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 26 July 2026:
1. Keynote speech by Chinese President Xi Jinping at opening ceremony of 2026 World AI Conference
We human beings must answer the questions posed by our times: How to get along with thinking machines? How to ensure security when algorithm is part of decision making? How to tackle ethical challenges by technologies through adaptive governance? How to realize AI for all when the divide keeps widening? These questions demand serious consideration and real answers from the whole international community.
In China’s view, all countries should take a people-centered approach and develop AI for the positive and for good. We should ensure that AI is an important driver for shared prosperity and common security. We should join hands to build a just and equitable system for global AI governance. To this end, I wish to share four observations.
First, we should adhere to the principle of openness and win-win and boost innovation-driven development. As a new engine of world economic growth and an accelerator for the shift of growth drivers, AI is moving from the digital world into the physical world. We should seize this rare, historic opportunity to encourage open source, openness, collaboration and sharing. We should facilitate technological innovation, industrial development and scenario-based application of AI. We should make coordinated advances in the transformation and upgrade of traditional industries, the cultivation and growth of emerging industries and forward-looking planning for future industries, so that all sectors and businesses can benefit from AI.
Second, we should strengthen risk-awareness and ensure that AI is secure and controllable. AI should be a trusted tool for humanity. We should take seriously the various types of inherent and secondary risks that AI may trigger. We should put in place laws and regulations, technological monitoring, early warning and emergency response systems in order to strengthen the line of security, prevent abuses and malicious use, and ensure that AI is always under human control. In the meantime, we should jointly oppose overstretching the national security concept in the field of AI and placing one country’s security over that of others.
2. The AI Bubble? – Nothing Linear?
Five very different numbers get quoted in this debate, and they blur together constantly:
- $725B — 2026 AI capex from the big four hyperscalers alone (Microsoft, Google, Amazon, Meta).
- ~$1.5T — estimated total AI infrastructure spend across the whole industry in 2026.
- ~$3T — Sequoia’s David Cahn’s estimate of the annual revenue AI must eventually earn to justify that build-out.
- ~$50–100B — roughly what the largest AI firms actually earn today: OpenAI ~$25B run-rate, Anthropic ~$30–47B ARR, plus Google/Microsoft AI revenue.
- The gap — the distance between the last two is the entire debate. It is the widest it has ever been, and widening…
…The three genuine choke points:
Advanced AI chips. Nvidia holds roughly 80% of the AI-accelerator market, and TSMC is effectively the sole advanced-node foundry able to fabricate the leading chips — virtually every leading chip, Nvidia’s, AMD’s, and hyperscalers’ custom silicon alike, passes through it. TSMC’s advanced-node capacity and CoWoS packaging are reportedly booked out for years, with Nvidia taking a large share. A single geopolitical or manufacturing disruption in Taiwan would immediately choke the entire industry’s ability to train or serve frontier models. Lesson: sovereign AI ultimately needs sovereign or allied access to advanced chip fabrication — software alone cannot route around this bottleneck.
Electricity & data-centre capacity. The IEA projects data-centre electricity demand roughly doubling from ~415 TWh (2024) to ~945 TWh by 2030, with AI the dominant driver. In some US states and countries (Ireland, Virginia) data centres already draw a fifth or more of total electricity. Power availability, not chip supply, is increasingly the binding constraint on how fast new capacity can be built. AI’s physical footprint — power, water, land — is becoming as strategically important as its digital one, and draws the same local political resistance as any heavy industry.
Frontier model training. Roughly six labs — OpenAI, Google, Anthropic, xAI, DeepSeek, Zhipu — can credibly compete at the very top, because training a genuinely frontier model costs tens to hundreds of millions in compute alone. But this choke point compresses faster than the physical ones: open-weight releases (Llama, DeepSeek, Qwen, GLM) have shown near-frontier capability at a fraction of the presumed cost. Algorithmic efficiency keeps lowering the compute needed to reach yesterday’s frontier.
A fourth layer, high-quality training data, is a contested rather than settled choke point: the freely-scrapeable internet text that trained the first LLMs is finite and non-renewing, and a fast-growing share of new web content is itself AI-generated — raising the risk of models training on their own degraded output (”model collapse”). Licensing disputes and lawsuits are actively reshaping what data can legally be used, pushing labs toward paid licensing, synthetic data and proprietary enterprise data. Data is shifting from an abundant free input to a scarce, litigated, paid-for one — a structural change in the industry’s cost base most forecasts do not yet price in…
…Part 5 · The bull case — why this may not be a bubble
The technology genuinely works. Unlike some past manias, the product is real and used daily by hundreds of millions. Generative AI reached an estimated 53% adoption among the general population within about three years — faster than the PC or the internet.
Revenue is growing very fast. The leading labs are scaling revenue at rates rarely seen: Anthropic went from roughly $1B to a ~$30–47B run-rate inside 18 months, and OpenAI reached a ~$25B run-rate. Fast growth can, in principle, close even a very wide gap.
The spenders are the strongest companies in history. Unlike the debt-laden telecom builders of 2000, the biggest AI spenders are among the most profitable firms ever. Nvidia earned roughly $120B in net income last year; Microsoft, Google and Amazon are cash machines. They will not simply collapse if returns are slow.
Valuations are rich, but not insane. The NASDAQ-100’s forward P/E is around 26x today, versus roughly 60x at the 2000 dot-com peak. Expensive, and concentrated in a few names — but not the pure fantasy of the last great tech bubble.
Individuals are extracting real value now. Even where enterprise ROI is unclear, the estimated value of generative-AI tools to US consumers reached about $172B annually by early 2026, with median value per user roughly tripling year over year.
Part 6 · The bear case — why it may be a bubble…
…Big-four AI capex has gone from ~$90B (2020) to $147B (2023) to $410B (2025) to $725B (2026). The gap itself has widened from a $200B question (2023) to $600B (2024) to an estimated ~$800B+ annual gap in 2026 — Allianz pegs the capex-vs-revenue divergence at ~46%, exceeding the ~32% seen in the 2001 telecom bubble.The cash-flow squeeze. PIMCO estimates Big Tech capex will consume ~94% of operating cash flow over the next two years, up from ~40% in 2023 — meaning for every $100 earned, only ~$6 is left for dividends, buybacks, salaries and everything else. For the first time, the hyperscalers are also leaning on debt: the big five raised roughly $108B in new debt in 2025.
Enterprise ROI — the crack in the revenue side. The core bull rebuttal is “enterprises will pay, because AI makes them radically more efficient.” The data from the industry’s own consultants is, so far, harsher. MIT Project NANDA found 95% of enterprise GenAI pilots showed zero measurable P&L impact (2025, 300 deployments). BCG’s “AI at Scale” survey of 1,800 executives found only ~26% of companies are generating meaningful financial value from AI. S&P Global found 42% of companies abandoned most AI projects in 2025 — more than double the prior year. McKinsey’s 2026 State of AI found fewer than 20% of pilots reach enterprise-scale production.
Most failures are not the technology’s fault — MIT found ~70% of the work to make AI pay off is process and workflow redesign, not the model. But from a market standpoint the cause barely matters: if the ROI is not showing up in the P&L, the revenue needed to justify the capex is not showing up either…
…The technology can be completely real and the bubble can still burst. That is not a contradiction — it is the normal pattern of every industrial build-out in history. The pattern runs four steps: high returns attract capital; capital keeps flowing until overcapacity is built; overcapacity triggers collapse; survivors inherit the wreckage cheaply and, when demand finally catches up, make fortunes on assets others paid to build.
3.Maybe Intelligence Ain’t All That – Clifford Sosin
Superintelligence arrived. You probably didn’t notice, because it turned out to be kind of incremental…
…We hold a thin scatter of facts about the world, and intelligence or reasoning is whatever fills the space between them. It extrapolates between things we already know, and LLMs are incredible at that. In studying our words, they learned the structure of our thoughts. We will soon have it in unlimited supply, at 200 IQ, for close to nothing.
Where the space between the facts behaves well, this is close to godlike. Real estate law isn’t hard. We made it up, it’s internally consistent, and a model holding every statute and ruling should beat any lawyer for free. Coding, math and most administrative work are similarly benign. What makes them easy is that they have relatively smooth solution spaces and are tractably verifiable.
Most of what matters doesn’t behave like that. The universe is mostly the emergent behavior of complex systems. Stir cream into coffee. Watch a storm build out of nothing but atmospheric temperature differences and water vapor. The local rules are simple and the tornado is not. In systems like these, no amount of reasoning delivers the answer, because there’s no shortcut hiding in the gaps. You have to run the thing. And running it has a ceiling of its own, since small errors compound, which is why no supercomputer will ever give you a clean two-week forecast. The smallest object that can perfectly simulate the universe is the universe. Human systems are similarly complex…
…An AI will hand you a genuinely clever design for a jet turbine blade. It might be far more likely to work than anything your engineers came up with. It’ll still probably fail, because that’s the base rate at the edge of what anyone knows. The only way to find out is to build the blade and try to break it.
That’s the real limit on learning, and it doesn’t care how smart you are. Coming up with ideas was never the hard part. The hard part is how fast reality answers them.
4. A Stock Certificate From 1941 Taught Me More About AI Than Anyone from OpenAI – Francis Huang
But I think the railroad story, if you take it seriously, tells you six things about what happens next.
First, the technology will work. That part isn’t in doubt. Railroads worked. AI works. The question was never “will trains move faster than horses?” or “will language models generate useful output?” The answer to both was always yes. The technology question is settled. Everything that follows is a finance question and an ownership question, and those are much harder.
Second, the buildout will be larger than anyone currently projects. In 1850, America had 8,879 miles of track. By 1860, it had 30,626. Nobody in 1850 would have believed that number. AI infrastructure spending has quadrupled since 2022 and shows no sign of slowing. McKinsey’s $6.7 trillion projection for 2030 might end up being low. When a technology changes the cost structure of everything, the capital required to build it out has a way of exceeding every estimate, including the ones that already seemed crazy.
Third, the financial system will change in ways we can’t anticipate. Railroads didn’t just use the existing capital markets; they created new ones. Bond markets, equity markets, underwriting syndicates, credit analysis, bankruptcy law, corporate governance, even the concept of the limited liability corporation, all got reshaped or invented to handle railroad finance. AI will do the same. We don’t know what the new financial instruments will look like yet. Maybe compute futures. Maybe revenue-sharing tokens tied to model performance. Maybe something nobody has named yet. The railroad precedent says the instruments themselves will be part of the story.
Fourth, a crisis will come, and its trigger will be something nobody is watching right now. In 1873, it was a Viennese real estate bubble. In 1893, it was the cumulative effect of rate wars nobody thought would last. Whatever hits AI won’t be “AI doesn’t work.” It’ll come from some adjacent system that’s quietly entangled with the AI buildout in ways nobody has mapped. A chip supply disruption. A sovereign debt crisis in a country that’s heavily invested in AI infrastructure. An energy bottleneck. Something nobody is talking about at Davos or on the All-In Podcast. The black swan, by definition, is the one you’re not looking for.
Fifth, the people who build it and the people who own it long-term will be different. The merchants of Boston who funded the Boston & Worcester Railroad in 1831 were not the Vanderbilt family that controlled it in 1900, and neither of them were CSX, which runs it today. OpenAI’s current investors, Anthropic’s current investors, the hyperscalers currently spending $700 billion a year: these may or may not be the entities that own and profit from AI infrastructure in 2050. Morgan got rich not by building railroads but by picking up the pieces after other people’s railroads collapsed. Someone will play that role in AI. We don’t know their name yet.
Sixth, and this is the one I keep coming back to: the infrastructure will outlast everyone’s financial projections. The tracks the Boston & Albany laid in the 1840s are still carrying freight in 2026. Data centers being built today will run computation, in some form, for decades after the companies that built them have been restructured, merged, acquired, or dissolved. The physical layer endures. The capital layer above it churns.
5. Why Are Berries Everywhere, in Every Season? Driscoll’s – Julia Moskin
In just the last decade, berries have completed the journey from fragile, local, seasonal treat to worldwide refrigerator staple and marketing juggernaut. Global production has tripled since 2000, according to research from the U.N.’s Food and Agriculture Organization, and still cannot keep up with demand. In sales and volume, berries are the fastest-growing category in American produce, according to data from the U.S. Department of Agriculture.
Most of that growth has been driven by Driscoll’s, a $7 billion California company that began as a multifamily farm in 1904, patented its first strain of strawberries in 1958 and is still controlled by family members. In 1989, its board made what the company calls the Meadowood Declaration, a resolution that seemed preposterous at the time: to make all four berries available, in every season, in every part of the world.
Today the company is the undisputed global market leader, shipping four billion containers of highly perishable fruit across 60 countries each year…
…According to Circana, a market research firm, Driscoll’s is now the second-highest-earning brand in American supermarkets, behind only Coca-Cola…
…Instead of owning land, the company owns the genetic material of its berries and the knowledge of how best to plant, pick and transport them. It subcontracts with farmers around the world to grow those breeds according to its specifications, then handles sales and distribution after harvest.
But global access to berries has a cost, measured in metrics like water consumption, pollution, pesticides and labor practices. Driscoll’s has come under fire on all four fronts…
…Inside a nearby laboratory, where two full-time sensory scientists make their assessments, 210 raspberry varieties were laid out in a grid of plastic pints. Some had been bred for visual appeal, with more shapely shoulders, uniform drupelets and less “hair” (the thin red styles that sprout where the berry is pollinated). Others were developed to maximize yield, with fewer thorns and better “plant architecture” — tall, fluffy stalks that make the berries easy to pick. Each cultivar is tested for qualities like P.S.I., the interior pressure that determines whether a berry will yield to the teeth with an explosive, juicy pop.
Out of those 210 strains, said Kyle Rak, the company’s chief raspberry scientist, perhaps two will make it to market…
…Many Driscoll’s berries are no longer planted in soil, but grown in pots filled with carefully balanced mixtures of organic materials like coconut fiber and moss. This system of substrate farming was developed over centuries in the Netherlands to produce maximum yields from minimal land.
It requires a substantial start-up investment by Driscoll’s growers, who also absorb the costs of ever-shifting factors like labor, weather, equipment and rent. The company provides seed plants and “inputs” like soil treatments, along with technical support and marketing dollars. After harvest, the company retrieves the filled clamshells, then compensates the grower according to the price those berries command. According to Driscoll’s, growers receive 75 to 80 percent of the revenue…
…Demand for berries has exploded in the United States because of overlapping recent trends: more snacking and the rise of “functional” foods that promise specific health benefits…
…Berries are shifting entire economies.
In 2023, they became Mexico’s most lucrative agricultural export, surpassing avocados, beer and tequila. On Moldovan plantations and in Andean highlands, growers of low-margin crops like sugar cane and corn have switched to berries, which command premium prices.
In 2025, China overtook the United States as the world’s largest blueberry producer. Driscoll’s, the first foreign berry company allowed to operate there, now has about 8,000 acres under cultivation.
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, Microsoft, and TSMC. Holdings are subject to change at any time.