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 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 06 September 2026:
1. X post on AI demand – Philippe Lemoine
Dwarkesh’s basic intuition seems to be that AI will soon be able to fully automate many white-collar jobs, so to get a sense of how much revenue OpenAI/Anthropic can generate by replacing them, you can just look at the wage bills for those workers, but that’s not how things work.
First, if AI can fully automate most tasks performed by those workers but even a handful of them prove resistant to automation, AI will still generate a lot of value but it will presumably be captured mostly by the remaining workers, their firms or their customers and the current wage bill for those workers will massively overestimate the future revenue of frontier labs.
In that case, demand for AI may not explode in the way Dwarkesh and Dylan expect, because it will be bottlenecked by humans in production.
But let’s assume, as Dwarkesh and Dylan presumably do, that AI capabilities will improve so much that it can automate every task those workers perform and their jobs can be fully automated. It still doesn’t follow that demand will be as high as they think.
Even when it has become technically possible to replace most white-collar jobs with AI in principle, before it can happen in practice, there will be enormous transition costs and frictions as firms will have to undergo massive reorganization. This will be very slow even if politics doesn’t interfere, which it will.
Dwarkesh’s argument implicitly assumes that frontier AI companies will be able to charge about the same for tokens that provide services equivalent to that of the worker being replaced, but as long as there is enough competition, the prices they can charge will be pushed down and if he’s right about how fast capabilities will improve they may fall very quickly…
…Now, even if prices do fall rapidly, this could in theory be compensated by increased demand in volume. But there is only so much legal, managerial, engineering, etc. services that people want, even if prices fall.
The price elasticity of the demand for those services will eventually fall and, even if improved capabilities shift the demand curve outward, this will only generate so much demand because no matter how good your lawyer gets and how low his rates fall, at some point you don’t need more of his services…
…the thing about AI agents is that even if they reach the point where they can fully replace human workers on the production side, they can’t replace them on the consumption side because they don’t consume anything except compute and a few B2B services.
The income will still go to someone, but they will probably have a lower propensity to consume and different spending patterns, so there is a macroeconomic problem here about where the ultimate demand for the output of consumer goods and services produced by the customers of OpenAI and Anthropic will come from.
2. AI Semiconductor Endgame 2026 (III) – Fin
Hardware/infrastructure revenue growth in the internet revolution had exponential growth along only one dimension of penetration: user count
Because everyone was on a subscription — $50 a month for internet access, and there’s nowhere to upsell from there. It’s a flat-rate service. Looking at ten times as many web pages doesn’t make you pay ten times as much. A flat-rate service has no second dimension of exponential growth; once it saturates, it saturates…
…AI hardware demand, by contrast, comes from the exponential growth of tokens. The reason this round of AI is so hardware-hungry is fundamentally that the ceiling is determined by two dimensions: exponential growth in user count/penetration, and exponential growth in usage per person
These two dimensions of exponential growth multiply, and both convert into a single unit of account: tokens — which raises the infrastructure ceiling by far, far more
Where does each of the two dimensions stand in our era?
The first dimension, user penetration, crossed 50% a few months ago — meaning the derivative of the sigmoid is about to inflect…
…The second dimension, tokens per person, is still early in 2026. The median AI spend inside US enterprises is currently $12 per person per month. Long term, that number could plausibly reach at least 10% of a white-collar salary — roughly $1,000 a month. There are nearly two orders of magnitude in between…
…This should be historically rare: compute/infrastructure demand created by a relay of sigmoid curves across two dimensions, and with it a semiconductor supercycle
But all exponential growth is an illusion before the ceiling arrives, and this round of AI buildout is no exception — the illusion just lasts much longer, because there’s an extra dimension of exponential growth taking the baton…
…Right now, broadly defined coding tasks account for roughly 60–70% of all ARR revenue. If you count only the narrow developer definition it’s maybe 40%; the other 20–30% is likely tasks from other industries converted into a coding-agent execution structure…
…What takes the baton from developer coding is not some other isolated mega-scenario. The fastest-growing thing right now is precisely “non-programmers using coding infrastructure to complete non-coding product tasks”, coding is still the default execution kernel, but it isn’t necessarily the product category the user sees…
…As of June 2026, among OpenAI’s enterprise customers, Codex already accounts for 64% of combined Codex + ChatGPT output tokens. Since February, Codex weekly active users have grown 108x in legal, 41x in sales and recruiting, 26x in marketing — and only 5x in engineering
Looking at Anthropic’s revenue mix by vertical, narrow developer/software-company coding is only 40%, financial services/insurance is over 20%, and legal, pharma/life sciences, and consumer/retail e-commerce are all very substantial, each on the order of 10%.
The logic of software eating the world, once LLMs lowered the barrier to coding by a hundred times, has even intensified into coding eating the world, or put another way, coding is the modality agents are most fluent in, and agents are using coding to spread their utilization into all knowledge work…
…From the broad agent-coding angle, I don’t currently see a hard ceiling on large-model/agent demand. We’re still in the phase where the more you use it, the more efficiency you gain — the marginal positive return is still large, and the main bottleneck is how much budget revenue/profit can support. Willingness and use cases aren’t the problem…
…Inference is simply too profitable right now, so overbuild is an inevitable outcome, because everyone wants to grab a piece of the inference market, every player is in the arms race, nobody can stomach losing share and failing to ride the wave of high-speed growth, and no one is going to back down
Closed-source inference gross margins clear 70% easily (even after the CSP takes its cut), and token factories hosting open-source models and selling tokens run 60%+ gross margins as well (DeepSeek’s GM is beyond 80%), GPU prices keep rising, rent for 1GW is heading toward $20B, and even just selling short-term GPU cluster contracts carries 60%+ gross margin. This AI infrastructure market is so profitable and so tempting that anyone with distribution and technology is piling in as a neocloud, and because there’s generally a one- to two-year gap between build plans and delivery, overbuild is certain to happen — it’s only a question of when
“A question of when” means that at current demand and current compute plans, overbuild isn’t visible for at least two years. Any step-function improvement in models (reasoning -> agent, for instance), or any innovation in the application paradigm that sustains AI spending growth, keeps extending the overbuild timeline further out
When those extensions can no longer keep up with the pace of the buildout, overbuild will slowly start to surface
3. The Teaser Period: Why the AI Boom Is Hitting a Reset Wall – Groundbreaker
Nothing looked wrong in the summer of 2006. Home prices had risen for the better part of a decade. Delinquencies were near historic lows. Credit spreads were tight, the ratings held, and the securitization machine hummed. If you had asked a hundred people on a trading desk whether the American mortgage market was months from seizing, most would have laughed.
Millions of subprime borrowers were, at that moment, paying the low introductory rate on a two-year adjustable rate mortgage – the 2/28 ARM. A low fixed-rate for two years, then the rate reset to a payment 30% to 50% higher. During those first two years the loan performed beautifully: the borrower paid, the servicer collected, and the bond paid its coupon. Nothing looked wrong because the whole complex – housing, mortgages, securitization – was sitting inside the teaser period.
Every ARM reset was known, dated, and contractually inevitable from the moment of origination. Aggregate those reset schedules and you get the most damning exhibit of the era: the reset wall. Roughly a trillion dollars of adjustable-rate mortgages were contractually set to reset across 2007 and 2008 – thirty to forty billion dollars a month at the peak…
…The parallel to 2006 is exact and it explains the single most-cited absurdity of this cycle: How does OpenAI, a company with some $40 billion of run-rate revenue, sign $1.4 trillion of compute commitments? The same way a household with $60,000 of income signed a $600,000 mortgage: because the terms at signing do not require the payment yet, and because everyone at the table – borrower, lender, and the market – believes the growth will arrive before the payment does…
…The compute commencement wall can be made visible in exactly the way the reset wall was visible in 2007 – from disclosed contracts and delivery schedules…
…A frontier lab signs a $12 billion, ten-year capacity commitment with a compute provider. The contract is take-or-pay, meaning the lab commits to payments once the capacity is delivered, and delivery requires a campus that does not yet exist: two years of construction, procurement, and power work stand between signature and completion.
Now look at what each party’s financial statements show during the two-year teaser.
The seller – a hyperscaler or neocloud – books the arrangement into RPO or contracted backlog on day one – the full $12 billion, disclosed, quoted, and celebrated. The market values it as contractual future revenue. Meanwhile the seller’s cash flow statement hemorrhages: the campus is being built, so capex runs far ahead of receipts. Booked backlog rises; reported earnings feel none of the buildout; financing frequently sits off-balance sheet.
The buyer – a frontier lab like OpenAI or Anthropic – announces access to the compute it needs to pursue its scaling roadmap, and its private valuation reprices on the announcement. The commitment is a future obligation, disclosed – if at all – deep in a contractual-obligations footnote or, for the private labs, nowhere public. No expense hits the P&L because no service is being received. A lab that has committed tens of billions across multiple providers carries a cost structure that reflects only its commenced capacity.
The market sees a seller with explosive backlog and a buyer with secured compute capacity, and prices both as growth stories. Nobody is lying. Every number is GAAP-clean. The structure simply guarantees that during the teaser period, the system’s reported economics and its committed economics diverge by the full value of everything signed and not yet commenced…
…In residential credit, the interval between origination boom and reset wall was twenty-four months, because that was the teaser’s term. In compute, the interval is the construction timeline – twenty-four to thirty-six months. The 2025–26 signing boom therefore mathematically guarantees a 2027–28 commencement boom, exactly as 2005–06 originations guaranteed 2007–08 resets…
…This is what it means to say we are in the teaser period. The booked figure is enormous; the billed figure is a fraction of it and only beginning to turn up. Everything about the present looks like strength. The obligations that will govern 2027 and 2028 are already signed, already dated, and already sitting in RPO. What has not happened yet is the conversion – the moment booked becomes billed and the take-or-pay clock starts running regardless of the revenue and the counterparty’s ability to pay…
…The claim is not that commencement causes a lab to fail. It is that commencement is the date on which a pre-existing mismatch – fixed obligation against assumed revenue – becomes cash-due, and that, as in 2008, the mismatch is visible in the fundamentals well before the date makes it unavoidable. You do not need demand to fall. You need it only to decelerate below the rate the booked compute was underwritten to.
The compute contracts commencing in 2025 and early 2026 cleared, or very nearly cleared, the required growth rate. This is the crucial point, and it is the reason there is no alarm anywhere in the system: the early vintages worked.
They worked the way the 2005 and 2006 subprime resets worked. The collateral appreciated fast enough. The refinancing happened. Everyone who signed was vindicated, and vindication is the input to the next round of underwriting. Success in the early vintages is the mechanism that manufactures the late ones…
…For committed compute merely to equal revenue in 2027 – not to be comfortably covered, simply to reach parity, before a single dollar is spent on wages, research, sales, or tax – revenue would have to compound at 217% annually off the 2025 base. The dashed line at 100% represents revenue doubling every single year and sustaining it, which no company at this scale of revenue has ever done for a multi-year stretch. The obligation is accelerating at more than double the rate of the best case for the cash flow meant to cover it.
The obligation curve is contractually fixed and steep – it ramps according to a defined construction timeline. The revenue curve is a growth rate. If the growth rate rolls over – the two curves cross. That is the reckoning: not a demand collapse, but a demand deceleration meeting a cost schedule that was set in a more optimistic year.
4. 90% of executives say AI hasn’t boosted productivity. Some are still cutting jobs – Mark Ma and The Conversation
One Atlanta Federal Reserve study found that about 90% of executives believe AI has not yet boosted productivity at their companies. Other evidence suggests that the broader increase in productivity seen since 2021 is more likely due to remote work or factors other than AI, like downsizing in sectors such as technology…
…Managers at publicly traded firms typically make decisions based on whether a new investment improves short-term profitability and share price. So after investing heavily in AI, managers face pressure to show a strong financial return. The expectation is that if AI makes employees more efficient, the company will need fewer of them to complete the same work.
As a result, a quick way for managers to help businesses realize that anticipated return is by cutting headcount and lowering labor costs. Some of the companies we studied even started to lay off employees before pouring money into AI, as a way to free up capital for future AI investments…
…To see how employees perceive and react to AI adoption, we analyzed millions of employee-satisfaction reviews on the workplace review site Glassdoor.com. By identifying and analyzing AI-related comments, we found them to be much more negative than the overall tone of employee reviews.
This negativity reflects widespread concerns over corporate AI adoption and anti-AI sentiment among workers. At the same time, there’s a strong association between employee sentiment toward AI and firm productivity based on the employer’s financial information. This suggests that anti-AI sentiment among workers actually lowers productivity and offsets the potential efficiency gains caused by AI…
…What businesses need to understand, I believe, is that managing how employees feel is key to unlocking AI’s benefits. In turn, that means creating an environment where workers feel that AI is working with them, not against them.
5. Prediction: AI will collapse – wordgrammer
The modern AI discourse is making a similar mistake to saying “everyone will have a website” back in 1995.
I hear stuff like:
- People want personal software
- Everyone will build a mobile app
- The market for software is infinite
No.
You are directionally correct, but you are missing something big.
Every human will use agents. But less than 1% will use coding agents. (In the same way every human has a “personal page” through social media, but less than 1% have a personal website.)
You need an “idiotproof” sandbox. Your consumer agent, as well as agents for enterprise use cases, must have a 0% chance of failure. Your consumer agent must be usable:
- While drunk In bed
- Having sex
- On the toilet
- During lunch break
- In the background while working on homework
- While your brain is 99% preoccupied with a difficult task at work
It must be impossible for AI to: make purchases you did not intend, send messages you did not intend, and access information you did not want to share. It cannot be “difficult” for these to happen. It must be impossible. (In the same way that YouTube videos have a 0% chance of phishing, cross-site scripting, malware, and viruses.)
No amount of models getting better will give them a 0% chance of failure. To get to a 0% chance of failure, you need to be creative. You need to innovate. You need to radically rethink your abstractions.
Every human has a “personal page” through social media, but less than 1% has a personal website. What is the same shift for AI?
Every human will use _____ agents, but less than 1% use coding agents. What is _____?
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 no vested interest in any companies mentioned. Holdings are subject to change at any time.