can be seen in Deep Research; Deep Research was launched in Gemini Advanced in December and is being rolled out to Android users globally; the consumer Gemini app debuted on iOS in November 2024 and has seen great product momentum; Project Mariner and Project Astra are AI agent products currently being tested and they will appear in the Gemini app sometime in 2025; Gemini and Google’s video and image generation models consistently excel in industry leaderboards and benchmarks; 4.4 million developers are using Gemini models today, double from just six months ago; Google has 7 products with over 2 billion users each, and all 7 products use Gemini; all Google Workspace business and enterprise customers were recently given access to all of Gemini’s AI capabilities; Gemini 2.0 Flash is one of the most capable models people can access for free; management’s current intention for monetisation of Gemini is through subscriptions and improving the user experience, but they have an eye on advertising revenues

In December, we unveiled Gemini 2.0, our most capable AI model yet, built for the agent era. We launched an experimental version of Gemini 2.0 Flash, our workhorse model with low-latency and enhanced performance. Flash has already rolled-out to the Gemini app, and tomorrow we are making 2.0 Flash generally available for developers and customers, along with other model updates…

…Late last year, we also debuted our experimental Gemini 2.0 Flash Thinking model. The progress to scale thinking has been super-fast and the review so-far have been extremely positive. We are working on even better thinking models and look-forward to sharing those with the developer community soon.

Gemini 2.0’s advances in multimodality and native tool use enable us to build new agents that bring us closer to our vision of a universal assistant. One early example is deep research. It uses agent capabilities to explore complex topics on your behalf and give key findings along with sources. It launched in Gemini Advanced in December and is rolling out to Android users all over the world.

We are seeing great product momentum with our consumer Gemini app, which debuted on iOS last November.

…We have opened up trusted tester access to a handful of research prototypes, including Project Mariner, which can understand and reason across information on a browser screen to complete tasks and Project Astra. We expect to bring features from both to the Gemini app later this year…

…Veo2, our state-of-the-art video generation model and Imagine3, our highest-quality text image model. These generative media models as well as Gemini consistently top industry leaderboards and score top marks across industry benchmarks. That’s why more than 4.4 million developers are using our Gemini models today, double the number from just six months ago…

…We have seven products and platforms with over 2 billion users and all are using Gemini…

…We recently gave all Google Workspace business and enterprise customers access to all of our powerful Gemini AI capabilities to help boost their productivity…

…2.0 Flash. I mean, I think that’s one of the most capable models you can access at the free tier…

…[Question] How should we think about the future monetization opportunity of Gemini? Today, it’s really a premium subscription offering or a free offering. Over time, do you see an ad component?

[Answer] On the monetization side, obviously, for now, we are focused on a free tier and subscriptions. But obviously, as you’ve seen in Google over time, we always want to lead with user experience. And we do have very good ideas for native ad concepts, but you’ll see us lead with the user experience. And — but I do think we’re always committed to making the products work and reach billions of users at scale. And advertising has been a great aspect of that strategy. And so, just like you’ve seen with YouTube, we’ll give people options over time. But for this year, I think you’ll see us be focused on the subscription direction.

Google Cloud’s AI developer platform, Vertex AI, had a 5x year-on-year increase in customers in 2024 Q4; Vertex AI offers more than 200 foundation models form Google; Vertex AI’s usage grow 20x in 2024

Our AI developer platform, Vertex AI, saw a 5x increase in customers year-over-year with brands like, International and WPP building new applications and benefiting from our 200 plus foundation models. Vertex usage increased 20x during 2024 with particularly strong developer adoption of Gemini Flash, Gemini 2.0, and most recently VEO.

Alphabet’s management will be introducing Veo2, Google’s video generation model, for creators in Youtube in 2025; advertisers around the world can now promote Youtube creator videos and ad campaigns across all AI-powered campaign types and Google ads

Expanding on our state-of-the-art video generation model, we announced Veo2, which creates incredibly high-quality video in a wide range of subjects and styles. It’s been inspiring to see how people are experimenting with it. We’ll make it available to creators on YouTube in the coming months…

…. All advertisers globally can now promote YouTube creator videos and ad campaigns across all AI-powered campaign types and Google Ads, and creators can tag partners in their brand videos…

Alphabet’s management announced the first beta of Android 16 in January 2025; there will be deeper Gemini integration for the new Samsung Galaxy S25 smartphone series; Alphabet has announced Android XR, the first Android platform for the Gemini era

Last month, we announced the first beta of Android 16 plus new Android updates, including a deeper Gemini integration coming to the new Samsung Galaxy S25 series. We also recently-announced Android XR, the first Android platform built for the Gemini era. Created with Samsung and Qualcomm, Android XR is designed to power an ecosystem of next-generation extended reality devices like headsets and glasses.

Waymo is now serving more than 150,000 trips per week (was 150,000 in 2024 Q3); Waymo is expanding in new markets in the USA this year and in 2026; Waymo will soon be launched in Tokyo; Waymo is developing its 6th-gen driver, which will significantly reduce hardware costs

It’s now averaging over 150,000 trips each week and growing. Looking ahead, Waymo will be expanding its network and operations partnerships to open up new markets, including Austin and Atlanta this year and Miami next year. And in the coming weeks, Waymo One vehicles will arrive in Tokyo for their first international road trip. We are also developing the sixth-generation Waymo driver, which will significantly lower hardware costs.

Alphabet’s management introduced a new Google shopping experience, infused with AI, in 2024 Q4, and there was 13% more daily active users in Google shopping in December 2024 compared to a year ago; the new Google Shopping experience helps users speed up their shopping research

Google is already present in over half of journeys where a new brand, product or retailer are discovered by offering new ways for people to search, we’re expanding commercial opportunities for our advertisers…

…Retail was particularly strong this holiday season, especially on Black Friday and Cyber Monday, which each generated over $1 billion in ad revenue. Interestingly, despite the U.S. holiday shopping season being the shortest since 2019, retail sales began much earlier in October, causing the season to extend longer than anticipated.

People shop more than 1 billion times a day across Google. Last quarter, we introduced a reinvented Google shopping experience, rebuilt from the ground up with AI. This December saw roughly 13% more daily active users in Google shopping in the U.S., compared to the same period in 2023…

…The new Google Shopping experiences specifically to your question, users to really intelligently show the most relevant products, helping to speed up and simplify your research. You get an AI-generated brief with top things to consider for your search plus maybe products that meet your needs. So, shoppers very often want low prices. So, the new page not only includes like deal-finding tools like price comparison, price insights, price tracking throughout. But it’s also a new and dedicated personalized deals page, which can browse deals for you, and all this is really built on the backbone of AI.

Shoppers can now take a photo and use Lens to quickly find information about the product; Lens is now used in over 20 billion visual searches per month (was over 20 billion in 2024 Q3); majority of Lens searches are incremental

Shoppers can now take a photo of a product and using Lens quickly find information about the product, reviews, similar products and where they can get it for a great price. Lens is used for over 20 billion visual search queries every month and the majority of these searches are incremental.

Alphabet’s management continues to infuse AI capabilities into Google’s advertising business; Petco used Demand Gen campaigns to achieve a 275% increase in return on ad spend and a 74% increase in click-through rates compared to social benchmarks; Youtube Select Creator Takeovers is now generally available in the US and will be rolled out across the world; PMax was recently strengthened with new controls and easier reporting functions; Event Ticket Center used PMax and saw a 5x increase in production of creative assets, driving a 300% increase in conversions compared to using manual assets; Meridian, Google’s marketing mix model, was recently made generally available and it delivers 17% higher return on advertising spend on Youtube compared to manual campaigns

We continue investing in AI capabilities across media buying, creative and measurement. As I’ve said before, we believe that AI will revolutionize every part of the marketing value chain.

And over the past quarter, we’ve seen how our customers are increasingly focusing on optimizing the use of AI. As an example, [ Petco ], used Demand Gen campaigns across targeting, creative generation and bidding to find new pet parent audiences across YouTube. They achieved a 275% higher return on ad spend and a 74% higher click-through rate than their social benchmarks.

On media buying, we made YouTube Select Creator Takeovers generally available in the U.S. and will be expanding to more markets this year. Creators know their audience the best and creator takeovers help businesses connect with consumers through authentic and relevant content.

Looking at Creative, we introduced new controls and made reporting easier in PMax, helping customers better understand and reinvest into their best-performing assets. Using asset generation in PMax, Event Ticket Center achieved a 5x increase in production of creative assets saving time and effort. They also increased conversions by 300% compared to the previous period when they used manual assets…

…Last week, we made Meridian, our marketing mix model, generally available for customers, helping more business reinvest into creative and media buying strategies that they know work. Based on the Nielsen meta analysis of marketing mix models, on average, Google AI-powered video campaigns on YouTube delivered 17% higher return on advertising spend than manual campaigns.

Sephora used Demand Gen Shorts-only channel for advertising that drove an 82% increase in searches for Sephora Holiday

Sephora used demand gen Shorts-only channel to boost traffic and brand searches for the holiday gift guide campaign and leverage greater collaborations to find the best gift. This drove an 82% relative uplift in searches for Sephora holiday.

Citi is using Google Cloud for its generative AI initiatives across customer service, document summarisation, and search

Another expanding partnership is with Citi, who is modernizing its technology infrastructure with Google Cloud to transform employee and customer experiences. Using Google Cloud, it will improve its digital products, streamline employee workflows and use advanced high-performance computing to enable millions of daily computations. This partnership also fuels Citi’s generate AI initiatives across customer service, document summarization and search to reduce manual processing.

Google Cloud had 30% revenue growth in 2024 Q4 (was 35% in 2024 Q3) driven by growth in core GCP products, AI infrastructure, and generative AI solutions; operating margin was 17.5% (was 17% in 2024 Q3 and was 9.4% in 2023 Q4); GCP grew at a much higher rate than Google Cloud overall; Google Cloud had more AI demand than capacity in 2024 Q4; management is thinking about Google’s capital intensity, but they want to invest because they are seeing strong AI demand both internally and externally; the capex Google is making can be repurposed across its different businesses

Turning to the Google Cloud segment, which continued to deliver very strong results this quarter. Revenue increased by 30% to $12 billion in the fourth quarter, reflecting growth in GCP, across core GCP products, AI infrastructure, and generative AI solutions. Once again, GCP grew at a rate that was much higher than cloud overall. Healthy Google Workspace growth was primarily driven by increase in average revenue per seat. Google Cloud operating income increased to $2.1 billion and operating margin increased from 9.4% to 17.5%…

…We do see and have been seeing very strong demand for our AI products in the fourth quarter in 2024. And we exited the year with more demand than we had available capacity. So, we are in a tight supply demand situation, working very hard to bring more capacity online…

…[Question] How do you think about long-term capital intensity for this business?

[Answer] On the first one, certainly, we’re looking ahead, but we’re managing very responsibly. It was a very rigorous, even internal governance process, looking at how do we allocate the capacity and what would we need to support the customer demand externally, but also across the Google — the Alphabet business. And as you’ve seen in the comment I’ve just made on Cloud, we do have demand that exceeds our available capacity. So, we’ll be working hard to address that and make sure we bring more capacity online. We do have the benefit of having a very broad business, and we can repurpose capacity, whether it’s through Google Services or Google Cloud to support, as I said, whether it’s search or GDM, or Google Cloud customers, we can do that in a more efficient manner.

Alphabet’s management thinks Google’s AI models are in the lead when compared to DeepSeek’s, and this is because of Google’s full-stack development

If you look at one of the areas in which the Gemini model shines is the Pareto frontier of cost, performance, and latency. And if you look at all three attributes, I think we are — we lead this period of frontier. And I would say both our 2.0 Flash models, our 2.0 Flash thinking models, they are some of the most efficient models out there, including comparing to DeepSeek’s V3 and R1. And I think a lot of it is our strength of the full stack development, end-to-end optimization, our obsession with cost per query.

Alphabet’s management has seen the proportion of AI spend on inference growing over the last 3 years when compared to training; management thinks reasoning AI models will accelerate this trend

A couple of things I would say are if you look at the trajectory over the past three years, the proportion of the spend toward inference compared to training has been increasing, which is good because, obviously, inferences to support businesses with good ROIC…

…I think the reasoning models, if anything, accelerates that trend because it’s obviously scaling upon inference dimension as well.

Alphabet’s management thinks that AI agents and Google Search are not competing in a zero-sum game

[Question] With your own project Mariner efforts and a competitor’s recent launch, it seems there’s suddenly really strong momentum on AI consumer agents and kind of catching up to that old Google Duplex Vision. I think when you look a few years ahead, where do you see consumer agents going? And really, what does it mean to Google Search outside of Lens? Is there room for both to flourish?

[Answer] Gemini 2.0 was definitely built with the view of enabling more agentic use cases. And so, I actually — we are definitely seeing progress inside. And I think we’ll be able to do more agentic experiences for our users. Look, I actually think all of this expands the opportunity space. I think it — historically, we’ve had information use cases, but now you can have — you can act on your information needs in a much deeper way. It’s always been our vision when we have talked about Google Assistant, etc. So, I think the opportunity space expands. I think there’s plenty of it, feels very far from a zero-sum game. There’s plenty of room, I think, for many new types of use cases to flourish. And I think for us, we have a clear sense of additional use cases we can start to tackle for our users in Google Search.

Alphabet’s management has been passing on cost differentiations arising from Google Cloud’s end-to-end stack approach to customers

Part of the reason we have taken the end-to-end stack approach is so that we can definitely drive a strong differentiation in end-to-end optimizing and not only on a cost but on a latency basis, on a performance basis. Be it the Pareto frontier we mentioned, and I think our full stack approach and our TPU efforts all play give a meaningful advantage. And we plan — you already see that. I know you asked about the cost, but it’s effectively captured when we price outside, we pass on the differentiation. 

Amazon (NASDAQ: AMZN)

AWS grew 19% year-on-year in 2024 Q4, and is now at a US$115 billion annualised revenue run rate; management expects lumpiness in AWS’s growth in the next few years, but is incredibly optimistic about AWS’s growth; management thinks the future will be one where (a) every app is infused with generative AI that has inference as a core building block, and (b) companies will have AI agents accomplishing tasks and interacting with each other; management believes this future will be built on the cloud, and mostly on AWS; the shift by enterprises from on-premises to the cloud, which is a non-AI activity, continues for AWS; AWS continues to innovate in non-AI areas; AWS’s growth in 2024 Q4 was driven by both generative AI and non-generative AI offerings; AWS had a massive 48% year-on-year jump in operating income in 2024 Q4, helped partly by an increase in estimated useful life of servers that started in 2024; management sees AWS being capable of faster growth today if not for supply constraints; the constraints relate to (1) chips from 3rd-party partners (most likely referring to NVIDIA), (2) AWS’s own Trainium chips, (3) power for data centers, and (4) other supply chain components; management sees the AWS constraints starting to relax in 2025 H2; AWS’s AI services come with lower margins right now, but management thinks the AI-related margin will over time be on par with the non-AI margin

In Q4, AWS grew 19% year-over-year and now has a $115 billion annualized revenue run rate. AWS is a reasonably large business by most folks’ standards. And though we expect growth will be lumpy over the next few years as enterprise adoption cycles, capacity considerations and technology advancements impact timing, it’s hard to overstate how optimistic we are about what lies ahead for AWS’ customers and business…

…While it may be hard for some to fathom a world where virtually every app has generative AI infused in it, with inference being a core building block just like compute, storage and database, and most companies having their own agents that accomplish various tasks and interact with one another, this is the world we’re thinking about all the time. And we continue to believe that this world will mostly be built on top of the cloud with the largest portion of it on AWS…

…While AI continues to be a compelling new driver in the business, we haven’t lost our focus on core modernization of companies’ technology infrastructure from on-premises to the cloud. We signed new AWS agreements with companies, including Intuit, PayPal, Norwegian Cruise Line Holdings, Northrop Grumman, The Guardian Life Insurance Company of America, Reddit, Japan Airlines, Baker Hughes, The Hertz Corporation, Redfin, Chime Financial, Asana, and many others. Consistent customer feedback from our recent AWS re:Invent gathering was appreciation that we’re still inventing rapidly in non-AI key infrastructure areas like storage, compute, database and analytics…

…During the fourth quarter, we continued to see growth in both generative AI and non-generative AI offerings as companies turn their attention to newer initiatives, bring more workloads to the cloud, restart or accelerate existing migrations from on-premise to the cloud, and tap into the power of generative AI…

…AWS reported operating income of $10.6 billion, an increase of $3.5 billion year-over-year. This is a result of strong growth, innovation in our software and infrastructure to drive efficiencies, and continued focus on cost control across the business. As we’ve said in the past, we expect AWS operating margins to fluctuate over time driven in part by the level of investments we’re making. Additionally, we increased the estimated useful life of our servers starting in 2024, which contributed approximately 200 basis points to the AWS margin increase year-over-year in Q4…

……It is true that we could be growing faster, if not for some of the constraints on capacity. And they come in the form of, I would say, chips from our third-party partners, come a little bit slower than before with a lot of midstream changes that take a little bit of time to get the hardware actually yielding the percentage-healthy and high-quality servers we expect. It comes with our own big new launch of our own hardware and our own chips and Trainium2, which we just went to general availability at re:Invent, but the majority of the volume is coming in really over the next couple of quarters, the next few months. It comes in the form of power constraints where I think the world is still constrained on power from where I think we all believe we could serve customers if we were unconstrained. There are some components in the supply chain, like motherboards, too, that are a little bit short in supply for various types of servers…

…I predict those constraints really start to relax in the second half of ’25…

…At the stage we’re in right now, AI is still early stage. It does come originally with lower margins and a heavy investment load as we’ve talked about. And in the short term, over time, that should be a headwind on margins. But over the long term, we feel the margins will be comparable in non-AI business as well.

Amazon’s management sees NVIDIA being an important partner of AWS for a long time; management does not see many large-scale generative AI apps existing right now; when generative AI apps reach scale, their costs to operate can rise very quickly, and management believes this will drive customers to demand better price performance from chips, which is why AWS built its custom AI chips; Trainium 2, AWS’s custom AI chip, was launched in December 2024; EC2 instances powered by Trainium 2 is 30%-40% more price performant than instances powered by other GPUs; important technology companies such as Adobe, Databricks, and Qualcomm have seen impressive results after testing Trainium 2; Anthropic is building its future frontier models on Trainium 2; AWS is collaborating with Anthropic on Project Rainier, which is a cluster of a few hundred thousand Trainium 2 chips that have 5x the exaflops Anthropic used to train its current set of models; management is already 

Most AI compute has been driven by NVIDIA chips, and we obviously have a deep partnership with NVIDIA and will for as long as we can see into the future. However, there aren’t that many generative AI applications of large scale yet. And when you get there, as we have with apps like Alexa and Rufus, cost can get steep quickly. Customers want better price performance and it’s why we built our own custom AI silicon. Trainium2 just launched at our AWS re:Invent Conference in December. And EC2 instances with these chips are typically 30% to 40% more price performant than other current GPU-powered instances available. That’s very compelling at scale. Several technically-capable companies like Adobe, Databricks, Poolside and Qualcomm have seen impressive results in early testing of Trainium2. It’s also why you’re seeing Anthropic build their future frontier models on Trainium2. We’re collaborating with Anthropic to build Project Rainier, a cluster of Trainium2 UltraServers containing hundreds of thousands of Trainium2 chips. This cluster is going to be 5x the number of exaflops as the cluster that Anthropic used to train their current leading set of cloud models. We’re already hard at work on Trainium3, which we expect to preview late in ’25 and defining Trainium4 thereafter.

Building outstanding performant chips that deliver leading price performance has become a core strength of AWS’, starting with our Nitro and Graviton chips in our core business and now extending to Trainium and AI and something unique to AWS relative to other competing cloud providers.

Amazon’s management has seen Amazon SageMaker AI, AWS’s fully-managed AI service, become the go-to service for AI model builders; SageMaker’s HyperPod automatically splits training workloads across many AI accelerators and prevents interruptions, saving training time up tp 40%; management recently released new features for SageMaker, such as the ability to prioritise which workloads to receive capacity when budgets are reached; the latest version of SageMaker is able to integrate all of AWS’s data analytics and AI services into one surface

I won’t spend a lot of time in these comments on Amazon SageMaker AI, which has become the go-to service for AI model builders to manage their AI data, build models, experiment and deploy these models, except to say that SageMaker’s HyperPod capability, which automatically splits training workloads across many AI accelerators, prevents interruptions by periodically saving checkpoints, and automatically repairing faulty instances from their last saved checkpoint and saving training time by up to 40%. It continues to be a differentiator, received several new compelling capabilities at re:Invent, including the ability to manage costs at a cluster level and prioritize which workloads should receive capacity when budgets are reached, and is increasingly being adopted by model builders…

…There were several key launches customers were abuzz about, including Amazon Aurora DSQL, our new serverless distributed SQL database that enables applications with the highest availability, strong consistency, PostgreS compatibility and 4x faster reads and writes compared to other popular distributed SQL databases; Amazon S3 tables, which make S3 the first cloud object store with fully managed support for Apache Iceberg for faster analytics; Amazon S3 Metadata, which automatically generates queryable metadata, simplifying data discovery, business analytics, and real-time inference to help customers unlock the value of their data in S3; and the next generation of Amazon SageMaker, which brings together all of the data analytics services and AI services into one interface to do analytics and AI more easily at scale.

Amazon Bedrock is AWS’s fully-managed service for developers to build generative AI applications by leverage on frontier models; management recently introduced more than 100 popular emerging models on Bedrock, including DeepSeek’s R1 models; management recently introduced new features to Bedrock to help customers lower cost and latency in inference workloads; management is seeing Bedrock resonate strongly with customers; management recently released Amazon’s own Nova family of frontier models on Bedrock; customers are starting to experiment with DeepSeek’s models

Amazon Bedrock is our fully managed service that offers the broadest choice of high-performing foundation models with the most compelling set of features that make it easy to build a high-quality generative AI application. We continue to iterate quickly on Bedrock announcing Luma AI poolside and over 100 other popular emerging models to Bedrock at re:Invent. In short order, we also just added DeepSeek’s R1 models to Bedrock and SageMaker…

…We delivered several compelling new Bedrock features at re:Invent, including prompt caching, intelligent prompt routing and model distillation, all of which help customers achieve lower cost and latency in their inference. Like SageMaker AI, Bedrock is growing quickly and resonating strongly with customers…

…We also just launched Amazon’s own family of frontier models in Bedrock called Nova…

…We moved so quickly to make sure that DeepSeek was available both in Bedrock and in SageMaker faster than you saw from others. And we already have customers starting to experiment with that.

The Nova family has comparable intelligence with other leading AI models, but also offers lower latency and price, and integration with important Bedrock features; many large enterprises, including Palantir, Deloitte, and SAP, are already using Nova

We also just launched Amazon’s own family of frontier models in Bedrock called Nova. These models compare favorably in intelligence against the leading models in the world but offer lower latency; lower price, about 75% lower than other models in Bedrock; and are integrated with key Bedrock features like fine-tuning, model distillation, knowledge bases of RAG and agentic capabilities. Thousands of AWS customers are already taking advantage of the capabilities and price performance of Amazon Nova models, including Palantir, Deloitte, SAP, Dentsu, Fortinet, Trellix, and Robinhood, and we’ve just gotten started.

Amazon’s management still thinks Amazon Q is the most capable AI-powered software development assistant; early testing shows that Amazon Q can now shorten a multi-year mainframe migration by 50%

Amazon Q is the most capable generative AI-powered assistant for software development and to leverage your own data…

…We obliged with our recent deliveries of Q Transformations that enable moves from Windows.NET applications to Linux, VMware to EC2, and accelerates mainframe migrations. Early customer testing indicates that Q can turn what was going to be a multiyear effort to do a mainframe migration into a multi-quarter effort, cutting by more than 50% the time to migrate mainframes. This is a big deal and these transformations are good examples of practical AI.

Amazon’s management expects capital expenditures of around US$105 billion for the whole of 2025 (was around $75 billion in 2024); the capex in 2025 will be for AWS as well as the retail business, but will primarily be for AWS’s AI infrastructure; reminder that the faster AWS grows, the faster Amazon needs to invest capital for hardware; management will only spend on capex if they see significant signals of demand; management thinks AI is a once-in-a-lifetime business opportunity, and that it’s a good sign on the long-term growth opportunities AWS has when capex is expanding

Capital investments were $26.3 billion in the fourth quarter, and we think that run rate will be reasonably representative of our 2025 capital investment rate. Similar to 2024, the majority of the spend will be to support the growing need for technology infrastructure. This primarily relates to AWS, including to support demand for our AI services, as well as tech infrastructure to support our North America and International segments. Additionally, we’re continuing to invest in capacity for our fulfillment and transportation network to support future growth. We’re also investing in same-day delivery facilities and our inbound network as well as robotics and automation to improve delivery speeds and to lower our cost to serve. These capital investments will support growth for many years to come…

…The vast majority of that CapEx spend is on AI for AWS. The way that AWS business works and the way the cash cycle works is that the faster we grow, the more CapEx we end up spending because we have to procure data center and hardware and chips and networking gear ahead of when we’re able to monetize it. We don’t procure it unless we see significant signals of demand. And so when AWS is expanding its CapEx, particularly in what we think is one of these once-in-a-lifetime type of business opportunities like AI represents, I think it’s actually quite a good sign, medium to long term, for the AWS business…

…We also have CapEx that we’re spending this year in our Stores business, really with an aim towards trying to continue to improve the delivery speed and our cost to serve. And so you’ll see us expanding the number of same-day facilities from where we are right now. You’ll also see us expand the number of delivery stations that we have in rural areas so we can get items to people who live in rural areas much more quickly, and then a pretty significant investment as well on robotics and automation so we can take our cost to serve down and continue to improve our productivity.

Amazon’s management completed a useful life study for its servers and network equipment in 2024 Q4 and has decreased the useful life estimate; management early retired some servers and network equipment in 2024 Q4; the decrease in useful life estimate and the early retirement will lower Amazon’s operating income, primarily in the AWS segment

In Q4, we completed a useful life study for our servers and network equipment, and observed an increased pace of technology development, particularly in the area of artificial intelligence and machine learning. As a result, we’re decreasing the useful life for a subset of our servers and network equipment from 6 years to 5 years, beginning in January 2025. We anticipate this will decrease full year 2025 operating income by approximately $700 million. In addition, we also early retired a subset of our servers and network equipment. We recorded a Q4 2024 expense of approximately $920 million from accelerated depreciation and related charges and expect this will also decrease full year 2025 operating income by approximately $600 million. Both of these server and network equipment useful life changes primarily impact our AWS segment.

Amazon’s management sees AI as the biggest opportunity since cloud and the internet

From our perspective, we think virtually every application that we know of today is going to be reinvented with AI inside of it and with inference being a core building block, just like compute and storage and database. If you believe that, plus altogether new experiences that we’ve only dreamed about are going to actually be available to us with AI, AI represents, for sure, the biggest opportunity since cloud and probably the biggest technology shift and opportunity in business since the Internet.

Amazon’s management has been impressed with DeepSeek’s innovations

I think like many others, we were impressed with what DeepSeek has done, I think in part impressed with some of the training techniques, primarily in flipping the sequencing of reinforcement learning being earlier and without the human-in-the-loop. We thought that was interesting ahead of the supervised fine-tuning. We also thought some of the inference optimizations they did were also quite interesting

Amazon’s management’s core belief remains that generative AI apps will use multiple models and different customers will use different AI models for different workloads 

You have a core belief like we do that virtually all the big generative AI apps are going to use multiple model types, and different customers are going to use different models for different types of workloads.

Amazon’s management thinks that the cheaper AI inference becomes, the more inference spending there will be; management believes that the cost of AI inference will fall substantially over time

Sometimes people make the assumptions that if you’re able to decrease the cost of any type of technology component, in this case, we’re really talking about inference, that somehow it’s going to lead to less total spend in technology. And we have never seen that to be the case. We did the same thing in the cloud where we launched AWS in 2006, where we offered S3 object storage for $0.15 a gigabyte and compute for $0.10 an hour, which, of course, is much lower now many years later, people thought that people would spend a lot less money on infrastructure technology. And what happens is companies will spend a lot less per unit of infrastructure, and that is very, very useful for their businesses, but then they get excited about what else they could build that they always thought was cost prohibitive before, and they usually end up spending a lot more in total on technology once you make the per unit cost less. And I think that is very much what’s going to happen here in AI, which is the cost of inference will substantially come down. What you heard in the last couple of weeks, DeepSeek is a piece of it, but everybody is working on this. I believe the cost of inference will meaningfully come down. I think it will make it much easier for companies to be able to infuse all their applications with inference and with generative AI.

Amazon’s management currently sees 2 main ways that companies are getting value out of AI; the 1st way is through productivity and cost savings, and it is the lowest-hanging fruit; the 2nd way is by building new experiences

There’s kind of two macro buckets of how we see people, both ourselves inside Amazon as well as other companies using AWS, how we see them getting value out of AI today. The first macro bucket, I would say, is really around productivity and cost savings. And in many ways, this is the lowest-hanging fruit in AI…

….I’d say the other big macro bucket are really altogether new experiences.

Amazon has built a chatbot with generative AI and it has lifted customer satisfaction by 500 basis points; Amazon has built a generative AI application for 3rd-party sellers to easily fill up their product detail pages; Amazon has built generative AI applications for inventory management that improve inventory forecasting by 10% and regional predictions by 20%; the brains of Amazon’s robotics are infused with generative AI

If you look at customer service and you look at the chatbot that we’ve built, we completely rearchitected it with generative AI. It’s delivering. It already had pretty high satisfaction. It’s delivering 500 basis points better satisfaction from customers with the new generative AI-infused chatbot.

If you look at our millions of third-party selling partners, one of their biggest pain points is, because we put a high premium on really organizing our marketplace so that it’s easy to find things, there’s a bunch of different fields you have to fill out when you’re creating a new product detail page, but we’ve built a generative AI application for them where they can either fill in just a couple of lines of text or take a picture of an image or point to a URL, and the generative AI app will fill in most of the rest of the information they have to fill out, which speeds up getting selection on the website and easier for sellers.

If you look at how we do inventory management and trying to understand what inventory we need, at what facility, at what time, the generative AI applications we’ve built there have led to 10% better forecasting on our part and 20% better regional predictions.

In our robotics, we were just talking about the brains in a lot of those robotics are generative AI-infused that do things like tell the robotic claw what’s in a bin, what it should pick up, how it should move it, where it should place it in the other bin that it’s sitting next to. So it’s really in the brains of most of our robotics.

Amazon’s Rufus is an AI-infused shopping assistant that is growing significantly; users can take a picture of a product with Amazon Lens and have the service surface the exact item through the use of AI; Amazon is using AI to know the relative sizing of clothes and shoes from different brands so that it can recommend the right sizes to shoppers; Amazon is using AI to improve the viewing experience of sporting events; Rufus provides a significant improvement to the shopping experience for shoppers and management expects the usage of Rufus to increase throughout 2025

You see lots of those in our retail business, ranging from Rufus, which is our AI-infused shopping assistant, which continues to grow very significantly; to things like Amazon Lens, where you can take a picture of a product that’s in front of you, you check it out in the app, you can find it in the little box at the top, you take a picture of an item in front of you, and it uses computer vision and generative AI to pull up the exact item in search result; to things like sizing, where we basically have taken the catalogs of all these different clothing manufacturers and then compare them against one another so we know which brands tend to run big or small relative to each other. So when you come to buy a pair of shoes, for instance, it can recommend what size you need; to even what we’re doing in Thursday Night Football, where we’re using generative AI for really inventive features like it sends alerts where we predict which players are going to put quarterback or defensive vulnerabilities, where we were able to show viewers what area of the field is vulnerable…

…I do think that Rufus, if you look at how it impacts the customer experience and if you actually use it month-to-month, it continues to get better and better. If you’re buying something and you’re on our product detail page, our product detail pages provide so much information that sometimes it’s hard, if you’re trying to find something quickly, to scroll through and find that little piece of information. And so we have so many customers now who just use Rufus to help them find a quick fact about a product. They also use Rufus to figure out how to summarize customer reviews so they don’t have to read 100 customer reviews to get a sense of what people think about that product. If you look at the personalization, really, most prominently today, your ability to go into Rufus and ask what’s happened to an order or what did I just order or can you pull up for me this item that I ordered 2 months ago, the personalization keeps getting much better. And so we expect throughout 2025, that the number of occasions where you’re not sure what you want to buy and you want help from Rufus are going to continue to increase and be more and more helpful to customers.

Amazon has around 1,000 generative AI applications that it has built or is building

We’ve got about 1,000 different generative AI applications we’ve either built or in the process of building right now.

Apple (NASDAQ: AAPL)

Apple Intelligence was first released in the USA in October 2024, with more features and countries introduced in December 2024; Apple Intelligence will be rolled out to even more countries in April 2025; management sees Apple Intelligence as a breakthrough for privacy in AI; SAP is using Apple Intelligence in the USA to improve the employee as well as customer experience; the Apple Intelligence features that people are using include Writing Tools, Image Playground, Genmoji, Visual Intelligence, Clean Up, and more; management has found Apple Intelligence’s email summarisation feature to be very useful; management thinks that different users will find their own “killer feature” within Apple Intelligence

In October, we released the first set of Apple Intelligence features in U.S. English for iPhone, iPad and Mac, and we rolled out more features and expanded to more countries in December.

Now users can discover the benefits of these new features in the things they do every day. They can use Writing Tools to help find just the right words, create fun and unique images with Image Playground and Genmoji, handle daily tasks and seek out information with a more natural and conversational Siri, create movies of their memories with a simple prompt and touch up their photos with Clean Up. We introduced visual intelligence with Camera Control to help users instantly learn about their surroundings. Users can also seamlessly access ChatGPT across iOS, iPadOS and macOS.

And we were excited to recently begin our international expansion with Apple Intelligence now available in Australia, Canada, New Zealand, South Africa and the U.K. We’re working hard to take Apple Intelligence even further. In April, we’re bringing Apple Intelligence to more languages, including French, German, Italian, Portuguese, Spanish, Japanese, Korean and simplified Chinese as well as localized English to Singapore and India. And we’ll continue to roll out more features in the future, including an even more capable Siri.

Apple Intelligence builds on years of innovations we’ve made across hardware and software to transform how users experience our products. Apple Intelligence also empowers users by delivering personal context that’s relevant to them. And importantly, Apple Intelligence is a breakthrough for privacy in AI with innovations like Private Cloud Compute, which extends the industry-leading security and privacy of Apple devices into the cloud…

…We’re excited to see leading enterprises such as SAP leverage Apple Intelligence in the U.S. with features like Writing Tools, summarize and priority notifications to enhance both their employee and customer experiences…

…In terms of the features that people are using, they’re using all of the ones that I had referenced in my opening comments, from Writing Tools to Image Playground and Genmoji, to visual intelligence and more. And so we see all of those being used. Clean Up is another one that is popular, and people love seeing that one demoed in the stores as well…

…I know from my own personal experience, once you start using the features, you can’t imagine not using them anymore. I know I get hundreds of e-mails a day, and the summarization function is so important…

…[Question] Do you guys see the upgraded Siri expected in April as something that will, let’s say, be the killer application among the suite of features that you have announced in Apple Intelligence?

[Answer] I think the killer feature is different for different people. But I think for most, they’re going to find that they’re going to use many of the features every day. And certainly, one of those is the — is Siri, and that will be coming over the next several months.

Many customers are excited about the iPhone 16 because of Apple Intelligence; the iPhone 16’s year-on-year performance was stronger in countries where Apple Intelligence was available compared to countries where Apple Intelligence was not available

Our iPhone 16 lineup takes the smartphone experience to the next level in so many ways, and Apple Intelligence is one of many reasons why customers are excited…

…We did see that the markets where we had rolled out Apple Intelligence, that the year-over-year performance on the iPhone 16 family was stronger than those where Apple Intelligence was not available…

Apple’s management thinks the developments in the AI industry brought on by DeepSeek’s emergence is a positive for Apple

[Question] There’s a perception that you’re a big beneficiary of lower cost of compute. And I was wondering if you could give your worldly perspective here on the DeepSeek situation.

[Answer] In general, I think innovation that drives efficiency is a good thing. And that’s what you see in that model. Our tight integration of silicon and software, I think, will continue to serve us very well.

Arista Networks (NYSE: ANET)

Cloud and AI titans were a significant contributor to Arista Networks’ revenue in 2024; management considers Oracle an AI titan too

Now shifting to annual sector revenue for 2024. Our cloud and AI titans contributed significantly at approximately 48%, keeping in mind that Oracle is a new member of this category.

Arista Networks’ core cloud AI and data center products are built off its extensible OS (operating system) and goes up to 800 gigabit Ethernet speeds

Our core cloud AI and data center products are built off a highly differentiated, extensible OS stack and is successfully deployed across 10, 25, 100, 200, 400 and 800 gigabit Ethernet speeds. It delivers power efficiency, high availability, automation and agility as the data centers demand, insatiable bandwidth capacity and network speeds for both front-end and back-end storage, compute and AI zones.

Arista Networks’ management expects the company’s 800 gigabit Ethernet switch to emerge as an AI back-end cluster in 2025

We expect 800 gigabit Ethernet to emerge as an AI back-end cluster in 2025.

Arista Networks’ management is still optimistic that AI revenues will reach $1.5 billion in 2025, including $750 million in AI back-end clusters; the $750 million in revenue from AI back-end clusters will have a major helping hand from 3 of the 5 major AI trials Arista Networks is working on that are rolling out a cumulative 100,000 GPUs in 2025 (see more below)

We remain optimistic about achieving our AI revenue goal of $1.5 billion in AI centers, which includes the $750 million in AI back-end clusters in 2025…

…[Question] You are reiterating $750 million AI back-end sales this year despite the stalled or the fifth customer. Can you talk about where is the upside coming from this year? Is it broad-based or 1 or 2 customers?

[Answer] We’re well on our way and 3 customers deploying a cumulative of 100,000 GPUs is going to help us with that number this year. And as we increased our guidance to $8.2 billion, I think we’re going to see momentum both in AI, cloud and enterprises. I’m not ready to break it down and tell you which where. I think we’ll see — we’ll know that much better in the second half. But Chantelle and I feel confident that we can definitely do the $8.2 billion that we historically don’t call out so early in the year. So having visibility if that helps.

Arista Networks is building some of the world’s greatest Arista AI centers at production scale and it’s involved with both the back-end clusters and front-end networks; Arista Networks’ management sees the data traffic flow of AI workloads as having significant differences from traditional cloud workloads and Arista AI centers can seamlessly connect to the front end compute storage with its backend Ethernet portfolio; Arista’s AI networking portfolio consists of 3 families and over 20 Etherlink switches

Networking for AI is also gaining traction as we move into 2025, building some of the world’s greatest Arista AI centers at production scale. These are constructed with both back-end clusters and front-end networks…

…The fidelity of the AI traffic differs greatly from cloud workloads in terms of diversity, duration and size of flow. Just one slow flow can flow the entire job completion time for a training workload. Therefore, Arista AI centers seamlessly connect to the front end of compute storage WAN and classic cloud networks with our back-end Arista Etherlink portfolio. This AI accelerated networking portfolio consists of 3 families and over 20 Etherlink switches, not just 1 point switch.

Arista Networks’ management’s AI for Networking strategy is doing well and it includes software that have superior AI ops

Our AI for networking strategy is also doing well, and it’s about curating the data for higher-level network functions. We instrument our customer’s networks with our published subscribed state Foundation with our software called Network Data Lake to deliver proactive, predictive and prescriptive platforms that have superior AI ops with A care support and product functions.

Arista Networks’ management is still committed to 4 of the 5 major AI trials that they have been discussing in recent earnings calls; the remaining AI trial is still stalled and the customer is not a Cloud Titan and is waiting for funding; 3 of the 4 trials that are active are expected to roll out a cumulative 100,000 GPUs in 2025 and they are all waiting for the next-generation NVIDIA GPU; Arista Networks’ management expects to do very well on the back-end with those 3 trials; the remaining trial of the 4 active trials is migrating from Infiniband to Ethernet to test the viability of Ethernet, and Arista Networks’ management expects to enter production in 2026

I want to say Arista is still committed to 4 out of our 5 AI clusters that I mentioned in prior calls, but just one is a little bit stalled. It is not a Cloud Titan. They are awaiting GPUs and some funding too, I think. So I hope they’ll come back next year, but for this year, we won’t talk about them. But the remaining 4, let me spend some — jgive you some color, 3 out of the 4 customers are expected to this year rolled out a cumulative of 100,000 GPUs. So we’re going to do very well with 3 of them on the back end. And you can imagine, they’re all pretty much one major NVIDIA class of GPU — it’s — they will be waiting for the next generation of GPUs. But independent of that, we’ll be rolling out fairly large numbers. On the fourth one, we are migrating right now from InfiniBand to proving that Ethernet is a viable solution, so we’re still — they’ve historically been InfiniBand. And so we’re still in pilot and we expect to go into production next year. We’re doing very well in 4 out of 4, the Fifth one installed and 3 out of the 4 expected to be 100,000 GPUs this year.

Arista Networks thinks the market for AI networking is large enough that there will be enough room for both the company and other whitebox networking manufacturers; management also thinks Arista Networks’ products have significant differentiation from whitebox products, especially in the AI spine in a typical leaf-spine network, because Arista Networks’ products can automatically provide an alternate connection when a GPU in the network is in trouble

[Question] Can you maybe share your perspective that when it comes to AI network especially the back-end networks, how do you see the mix evolving white box versus OEM solution?

[Answer] This TAM is so huge and so large. We will always coexist with white boxes and operating systems that are non-EOS, much like Apple coexists on the iPhone with other phones of different types. When you look at the back end of an AI cluster, there are typically 2 components, the AI lead and the AI spine. The AI lead connects to the GPUs and therefore, is the first, if you will, point of connection. And the AI spine aggregates all of these AI leads. Almost in all the back-end examples we’ve seen, the AI spine is generally 100% Arista-branded EOS. You’ve got to do an awful lot of routing, scale, features, capabilities that are very rich that would be difficult to do in any other environment. The AI leads can vary. So for example, the — let’s take the example of the 5 customers I mentioned a lot, 3 out of the 5 are all EOS in the [indiscernible] spine. 2 out of the 5 are kind of hybrids. Some of them have some form of SONic or FBOSS. And as you know, we co-develop with them and coexist in a number of use cases where it’s a real hybrid combination of EOS and an open OS. So for most part, I’d just like to say that white box and Arista will coexist and will provide different strokes for different folks…

…A lot of our deployments right now is 400 and 800 gig, and you see a tremendous amount of differentiation, not only like I explained to you in scale and routing features, but cost and load balancing, AI visibility and analytics at real time, personal queuing, congestion control, visibility and most importantly, smart system upgrade because you sort of want your GPUs to come down because you don’t have the right software to accelerate so that the network provides the ideal foundation that if the GPU is in trouble, we can automatically give a different connection and an alternate connection. So tremendous amount of differentiation there and even more valid in a GPU which costs typically 5x as much as a CPU…

…When you’re buying these expensive GPUs that cost $25,000, they’re like diamonds, right? You’re not going to string a diamond on a piece of thread. So first thing I want to say is you need a mission-critical network, whether you want to call it white box, blue box, EOS or some other software, you’ve got to have mission-critical functions, analytics, visibility, high availability, et cetera. As I mentioned, and I want to reiterate, they’re also typically a leaf spine network. And I have yet to see an AI spine deployment that is not EOS-based. I’m not saying it can’t happen or won’t happen. But in all 5 major installations, the benefit of our EOS features for high availability for routing, for VXLAN, for telemetry, our customers really see that. And the 7800 is the flagship AI spine product that we have been deploying last year, this year and in the future. Coming soon, of course, is also the product we jointly engineered with Meta, which is the distributed [Ecolink] switch. And that is also an example of a product that provides that kind of leaf spine combination, both with FBOSS and EOS options in it. So in my view, it’s difficult to imagine a highly resilient system without Arista EOS in AI or non-AI use cases.

On the leaf, you can cut corners. You can go with smaller buffers, you may have a smaller installation. So I can imagine that some people will want to experiment and do experiment in smaller configurations with non-EOS. But again, to do that, you have to have a fairly large staff to build the operations for it. So that’s also a critical element. So unless you’re a large Cloud Titan customer, you’re less likely to take that chance because you don’t have the staff.

Arista Networks’ management is seeing strong demand from its Cloud Titan customers

Speaking specifically to Meta, we are obviously in a number of use cases in Meta. Keep in mind that our 2024 Meta numbers is influenced by more of their 2023 CapEx, and that was Meta’s year of efficiency where their CapEx was down 15% to 20%. So you’re probably seeing some correlation between their CapEx being down and our revenue numbers being slightly lower in ’24. In general, I would just say all our cloud titans are performing well in demand, and we shouldn’t confuse that with timing of our shipments. And I fully expect Microsoft and Meta to be greater than 10% customers in a strong manner in 2025 as well. Specific to the others we added in, they’re not 10% customers, but they’re doing very well, and we’re happy with their cloud and AI use cases.

Arista Networks’ management thinks the emergence of DeepSeek will lead to AI development evolving from back-end training that’s concentrated in a handful of users, to being distributed more widely across CPUs and GPUs; management also thinks DeepSeek’s emergence is a positive for Arista Networks because DeepSeek’s innovations can drive the AI industry towards a new class of CPUs, GPUs, AI accelerators and Arista Networks is able to scale up network for all kinds of XPUs

DeepSeek certainly deep fixed many stocks, but I actually see this as a positive because I think you’re now going to see a new class of CPUs, GPUs, AI accelerators and where you can have substantial efficiency gains that go beyond training. So that could be some sort of inference or mixture of experts or reasoning and which lowers the token count and therefore, the cost. So what I like about all these different options is Arista can scale up network for all kinds of XPUs and accelerators. And I think the eye-opening thing here for all of our experts who are building all these engineering models is there are many different types and training isn’t the only one. So I think this is a nice evolution of how AI will not just be a back-end training only limited to 5 customers type phenomenon, but will become more and more distributed across a range of CPUs and GPUs.

Arista Networks’ management thinks hyper-scale GPU clusters, such as Project Stargate, will drive the development of vertical rack integration in the next few years and Andy Bechtolsheim, an Arista Networks co-founder, is personally involved in these projects

If you look at how we have classically approached GPUs and connected libraries, we’ve largely looked at it as 2 separate building blocks. There’s the vendor who provides the GPUs and then there’s us who provides the scale-out networking. But when you look at Stargate and projects like this, I think you’ll start to see more of a vertical rack integration where the processor, the scale up, the scale out and all of the software to provide a single point of control and visibility starts to come more and more together. This is not a 2025 phenomenon, but definitely in ’26 and ’27, you’re going to see a new class of AI accelerators for — and a new class of training and inference, which is extremely different than the current more pluggable label type of version. So we’re very optimistic about it.

Andy Bechtolsheim is personally involved in the design of a number of these next-generation projects and the need for this type of shall we say, pushing Moore’s Law of improvements in density of performance that we saw in the 2000s is coming back, and you can boost more and more performance per XPU, which means you have to boost the network scale from 800 gig to 1.16.

Arista Networks’ management sees a $70 billion total addressable market in 2028, of which roughly a third is related to AI

[Question] If you can talk to the $70 billion TAM number for 2028, how much is AI?

[Answer] On the $70 billion TAM in 2028, I would roughly say 1/3 is AI, 1/3 is data center and cloud and 1/3 is campus and enterprise. And obviously, absorbed into that is routing and security and observability. I’m not calling them out separately for the purpose of this discussion.

Arista Networks’ management sees co-packaged optics (CPO) as having weak adoption compared to co-packaged copper (CPC) because CPO has been experiencing field failures

Co-packaged optics is not a new idea. It’s been around 10 to 20 years. So the fundamental reason, let’s go through why co-packaged optics has had a relatively weak adoption so far is because of field failures and most of it is still in proof of concept today. So going back to networking, the most important attribute of a network switch is reliability and troubleshooting. And once you solder a co-packaged optics on a PCB, you lose some of that flexibility and you don’t get the serviceability and manufacturing. That’s been the problem. Now a number of alternatives are emerging, and we’re a big fan of co-packaged copper as well as pluggable optics that can complement this like linear drive or LTO as we call it.

Now we also see that if co-packaged optics improves some of the metrics it has right now. For example, it has a higher channel count than the industry standard of 8-channel pluggable optics, but we can do higher channel pluggable optics as well. So some of these things improve, we can see that both CPC and CPO will be important technologies at 224 gig or even 448 gig. But so far, our customers have preferred a LEGO approach that they can mix and match pluggable switches and pluggable optics and haven’t committed to soldering them on the PCB. And we feel that will change only if CPO gets better and more reliable. And I think CPC can be a nice alternative to that.

Arista Networks’ management is seeing customers start moving towards actual use-cases for AI, but the customers are saying that these AI projects take time to implement

For the AI perspective, speaking with the customers, it’s great to move from kind of a theory to more specific conversation, and you’re seeing that in the banks and some of the higher tier Global 2000, Fortune 500 companies. And so they’re moving from theory to actual use cases they’re speaking to. And the way they describe it is it takes a bit of time. They’re working mostly with cloud service providers at the beginning, kind of doing some training and then they’re deciding whether they bring that on-prem and inference. So they’re making those decisions.

Arista Networks’ management is seeing a new class of Tier 2 specialty AI cloud providers emerge

We are seeing a new class of Tier 2 specialty cloud providers emerge that want to provide AI as a service and want to be differentiated there. And there’s a whole lot of funding, grant money, real money going in there. So service providers, too early to call. But Neo clouds and specialty providers, yes, we’re seeing lots of examples of that.

The advent of AI has accelerated the speed-transitions in networking data switches, but there’s still going to be a long runway for Arista Networks’ 400 gig and 800 gig products, with 1.6 tera products being deployed in a measured way

The speed transitions because of AI are certainly getting faster. It used to take when we went from 200 gig, for example, at Meta or 100 gig in some of our Cloud Titans to 400, that speed transition typically took 3 to 4, maybe even 5 years, right? In AI, we see that cycle being almost every 2 years…

…2024 was the year of real 400 gig. ’25 and ’26, I would say, is more 800 gig. And I really see 1.6T coming into the picture because we don’t have chips yet, maybe in what do you say, John, late ’26 and real production maybe in ’27. So there’s a lot of talk and hype on it, just like I remember talk and hype on 400 gig 5 years ago. But I think realistically, you’re going to see a long runway for 400 and 800 gig. Now as we get into 1.6T, part of the reason I think it’s going to be measured and thoughtful is many of our customers are still awaiting their own AI accelerators or NVIDIA GPUs, which with liquid cooling that would actually push that kind of bandwidth. So new GPUs will require new bandwidth, and that’s going to push it out a year or 2.

Arista Networks’ management sees a future where the market share between NVIDIA GPUs and custom AI accelerators (ASICs) is roughly evenly-split, but Arista Networks’ products will be GPU-agnostic

[Question] There’s been a lot of discussion over the last few months between the general purpose GPU clusters from NVIDIA and then the custom ASIC solutions from some of your popular customers. I guess just in your view, over the longer term, does Arista’s opportunity differ across these 2 chip types?

[Answer] I think I’ve always said this, you guys often spoke about NVIDIA as a competitor. And I don’t see it that way. I see that — thank you, NVIDIA. Thank you, Jensen, for the GPUs because that gives us an opportunity to connect to them, and that’s been a predominant market for us. As we move forward, we see not only that we connect to them, but we can connect to AMD GPUs and built in in-house AI accelerators. So a lot of them are in active development or in early stages. NVIDIA is the dominant market share holder with probably 80%, 90%. But if you ask me to guess what it would look like 2, 3 years from now, I think it could be 50-50. So Arista could be the scale-out network for all types of accelerators. We’ll be GPU agnostic. And I think there’ll be less opportunity to bundle by specific vendors and more opportunity for customers to choose best-of-breed. 

ASML (NASDAQ: ASML)

AI will be the biggest driver of ASML’s growth and management sees customers benefiting very strongly from it; management thinks ASML will hit the upper end of the revenue guidance range for 2025 if its customers can bring on additional AI-related capacity during the year, but there are also risks that could result in only the lower end of the guidance coming true

We see total revenue for 2025 between €30 billion and €35 billion and the gross margin between 51% and 53%. AI is the clear driver. I think we started to see that last year. In fact, at this point, we really believe that AI is creating a shift in the market and we have seen customers benefiting from it very strongly…

…If AI demand continues to be strong and customers are successful in bringing on additional capacity online to support that demand, there is potential opportunity towards the upper end of our range. On the other hand, there are also risks related to customers and geopolitics that could drive results towards the lower end of the range.

ASML’s management is still very positive on the long-term outlook for ASML, with AI being a driver for growth; management expects AI to create a shift in ASML’s end-market products towards more HPC (high performance computing) and HBM (high bandwidth memory), which requires more advanced logic and DRAM, which in turn needs more critical lithography exposures

I think our view on the long term is still, I would say, very positive…

…Looking longer term, overall the semiconductor market remains strong with artificial intelligence creating growth but also a shift in market dynamics as I highlighted earlier. These dynamics will lead to a shift in the mix of end market products towards more HPC and HBM which requires more advanced logic and DRAM. For ASML, we anticipate that an increased number of critical lithography exposures for these advanced Logic and Memory processes will drive increasing demand for ASML products and services. As a result, we see a 2030 revenue opportunity between 44 billion euros and 60 billion euros with gross margins expected between 56 percent and 60 percent, as we presented in Investor Day 2024.

ASML’s management is seeing aggressive capacity addition among some DRAM memory customers because of demand for high bandwidth memory (HBM), but apart from HBM, other DRAM memory customers have a slower recovery

 I think that high-bandwidth memory is driving today, I would say, also an aggressive capacity addition, at least for some of the customer. I think on the normal DRAM, I would say, my comment is similar to the one on mobile [ photology ] before. I think there are also nothing spectacular, but there is some recovery, which also called for more capacity. So that’s why we still see DRAM pretty strong in 2025.

Datadog (NASDAQ: DDOG)

Datadog launched LLM Observability in 2024; management continues to see increased interest in next-gen AI; 3,500 Datadog customers at the end of 2024 Q4 used 1 or more Datadog AI integrations; when it comes to AI inference, management is seeing most customers using a 3rd-party AI model through an API or a 3rd-party inference platform, and these customers want to observe whether the model is doing the right thing, and this need is what LLM Observability is serving; management is seeing very few customers running the full AI inference stack currently, but they think this could happen soon and it would be an exciting development

We launched LLM observability, in general availability to help customers evaluate, safely deploy and manage their models in production, and we continue to see increased interest in next-gen AI. At the end of Q4, about 3,500 customers use 1 or more Datadog AI integrations to send this data about their machine learning, AI, and LLM usage…

…On the inference side, the — mostly still what customers do is they use a third-party model either through an API or through a third-party inference platform. And what they’re interested in is measuring whether that model is doing the right thing. And that’s what we serve right now with LLM observability, for example, as well, we see quite a bit of adoption that does not come largely from the AI-native companies. So that’s what we see today.

In terms of operating the inference stack fully and how we see relatively few customers with that yet, we think that’s something that’s going to come next. And by the way, we’re very excited by the developments we see in the space. So it looks like there is many, many different options that are going to be viable for running your AI inference. There’s a very healthy set of commercial API-gated services. There’s models that you can install in the open source. There are models in the open source today that are rivalling in quality with the best closed API models. So we think the ecosystem is developing into a rich diversification that will allow customers to have a diversity of modalities for using AI, which is exciting. 

AI-native customers accounted for 6% of Datadog’s ARR in 2024 Q4 (was 6% 2024 Q3); AI-native customers contributed 5 percentage points to Datadog’s year-on-year growth in 2024 Q4, compared to 3 percentage points in 2023 Q4; among customers in the AI-native cohort, management has seen optimisation of usage and volume discounts related to contracts in 2024 Q4, and management thinks these customers will continue to optimise cloud and observability usage in the future; the dynamics with the AI-native cohort that happened in 2024 Q4 was inline with management’s expectations

We continue to see robust contribution from AI native customers who represented about 6% of Q4 ARR roughly the same as the quarter — as last quarter and up from about 3% of ARR in the year-ago quarter. AI native customers contributed about 5 percentage points of year-over-year revenue growth in Q4 versus 4 points in the last quarter and about 3 points in the year-ago quarter. So we saw strong growth from AI native customers in Q4. We believe that adoption of AI will continue to benefit Datadog in the long term. Meanwhile, we did see some optimization and volume discounts related to contract renewals in Q4. We remain mindful that we may see volatility in our revenue growth on the backdrop of long-term volume growth from this cohort as customers renew with us on different terms, and as they may choose to optimize cloud and observability usage. ..

… [Question] I’m trying to understand if the AI usage and commits are kind of on the same trajectory that they were on or whether you feel that there are some oscillations there.

[Answer] What happened during the quarter is pretty much what we thought would happen when we discussed it in the last earnings call. When you look at the AI cohort, we definitely saw some renewals with higher commit, better terms and optimization usage all at the same time, which is fairly typical, which typically happens with larger end customers in particular is at the time of renewal, customers are going to trying and optimize what they can. They’re going to get better prices from us, up their commitments and we might see a flat or down a month or quarter after that, with a still sharp growth from the year before and growth to come in the year to come. So that’s what we typically see. When you look at the cohort as a whole, even with that significant renewal optimization and better unit economics this quarter is wholly stable, this quarter as a whole is stable quarter-to-quarter in its revenue and it’s growing a lot from the quarter before, even with all that.

Datadog’s management sees some emerging opportunities in Infrastructure Monitoring that are related to the usage of GPUs (Graphics Processing Units), but the opportunities will only make sense if there is broad usage of GPUs by a large number of customers, which is not happening today

There’s a number of new use cases that are emerging that are related to infrastructure that we might want to cover. Again, we — when I say they’re emerging, they’re actually emerging, like we still have to see what the actual need is from a large number of customers. I’m talking in particular about infrastructure concerns around GPU management, GPU optimization, like there’s quite a lot going on there that we can potentially do. But we — for that, we need to see broad usage of the raw GPUs by a large number of customers as opposed to usage by a smaller number of native customers, which is mostly what we still see today.

Datadog’s management thinks it’s hard to tell where AI agents can be the most beneficial for Datadog’s observability platform because it’s still a very nascent field and management has observed that things change really quickly; when management built LLM Observability, the initial use cases were for AI chatbots and RAG (retrieval augmented generation), but now the use cases are starting to shift towards AI agents

[Question] Just curious, when we think about agents, which parts of the core observability platform that you think are most relevant or going to be most beneficial to your business as you start to monitor those?

[Answer] It’s a bit hard to tell because it’s a very nascent field. So my guess is in a year if we probably look different from what it looks like today. Just like this year, it looks very different from what it looks like last year. What we do see, though, is that — so when we built — we started building our LLM Observability product, most of the use cases we saw there from customers were chatbot in nature or RAG in nature, trying to access information and return the information. Now we see more and more customers building agents on top of that and sending the data from their agents. So we definitely see a growing trend there of adoption and the LLM Observability product is a good level of abstraction, at least for the current iteration of these agents to get them. So that’s what we can see today.

Datadog’s management sees AI touching many different areas of Datadog, such as how software is being written and deployed, how customer-support is improved, and more

What’s fascinating about the current evolution of AI, in particular, is that it touches a lot of the different areas of the business. The first area for company like ours the first area to be transformed is really the way software is being built. What engineers use, how they write software, how they debug software, how do they also operate systems. And part of that is outside tooling we’re using for writing software. Part of that is dogfooding, or new products for incident resolution and that sort of thing. So that’s the first area. There’s a number of other areas that are going to see large improvements in productivity. Typically, everything that has to do with supporting customers, helping with onboarding and helping troubleshoot issues like all of that is in acceleration. In the end, we expect to see improvements everywhere, from front office to back office.

Fiverr (NYSE: FVRR)

Fiverr’s management launched Fiverr Go in February 2025, an open platform for personalised AI tools designed to give creators full control over their creative processes and rights; Fiverr Go enables freelancers to build personalised AI models (there was a presentation on this recently) without having to know AI engineering; Fiverr Go is designed to be personalised for the creator, so the creator becomes more important compared to the AI technology; Fiverr Go is generative AI technology with human accountability (will be interesting to see if Fiverr Go is popular; people can create designs/images with other AI models, so customers who use Fiverr Go are those who need the special features that Fiverr Go offers); Fiverr Go generates content that is good enough for mission critical business tasks, unlike what’s commonly happening with other AI-generated content; Fiverr Go is no different from a direct order from the freelancer themself, except it is faster and easier for buyers; Fiverr Go has personalised AI assistants for freelancers; Fiverr Go has an open developer platform for 3rd-party developers to build generative AI apps

Fiverr Go is an open platform for personalized AI tools that include the personalized AI assistant and the AI creation model. Different from other AI platforms that often exploit human creativity without proper attribution or compensation, Fiverr Go is uniquely designed to reshape this power dynamic by giving creators full control over their creative process and rights. It also enables freelancers to build personalized AI models without the need to collect training data sets or understand AI engineering, thanks to Fiverr’s unparalleled foundation of over 6.5 billion interactions and nearly 150 million transactions on the marketplace and most importantly, it allows freelancers to become a one-person production house, making more money while focusing on the things that matter. By giving freelancers control over configuration, pricing and creative rights and leveling the playing field of implementing AI technology, Fiverr Go ensures that creators remain at the center of the creative economy. It decisively turned the power dynamic between humans and AI towards the human side…

…For customers, Fiverr Go is also fundamentally different from other AI platforms. It is GenAI with human accountability. AI results often feel unreliable, generic and very hard to edit. What is good enough for a simple question and answer on ChatGPT does not cut it for business mission-critical tasks. In fact, many customers come to Fiverr today with AI-generated content because they miss the confidence that comes from another human eye and talent, helping them perfect the results for their needs. Fiverr Go eliminates all of this friction and frustration. Every delivery on Fiverr Go is backed by the full faith of the creator behind it with an included revision as the freelancer defines. This means that the quality and service you get from Fiverr Go is no different from a direct order from the freelancers themselves, except for a faster, easier and more delightful experience. The personalized AI assistant on Fiverr Go can communicate with potential clients when the freelancer is away or busy, handle routine tasks and provide actionable business insights, effectively becoming the creator’s business partner. It often feels smarter than an average human assistant because it’s equipped with all the history of how the freelancer works as well as knowledge of trends and best practices on the Fiverr marketplace…

…We’ve also announced an open developer platform on Fiverr Go to allow AI specialists and model developers to build generative AI applications across any discipline. We provide them with an ecosystem to collaborate with domain experts on Fiverr and the ability to leverage Fiverr’s massive data assets so that these applications can solve real-world problems and most important of all, Fiverr provides them an avenue to generate revenue from those applications through our marketplace…

…So from our experience with AI, what we come to learn is that a lot of the creation process using AI is very random and take you through figuring out what are the best tools because there’s thousands of different options around AI. And each one operates slightly differently. And you need to master each one of them. And you need to become a prompt engineer. And then editing is extremely, extremely hard. Plus you don’t get the feedback that comes from working with a human being that can actually look at the creation from a human eye and give you a sense if this is actually capturing what you’re trying to do. It allows us or allows freelancers to design their own model in a way that rewards them but remains extremely accurate to their style, allowing customers to get the results they expect to get because they see the portfolio of their freelancer, like the style of writing or design or singing or narration, and they can get exactly this. So we think that, that combination and that confidence that comes from the fact that the creator itself is always there.

The AI personal assistant in Fiverr Go can help to respond to customer questions based on individual freelancers; the first 3 minutes after a buyer writes to a freelancer is the most crucial time for conversion and this is where the AI assistant can help; there are already thousands of AI assistants running on Fiverr Go, converting customers

Fiverr Go is actually a tool for conversion. That’s the entire idea because we know that customers these days expect instant responses and instant results. And as a result of that, we designed those 2 tools, the AI personal assistant, which is able to answer customer questions immediately even if the freelancer is away or busy. We know that the first 3 minutes after a customer writes to a freelancer are the most crucial time for conversion and this is why we designed this tool. And this tool is essentially encapsulating the entire knowledge of the freelancer and basing itself on it, being able to address any possible question and bring it to conversion…

…It’s fresh from yesterday, but we have many thousands of assistants running on the system, converting customers already, which is an amazing sign.

Fiverr Go is a creator tool that can create works based off freelancers’ style and allows customers to get highly-accurate samples of a freelancers’ work to lower friction in selecting freelancers

When we think about the creation model, the creation model allows customers to get the confidence that this is the freelancer, this is the style that they’re looking for, because now instead of asking a freelancer for samples, waiting for it, causing the freelancer to essentially work for free, they can get those samples right away. Now the quality of these samples is just mind-blowing. The level of accuracy that these samples produce are exact match with the style of the freelancer, which gives the customer the confidence that if they played and liked it, this is the type of freelancer that they should engage with.

The Fiverr Go open developer platform is essentially an app store for AI apps; the open developer platform allows developers to train AI models on Fiverr’s transactional data set, which is probably the largest dataset of its kind in existence

Now what we’re doing with this is actually we’re opening up the Go platform to outside developers. Think about it as an app store in essence. So what we’re doing is we’re allowing them to develop models, APIs, workflows, but then train those models on probably the biggest transactional data set in existence today that we hold so that they can actually help us build models that freelancers can join — can enjoy from. And we believe that by doing so and giving those developers incentives to do so because every time their app is going to be used for a transaction, they’re going to make money out of it.

Fiverr Go’s take rate will be the same for now and management will learn as they go

[Question] Would your take rate be different in Fiverr Go?

[Answer] For now, the take rate remains the same for Go. And as we roll it out and as we see usage, we will figure out what to do or what’s the right thing to do. For now, we treat it as a normal transaction with the same take rate.

Mastercard (NYSE: MA)

Mastercard closed the Recorded Future acquisition in 2024 Q4 (Recorded Future provides AI-powered solutions for real-time visibility into potential threats related to fraud); Recorded Future has been deploying AI for over a decade, just like Mastercard has; Recorded Future uses AI to analyse threat data across the entire Internet; the acquisition of Recorded Future improves Mastercard’s cybersecurity solutions

Our diverse capabilities in payments and services and solutions, including the acquisition of Recorded Future this quarter set us apart…

…Recorded Future is the world’s largest threat intelligence company with more than 1,900 customers across 75 countries. Customers include over 50% of the Fortune 100 and government agencies in 45 countries, including more than half of the G20. We’ve been deploying AI at scale for well over a decade, so has Recorded Future. They leverage AI-powered insights to analyze threat data from every corner of the Internet and customers gain real-time visibility and actionable insights to proactively reduce risks. We now have an even more robust set of powerful intelligence, identity, dispute, fraud and scan prevention solutions. Together, these uniquely differentiated technologies will enable us to create smarter models, distribute these capabilities more broadly and help our customers anticipate threats before cyber-attacks can take place. That means better protection for governments, businesses, banks, consumers the entire ecosystem and well beyond the payment transactions. We’re also leveraging our distribution at scale to deepen market penetration of our services and solutions.

Meta Platforms (NASDAQ: META)

Meta’s management expects Meta AI to be the leading AI assistant in 2025, reaching more than 1 billion people; Meta AI is already the most-used AI assistant in the world with more than 700 million monthly actives; management believes Meta AI is at a scale that allows it to develop a durable long-term advantage; management has an exciting road map for Meta AI in 2025 that focuses on personalisation; management does not believe that there’s going to be only one big AI that is the same for everyone; there are some fun surprises for Meta AI in 2025 that management has up their sleeves; Meta AI can now remember certain details of people’s prior queries; management sees a few possible paths for Meta AI’s monetisation, but their focus right now is just on building a great user experience; WhatsApp has the strongest Meta AI usage, followed by Facebook; people are using Meta AI on WhatsApp for informational, educational, and emotional purposes 

 In AI, I expect that this is going to be the year when a highly intelligent and personalized AI assistant reaches more than 1 billion people, and I expect Meta AI to be that leading AI assistant. Meta AI is already used by more people than any other assistant. And once a service reaches that kind of scale, it usually develops a durable long-term advantage.

We have a really exciting road map for this year with a unique vision focused on personalization. We believe that people don’t all want to use the same AI. People want their AI to be personalized to their context, their interests, their personality, their culture, and how they think about the world. I don’t think that there’s just going to be one big AI that everyone uses that does the same thing. People are going to get to choose how their AI works and what it looks like for them. I continue to think that this is going to be one of the most transformative products that we’ve made, and we have some fun surprises that I think people are going to like this year…

… Meta AI usage continues to scale with more than 700 million monthly actives. We’re now introducing updates that will enable Meta AI to deliver more personalized and relevant responses by remembering certain details from people’s prior queries and considering what they engage with on Facebook and Instagram to develop better intuition for their interest and preferences…

…Our initial focus for Meta AI is really about building a great consumer experience, and that’s frankly, where all of our energies are kind of directed to right now. There will, I think, be pretty clear monetization opportunities over time, including paid recommendations and including a premium offering, but really not where we are focused in terms of the development of Meta AI today…

…WhatsApp continues to see the strongest Meta AI usage across our family of apps. People there are using it most frequently for information seeking and educational queries along with emotional support use cases. Most of the WhatsApp engagement is in one-on-one threads, though we see some usage in group messaging. And on Facebook, which is the second largest driver of Meta AI engagement, we’re seeing strong engagement from our feed deep dives integration that lets people ask Meta AI questions about the content that is recommended to that. So across, I would say, all query types, we continue to see signs that Meta AI is helping people leverage our apps for new use cases. We talked about information gathering, social interaction and communication Lots of people use it for humor and casual conversation. They use it for writing and editing research recommendations. 

Meta’s management thinks Llama will become the most advanced and widely-used AI model in 2025; Llama 4 is making great progress; Meta has a reasoning Llama model in the works; management’s goal for Llama 4 is for it be the leading AI model; Llama 4 is built to be multi-modal and agentic; management expects Llama 4 to unlock a lot of new use cases

I think this will very well be the year when Llama and open-source become the most advanced and widely used AI models as well. Llama 4 is making great progress in training, Llama 4 Mini is doing with pretraining and our reasoning models and larger model are looking good too. 

Our goal with Llama 3 was to make open source competitive with closed models. And our goal for Llama 4 is to lead. Llama 4 will be natively multimodal. It’s an omni model, and it will have agenetic capabilities. So it’s going to be novel, and it’s going to unlock a lot of new use cases.

Meta’s management thinks it will be possible in 2025 to build an AI engineering agent that is as capable as a human mid-level software engineer; management believes that the company that builds this AI engineering agent first will have a meaningful advantage in advancing AI research; Meta already has internal AI coding assistants, powered by Llama; management has no plan to release the AI engineer as an external product anytime soon, but sees the potential for it in the longer-term; management does not expect the AI engineer to be extremely widely deployed in 2025, with the dramatic changes happening in 2026 and beyond

I also expect that 2025 will be the year when it becomes possible to build an AI engineering agent that has coding and problem-solving abilities of around a good mid-level engineer. And this is going to be a profound milestone and potentially one of the most important innovations in history, like as well as over time, potentially a very large market. Whichever company builds this first, I think it’s going to have a meaningful advantage in deploying it to advance their AI research and shape the field…

…As part of our efficiency focus over the past 2 years, we’ve made significant improvements in our internal processes and developer tools and introduce new tools like our AI-powered coding assistant, which is helping our engineers write code more quickly. Looking forward, we expect that the continuous advancements in Llama’s coding capabilities will provide even greater leverage to our engineers, and we are focused on expanding its capabilities to not only assist our engineers in writing and reviewing our code but to also begin generating code changes to automate tool updates and improve the quality of our code base…

…And then the AI engineer piece, I’m really excited about it. I mean, I don’t know that that’s going to be an external product anytime soon. But I think for what we’re working on, our goal is to advance AI research and advance our own development internally. And I think it’s just going to be a very profound thing. So I mean that’s something that I think will show up through making our products better over time. But — and then as that works, there will potentially be a market opportunity down the road. But I mean, for now and this year, we’re really — I think this is — I don’t think you’re going to see this year like an AI engineer that is extremely widely deployed, changing all of development. I think this is going to be the year where that really starts to become possible and lays the groundwork for a much more dramatic change in ’26 and beyond.

The Ray-Ban Meta AI glasses are a big hit so far but management thinks 2025 will be the pivotal year to determine if the AI glasses can be on a path towards being the next computing platform and selling hundreds of millions, or more, units; management continues to think that glasses are the perfect form factor for AI; management is optimistic about AI glasses, but there’s still uncertainty about the long-term trajectory

Our Ray-Ban Meta AI glasses are a real hit. And this will be the year when we understand the trajectory for AI glasses as a category. Many breakout products in the history of consumer electronics have sold 5 million to 10 million units and they’re third generation. This will be a defining year that determines if we’re on a path towards many hundreds of millions and eventually billions of AI glasses and glasses being the next computing platform like we’ve been talking about for some time or if this is just going to be a longer grind. But it’s great overall to see people recognizing that these glasses are the perfect form factor for AI as well as just great stylish glasses…

…There are a lot of people in the world who have glasses. It’s kind of hard for me to imagine that a decade or more from now, all the glasses aren’t going to basically be AI glasses as well as a lot of people who don’t wear glasses today, finding that to be a useful thing. So I’m incredibly optimistic about this…

…But look, the Ray-Ban Metas were hit. We still don’t know what the long-term trajectory for this is going to be. And I think we’re going to learn a lot this year. 

Meta will bring ~1 gigawatt of AI data center capacity online in 2025 and is building an AI data center that is at least 2 gigawatts in capacity; management intends to fund the investments through revenue growth that is driven by its AI advances; most of Meta’s new headcount growth will go towards its AI infrastructure and AI advances; management expects compute will be very important for the opportunities they want Meta to pursue; management is simultaneously growing Meta’s capacity and increasing the efficiency of its workloads; Meta is increasing the useful lives of its non-AI and AI servers to 5.5 years (from 4-5 years previously), which will lead to lower depreciation expenses per year; Meta started deploying its own MTIA (Meta Training and Inference Accelerator) AI chips in 2024 for inference workloads; management expects to ramp up MTIA usage for inference in 2025 and for training workloads in 2026; management will continue to buy third-party AI chips (likely referring to NVIDIA), but wants to use in-house chips for unique workloads; management thinks MTIA helps Meta achieve greater compute efficiency and performance per cost and power; management has been thinking about the balance of compute used in pre-training versus inference, but this does not mean that Meta will need less compute; management thinks that inference-time compute (or test-time compute) scaling will help Meta deliver a higher quality of service and that Meta has a strong business model to support the delivery of inference-time compute scaling; management believes that investing heavily in AI infrastructure is still going to be a strategic advantage over time, but it’s possible the reverse may be true in the future; management thinks it’s too early to tell what the long-run capacity intensity will look like

I announced last week that we expect to bring online almost a gigawatt of capacity this year. And we’re building a 2 gigawatt and potentially bigger AI data center that is so big that it will cover a significant part of Manhattan if we were placed there. We’re planning to fund all of this by, at the same time, investing aggressively in initiatives that use these AI advances to increase revenue growth…

…That’s what a lot of our new headcount growth is going towards and how well we execute on this will also determine our financial trajectory over the next few years…

…We expect compute will be central to many of the opportunities we’re pursuing as we advance the capabilities of Llama, drive increased usage of generative AI products and features across our platform and fuel core ads and organic engagement initiatives. We’re working to meet the growing capacity needs for these services by both scaling our infrastructure footprint and increasing the efficiency of our workloads…

…Our expectation going forward is that we’ll be able to use both our non-AI and AI [indiscernible] servers for a longer period of time before replacing them, which we estimate will be approximately 5.5 years. This will deliver savings in annual CapEx and resulting depreciation expense, which is already included in our guidance.

Finally, we’re pursuing cost efficiencies by deploying our custom MTIA silicon in areas where we can achieve a lower cost of compute by optimizing the chip to our unique workloads. In 2024, we started deploying MTIA to our ranking and recommendation inference workloads for ads and organic content. We expect to further ramp adoption of MTIA for these use cases throughout 2025, before extending our custom silicon efforts to training workloads for ranking and recommendations next year…

…We expect that we are continuing to purchase third-party silicon from leading providers in the industry. And we are certainly committed to those long-standing partnerships, but we’re also very invested in developing our own custom silicon for unique workloads, where off-the-shelf silicon isn’t necessarily optimal and specifically because we’re able to optimize the full stack to achieve greater compute efficiency and performance per cost and power because our workloads might require a different mix of memory versus network, bandwidth versus compute and so we can optimize that really to the specific needs of our different types of workloads.

Right now, the in-house MTIA program is focused on supporting our core ranking and recommendation inference workloads. We started adopting MTIA in the first half of 2024 for core ranking and recommendations in [indiscernible]. We’ll continue ramping adoption for those workloads over the course of 2025 as we use it for both incremental capacity and to replace some GPU-based servers when they reach the end of their useful lives. Next year, we’re hoping to expand MTIA to support some of our core AI training workloads and over time, some of our Gen AI use cases…

…There’s already sort of a debate around how much of the compute infrastructure that we’re using is going to go towards pretraining versus as you get more of these reasoning time models or reasoning models where you get more of the intelligence by putting more of the compute into inference, whether just will mix shift how we use our compute infrastructure towards that. That was already something that I think a lot of the — the other labs and ourselves were starting to think more about and already seemed pretty likely even before this, that — like of all the compute that we’re using, that the largest pieces aren’t necessarily going to go towards pre-training. But that doesn’t mean that you need less compute because one of the new properties that’s emerged is the ability to apply more compute at inference time in order to generate a higher level of intelligence and a higher quality of service, which means that as a company that has a strong business model to support this, I think that’s generally an advantage that we’re now going to be able to provide a higher quality of service than others who don’t necessarily have the business model to support it on a sustainable basis…

…I continue to think that investing very heavily in CapEx and infra is going to be a strategic advantage over time. It’s possible that we’ll learn otherwise at some point, but I just think it’s way too early to call that…

…I think it is really too early to determine what long-run capital intensity is going to look like. There are so many different factors. The pace of advancement in underlying models, how efficient can they be? What is the adoption and use case of our Gen AI products, what performance gains come from next-generation hardware innovations, both our own and third party and then ultimately, what monetization or other efficiency gains our AI investments unlock. 

In 2024 H2, Meta introduced a new machine learning system for ads ranking, in partnership with Nvidia, named Andromeda; Andromeda has enabled a 10,000x increase in the complexity of AI models Meta uses for ads retrieval, driving an 8% increase in quality of ads that people see; Andromeda can process large volumes of ads and positions Meta well for a future where advertisers use the company’s generative AI tools to create and test more ads

In the second half of 2024, we introduced an innovative new machine learning system in partnership with NVIDIA called Andromeda. This more efficient system enabled a 10,000x increase in the complexity of models we use for ads retrieval, which is the part of the ranking process where we narrow down a pool of tens of millions of ads to the few thousand we consider showing someone. The increase in model complexity is enabling us to run far more sophisticated prediction models to better personalize which ads we show someone. This has driven an 8% increase in the quality of ads that people see on objectives we’ve tested. Andromeda’s ability to efficiently process larger volumes of ads also positions us well for the future as advertisers use our generative AI tools to create and test more ads.

Advantage+ has surpassed a $20 billion annual revenue run rate and grew 70% year-on-year in 2024 Q4; Advantage+ will now be turned on by default for all campaigns that optimise for sales, app, or lead objectives; more than 4 million advertisers are now using at least one of Advantage+’s generative AI ad creative tools, up from 1 million six months ago; Meta’s first video generation tool, released in October, already has hundreds of thousands of advertisers using it monthly

 Adoption of Advantage+ shopping campaigns continues to scale with revenues surpassing a $20 billion annual run rate and growing 70% year-over-year in Q4. Given the strong performance and interest we’re seeing in Advantage+ shopping and our other end-to-end solutions, we’re testing a new streamlined campaign creation flow. So advertisers no longer need to choose between running a manual or Advantage+ sales or app campaign. In this new setup, all campaigns optimizing for sales, app or lead objectives will have Advantage+ turned on from the beginning. This will allow more advertisers to take advantage of the performance Advantage+ offers while still having the ability to further customize aspects of their campaigns when they need to. We plan to expand to more advertisers in the coming months before fully rolling it out later in the year.

Advantage+ Creative is another area where we’re seeing momentum. More than 4 million advertisers are now using at least one of our generative AI ad creative tools, up from 1 million six months ago. There has been significant early adoption of our first video generation tool that we rolled out in October, Image Animation, with hundreds of thousands of advertisers already using it monthly.

Meta’s management thinks the emergence of DeepSeek makes it even more likely for a global open source standard for AI models to develop; the presence of DeepSeek also makes management think it’s important that the open source standard be made in America and that it ’s even more important for Meta to focus on building open source AI models; Meta is learning from DeepSeek’s innovations in building AI models; management currently does not have a strong opinion on how Meta’s capex plans for AI infrastructure will change because of the recent news with DeepSeek 

I also just think in light of some of the recent news, the new competitor DeepSeek from China, I think it also just puts — it’s one of the things that we’re talking about is there’s going to be an open source standard globally. And I think for our kind of national advantage, it’s important that it’s an American standard. So we take that seriously, and we want to build the AI system that people around the world are using and I think that if anything, some of the recent news has only strengthened our conviction that this is the right thing for us to be focused on…

…I can start on the DeepSeek question. I think there’s a number of novel things that they did that I think we’re still digesting. And there are a number of things that they have advances that we will hope to implement in our systems. And that’s part of the nature of how this works, whether it’s a Chinese competitor or not…

…It’s probably too early to really have a strong opinion on what this means for the trajectory around infrastructure and CapEx and things like that. There are a bunch of trends that are happening here all at once.

Meta’s capex in 2025 is going to grow across servers, data centers, and networking; within each of servers, data centers, and networking, management expects growth in both AI and non-AI capex; management expects most of the AI-related capex in 2025 to be directed specifically towards Meta’s core AI infrastructure, but the infrastructure Meta is building can support both AI and non-AI workloads, and the GPU servers purchased can be used for both generative AI and core AI purposes

[Question] As we think about the $60 billion to $65 billion CapEx this year, does the composition change much from last year when you talked about servers as the largest part followed by data centers and networking equipment. And how should we think about that mix between like training and inference

[Answer] We certainly expect that 2025 CapEx is going to grow across all 3 of those components you described.

Servers will be the biggest growth driver that remains the largest portion of our overall CapEx budget. We expect both growth in AI capacity as we support our gen AI efforts and continue to invest meaningfully in core AI, but we are also expecting growth in non-AI capacity as we invest in the core business, including to support a higher base of engagement and to refresh our existing servers.

On the data center side, we’re anticipating higher data center spend in 2025 to be driven by build-outs of our large training clusters and our higher power density data centers that are entering the core construction phase. We’re expecting to use that capacity primarily for core AI and non-AI use cases.

On the networking side, we expect networking spend to grow in ’25 as we build higher-capacity networks to accommodate the growth in non-AI and core AI-related traffic along with our large Gen AI training clusters. We’re also investing in fiber to handle future cross-region training traffic.

And then in terms of the breakdown for core versus Gen AI use cases, we’re expecting total infrastructure spend within each of Gen AI, non-AI and core AI to increase in ’25 with the majority of our CapEx directed to our core business with some caveat that, that is — that’s not easy to measure perfectly as the data centers we’re building can support AI or non-AI workloads and the GPU-based servers, we procure for gen AI can be repurposed for core AI use cases and so on and so forth.

---

Disclaimer: The Good Investors is the personal investing blog of two simple guys who are passionate about educating Singaporeans about stock market investing. By using this Site, you specifically agree that none of the information provided constitutes financial, investment, or other professional advice. It is only intended to provide education. Speak with a professional before making important decisions about your money, your professional life, or even your personal life. I have a vested interest in Alphabet, Amazon, Apple, ASML, Coupang, Datadog, Fiverr, Mastercard, and Meta Platforms. Holdings are subject to change at any time.

I’ve long been sector-agnostic when it comes to the companies I’m interested in because I believe that great companies – and thus, great stocks – can come from anywhere. 

My belief was formed because of something I learnt more than a dozen years ago about the US-based airline, Southwest Airlines. Ned Davis Research was tasked by CNN’s MONEY Magazine in 2002 to find the five US-listed stocks with the highest returns over the past 30 years. The winner was Southwest Airlines with its annualised return of 26% from 1972 to 2002; a $1,000 investment at the start of the period would have become $1 million by the end. What is noteworthy here is airlines were widely regarded back then as businesses with horrendous economics. In Berkshire Hathaway’s 2007 annual shareholders’ letter, Warren Buffett wrote (emphasis is mine): 

“The worst sort of business is one that grows rapidly, requires significant capital to engender the growth, and then earns little or no money. Think airlines. Here a durable competitive advantage has proven elusive ever since the days of the Wright Brothers. Indeed, if a farsighted capitalist had been present at Kitty Hawk, he would have done his successors a huge favor by shooting Orville down.

The airline industry’s demand for capital ever since that first flight has been insatiable. Investors have poured money into a bottomless pit, attracted by growth when they should have been repelled by it. And I, to my shame, participated in this foolishness when I had Berkshire buy U.S. Air preferred stock in 1989. As the ink was drying on our check, the company went into a tailspin, and before long our preferred dividend was no longer being paid. But we then got very lucky. In one of the recurrent, but always misguided, bursts of optimism for airlines, we were actually able to sell our shares in 1998 for a hefty gain. In the decade following our sale, the company went bankrupt. Twice.” 

And yet, it was an airline that topped the charts in 2002 for the best-performing US stock in the past 30 years. The timeframe of 30 years is also sufficiently long, such that Southwest Airlines’ gains had to be the result of its business’s excellent long-term performance, and not some fortunate short-term hiccup in the fortunes of its business or its stock price.

A recent study from the highly-regarded investment researcher Michael Maubossin, titled Measuring the Moat: Assessing the Magnitude and Sustainability of Value Creation, bolsters my belief. He found that differences in the return on invested capital (ROIC) between industries is lower than the differences in ROICs of companies within industries. In Mauboussin’s data-set, the industry with the highest median ROIC from 1963 to 2023 is Personal Care Products at around 18%. But within Personal Care Products, the companies have ROICs ranging from a low of around 5% to a high of around 40%. Meanwhile, the Wireless Telecom Services industry has one of the lowest median ROICs at around 1%. Yet, the companies within have ROICs ranging from just below 40% to deeply negative figures. Said another way, the best company in a poor industry (Wireless Telecom Services) still has an excellent business that performs significantly better than the median company in a great industry (Personal Care Products). 

I continue to believe that excellent investing opportunities can be found everywhere, so I will, for the foreseeable future, remain sector-agnostic. Sometimes, a gleaming diamond can be found amongst a lump of coal for those with the ability to spot a true bargain.

---

Disclaimer: The Good Investors is the personal investing blog of two simple guys who are passionate about educating Singaporeans about stock market investing. By using this Site, you specifically agree that none of the information provided constitutes financial, investment, or other professional advice. It is only intended to provide education. Speak with a professional before making important decisions about your money, your professional life, or even your personal life. I currently have no vested interest in any company mentioned. Holdings are subject to change at any time.

We’ve constantly been sharing a list of our recent reads in our weekly emails for The Good Investors.

Do subscribe for our weekly updates through the orange box in the blog (it’s on the side if you’re using a computer, and all the way at the bottom if you’re using mobile) – it’s free!

But since our readership-audience for The Good Investors is wider than our subscriber base, we think sharing the reading list regularly on the blog itself can benefit even more people. The articles we share touch on a wide range of topics, including investing, business, and the world in general. 

Here are the articles for the week ending 12 January 2025:

1. The art of outlasting: What we can learn from timeproof Japanese businesses – Eric Markowitz

Japan is home to an extraordinary number of shinise, or long-established businesses. A 2008 study found that Japan had over 21,000 companies older than 100 years, including more than 3,000 that had crossed the 200-year mark. These firms are not just historical artifacts — they are vibrant examples of how to endure and thrive in a rapidly changing world. Their strategies — balancing tradition with adaptability, patience with practicality — are a masterclass in long-term thinking that today’s entrepreneurs and executives would be wise to study…

…What ties these stories together is an approach to business that’s almost rebellious in its patience. While the modern world glorifies disruption and speed, Japan’s ancient companies remind us that longevity is often about playing the long game. It’s about building something so solid, so aligned with its environment, that it can weather any storm. But let’s not romanticize this too much. Strip away the poetry of water metaphors and ancient traditions, and you’ll find ruthless pragmatism at the core of these businesses’ survival.

When Japan’s post-war construction boom faded, Kongo Gumi didn’t just stick to temples — they pivoted hard into office buildings and apartments while maintaining their temple maintenance business as a hedge. During the lean years of the 1990s recession, Hōshi Ryokan cut costs to the bone while refusing to lay off staff, with family members taking deep pay cuts to keep their centuries-old workforce intact. Okaya transformed from selling samurai swords to becoming a global steel trader, making calculated bets on new technologies and markets while keeping their supply chain relationships rock solid.

These companies didn’t just drift through history — they clawed their way through wars, depressions, and cultural upheavals, making brutal choices about what to preserve and what to sacrifice. Their longevity wasn’t achieved through Zen-like detachment, but through gritted teeth and white-knuckled adaptability.

2. Notes on China – Dwarkesh Patel

I got quite mixed messages about the state of public opinion in China. This is to be expected in a society where you can’t establish common knowledge. One person told me that the new generation is quite nationalist, unlike the older reform generation which personally experienced the catastrophes of Mao and the tangible benefits of liberalization. He made the rather insightful point that this tilt in Chinese public opinion increasingly gives lie to the American talking point, “We’re against the CCP, not the Chinese people.” In fact, he went on to say that the current regime is way more liberal than what would result from an election in China.

Another person told me that these Chinese nationalists were only a vocal minority, similar to the wokes in America circa 2020. While they make up only about 10% of the population, they aggressively shout down others on Weibo (China’s Twitter equivalent). Most people find them annoying but feel uncomfortable confronting them directly. This matches what a student who graduated from a top university there told me – the vast majority of his classmates are simply apolitical. And in our own interactions with locals, we saw little evidence of widespread nationalism. In fact, when my Chinese-speaking trip mate (who could actually speak Chinese) would mention he was from the UK to taxi drivers, they would often respond enthusiastically: “Oh wonderful, we love the UK!”…

…We chatted up quite a lot of young people on night life streets. I was struck by how many young people expressed feeling stressed or overwhelmed. We met a musician in Chengdu who was writing songs about youth anxiety. We chatted up some modeling school students – even they complained about the intense pressure they felt. We met a guy who had studied in Australia but returned to China during COVID. He explained that many of his friends with prestigious degrees are moving away from Shanghai and Beijing – Yes, the pay there can be twice as high as in second or third tier cities. But the competitiveness is insane. And in order to actually land the high skilled positions, they have to work truly insane hours (9-9-6 is not a myth). He said that many of his friends were opting for these less ambitious lower-paying careers in smaller cities, where the rent is lower and the pressure is manageable…

…I’m still puzzled by how China can have both a demographic collapse and massive youth unemployment. You’d think with fewer young people being born, the ones who are around would be in high demand. One explanation I heard while there is that there are plenty of menial jobs available, but today’s educated youth – who’ve gone through high school and college – just won’t take the low-skilled positions their parents and grandparents did. Meanwhile, there’s a real shortage of the high-skilled jobs that would actually match their education and aspirations. It’s a mismatch between the jobs available and the jobs young people feel qualified for and willing to do…

…The biggest surprise from talking to Chinese VCs people at AI labs was how capital constrained they felt. Moonshot AI, one of China’s leading AI labs, raised $1 billion at a $3 billion valuation. Meanwhile, just xAI’s new cluster alone will cost $3-4 billion.

The tech ecosystem feels quite shell shocked from the 2021 crackdown. One VC half-jokingly asked if I could help him get his money out of China. If you keep your money in China, you’re basically stuck choosing between terrible options. You can either accept a measly 2% yield from state banks, or throw it into China’s perpetually struggling stock market. This helps explain why valuations for Chinese companies are chronically low – the exit opportunities just suck. Even if you build (or invest in) something great, there’s no guarantee the company will be able to raise the next round. And even if you do raise again and succeed, the government might randomly cancel your IPO. And even if you somehow make it to the public markets, Chinese equities have been performing terribly anyways. It’s a good reminder of how easy it is to completely wreck an innovation ecosystem that depends on risk-taking investors.

3. Is AI progress slowing down? – Arvind Narayanan and Sayash Kapoor

To be clear, there is no reason to doubt the reports saying that many AI labs have conducted larger training runs and yet not released the resulting models. But it is less clear what to conclude from it. Some possible reasons why bigger models haven’t been released include:

• Technical difficulties, such as convergence failures or complications in achieving fault tolerance in multi-datacenter training runs.
• The model was not much better than GPT-4 class models, and so would be too underwhelming to release.
• The model was not much better than GPT-4 class models, and so the developer has been spending a long time trying to eke out better performance through fine tuning.

To summarize, it’s possible that model scaling has indeed reached its limit, but it’s also possible that these hiccups are temporary and eventually one of the companies will find ways to overcome them, such as by fixing any technical difficulties and/or finding new data sources…

…Industry leaders don’t have a good track record of predicting AI developments. A good example is the overoptimism about self-driving cars for most of the last decade. (Autonomous driving is finally real, though Level 5 — full automation — doesn’t exist yet.) As an aside, in order to better understand the track record of insider predictions, it would be interesting to conduct a systematic analysis of all predictions about AI made in the last 10 years by prominent industry insiders.

There are some reasons why we might want to give more weight to insiders’ claims, but also important reasons to give less weight to them. Let’s analyze these one by one. It is true that industry insiders have proprietary information (such as the performance of as-yet-unreleased models) that might make their claims about the future more accurate. But given how many AI companies are close to the state of the art, including some that openly release model weights and share scientific insights, datasets, and other artifacts, we’re talking about an advantage of at most a few months, which is minor in the context of, say, 3-year forecasts.

Besides, we tend to overestimate how much additional information companies have on the inside — whether in terms of capability or (especially) in terms of safety. Insiders warned for a long time that “if only you know what we know…” but when whistleblowers finally came forward, it turns out that they were mostly relying on the same kind of speculation that everyone else does.

Another potential reason to give more weight to insiders is their technical expertise. We don’t think this is a strong reason: there is just as much AI expertise in academia as in industry. More importantly, deep technical expertise isn’t that important to support the kind of crude trend extrapolation that goes into AI forecasts. Nor is technical expertise enough — business and social factors play at least as big a role in determining the course of AI. In the case of self-driving cars, one such factor is the extent to which societies tolerate public roads being used for experimentation. In the case of large AI models, we’ve argued before that the most important factor is whether scaling will make business sense, not whether it is technically feasible…

…As an example, Sutskever had an incentive to talk up scaling when he was at OpenAI and the company needed to raise money. But now that he heads the startup Safe Superintelligence, he needs to convince investors that it can compete with OpenAI, Anthropic, Google, and others, despite having access to much less capital. Perhaps that is why he is now talking about running out of data for pre-training, as if it were some epiphany and not an endlessly repeated point.

To reiterate, we don’t know if model scaling has ended or not. But the industry’s sudden about-face has been so brazen that it should leave no doubt that insiders don’t have any kind of crystal ball and are making similar guesses as everyone else, and are further biased by being in a bubble and readily consuming the hype they sell to the world…

…Inference scaling is useful for problems that have clear correct answers, such as coding or mathematical problem solving. In such tasks, at least one of two related things tend to be true. First, symbolic reasoning can improve accuracy. This is something LLMs are bad at due to their statistical nature, but can overcome by using output tokens for reasoning, much like a person using pen and paper to work through a math problem. Second, it is easier to verify correct solutions than to generate them (sometimes aided by external verifiers, such as unit tests for coding or proof checkers for mathematical theorem proving).

In contrast, for tasks such as writing or language translation, it is hard to see how inference scaling can make a big difference, especially if the limitations are due to the training data. For example, if a model works poorly in translating to a low-resource language because it isn’t aware of idiomatic phrases in that language, the model can’t reason its way out of this.

The early evidence we have so far, while spotty, is consistent with this intuition. Focusing on OpenAI o1, it improves compared to state-of-the-art language models such as GPT-4o on coding, math, cybersecurity, planning in toy worlds, and various exams. Improvements in exam performance seem to strongly correlate with the importance of reasoning for answering questions, as opposed to knowledge or creativity: big improvements for math, physics and LSATs, smaller improvements for subjects like biology and econometrics, and negligible improvement for English.

Tasks where o1 doesn’t seem to lead to an improvement include writing, certain cybersecurity tasks (which we explain below), avoiding toxicity, and an interesting set of tasks at which thinking is known to make humans worse…

…We think there are two reasons why agents don’t seem to benefit from reasoning models. Such models require different prompting styles than regular models, and current agentic systems are optimized for prompting regular models. Second, as far as we know, reasoning models so far have not been trained using reinforcement learning in a setting where they receive feedback from the environment — be it code execution, shell interaction, or web search. In other words, their tool use ability is no better than the underlying model before learning to reason…

…The furious debate about whether there is a capability slowdown is ironic, because the link between capability increases and the real-world usefulness of AI is extremely weak. The development of AI-based applications lags far behind the increase of AI capabilities, so even existing AI capabilities remain greatly underutilized. One reason is the capability-reliability gap — even when a certain capability exists, it may not work reliably enough that you can take the human out of the loop and actually automate the task (imagine a food delivery app that only works 80% of the time). And the methods for improving reliability are often application-dependent and distinct from methods for improving capability. That said, reasoning models also seem to exhibit reliability improvements, which is exciting.

Here are a couple of analogies that help illustrate why it might take a decade or more to build products that fully take advantage of even current AI capabilities. The technology behind the internet and the web mostly solidified in the mid-90s. But it took 1-2 more decades to realize the potential of web apps. Or consider this thought-provoking essay that argues that we need to build GUIs for large language models, which will allow interacting with them with far higher bandwidth than through text. From this perspective, the current state of AI-based products is analogous to PCs before the GUI.

4. Waymo still doing better than humans at preventing injuries and property damage – Andrew J. Hawkins

The study is the product of the collaboration between Waymo and insurer Swiss Re, which analyzed liability claims related to collisions from 25.3 million fully autonomous miles driven by Waymo in four cities: Phoenix, San Francisco, Los Angeles, and Austin. They then compared those miles to human driver baselines, which are based on Swiss Re’s data from over 500,000 claims and over 200 billion miles traveled.

They found that the performance of Waymo’s vehicles was safer than that of humans, with an 88 percent reduction in property damage claims and a 92 percent reduction in bodily injury claims. Across 25.3 million miles, Waymo was involved in nine property damage claims and two bodily injury claims. The average human driving a similar distance would be expected to have 78 property damage and 26 bodily injury claims, the company says.

Waymo’s vehicles also performed better when compared to new vehicles equipped with all the latest safety tech, including automatic emergency braking, lane-keep assist, and blind spot detection. When compared to this group, Waymo’s autonomous driving system showed an 86 percent reduction in property damage claims and a 90 percent reduction in bodily injury claims.

5. SITALWeek #454 – Brad Slingerlend

I think we are approaching the point where we can start to estimate the value of AI for developers and the companies/consumers who are going to buy the next wave of innovative applications. I think the salient question for AI (and, frankly, humanity!) is: How much AI reasoning can you get for a human-equivalent salary? In other words, for a certain salary, how much compute power will it take to match or outperform a human (assuming the AI can collaborate with other humans/AIs using the same methods and tools a human would)…

… LLMs are shifting from a pure token-in/token-out model to a test-time scaling model, which may offer us better inroads for estimating costs. Essentially, they are thinking harder before spitting out a reply; thus, rather than just predicting the next words in a response using a probability model (see You Auto-Complete Me), they are doing some deep thinking to arrive at more accurate, useful answers. This is a major leap in capability that comes with a major leap in cost. OpenAI raised prices for their o1 model to $200/mo (Pro subscription) from $20 (Plus subscription). For developers, use of o1’s advanced reasoning API comes at 3-4x the cost of their “general purpose” GPT-4o. If o1 were priced at a typical Western office worker wage of $40/hr, the reasoning of the model would equate to around 5 hours of work per month. We also don’t know if the $200/mo price point is profitable for OpenAI or if they are just relying on Microsoft to further subsidize their business model (which brings us back to the principal-agent problem I started this section off with). So, all of my hand waving here seems to imply you can get a decent amount of human-equivalent reasoning for an amount of money in the realm of human labor cost. If true, after a few more years of advancements in semiconductors and AI models, we should have markedly affordable “human reasoning as a service”, an explosion in demand, and a wide range of outcomes for how much human supervision of AI will be required (it may be that human jobs stay relatively flat, but each human is 2x productive, then 4x, etc.).

Following this logic, at current AI reasoning costs, companies would need to lay off one human for every AI human equivalent they hire and would probably lose more skill/knowledge than they gain. In other words, based on my attempts to guess the cost of replacing human reasoning, today’s AI offerings aren’t likely compelling enough. In a couple years, however, maybe you will be able to lay off one human and hire a handful of AIs, which, by collaborating with each other and humans, may yield superior results. Even today, extremely high-value tasks, such as in-depth research or stock market predictions, may be able to take advantage of the high-cost test-time scaling AI models. And, if any of this math is in the realm of reason, you can easily see that AI may not require such high-value-add applications to be cost effective in the near to medium future. The proof will come within the next couple of years as today’s entrepreneurs develop the next generation of apps leveraging LLMs and overtaking human capabilities: If these apps are at price points that outcompete human employees, a significant wave of change could come much faster to society. 

---

Disclaimer: The Good Investors is the personal investing blog of two simple guys who are passionate about educating Singaporeans about stock market investing. By using this Site, you specifically agree that none of the information provided constitutes financial, investment, or other professional advice. It is only intended to provide education. Speak with a professional before making important decisions about your money, your professional life, or even your personal life. We currently have a vested interest in Alphabet (parent of Google and Waymo) and Microsoft. Holdings are subject to change at any time.

We’ve constantly been sharing a list of our recent reads in our weekly emails for The Good Investors.

Do subscribe for our weekly updates through the orange box in the blog (it’s on the side if you’re using a computer, and all the way at the bottom if you’re using mobile) – it’s free!

But since our readership-audience for The Good Investors is wider than our subscriber base, we think sharing the reading list regularly on the blog itself can benefit even more people. The articles we share touch on a wide range of topics, including investing, business, and the world in general. 

Here are the articles for the week ending 29 December 2024:

1. Quantum Computers Cross Critical Error Threshold – Ben Brubaker

In the 1990s, researchers worked out the theoretical foundations for a way to overcome these errors, called quantum error correction. The key idea was to coax a cluster of physical qubits to work together as a single high-quality “logical qubit.” The computer would then use many such logical qubits to perform calculations. They’d make that perfect machine by transmuting many faulty components into fewer reliable ones…

…This computational alchemy has its limits. If the physical qubits are too failure-prone, error correction is counterproductive — adding more physical qubits will make the logical qubits worse, not better. But if the error rate goes below a specific threshold, the balance tips: The more physical qubits you add, the more resilient each logical qubit becomes.

Now, in a paper(opens a new tab) published today in Nature, Newman and his colleagues at Google Quantum AI have finally crossed the threshold. They transformed a group of physical qubits into a single logical qubit, then showed that as they added more physical qubits to the group, the logical qubit’s error rate dropped sharply…

…At first, many researchers thought quantum error correction would be impossible. They were proved wrong in the mid-1990s, when researchers devised simple examples of quantum error-correcting codes. But that only changed the prognosis from hopeless to daunting.

When researchers worked out the details, they realized they’d have to get the error rate for every operation on physical qubits below 0.01% — only one in 10,000 could go wrong. And that would just get them to the threshold. They would actually need to go well beyond that — otherwise, the logical qubits’ error rates would decrease excruciatingly slowly as more physical qubits were added, and error correction would never work in practice…

…That variation, called the surface code, is based on two overlapping grids of physical qubits. The ones in the first grid are “data” qubits. These collectively encode a single logical qubit. Those in the second are “measurement” qubits. These allow researchers to snoop for errors indirectly, without disturbing the computation.

This is a lot of qubits. But the surface code has other advantages. Its error-checking scheme is much simpler than those of competing quantum codes. It also only involves interactions between neighboring qubits — the feature that Preskill found so appealing.

In the years that followed, Kitaev, Preskill and a handful of colleagues fleshed out the details(opens a new tab) of the surface code. In 2006, two researchers showed(opens a new tab) that an optimized version of the code had an error threshold around 1%, 100 times higher than the thresholds of earlier quantum codes. These error rates were still out of reach for the rudimentary qubits of the mid-2000s, but they no longer seemed so unattainable…

…Fowler, Martinis and two other researchers wrote a 50-page paper(opens a new tab) that outlined a practical implementation of the surface code. They estimated that with enough clever engineering, they’d eventually be able to reduce the error rates of their physical qubits to 0.1%, far below the surface-code threshold. Then in principle they could scale up the size of the grid to reduce the error rate of the logical qubits to an arbitrarily low level. It was a blueprint for a full-scale quantum computer…

…When you put the theory of quantum computing into practice, the first step is perhaps the most consequential: What hardware do you use? Many different physical systems can serve as qubits, and each has different strengths and weaknesses. Martinis and his colleagues specialized in so-called superconducting qubits, which are tiny electrical circuits made of superconducting metal on silicon chips. A single chip can host many qubits arranged in a grid — precisely the layout the surface code demands.

The Google Quantum AI team spent years improving their qubit design and fabrication procedures, scaling up from a handful of qubits to dozens, and honing their ability to manipulate many qubits at once. In 2021, they were finally ready to try error correction with the surface code for the first time. They knew they could build individual physical qubits with error rates below the surface-code threshold. But they had to see if those qubits could work together to make a logical qubit that was better than the sum of its parts. Specifically, they needed to show that as they scaled up the code — by using a larger patch of the physical-qubit grid to encode the logical qubit — the error rate would get lower.

They started with the smallest possible surface code, called a “distance-3” code, which uses a 3-by-3 grid of physical qubits to encode one logical qubit (plus another eight qubits for measurement, for a total of 17). Then they took one step up, to a distance-5 surface code, which has 49 total qubits. (Only odd code distances are useful.)

In a 2023 paper(opens a new tab), the team reported that the error rate of the distance-5 code was ever so slightly lower than that of the distance-3 code. It was an encouraging result, but inconclusive — they couldn’t declare victory just yet…

…At the beginning of 2024, they had a brand-new 72-qubit chip, code-named Willow, to test out. They spent a few weeks setting up all the equipment needed to measure and manipulate qubits…

…Then a graph popped up on the screen. The error rate for the distance-5 code wasn’t marginally lower than that of the distance-3 code. It was down by 40%. Over the following months, the team improved that number to 50%: One step up in code distance cut the logical qubit’s error rate in half…

…The team also wanted to see what would happen when they continued to scale up. But a distance-7 code would need 97 total qubits, more than the total number on their chip. In August, a new batch of 105-qubit Willow chips came out…

…When the group returned the following morning, they saw that going from a distance-5 to a distance-7 code had once again cut the logical qubit’s error rate in half. This kind of exponential scaling — where the error rate drops by the same factor with each step up in code distance — is precisely what the theory predicts. It was an unambiguous sign that they’d reduced the physical qubits’ error rates well below the surface-code threshold…

…At the same time, researchers recognize that they still have a long way to go. The Google Quantum AI team only demonstrated error correction using a single logical qubit. Adding interactions between multiple logical qubits will introduce new experimental challenges.

Then there’s the matter of scaling up. To get the error rates low enough to do useful quantum computations, researchers will need to further improve their physical qubits. They’ll also need to make logical qubits out of something much larger than a distance-7 code. Finally, they’ll need to combine thousands of these logical qubits — more than a million physical qubits.

2. History: Kodak & Fujifilm – Find Value

Ultimately, Kodak couldn’t adapt to the changing world and filed for bankruptcy in 2012.

In the game for over 100 years, Kodak survived two World Wars and the Great Depression and helped humans photograph the moon and Mars. Like Coca-Cola and McDonald’s, it used to be one of the most recognized brands in the world…

…Faced with a sharp decline in sales from its cash cow product, Fujifilm acted swiftly and changed its business through innovation and external growth. Under Shigetaka Komori (President in 2000), Fujifilm quickly carried out massive reforms. In 2004, Komori came up with a six-year plan called VISION75.

The management restructured its film business by downscaling the production lines and closing redundant facilities. In the meantime, the R&D departments moved to a newly built facility to unify the research efforts and promote better communication and innovation culture among engineers.

Realizing that the digital camera business would not replace the lucrative film due to the low margins, Fujifilm performed a massive diversification based on capabilities and innovation.

Even before launching the VISION75 plan, Komori had taken stock of their technologies and compared them with the demand of the international market. After which the R&D team came up with a chart listing the all existing in-house technologies that could match future markets.

For instance, Fujifilm was able to predict the boom of LCD screens and invested heavily in this market. Leveraging on photo film technology, they created FUJITAC, a variety of high-performance films essential for making LCD panels for TV, computers, and smartphones. Today, FUJITAC owns 70% of the market for protective LCD polarizer films.

Fujifilm also targeted unexpected markets like cosmetics. The rationale behind cosmetics comes from 70 years of experience in gelatin, the chief ingredient of photo film which is derived from collagen. Human skin is 70% collagen. Fujifilm also possessed deep knowledge in oxidation, a process connected both to the aging of human skin and to the fading of photos over time.

When promising technologies didn’t exist internally, Fujifilm proceeded by mergers and acquisitions. Based on technological synergies, it acquired Toyoma Chemical in 2008 to enter the drug business. Delving further into the healthcare segment, Fujifilm also brought a radio-pharmaceutical company now called Fujifilm RI Pharma. It also reinforced its position in existing joint ventures such as Fuji-Xerox which became a consolidated subsidiary in 2001 after Fujifilm purchased an additional 25% share in this partnership.

Fast forward 9 years after the peak of film sales, in 2010, Fujifilm was a new company. In 2000, 60% of sales and 70% of profits came from the film ecosystem, compare this to 2010 where the “Imaging segment” accounted for less than 16% of sales. Fujifilm managed to emerge victorious through a restructuring and diversification strategy…

…Unlike Fujifilm which recognized early on that photography was a doomed business and tackled new markets with a completely different portfolio, Kodak made multiple wrong moves and persisted in the decaying film industry.

It was not that Kodak didn’t want to change, it tried hard, but it did it wrong. Kodak’s management didn’t fully recognize that the rise of digital imaging would have dire consequences for the future of photo printing. It tried to replicate the film print business model in the digital world. In 2004, Facebook was launched, and people are just not going to print pictures anymore.

Interestingly, Kodak understood the impact of digitalization and predicted that pictures would be shared online. They acquired a photo-sharing website called Ofoto in 2001. Unfortunately, the company used Ofoto to make people print digital pictures. They failed in realizing that online photo sharing was the new business, not just a way to expand printing sales…

…While Fujifilm invested heavily in the pharmaceutical and healthcare sector to reduce its exposure to the challenging photo industry, Kodak sold its highly profitable Healthcare Imaging branch in 2007 to put more resources into its losing consumer camera division.

3. One Bed, Two Dreams: Building Silicon Valley Bank in China with Ken Wilcox (Transcript here) – Bernard Leong and Ken Wilcox

Wilcox: In the US, banks sometimes fail. When I started my career 40 years ago in banking, we had 18,000 banks. Today we have about 5,000. What happened to all of them? Where did 13,000 banks go? Some of them got acquired, but many of them failed. When a bank makes too many bad loans, the Federal Reserve causes it to fail and it disappears. In China, banks don’t fail. First of all, banks are fundamentally owned by the government and when they make too many bad loans, they don’t typically fail. Usually the government, the regulators, come and somebody gets arrested and the government re-capitalizes the bank. It’s often very quiet – it’s not even necessarily announced to the world – and the bank keeps on going. What does that mean? That means that Chinese banks can take more risk than US banks can. In the US, we had almost no competitors because everybody thought “Lending to technology companies is way too risky, so we’ll just let Silicon Valley Bank do it. None of the rest of us will try.” In China, many, many, many banks want to copy us and do the same thing, because they’re not worried about what happens if we lose too much money. So that’s another big difference there…

…Wilcox: After I’d been there for several months, it occurred to me one day that my main conversation partner, the guy who is the Chairman, who was from Shanghai Pudong Development Bank, it occurred to me that he actually wears three hats. The only hat I wear is banker / businessman. But he had a banker / businessman hat, and he had a party hat, and he had a government hat. Then I started to wonder, when I’m talking with him, which hat is he wearing? It took me a long time before I figured out he doesn’t even think he has three hats. He thinks they’re all the same hat, so he’s not even thinking about it the same way I was. So I think that’s quite confusing. 

It’s also confusing when people find out, when a US company comes to China and finds out that it’s going to get a Party Committee in their organization. They get very confused because they don’t know what a Party Committee is. If you ask people in government or in the party, “What’s a Party Committee?” You say, “We’re going to have one , but I don’t understand what it is?” It’s hard for them to explain. You get multiple definitions and then you don’t know what is actually going to happen. Some people will tell me, “When you get a Party Committee, it’ll be so good because all the employees in your organization who are members of the party will have a place to gather once a month and discuss things.” Then somebody else says, “When you get a Party Committee, it’ll be so much easier because the Party Committee will help you put on social events for the employees, all of the employees.” But then somebody else told me, “No, when you get a Party Committee, it’ll be like another board, but a secret board. You won’t know who’s on it and they will influence what the real board does – or what I would call the real board.” Then other people told me, “Don’t pay any attention. That’s all silliness. There is no such thing as a Party Committee.” So it’s very, very confusing…

…Wilcox: I’ll give you the best example and that is that I believe based on the years I spent in China, that ultimately the main reason they wanted us in China – and they actually were very determined to get us to come to China. I remember that early on, a couple of years before my wife and I moved to China, I had a series of meetings with a very high-level government official who’s also got a lot of status in the party. He was saying to me, “Ken, we really want you to bring your bank to China. Your bank is more important than any bank we’ve ever met. You’re more important than – he explicitly said this – he says, You’re more important than Morgan Stanley and more important than Goldman Sachs. And by the way Ken, you’re one of the smartest Americans we’ve met.” So you think to yourself, “Well this is an exaggeration, but it does feel nice.” He obviously is going to help me get established in China. But what I didn’t realize is that the main reason they wanted us in China was so that they could study our business model and figure out how to copy it over time. That was something I wasn’t expecting, but I should have if I were less naive. If I were better prepared, I would have realized that was the intention. So the original title, the working title I had for my book, which I had to change because the publisher didn’t like it, my original title was, “One Bed, Two Dreams”, because that’s a phrase that most Chinese are familiar with. It explains why it didn’t work well, because my dream was working with all these Chinese technology companies and helping them do business with the rest of the world, and their dream was learning our business model.

The result was that when they gave us our license, they also told us that we would not be able to use Chinese currency for three years. That made it almost impossible to do business for the first three years. The people that said these things were both members of the government and members of the party. So I don’t know which one was talking. But they said, “We understand that you won’t be able to do much business for the first three years because the companies that you want to work with all want renminbi, they don’t want US dollars. But you can still be a good citizen. You can do what we would do, and that is we here in China help each other. So you can be helpful and prove that you care about China by teaching other banks your business model during the three years when you can’t really do much business. We’ll give you subsidy to help support you during the three years when you can’t earn much money because you can’t really do any business.” Then at the end of the three years when they gave us permission to use renminbi, they said to us, “We are so happy that you came to China and we really admire your business model and we admire it so much that we’re starting a bank of our own using your business model. Would you mind staying a little longer and being an advisor to this new bank that’s going to use your business model?” It felt like they were stealing my intellectual property but I’m not sure they thought of it that way…

…Wilcox: General Motors when it went over to China in 1985, the Chinese really didn’t have an auto industry. They wanted General Motors there not because they wanted General Motors to make a lot of money. It was because they wanted to learn about automobile manufacturing and because it took so long to build up the knowledge base, General Motors was welcome for about 30 years. But now General Motors is slowly losing market share and it’s probably going to withdraw from China. Then what will happen is China has made so much progress partially because they’re hardworking and smart, partially because they had General Motors there to learn from them, and then once General Motors retracts and goes back to the US, the auto industry in China will begin exporting and competing globally. I think actually the Chinese have done such a good job of first of all, learning from foreign automakers, but then on top of that, taking it further that the foreign automakers are in huge trouble. I think China’s automobile industry will dominate in the future. 

4. Weekend thoughts: crypto, mania, and reflexivity follow up – Andrew Walker

When I first saw the “BTC yield” metric, I thought it was pretty crazy. MSTR is trading for approaching 3x the value of their bitcoin; if they issue stock and use all the stock to buy bitcoin, of course it’s going to cause their bitcoin holdings per share to go up…. and even more so if they issue debt and use that to buy bitcoin and then judge themselves on a per share basis! Taken to its extreme2, if you thought BTC yield was truly the be all, end all of value creation, and the higher the BTC yield the better, then any company following a pure BTC yield strategy should lever themselves up to the maximum amount possible, no matter the terms, and use all of the proceeds to buy BTC. Obviously no one does that because it would be insanity and eventually banks would stop lending, but I illustrate that only to show that purely maximize BTC yield is clearly not value maximizing….

But, if you look at the fine print, BTC yield is even crazier than simply suggesting increasing BTC per share is the only value creation metric that matters. If you really look at the MSTR BTC yield table above or read their disclosures, you’ll notice that the BTC yield assumes that all of their convertible debt converts…

…So, go back to MSTR’s BTC yield table; they have a set of 2029 converts that convert at $672.40/share. Those are far, far out of the money (MSTR’s stock trades for ~$400/share as I write this)…. yet MSTR’s BTC yield assumes those converts are in the money / will convert for their BTC yield.

That is an insane assumption that casually assumes MSTR’s shares almost double3. And, again, by taking this assumption to its extreme, we can see how wild it is. Like all things, convert debt involves different trade offs; for example, you could get a higher strike price by taking on a higher interest rate (i.e. if your strike price is ~$670 at a 0% interest rate, you could probably push it up to $770 by taking on a 3% interest rate or $870 by taking on a 6% interest rate4). MSTR has issued all of this convert debt deals at 0% interest rates, which is a great pitch (“we’re borrowing for free, we don’t have to pay a carry to buy BTC, etc”)…. but if BTC yield is all that matters, MSTR could start issuing convertible debt with really high interest rates, which would jack that strike price of the convert up, thus decreasing dilution and increasing the BTC yield…

…MSTR fans would say “but raising converts with interest doesn’t make sense; it’s no longer free money / now it has a carry cost.” And I understand that argument…. but convertible debt isn’t free money either, and I just do this to highlight how insane BTC yield is as a be all / end all metric!…

…The BTC yield that all of these companies present assumes that their convert debt converts, and that is a big / crazy assumption…. but it’s interesting to think about what will happen in five years. There is, of course, a world where BTC goes to $250k (or higher) and all of these stocks moon. In that world, the converts will be well in the money, and all of this worry will sound silly…. but there is also a world where BTC stalls out or drops over the next few years, and that world is really interesting. All of these companies are raising converts with 5-7 year maturities, so if BTC doesn’t moon and the converts aren’t in the money, you’re going to have all of the BTC standard companies facing a maturity wall at the same time. What happens then? I doubt they can roll the converts at anything close to the same terms (remember, cheap converts require high volatility, and if the stocks have stalled out for five years vol is going to be a lot lower), so they’ll either need to sell a ton of equity to paydown the debt (which will be tough; there probably won’t be much enthusiasm for the stock, and I’m not sure the market would be able to absorb the hypothetical amount of stock they’d need to issue without some enthusiasm)…. or you’ll have a wave of BTC standard companies all looking to sell down some of their bitcoin to payoff converts at the exact same time.

5. Satya Nadella | BG2 (Transcript here)- Bill Gurley, Brad Gerstner, and Satya Nadella

Gerstner: Shifting maybe to enterprise AI, Satya. The Microsoft AI business has already reported to be about $10 billion. You’ve said that it’s all inference and that you’re not actually renting raw GPUs to others to train on, because your inference demand is so high. As we think about this, there’s a lot of skepticism out there in the world as to whether or not major workloads are moving. If you think about the key revenue products that people are using today and how it’s driving that inference revenue for you today, and how that may be similar or different from Amazon or Google, I’d be interested in that.

Nadella: The way for us this thing has played out is, you got to remember most of our training stuff with OpenAI is sort of more investment logic. It’s not in our quarterly results – it’s more in the other income, based on our investment.

Gerstner: Other income or loss right?

Nadella: That is right. That’s how it shows up. So most of the revenue or all the revenue is pretty much our API business or in fact, to your point, ChatGPT’s inference costs are there, so that’s a different piece. The fact is the big-hit apps of this era are ChatGPT, Co-Pilot, GitHub Co-Pilot, and the APIs of OpenAI and Azure OpenAI. In some sense, if you had to list out the 10 most hit apps, these would probably be in the four or five of them and so therefore that’s the biggest driver.

The advantage we have had, and OpenAI has had, is we’ve had two years of runway pretty much uncontested. To your point, Bill made the point about everybody’s awake and it might be. I don’t think there will be ever again maybe a two-year lead like this, who knows? It’s all you say that and somebody else drops some sample and suddenly blows the world away. But that said, I think it’s unlikely that that type of lead could be established with some foundation model. But we had that advantage, that was the great advantage we’ve had with OpenAI. OpenAI was able to really build out this escape velocity with ChatGPT.

But on the API side, the biggest thing that we were able to gain was.. Take Shopify or Stripe or Spotify. These were not customers of Azure, they were all customers of GCP or they were customers of AWS. So suddenly we got access to many, many more logos, who are all “digital natives” who are using Azure in some shape or fashion and so on. So that’s sort of one. When it comes to the traditional enterprise, I think it’s scaling. Literally it is people are playing with Co-Pilot on one end and then are building agents on the other end using Foundry. But these things are design wins and project wins and they’re slow, but they’re starting to scale. Again, the fact that we’ve had two years of runway on it, I think…

I like that business a lot more, and that’s one of the reasons why the adverse selection problems here would have been lots of tech startups all looking for their H100 allocations in small batches. Having watched what happened to Sun Microsystems in the dotcom, I always worry about that. You just can’t chase everybody building models. In fact, even the investor side, I think the sentiment is changing, which is now people are wanting to be more capital-light and build on top of other people’s models and so on and so forth. If that’s the case, everybody who was looking for H100 will not want to look for it more. So that’s what we’ve been selective on.

Gerstner: You’re saying for the others that training of those models and those model clusters was a much bigger part of their AI revenue versus yours? 

Nadella: I don’t know. This is where I’m speaking for other people’s results. It’s just I go back and say, “What are the other big-hit apps?” I don’t know what they are. What models do they run? Where do they run them? When I look at the DAU numbers of any of these AI products, there is ChatGPT, and then there is – even Gemini, I’m very surprised at the Gemini numbers, obviously I think it’ll grow because of all the inherent distribution. But it’s kind of interesting to say that they’re not that many. In fact, we talk a lot more about AI scale, but there is not that many hit apps. There is ChatGPT, Github Co-Pilot, there’s Co-Pilot, and there’s Gemini. I think those are the four I would say, in a DAU, is there anything else that comes to your mind?…

…Gurley: Satya, on the enterprise side, obviously the coding space is off to the races and you guys are doing well and there’s a lot of venture-backed players there. On some of the productivity apps, I have a question about the the Co-Pilot approach and I guess Marc Benioff’s been obnoxiously critical on this front, calling it Clippy 2 or whatever. Do you worry that someone might think first-principles AI from ground-up, and that some of the infrastructure, say in an Excel spreadsheet, isn’t necessary to know if you did a AI-first product. The same thing by the way could be said about the CRM right? There’s a bunch of fields and tasks that that may be able to be obfuscated for the user.

Nadella: It’s a very, very, very important question. The SaaS applications or biz apps, let me just speak of our own Dynamics thing. The approach at least we’re taking is, I think the notion that business applications exist, that’s probably where they’ll all collapse in the agent era. Because if you think about it, they are essentially CRUD databases with a bunch of business logic. The business logic is all going to these agents, and these agents are going to be multi-repo CRUD. They’re not going to discriminate between what the back-end is, they’re going to update multiple databases, and all the logic will be in the AI tier so to speak. Once the AI tier becomes the place where all the logic is, then people will start replacing the backends right? In fact it’s interesting, as we speak, I think we are seeing pretty high rates of wins on Dynamics backends and the agent use, an we are going to go pretty aggressively and try and collapse it all, whether it’s in customer service, whether it is in… 

By the way, the other fascinating thing that’s increasing is just not CRM, but even what we call finance and operations, because people want more AI-native biz app. That means the biz app, the logic tier, can be orchestrated by AI and AI agents. So in other words, Co-Pilot to agent to my business application should be very seamless.

Now in the same way, you could even say, “Why do I need Excel?” Interestingly enough, one of the most exciting things for me is Excel with Python, is like GitHub with Co-Pilot. So what we’ve done is, when you have Excel – by the way this would be fun for you guys – which is you should just bring up Excel, bring up Co-Pilot, and start playing with it. Because it’s no longer like – it is like having a data analyst, so it’s no longer just making sense of the numbers that you have. It will do the plan for you. It will literally – like how GitHub Co-Pilot Workspace creates the plan and then it executes the plan – this is like a data analyst who is using Excel as a sort of row/column visualization to do analysis scratch pad. So it kind of tools you. So the Co-Pilot is using Excel as a tool with all of its action space because it can generate and it has python interpreter. That is in fact a great way to reconceptualize Excel. At some point you could say, “I’ll generate all of Excel” and that is also true. After all, there’s a code interpreter, so therefore you can generate anything.\

So yes, I think there will be disruption. The way we are approaching, at least our M365 stuff is, one is build Co-Pilot as that organizing layer UI for AI, get all agents, including our own agents – you can say Excel is an agent to my Co-Pilot, Word is an agent, it’s kind of a specialized canvases, which is I’m doing a legal document, let me take it into Pages and then to Word and then have the Co-Pilot go with it, go into Excel and have the Co-Pilot go with it. That’s sort of a new way to think about the work in workflow…

…Gurley: Satya, there’s been a lot of talk about model scaling and obviously there was talk, historically about 10x-ing the cluster size that you might do, over and over again, not once and then twice. X.AI is still making noise about going in that direction. There was a podcast recently where they flipped everything on their head and they said “If we’re not doing that anymore, it’s way better because we can just move on to inference which is getting cheaper and you won’t have to spend all this capex. I’m curious, those are two views of the same coin. But what’s your view on LLM model scaling and training cost, and where we’re headed in the future?

Nadella: I’m a big believer in scaling laws I’ll first say. In fact, if anything, the bet we placed in 2019 was on scaling laws and I stay on that. In other words, don’t bet against scaling laws. But at the same time, let’s also be grounded on a couple of different things.

One is these exponentials on scaling laws will become harder, just because as the clusters become harder, the distributed computing problem of doing large scale training becomes harder. That’s one side of it. But I would just still say – and I’ll let the OpenAI folks speak for what they’re doing – but they are continuing to – pre-training I think is not over, it continues. But the exciting thing, which again OpenAI has talked about and Sam has talked about, is what they’ve done with o1. This Chain of Thought with autograding is just a fantastic. In fact, basically, it is test-time compute or inference-time compute as an another scaling law. You have pre-training, and then you have effectively this test-time sampling that then creates the tokens that can go back into pre-training, creating even more powerful models that then are running on your inference. So therefore, that’s I think a fantastic way to increase model capability.

The good news of test-time or inference-time compute is sometimes, running of those o1 models means… There’s two separate things. Sampling is like training, when you’re using it to generate tokens for your pre-training. But also customers, when they are using o1, they’re using more of your meters, so you are getting paid for it. Therefore, there is more of an economic model, so I like it. In fact, that’s where I said I have a good structural position with 60-plus data centers all over the world.

Gurley: It’s a different hardware architecture for one of those scaling versus the other, for the pre-training versus…

Nadella: Exactly. I think the best way to think about it is, it’s a ratio. Going back to Brad’s thing about ROIC, this is where I think you have to really establish a stable state. In fact, whenever I’ve talked to Jensen, I think he’s got it right, which is you want to buy some every year. Think about it, when you depreciate something over 6 years, the best way is what we have always done, which is you buy a little every year and you age it, you age it, you age it. You use the leading node for training and then the next year it goes into inference, and that’s sort of the stable state I think we will get into across the fleet for both utilization and the ROIC and then the demand meets supply.

Basically, to your point about everybody saying, “Have the exponentials stopped?” One of the other things is the economic realities will also stop, right? At some point everybody will look and say, “What’s the economically rational thing to do?” Which is, “Even if I double every year’s capability but I’m not able to sell that inventory,” and the other problem is the Winner’s Curse, which is – you don’t even have to publish a paper, the other folks have to just look at your capability and do either a distillation… It’s like piracy. You can sign off all kinds of terms of use, but like it’s impossible to control distillation. That’s one. Second thing is, you don’t even have to do anything, you just have to reverse engineer that capability and you do it in a more computer efficient way. So given all this, I think there will be a governor on how much people will chase. Right now a little bit of everybody wants to be first. It’s great, but at some point all the economic reality will set in on everyone and the network effects are at the app layer, so why would I want to spend a lot on some model capability with the network effects are all on the app?…

…Gurley: Does your answer to Brad’s question about the balancing of GPU ROI, does that answer the question as to why you’ve outsourced some of the infrastructure to Coreweave in that partnership that you have?

Nadella: That we did because we all got caught with the hit called ChatGPT. It was impossible. There’s no supply chain planning I could have done. None of us knew what was going to happen. What happened in November of ‘22, that was just a bolt from the blue, therefore we had to catch up. So we said, “We’re not going to worry about too much inefficiency.” That’s why whether it’s Coreweave or many others – we bought all over the place. That is a one time thing and then now it’s all catching up. That was just more about trying to get caught up with demand.

Gerstner: Are you still supply-constrained Satya?

Nadella: Power, yes. I am not chip supply-constrained. We were definitely constrained in ‘24. What we have told the street is that’s why we are optimistic about the first half of ‘25, which is the rest of our fiscal year and then after that I think we’ll be in better shape going into ‘26 and so on. We have good line of sight.

---

Disclaimer: The Good Investors is the personal investing blog of two simple guys who are passionate about educating Singaporeans about stock market investing. By using this Site, you specifically agree that none of the information provided constitutes financial, investment, or other professional advice. It is only intended to provide education. Speak with a professional before making important decisions about your money, your professional life, or even your personal life. We currently have a vested interest in Alphabet (parent of Google), Amazon, Meta Platforms (parent of Facebook), and Microsoft. Holdings are subject to change at any time.

We’ve constantly been sharing a list of our recent reads in our weekly emails for The Good Investors.

Do subscribe for our weekly updates through the orange box in the blog (it’s on the side if you’re using a computer, and all the way at the bottom if you’re using mobile) – it’s free!

But since our readership-audience for The Good Investors is wider than our subscriber base, we think sharing the reading list regularly on the blog itself can benefit even more people. The articles we share touch on a wide range of topics, including investing, business, and the world in general. 

Here are the articles for the week ending 15 December 2024:

1. SpaceX: Rocket Ship – Matt Reustle and Luke Ward

Luke

So if we take the CapEx part of that first, NASA estimated that the cost to develop the Falcon 9 from scratch would be about $4 billion. But SpaceX ended up doing it for about a tenth of that price. So to begin with, that’s an order of magnitude improvement in the level of investment required.

SpaceX gives you the prices for launches on their website. So $70 million per launch of a Falcon 9 flight—that’s already 20 times cheaper than the Space Shuttle was per kilogram into orbit. But the real kicker, as you point out, is the operating leverage that comes from having partial reusability…

…Starship is designed to be fully and rapidly reusable. So unlike Falcon 9, which is only partially reusable but also able to fly multiple times every day, it’s going to have a payload capacity that’s about 100 tons to orbit at the beginning, but probably rising to closer to 200 tons to orbit over time.

And Musk has suggested that a variable cost of around $10 million per launch is the ballpark figure which they’d be aiming for at scale in a steady state, ambitiously maybe even falling to $2 million—a figure which has been touted. If you believe those kinds of performance levels are feasible, that gets the cost down to around $10 per kilogram. That’s over 100 times cheaper than the Falcon 9 we’re talking about at the moment. And that would have a dramatic effect on what’s economically feasible for humanity to do in space…

…Matt

Satellites in Low Earth Orbit—there is quite a bit of history in terms of that being the obvious space use case, that having an existing economy. I think Starlink is an extension of that. Different, absolutely, but an extension of what was going on.

Are there brand new industries being unlocked or obvious things with line of sight that open up from a space economy perspective that you see either today or, when I say near future, you could extend that out however far you think is reasonable.

Luke

A lot of these options which SpaceX has to develop, brand new markets that don’t exist already, are a function ultimately of the cost curve. Take semiconductor manufacturing on Earth; at the moment, we spend billions of dollars per fab to recreate the conditions which are readily accessible in space for free, if you can get there.

And so there’s some point on the cost curve intersecting between the cost of building a fab and the cost of launching a fab or the equipment of a fab into orbit and operating there instead. Same can be said of pharmaceutical research. The crystallization structures which are able to happen in space are different from the ones which are able to happen under the influence of gravity.

So if you think about pricing on pharmaceuticals, extending patent lives, etc., if you can move the manufacturing or the research lab for cutting-edge pharmaceuticals into space, you could make high-value, low-volume products. Something which would really make sense to do and doesn’t require a huge technological innovation to happen.

The list can go on and on—artificial organs, for example, being able to manufacture perfectly spherical lenses. There’s lots and lots of things which could be made.

Maybe the way to think about that is that space-based manufacturing could be the next large market for this if the costs can continue to come down. Starship having the volume of an A380 or a 747—think of the equivalent size of factory that represents. And if that can be launched every single day and recovered every single day for $10 per kilogram, that could be a really compelling way to do quite a lot of manufacturing.

Incidentally, that’s something that Jeff Bezos really focuses on in his vision for space as opposed to Mars per se, is where we can move a lot of the heavy-polluting industry off the planet. And why don’t we turn Earth into this perfect nature reserve, and all these polluting aspects of manufacturing can go into orbit, which again is very compelling.

Probably needs a lot more innovation to deliver communications from orbit, but I’d say it’s maybe an inevitability if the cost gets to a low enough point. You think how much solar energy is available without the atmospheric attenuation, for example—you know, 24/7. There’s lots of compelling reasons why if it’s cheap enough, at some point a lot of these things probably should happen, not just could happen.

Matt

The solar energy point, great example of something that is an entirely different dynamic in space than on Earth. What would the other things be? Just out of curiosity, when you mentioned semiconductors or pharmaceuticals, is it just purely gravity? Are there other things that are happening in space or not happening in space that happen on Earth that would drive that difference?

Luke

There’s the vacuum conditions—so there isn’t an atmosphere—so the level of impurities which you need to get rid of for a vapor deposition machine, for example. You don’t have the same kind of challenges there of having to have this deep vacuum.

Then, arguably, in space, because you don’t have gravity, you could construct much larger structures there rather than construct them on the ground and then launch them.

So again, that volume constraint which we were talking about earlier, in terms of how big your payload is—if you’re able to get enough stuff up there and assemble it in space, as we did with the International Space Station, things can be much, much larger given the payload bay of Starship than they could with the Space Shuttle.

Matt

When you think about low Earth orbit versus geosynchronous orbit versus something like Mars—which I think was the original vision with Elon and SpaceX—how much does that change the economics when you extend out?

Is it orders of magnitude where it’s an exponential cost curve to go further out? Even just if we focus on the launch and use a satellite for an example, before we get into all the manufacturing dynamics, is there any way to contextualize that from a cost perspective?

Luke

The really good news here is that gravitational force decreases with the square of distance. So the biggest challenge is getting off the surface and into orbit. Once you’re there, from an energy point of view, it’s a lot easier to go anywhere else in the solar system.

So if you were to take Falcon 9 again as the example, for the same price, it can place 20 tons into low Earth orbit, or it can place 4 tons into Martian orbit. That’s despite the latter being over a million times further away. Now, this feeds into what I think is probably the biggest misconception about SpaceX and its Mars ambitions.

I’d say for most people, the idea of a commercial entity pursuing exploration is naive at best. But I’d argue that long-term investors should be absolutely ecstatic about SpaceX having this mission as a forcing function. Firstly, it’s the key to getting the best people in the world to come and work for the organization and allow it to innovate in a manner and speed that others simply can’t match. That’s a huge competitive advantage.

Secondly, the way to get more cargo to Mars is actually about figuring out how to get more cargo into orbit around Earth, because that’s where the cost is all concentrated. It’s all in that first initial leap off the surface of our planet. So rather than framing Starship as a system that makes it possible to get to other planets, think about it instead being a system that could make it enormously more profitable to operate a business in Earth orbit and unlock brand new commercial use cases there as well…

…Luke

When we talk to SpaceX, they’re still very much focused on the here and now in the next couple of years. They have ambitions for things which they could do, but the focus is very much on the core business: serving the core customers, serving Starlink, getting Starship to launch status. We’ll deal with the next things next.

They’ve got so many things which they could be doing at the moment. When we come to this, a lot of this is us hypothesizing of how that could evolve beyond information which they’ve given us. The trend which you’ve seen of them to be vertical integrators could be quite informative. It might be that they end up being the ones who are commercializing a lot of these other services.

Rather than having a customer paying them for it at substantial scale, it would make more sense for them to do it. Could you start seeing some of these aspects? If they get into space-based manufacturing, for example, could that be priced on a value-added basis rather than a subscription basis or a volume basis? Certainly seems possible. If you start running data centers in space because it’s easier to power or cool them, etc., could you start offering data storage and machine learning alongside Starlink connectivity?

The further you look out, the more and more wacky it can get, but it’s also potentially financially plausible as well. You maybe have to take a bit of inspiration from science fiction here, but it’s quite a common trope in some of these movies of these large mega-corporations—the Weyland-Yutani Corporation from the Alien movies, or the Resources Development Administration from the Avatar films—where one mega-corporation was able to dominate access to space early on and then ends up controlling the entire extrasolar economy because of the advantages it had at that really early stage…

…Luke

The human spaceflight at the moment definitely has been the preserve of the rich and famous, but at scale that becomes cheaper and cheaper. And if we are talking about launching, Starship could be used as much for sending cargo and people to other points on the planet rather than other points in space. And so one option that the government’s looking into is this notion of rocket cargo delivery. Starship would be able to deliver 200,000 kg anywhere on the planet within 40 minutes.

What does that do for sort of a rapid reaction force, and what does that do for next-day delivery? At some stage, it’s going to be feasible to put a lot of astronauts or paying passengers on something like that, and it will be a quicker and potentially more efficient way to do long-distance travel. These things really could get quite wild, but it could be plausible at some stage. Again, that’s not the reason to invest in the company today; that’s not the basis of what they’re doing, and it’s a lot of people getting excited about things.

But come back in 10 years, I’d be disappointed if you or I weren’t able to go into space at some point in our lifetime for the cost of a premium economy ticket or something like that.

2. Japan vs Big Tech – Daye Deng

Put simply, US big tech has grown so dominant that it’s singlehandedly blowing a hole in the trade balance of a nation as large as Japan…

…In 2023, Japan recorded JPY 5.5 trillion in so-called digital trade deficit. The Ministry of International Trade and Industry (MITI) projects this to grow to JPY 8 trillion by 2030, at which point it could surpass Japan’s annual import of crude oil.

Japan’s total goods and services trade deficit in 2023 was JPY 6 trillion, with the digital deficit accounting for JPY 5.5 trillion…

…Japan has been in a structural deficit for goods trade over the past two decades. This may come as a surprise to those who have held onto the old idea that Japan is an export powerhouse.

There are several reasons for the shift:

• Japanese firms have moved production overseas. This isn’t entirely negative since Japanese firms (and their profits) continue to grow, but it has contributed to a widening trade deficit.
• Japan’s loss of global competitiveness in certain industries, like chips and appliances, to rivals such as South Korea.
• Rising cost of imports driven by energy shocks, rising overseas inflation, and weak yen.

The third point deserves elaboration. Japan’s reliance on imported energy has long been a critical structural weakness. For example, following 2011 Fukushima nuclear disaster, Japan significantly reduced domestic nuclear energy production and increased its reliance on imported LNG, becoming a major contributor to trade deficit.

A similar pattern emerged post-Covid. Global oil and commodity prices surged. This was compounded by high rates of overseas inflation on general imports. On top, a historically weak yen made imports even more expensive…

…Since 2014, the Japanese government has been disclosing the digital deficit, which has grown 2.6-fold from 2014 to JPY 5.5 trillion in 2023. This is a net figure derived from JPY 9.2 trillion paid for digital services and JPY 3.7 trillion received from abroad…

…The picture is quite clear: on the services side, Japan is taking its hard-earned surplus from tourism and spending it all on paying for digital services.

How will this play out? While I’m personally bullish on the Japanese tourism industry, it still has natural growth constraints. However, there is no ceiling on how much Japan can continue to spend on digital services. In fact, digital services spend could accelerate given:

• Japan is already playing catch-up in the digital realm, and is behind other major countries in many key digital metrics.
• AI is poised to make Japan’s digital dependency crisis even worse, in a world where firms like Nvidia and those that are able to scale AI services (e.g. hyperscalers) dominate AI economics.

Without an AI champion of its own, Japan has few options if it wants to avoid being left behind in the new digital paradigm…

…Based on our discussion so far, does it surprise you that the Japanese yen has been weak?

“According to an analysis by Mizuho Research & Technologies, if the digital deficit doubles from the 2023 level by the end of March 2026, it will add another 5 to 6 yen of depreciation in the Japanese currency’s value against the dollar.”

– Nikkei Asian Review

Or let me put it another way — would you feel bullish about the currency of a country that relies on tourism as its primary growing surplus, while ultimately funneling all those earnings (and more) into paying for essential energy imports and ever-increasing digital spend on big tech?…

…In recent years we’ve seen how hard Japan has been trying to reclaim its position in the semiconductor industry. But do they only care about hardware and not its digital sovereignty? Will Japan continue to sit back and let US tech giants profit endlessly, or will it finally confront its position as a digital colony?

3. Guyana and the mystery of the largest ranch in the Americas – Swen Lorenz

Many mistakenly believe that Guyana is located in Africa – when it’s actually nestled right next to Venezuela…

…In 2015, ExxonMobil discovered oil off the coast of Guyana.

The discovery changed the course of the country. Long one of the poorest nations of the Western hemisphere, Guyana has since become the world’s fastest growing economy.

Since 2015, its GDP per capita has more than quintupled. In 2022 and 2023, its economy grew by 67% and 33%, respectively. Another stunner of a year is forecast for 2024, with 34% GDP growth.

The former British colony benefits from a large amount of oil wealth spread around a relatively small population of 800,000 people. Per head, there is twice as much oil as in Saudi Arabia. To put things in perspective, Guyana’s landmass is nearly as big as the UK, but it only has 1.2% of the UK’s population…

…Just a week ago, ExxonMobil reported that it had reached 500m barrels of oil produced in Guyana since output began in 2019. The goal is to lift production to 1.3m barrels per day by 2027, up from currently 650,000 barrels. In comparison, the UK’s North Sea produces just 1m barrels per day…

…Supporters of the country’s energy projects claim that they will bring untold riches to the population. Indeed, Guyana recently started to hand out cheques to its citizens, including the Guyanese diaspora of 400,000 people, who the government encourages to come back as it needs more labour to support the strong economic growth.

4. Capital, Compute & AI Scaling – Patrick O’Shaughnessy, Chetan Puttagunta, and Modest Proposal

Modest

Everyone knows the Mag 7 represent a larger percent of the S&P 500 today. But beyond that, I think thematically AI has permeated far broader into industrials, into utilities and really makes up, I would argue, somewhere between 40 and 45% of the market cap as a direct play on this. And if you even abstract to the rest of the world, you start bringing in ASML, you bring in TSMC, you bring in the entire Japanese chip sector. And so if you look at the cumulative market cap that is a direct play on artificial intelligence right now, it’s enormous…

… I think at the micro level this is a really powerful shift if we move from pre-training to inference time and there are a couple big ramifications.

One, it better aligns revenue generation and expenditures. I think that is a really, really beneficial outcome for the industry at large, which is in the pre-training world you were going to spend 20, 30, $40 billion on CapEx, train the model over 9 to 12 months, do post-training, then roll it out, then hope to generate revenue off of that in inference. In a test time compute scaling world you are now aligning your expenditures with the underlying usage of the model. So just from a pure efficiency and scalability on a financial side, this is much, much better for the hyperscalers.

I think a second big implication, again we have to say we don’t know that pre-training scaling is going to stop. But if you do see this shift towards inference time, I think that you need to start to think about how do you re-architect the network design? Do you need million chip super clusters in energy low-cost land locations or do you need smaller, lower-latency, more efficient inference-time data centers scattered throughout the country? And as you re-architect the network, the implications on power utilization, grid design?

A lot of the, I would say, narratives that have underpinned huge swaths of the investment world I think have to be rethought and I would say today because this is a relatively new phenomenon, I don’t believe that the public markets have started to grapple with what that potential new architecture looks like and how that may impact some of the underlying spend…

… Chetan

But at the moment, at this plateauing time, we’re starting to see these small teams catch up to the frontier. And what I mean by frontier is where are the state-of-the-art models, especially around text performing? We’re seeing these small teams of quite literally two to five people jumping to the frontier with spend that is not one order, but multiple orders of magnitude less than what these large labs were spending to get there.

I think part of what’s happened is the incredible proliferation of open-source models. Specifically, what Meta’s been doing with LLaMA has been an extraordinary force here. LLaMA 3.1 comes in three flavors, 405 billion, 70 billion, 8 billion. And then LLaMA 3.2 comes in 1 billion, 3 billion, 11 billion, and 90 billion.

And you can take these models, download them, put them on a local machine, you can put them in a cloud, you can put them on a server, and you can use these models to distill, fine-tune, train on top of, modify, et cetera, et cetera, and catch up to the frontier with pretty interesting algorithmic techniques.

And because you don’t need massive amounts of compute, or you don’t need massive amounts of data, you could be particularly clever and innovative about a specific vertical space, or a specific technique, or a particular use case to jump to the frontier very, very quickly…

…Chetan

The force of Llama today has been two things, and I think this has been very beneficial to Meta is one. The transformer architecture that Llama is using is a sort of standard architecture, but it has its own nuances.

And if the entire developer ecosystem that’s building on top of Llama is starting to just assume that that Llama 3 transformer architecture is the foundational and sort of standard way of doing things, it’s sort of standardizing the entire stack towards this Llama way of thinking, all the way from how the hardware vendors will support your training runs to the hyperscalers and on and on and on. And so standardizing on Llama itself is starting to become more and more prevalent.

And so if you were to start a new model company, what ends up happening is starting with Llama today is not only great because Llama is open source, it’s also extraordinarily efficient because the entire ecosystem is standardizing on that architecture…

…Modest

So I think the interesting part for OpenAI was because they just raised the recent round and there was some fairly public commentary around what the investment case was. You’re right, a lot of it oriented around the idea that they had escape velocity on the consumer side and that ChatGPT was now the cognitive reference and that over time they would be able to aggregate an enormous consumer demand side and charge appropriately for that and that it was much less a play on the enterprise API and application building.

And that’s super interesting if you actually play out what we’ve talked about when you look at their financials, if you take out training runs, if you take out the need for this massive upfront expenditure, this actually becomes a wildly profitable company quite quickly in their projections. And so in a sense it could be better.

Now then the question becomes what’s the defensibility of a company that is no longer step function advancing on the frontier?…

…Chetan

These products are truly, as a software investor, absolutely amazing.

They require a total rethinking from first principles on how these things are architected. You need unified data layers, you need new infrastructure, you need new UI and all this kind of stuff. And it’s clear that the startups are significantly advantaged against incumbent software vendors. And it’s not that the incumbent software vendors are standing still, it’s just that innovator’s dilemma in enterprise software is playing out much more aggressively in front of our eyes today than it is in consumer.

I think in consumer, the consumer players recognize it, are moving it, and are doing stuff about it. Whereas I think in enterprise, even if you recognize it, even if you have the desire to do something, the solutions are just not built in a way that is responsive to dramatic re-architecture. Now could we see this happening? Could a giant SaaS company just pause selling for two years and completely re-architect their application stack?

Sure, but I just don’t see that happening. And so if you just look at any sort of analysis on what’s happening on AI software spend, something like it’s 8x year-over-year growth between 2023 and 2024 on just pure spend. It’s gone from a couple of hundred million dollars to well over a billion in just a year’s time…

…Modest

If you listen to AWS, one of the fascinating things they say is they call AWS a logistics business.

I don’t think anyone externally would sort of look at cloud computing and say, oh yeah, that’s a logistics business. But their point is essentially what they have to do is they have to forecast demand and they have to build supply on a multi-year basis to accommodate it.

And over 20 years they’ve gotten extraordinarily good at what has happened in the last two years, and I talked about this last time, is you have had an enormous surge in demand hitting inelastic supply because you can’t build data center capacity in three weeks. And so if you get back to a more predictable cadence of demand where they can look at it and say, okay, we know now where the revenue generation is coming from.

It’s coming from test time, it’s coming from Chetan and his companies rolling out. Now we know how to align supply with that. Now it’s back to a logistics business. Now it’s not grab every mothballed nuclear site in the country and try to bring it online.

And so instead of this land grab, I think you get a more reasonable, sensible, methodical rollout of it maybe. And I actually would guess that if this path is right, that inference overtakes training much faster than we thought and gets much bigger than we may have suspected.

But I think the path there in the network design is going to look very different and it’s going to have very big ramifications for the people who were building the network, who were powering the network, who were sending the optical signals through the network. And all of that, I think, has not really started to come up in the probability-weighted distributions of a huge chunk of the public market.

And look, I think most people overly fixate on NVIDIA because they are sort of the poster child of this, but there are a lot of people downstream from NVIDIA that will probably suffer more because they have inferior businesses. NVIDIA is a wonderful business doing wonderful things. They just happen to have seen the largest surge in surplus. I think that there are ramifications far, far beyond who is making the bleeding edge GPU, even though I do think there will be questions about, okay, does this new paradigm of test time compute allow for customization at the chip level much more than it would have if we were only scaling on pre-train…

…Modest

If you think about a training exercise, you’re trying to utilize them at the highest possible percent for a long period of time. So you’re trying to put 50, 100,000 chips in a single location and utilize them at the highest rate possible for nine months. What’s left behind is a hundred thousand chip cluster that if you were to repurpose for inferencing is arguably not the most efficient build because inference is peaky and bursty and not consistent.

And so this is what I’m talking about that I just think from first principles you are going to rethink how you want to build your infrastructure to service a much more inference focused world than a training focused world. And Jensen has talked about the beauty of NVIDIA is that you leave behind this in place infrastructure that can then be utilized.

And in a sunk cost world you say, sure, of course if I’m forced to build a million chip supercluster in order to train a $50 billion model, I might as well sweat the asset when I’m done. But from first principles it seems clear you would never build a 350,000 chip cluster with 2 1/2 gigawatts of power in order to service the type of request that Chetan’s talking about.

And so if you end up with much more edge computing with low latency and high efficiency, what does that mean for optical networking? What does that mean for the grid? What does that mean for the need for on site power versus the ability to draw from the local utility?…

…Chetan

Semiconductor company called Cerebras, and they recently announced that inference on Llama 3.1 405 billion for Cerebras is it can generate 900-plus tokens per second, which is a dramatic order-of-magnitude increase. I think it’s like 70 or 75 times faster than GPUs for inference as an example. And so as we move to the inference world, the semiconductor layer, the networking layer, et cetera, there’s tons of opportunities for startups to really differentiate themselves…

…Modest

On a less sort of dramatic view, the way I think about this, there’s AlphaGo, which famously did that move that no one had ever seen, and I think it’s like move 37, everybody was super confused about, ended up winning. And another example I love is Noam Brown, because I like poker, talked about his poker bot confused—it was playing high stakes, no limit, and it continually over-bet dramatically larger sizes than pros had ever seen before.

And he thought the bot was making a mistake. And ultimately it destabilized the pros so much. Think about that. A computer destabilized humans in their approach that they have to some extent taken on over-betting now into their game.

And so those are two examples where if we think about pre-training being bounded by the data set that we’ve given it, if we don’t have synthetic data generation capabilities, here you have two examples where algorithms did something outside of the bounds of human knowledge. And that’s what’s always been confusing to me about this idea that LLMs on their own could get to superintelligence, is functionally they’re bounded by the amount of data we give them up front.

5. Will China Take Over the Global Auto Industry? – Brad Setser

China has, according to the New York Times, the capacity to produce over 40 million internal combustion engine (ICE) cars a year.

Goldman Sachs thinks China will also have the capacity to produce around 20 million electric vehicles by the end of 2024…

…China’s internal market is around 25 million cars, and not really growing —so rising domestic EV sales progressively frees up internal combustion engine capacity for export.   Domestic demand for traditional cars is likely to be well under 10 million cars next year given the enormous shift toward EVs now underway inside China…

…Historically, the autos market has been largely regional (setting aside trade in luxury cars, where volumes are smaller). Most cars sold in China were made in China, most cars sold in Europe are produced in Europe, most cars sold in the North America are produced in North America, and so on. The U.S. did import a few million cars, on net, from Asia, and China imported a million or so luxury cars from Europe, but those were the exceptions rather than the rule.

That could change, absent hefty restrictions on Chinese auto imports (like the 100 percent tariff the U.S. now levies on EVs imported from China).

The global market—with massive overcapacity in China’s internal combustion engine (ICE) sector, massive capacity expansion in China’s EV sector, effectively unlimited credit for Chinese manufacturing firms from China’s state banks, and a Chinese yuan that is weaker against the dollar than it was back in 2008—is pushing for global auto manufacturing to become more like global electronics manufacturing, with a concentration of global production in a single region and, for that matter, a single country…

…Overcapacity in China’s automotive sector is not, in fact, all that new.

China’s traditional automotive sector was dominated by the joint ventures (“JVs”) formed by the large foreign firms and their (typically state-owned) Chinese partners. Chinese auto demand took off after the global financial crisis, and global firms responded by massively expanding their Chinese production capacity – as only the German luxury markets were interested in paying the 25 percent tariff and supplying the Chinese market from abroad.

But demand growth eventually slowed, and by 2018, the Wall Street Journal was reporting that the Chinese market was oversupplied…

…China’s EV industry—like EV industries in the U.S. and Europe—initially received substantial state backing. Chinese EV manufactures benefitted from downstream subsidies that built out China’s battery and battery chemical industry, as well as access to the world’s cheapest steel.
EV firms benefitted from cheap state financing—both equity injections from a myriad of state-backed funds and loans from state banks who (still) have to meeting lending quotas.

Moreover, China was quite explicitly protectionist in the application of its “consumer” EV subsidies.

Only EVs that were on state lists of qualifying vehicles were eligible for the subsidy, and the subsidy was only provided to cars that were made in China…

…And initially, only cars that were made in China with a battery made in China by a Chinese firm qualified for the lists…

…The only exception to the basic rule that qualifying for the list required using a battery made in China by a Chinese firm only confirmed the broad pattern of discrimination: Chinese-owned Volvo was allowed to use a Korean battery in one of its early EVs.

State support has not disappeared in any way as China’s EV industry took off.   Looking at direct cash subsidies from the central government to the manufacturers misses the myriad of ways China, Inc helps out firms producing in China…

…Nio received a significant ($1.9 billion) equity investment from the City of Hefei and the Province of Anhui, helping to offset ongoing losses. That equity injection was on top of state support for a factory in Hefei, which The New York Times reports was effectively a gift from the local government.

“‘The local government provided the land and the building’, said Ji Huaqiang, Nio’s vice president for manufacturing. ‘Nio does not own the factory or the land — it is renting, but the factory was custom built for Nio’”

That kind of support explains how Nio managed to build out its EV capacity even when its existing factories weren’t really being used that much:

“Nio’s two factories give it the capacity to assemble 600,000 cars a year, even though its annual rate of sales this autumn [2023] is only about 200,000 cars. Nio is nonetheless already building a third plant.”…

...What’s even more striking is that the investments that built out China’s EV capacity came in a market that was already saturated with modern auto production capacity.  That kind of investment wouldn’t have taken place without state guidance and support, support that was intended both to develop an indigenous Chinese industry (See China 2025) and to support a green transition that would reduce Chinese dependence on import fossil energy. It was the result of policy driven by the central government and backed financially by all levels of government. It also worked, China is now the world leader in EVs and batteries…

…If the world’s global firms can only compete with Chinese firms by using Chinese batteries and Chinese parts, that will hollow out much of the automotive industries of Europe and North America—a European brand on a Chinese-made car with a Chinese battery and drive train won’t sustain the current European auto supply chain or current European employment in the auto industry.

---

Disclaimer: The Good Investors is the personal investing blog of two simple guys who are passionate about educating Singaporeans about stock market investing. By using this Site, you specifically agree that none of the information provided constitutes financial, investment, or other professional advice. It is only intended to provide education. Speak with a professional before making important decisions about your money, your professional life, or even your personal life. We currently have a vested interest in ASML, Meta, and TSMC. Holdings are subject to change at any time.

We’ve constantly been sharing a list of our recent reads in our weekly emails for The Good Investors.

Do subscribe for our weekly updates through the orange box in the blog (it’s on the side if you’re using a computer, and all the way at the bottom if you’re using mobile) – it’s free!

But since our readership-audience for The Good Investors is wider than our subscriber base, we think sharing the reading list regularly on the blog itself can benefit even more people. The articles we share touch on a wide range of topics, including investing, business, and the world in general. 

Here are the articles for the week ending 03 November 2024:

1. We Need to Have a Talk About “Bond Vigilantes” – Cullen Roche

In May the 10 year interest rate was as high as 4.75%. By September it was as low as 3.6%. Today it’s bounced back to 4.2%. And as rates tick higher in recent months there’s been a growing chorus about how “bond vigilantes” are going to teach the Fed a lesson. This has been especially loud coming from Stanley Druckenmiller, Paul Tudor Jones and Elon Musk. The basic narrative is that bond markets will teach the Fed and US government a lesson about reckless spending which will drive up interest rates and bankrupt the USA. Except there’s a huge problem in this narrative – the bond market has a lot less control over interest rates than this story would have you believe…

…Individuals go bankrupt. Large aggregated sectors do not. For example, the aggregated private sector cannot go bankrupt. The US government is a huge aggregated sector in the US economy. It does not go bankrupt. In the aggregate it can print money to fund all spending and it cannot run out of this money unless it borrows in a foreign currency, which it doesn’t do. Of course, it can cause wildly huge inflation. We know that after Covid, but comparing the Federal Government to an individual is just wrong…

…The Fed is the reserve monopolist. So, as I’ve explained in the past, they have absolute control over something like short-term interest rates. The market quite literally cannot force them to change this rate because the market cannot compete with the Fed. If a bank tries to set the short-term interest rate the Fed just comes in and smashes them with their bottomless pit of money…

…The best way to understand the yield curve is to think of the Fed as a dog walker who has absolute control of the leash at the handle and allows the dog to wander further out the leash. The Fed has absolute control over the handle (the Fed Funds Rate) and lets the 30 year wander from side to side, but still within a certain control. It might look like the dog is walking the Fed, but the Fed always has the ability to pull that leash in and grab that dog by the neck. This is what “yield curve control” would look like and if the Fed entered the market for 30 year bonds and started explicitly setting a target they could drive that rate to whatever they wanted. But they let it float a bit…

…I wrote nearly this exact same article 11 years ago in response to a WSJ op-ed in which Druckenmiller said the exact same thing. He said the USA was bankrupt and that the Fed was too loose…

… In my humble opinion, the error in this analysis is two-fold:

1. Assuming that interest payments are problematic – they are not because the Fed can control them with a dial and also because high rates put DOWNWARD pressure on inflation.
2. Assuming high inflation must result from government deficits. I think it’s absolutely true that large deficits put upward pressure on inflation. I’ve said this a billion times during Covid. But government spending is 23% of GDP. That’s the same level it was at in 1982! And while it’s a large portion of aggregate spending we should remember that 77% of spending is coming from OTHER sources. In most cases, it’s much more efficient sources such as the most efficient corporate machine the human race has ever seen (corporate America).

2. You’re Not Paranoid. The Market Is Out to Get You – Jason Zweig

Graham wasn’t only one of the best investors of all time; he may have been the wisest. His intellectual brilliance, six decades of investing and study of history gave him a profound understanding of human nature.

As he wrote: “The investor’s chief problem—and even his worst enemy—is likely to be himself.”…

…To be an intelligent investor doesn’t require a stratospheric IQ. It does require discipline and the ability to think for yourself.

As Graham pointed out, individual investors are “scarcely ever” forced to sell stocks or funds and—unlike professional portfolio managers who are continually measured against the market—are never compelled to care what other investors are doing.

That independence is your single most valuable asset, a luxury most professional investors can only dream of possessing. It’s what Graham called the “basic advantage” of the intelligent investor. But, he warned, “the investor who permits himself to be stampeded [by other people’s behavior]…is perversely transforming his basic advantage into a basic disadvantage.”

As I argue in the new edition of the book, it has never been harder to be a disciplined and independent investor. In today’s incessantly twitchy, infinitely networked markets, the siren song of smartphones, social media and streaming video can tempt you to trade more and copy the crowd.

After all, it often makes sense—and just feels right—to join the herd…

…Yet crowds aren’t always right, and their errors are contagious. What separates the wisdom from the madness of the crowd?

In 1907, the statistician Francis Galton described a contest at an agricultural fair in which nearly 800 visitors tried to guess the weight of an ox. Although many knew little or nothing about oxen and their guesses varied widely, their average estimate turned out to match the weight of the ox exactly.

Galton’s guessers had a variety of viewpoints, sought to win a prize for accuracy, didn’t know other people’s estimates and had to pay an entry fee. The sponsors of the contest collected and tallied all the guesses.

The judgments of that crowd were independent, confidential, diverse, incentivized and aggregated—and, therefore, remarkably accurate at estimating simple values.

But the judgments of today’s crowds are often the opposite of Galton’s…

…The weight of an ox doesn’t change with people’s estimates of it. However, if thousands of speculators decide a stock or cryptocurrency is worth $100,000, it will skyrocket—at least temporarily—even if it’s worthless.

Joining the crowd can change how you think, no matter how much you pride yourself on your independence. That’s especially insidious because it occurs subconsciously.

One recent study found that investors on social media are five times more likely to follow users who agree with them and will see nearly three times more messages they agree with than disagree with. Falling into such an echo chamber, the study showed, leads people to trade more—and earn lower returns.

Meanwhile, bucking the consensus engages circuits in the brain that generate pain and disgust. Experiments have shown that when you find out your peers disagree with you, your choices become up to three times more likely to match theirs, although you have no conscious awareness of being influenced…

…If you have views about which asset or investing strategy is right for you, write down your reasoning before you explore what some online group is saying. Take no action without reviewing your original rationale and determining that there’s a reasonable basis for changing it—grounded in independently verifiable evidence, not just the opinions of random people online.

Use a checklist to focus on the stability of the underlying business rather than share-price movements. Have I read the company’s financial reports? Do its executives admit mistakes, use conservative accounting and avoid hype? Have I written down at least three reasons why this is a good business that will be even better five years from now? What, exactly, do I understand about this company that most other investors are missing, and why?

3. Why the Fed Cut Rates and Mortgage Rates Jumped – Joe Weisenthal, Tracy Alloway, and Tom Graff

Joe (05:45):

But I actually have to refinance a mortgage in a couple of years. I could do it today, I guess, but I have to do it at some point. Alright. Government 30-year yields are 4.3%, 4.32% as we’re talking right now. I’ll probably want to get a 30-year fixed. Why can’t I just borrow at 4.32% if the government is already backstopping it?

Tom (06:05):

Well, so the key difference between a mortgage bond and a Treasury bond is that, in the United States, virtually all mortgages and all the ones that Fannie Mae and Freddie Mac back can be refinanced at any time without any penalty.

Joe (06:18):

Can’t I just promise not to? Well, I guess because I can always sell the house or something like that.

Tom (06:20):

Yeah. You can’t do that, Joe. And so, from an investor perspective, what that means is, if interest rates rise, no one refinances, everyone just stays where they are. Witness all the people kind of stuck in 2.5%, 3% mortgages right now, right? And so, those mortgages just stay outstanding and they might stay outstanding for 30 years for all we know, right?

Whereas if interest rates fall, you kind of don’t get any of the benefits. So if I buy a 30-year Treasury and interest rates drop, I can make 10%, 15%, 20% price appreciation as that happens. But in a mortgage bond, if interest rates fall, everyone just refinances, I just get all my money back at par, I’m no better off. And so you got to get paid for that, what — we’ll get into it —but what’s called negative convexity, you’ve got to get paid for that risk and that’s why there’s a spread between mortgage bonds and Treasury bonds…

…Tracy (09:53):

So talk to us about what goes into producing a mortgage rate. So if I want to buy a house and I go to a bank and I ask for a mortgage, what are the individual factors that go into the number that eventually gets quoted back to me?

Tom (10:07):

Okay. Yeah. So let’s assume for sake of argument, this is a loan that conforms to Fannie and Freddie’s standards. because That’s the ones we’re talking about here. So assuming that right? Your bank has to pay Fannie or Freddie a guarantee fee. The G-Fee. And that is based on your credit situation. So how much you’re putting down, what your credit score is, that sort of thing. And it’s all algorithmic. So they’re just typing it into a computer, Fannie and Freddie kicking back, here’s the rate, right?

Then they’re also going to think to themselves, okay, well where can I sell this mortgage? Right? What price am I going to get when I sell it in the open market? And that depends mostly on just what the general price is for the going rate for mortgages, but it might depend a little on your situation so we can get into it how certain kinds of mortgages command a bit more of a premium in the market than others and that will go into the rate you’re going to get quoted. And so every night the bank’s mortgage desk is sort of plugging in, hey, for mortgage like this, we’ll we’ll offer this rate for mortgage like that we’ll offer this rate and all these factors are going into that. So when your loan officer’s typing this into his computer, that’s what’s spitting out, right?

Joe (11:17):

Actually, let’s back up. What makes a mortgage conforming versus non-conforming?

Tom (11:21):

The biggest thing is the price. So the price relative to used to be a hard number, but now Fannie and Freddie do it relative to your sort of MSA or what, what your area? Yeah.

Joe (11:31):

So wait, above a certain price? Can you go into that a little further? Above a certain price, Fannie and Freddie just won’t back?

Tom (11:37):

Yeah. They’re just not backing it. And that has to do with their mandate from Congress to be about affordable housing…

…Tracy (25:56):

So I’m going to ask the question that I’m sure is on everyone’s minds per that Google trends chart, but when do mortgages come down?

Joe (26:04):

Yeah, or what will it take at least?

Tom (26:06):

Yeah, so, so we should, let’s, let’s talk about why they’ve risen since that Fed meeting, and then I think that’ll inform where they’re headed, right. So look, the 10-year Treasury is not a function of where the Fed is today. It’s a function of where people anticipate the Fed being in the next year, two, three. Beyond three, it’s sort of fuzzy, but like, you know, year two we sort of have a sense we can make a guess.

And so going into that September meeting, people started thinking themselves, boy Fed may cut 50 basis points in September, 50 basis points in November, maybe even 50 more basis points in December, right? If you pull up your WIRP chart on the Terminal, you can see this, right? If you go back to then, but since then what happened, we got a big jobs report the beginning of October. That was the September report, but came out in October.

And that was kind of a game changer because not only did we get a solid number for September, but it was huge upward revisions kind of erased what looked like a downward trend in hiring, right? Well now all of a sudden we’re like, boy, the Fed might be a lot closer to that neutral rate than we think, right? Eh, probably going to still cut in November, but maybe they’ll cut in December, maybe they won’t. But if they do, it’s certainly not going to be 50 basis points unless something changes.

And so that change in expectations has caused the tenure to rise. So commensurately the mortgage rate has risen, right? And so from that story you can say, all right, well it becomes pretty easy to see what’s going to cause mortgage rates to drop, the tenure needs to drop, right? And what’s going to cause the tenure to drop? Well, we’re going to need more Fed cuts priced in. Well what’s going to cause more Fed cuts to get priced in? We need the economy to get weaker.

4. An Interview with Hugo Barra About Orion and Meta’s AR Strategy – Ben Thompson and Hugo Barra

HB: Yeah, and this is worth taking a step back and talking about in a bit of detail, because there’s a few things that we don’t think about too much. The thing that annoys me having worked on smartphones for the last 15 plus years is that our smartphones make us work too hard. These workflows, these mobile app workflows are too repetitive, they’re not smart, they treat us generically.

It doesn’t make any sense that this world will continue for a lot longer and we know that AI is going to fundamentally change this. All apps are going to become agentic. Think of developers writing apps in their respective agents. Agents will make it possible to have much, much simpler workflows, which are highly, highly personalized. They’re still being rendered by the app, but the flows themselves are highly personalized, they have a much lower burden on users. Agents can do a lot of the prediction and anticipation and pre-thinking on a user’s behalf so that everything is boiled down to hopefully a small number of simple choices or no choices at all.

Oh, here you go, I have an analogy, you have to tell me if this fits what you’re going for here. So arguably the ultimate agentic experience that people experience right now, even though they don’t realize it, are their social media feeds, in that the feed is perfectly customized to serve up to you the entertainment that it thinks you want at every moment, and it actually turns out based on engagement numbers, it works pretty well, and while people claim they want a chronological timeline or whatever, that’s like saying you want a grid of apps and the reality is revealed preferences says that no, they don’t want that. Is that a good analogy for what you’re going for?

HB: I think that’s halfway there. I would say an agentic version of Instagram is going to be a little bit different. Instagram thinks it’s pretty smart, but it doesn’t have a lot of context from your life. As much as people say that Instagram listens to their conversations, that’s not true.

If only it did.

HB: Exactly. If only it did, it’d be great, but it of course doesn’t. So Instagram, to use an example, knows very little about you relatively speaking, about the broader context of your life. So it’s like a poor man’s agent that tries to represent your interests and serve what you want. A true agentic version of Instagram has an agent that represents what you want and can do a much better job ranking, filtering the content that you see at any point in time based on a bunch of other things, and it’s very tricky because Instagram can’t know about these things, because if they do, they will create this massive profile about you. So there’s a whole new architecture of the Internet that will have to be invented for these agentic capabilities to become unleashed because you have to keep your data…

…HB: Yeah, and this is where we get into I think the meat of the topic, which is what does a world of AR apps look like? What does it feel like to live in it? I’ve used this YouTube video called Hyper-Reality multiple times when I’ve given talks on AR. It’s completely absurd, it’s a world that we don’t want to live in, but it’s a joke, but it’s also not. So I always encourage people to go watch Hyper-Reality, it’s a beautiful artistic piece.

Before we get into what living in this AR world looks like, there’s a couple of things that I always like to talk about. One is that direct manipulation, which is what Apple brought to the world with multi-touch — we’ve had other forms in the past, but that’s really when it arrived — is genius and it has and will continue to exist, and direct manipulation when you’re literally touching something with your fingers has to be tactile, meaning it requires a physical surface. Pinch-to-zoom in midair isn’t nearly as useful as something on a tactile hard surface, so that’s the first thing.

The second thing is our arms get tired. This idea of midair computing is only really useful for quick actions. There’s this hilarious scene in Minority Report where Tom Cruise is probably sweating by making lots and lots of gestures in midair, and perhaps ahead of their time in their vision, but that’s not a thing, people don’t want to be computing in midair.

And look, if Tom Cruise can’t do it, none of us can do it.

HB: (laughing) Exactly. So anyway, we have to keep those things in mind. Direct manipulation is genius and your arms get tired, so there are three modalities of UI and UX in the Spatial Computing paradigm. The first are your tools, they’re like your utility belt, they’re things that walk with you wherever you go. They might be body-locked or in some cases head-locked, your notifications tray, your settings, your menu, etc., these things walk with you where you go, and you will access them through both 2D gestures and 3D gestures. But it’s all quick, it’s just how you get into the thing that you want to do.

Right, this is almost like the mechanical wristwatch of UI layers.

HB: Exactly right. So that’s your utility belt, we’re going to bring that with you everywhere.

The second thing are world-anchored apps. So it’s basically walking to your house and instantiating an app on your table sitting down and then playing with it. That app might be a 2D iPad style app, it might be a 2D app on a massive surface, it might be a little 3D app, like a tabletop app. Imagine calling an Uber using a tabletop 3D map that allows you to say exactly where you want to get picked up.

Right. You can pick up the car and put it on the map where you want it.

HB: You can put it on the map. So this is really useful because you can instantiate any app on any surface at any time.

Then the third thing, which is where it gets really exciting, are world-anchored virtual objects and maybe screens as well. So these are things that are just in the world. You walk into your house, you’ve got art on the walls, you’ve got maybe a control panel where ordinarily a light switch would be, and it allows you to do all sorts of things with your house because it’s not actually there, it’s just the wall. But you see something, you see a control panel on the wall that’s rendered for you and it will be agentic, etc., all that stuff.

That’s like real augmented reality because you are actually augmenting reality.

HB: Yes, this is real augmented reality. Think about annotating the world as well. You saw in the Orion demo the recipe thing where it annotates the ingredients and visually tracks them so if you walk away and look back, it’s still tracking them, they’re still there. This is crazy interesting stuff, and that’s where a lot of the new types of use cases are going to come from, and that’s it. Those are the three categories of UI in an augmented reality world…

…Yeah, this is the challenge here. You have a couple Apple points here, the one thing about linking it to the smartphone is, if you can offload all that compute and offload all that battery and offload all that connectivity into one device, it makes it a lot easier. I mean, you said for Apple, “Number eight, Apple will continue to slow-follow Meta on camera glasses and mixed reality headsets, but will be several years behind on AR glasses”.

HB: Yeah. I think that it’s a really easy win for Apple to launch a competitor to the Ray-Ban glasses. I mean, it’s a proven form factor, just do it and I think they’ll do a fantastic job at it. It makes total sense because of Apple Visual Intelligence. It’s just, just do it. So that was a rumor from Mark Gurman from I think last week, which I really believe in. The earbuds with cameras, I’m not so sure, but I do believe that camera glasses are a thing that makes sense for Apple to be building.

Now, the AR glasses though is not, in my opinion, a product that we’re going to see from Apple in much less than 10 years.

Wow.

HB: I think that one is going to take a very, very, very long time.

And why is that?

HB: I just think their product bar is going to be insanely high, and I think they’re going to have some hard architectural decisions to make. Is it attached to your iPhone as an accessory or is it a standalone thing with its own puck like Meta did? There’s a lot of trade-offs there that I think people don’t necessarily think about carefully. It is not easy for Apple, that’s my next point.

Number nine.

HB: To ship AR glasses as a smartphone accessory, because in practice they have significant cost margin, thermal envelope constraints on the iPhone because the iPhone is a single, super high volume product that needs to be a great product and a highly profitable smartphone, first and foremost, so as soon as you have to start to add more components to this thing to power AR glasses, you’re tasking your primarily profit center for the whole company and creating all these architectural constraints.

Couldn’t they just make an extra model like the AR model? But I guess then that ruins your TAM.

HB: Yeah, I think that’s like the worst of all worlds, in my opinion.

This is really interesting, 10 years does blow my mind because yeah, your thought immediately, let me restate your argument, make sure I get it, your thought immediately goes to Apple already has a smartphone, they can just do an accessory, but actually the issue is the smartphone is so successful and so profitable and so essential that, 100 million, is that in a quarter, whatever, all those smartphones can’t be compromised to support this because they’re so important.

HB: And the attach rate just doesn’t justify it.

That’s right.

HB: Look at the attach rate of Apple Watch, the attach rate of Apple Watch is still fairly small.

And yet if you did a separate model, you’re giving away your entire advantage so you’re stuck.

HB: Exactly. So I think it’s a harder trade-off space than people realize for Apple, and my guess is that this is a discussion that is highly unresolved.

5. Researchers say an AI-powered transcription tool used in hospitals invents things no one ever said –  Garance Burke and Hilke Schellmann

Tech behemoth OpenAI has touted its artificial intelligence-powered transcription tool Whisper as having near “human level robustness and accuracy.”

But Whisper has a major flaw: It is prone to making up chunks of text or even entire sentences, according to interviews with more than a dozen software engineers, developers and academic researchers…

…More concerning, they said, is a rush by medical centers to utilize Whisper-based tools to transcribe patients’ consultations with doctors, despite OpenAI’ s warnings that the tool should not be used in “high-risk domains.”

The full extent of the problem is difficult to discern, but researchers and engineers said they frequently have come across Whisper’s hallucinations in their work. A University of Michigan researcher conducting a study of public meetings, for example, said he found hallucinations in eight out of every 10 audio transcriptions he inspected, before he started trying to improve the model.

A machine learning engineer said he initially discovered hallucinations in about half of the over 100 hours of Whisper transcriptions he analyzed. A third developer said he found hallucinations in nearly every one of the 26,000 transcripts he created with Whisper.

The problems persist even in well-recorded, short audio samples. A recent study by computer scientists uncovered 187 hallucinations in more than 13,000 clear audio snippets they examined.

That trend would lead to tens of thousands of faulty transcriptions over millions of recordings, researchers said…

…The tool is integrated into some versions of OpenAI’s flagship chatbot ChatGPT, and is a built-in offering in Oracle and Microsoft’s cloud computing platforms, which service thousands of companies worldwide. It is also used to transcribe and translate text into multiple languages…

…Professors Allison Koenecke of Cornell University and Mona Sloane of the University of Virginia examined thousands of short snippets they obtained from TalkBank, a research repository hosted at Carnegie Mellon University. They determined that nearly 40% of the hallucinations were harmful or concerning because the speaker could be misinterpreted or misrepresented.

In an example they uncovered, a speaker said, “He, the boy, was going to, I’m not sure exactly, take the umbrella.”

But the transcription software added: “He took a big piece of a cross, a teeny, small piece … I’m sure he didn’t have a terror knife so he killed a number of people.”

A speaker in another recording described “two other girls and one lady.” Whisper invented extra commentary on race, adding “two other girls and one lady, um, which were Black.”

In a third transcription, Whisper invented a non-existent medication called “hyperactivated antibiotics.”

Researchers aren’t certain why Whisper and similar tools hallucinate, but software developers said the fabrications tend to occur amid pauses, background sounds or music playing…

…Over 30,000 clinicians and 40 health systems, including the Mankato Clinic in Minnesota and Children’s Hospital Los Angeles, have started using a Whisper-based tool built by Nabla, which has offices in France and the U.S.

That tool was fine-tuned on medical language to transcribe and summarize patients’ interactions, said Nabla’s chief technology officer Martin Raison.

Company officials said they are aware that Whisper can hallucinate and are addressing the problem.

It’s impossible to compare Nabla’s AI-generated transcript to the original recording because Nabla’s tool erases the original audio for “data safety reasons,” Raison said.

Nabla said the tool has been used to transcribe an estimated 7 million medical visits.

---

Disclaimer: The Good Investors is the personal investing blog of two simple guys who are passionate about educating Singaporeans about stock market investing. By using this Site, you specifically agree that none of the information provided constitutes financial, investment, or other professional advice. It is only intended to provide education. Speak with a professional before making important decisions about your money, your professional life, or even your personal life. We currently have a vested interest in Apple, Meta Platforms, and Microsoft. Holdings are subject to change at any time.