The Most Confusing Crash in History: Why US Tech Stocks Are Falling on Record Profits

On 23 July 2026, the American market started falling — and Think School calls it the most confusing crash in history. Normally stocks drop when companies fail or slow down, but these giants were posting record profits and cash. So why did Google lose $300 billion in market cap on a day it generated $39 billion in cash? This is the video's attempt to explain a fall that seems to defy every rule of how markets are supposed to work.

This post condenses the Think School video. The figures, forecasts, and framing are the creator's.

The Google Paradox

The video starts with what it calls the Google paradox, and to explain it introduces a fictional coffee chain called Brev. Brev has its best year ever — $500 million in cash from selling coffee — but its founder spends $700 million opening 400 new outlets. At year end, the bank account is down $200 million despite the company never losing a single dollar. That gap is free cash flow: cash from running the business minus cash spent building things.

Apply that to Alphabet, the video says, and the confusion evaporates. On paper Alphabet reported $112 billion in profit, its highest quarterly figure ever, and its operations generated $39 billion in cash — up 41% year on year. But it also spent $44.9 billion on AI infrastructure, leaving roughly –$5.9 billion in free cash flow, described as Alphabet's first negative free cash flow since 2004. And this isn't a one-off: the video cites $811 billion in total signed future purchase obligations. For years, it notes, raising AI spending sent tech stocks up because spending signaled confidence — but on 23 July the same news crashed them, with Alphabet down 7% and Amazon down 4%.

The Math That Doesn't Close

Why the sudden reversal? The video points to a JP Morgan calculation: because AI is a risky investment, investors would want at least a 10% return, which means AI needs to generate about $650 billion every year. Yet even generously assuming OpenAI at $25 billion, Anthropic at $47 billion and Gemini at $25 billion, the total comes to just $97 billion — and with a combined loss of $20–30 billion. The obvious question the video raises: how do loss-making companies keep signing half-a-trillion-dollar deals?

Vendor Financing and the Espresso Machine

The answer, the video argues, is a technique called vendor financing, and it uses a character named Dave to make it click. Dave sells $12,000 commercial espresso machines, but his 100 customers — new cafes — have no money and no credit. So he lets them take a machine for $1,000 and pay the rest only once they turn a profit. Overnight his revenue "explodes" to $1.2 million and newspapers call him a genius. But Dave hasn't made 100 sales; he's placed 100 bets that these cafes, barely selling 200 cups a day now, will together sell 20,000. If they fall short, his record revenue disappears and his investors are left holding nothing.

Scale that up, the video says, and you get Nvidia. It reports Nvidia is in talks to provide a roughly $250 billion financing guarantee to OpenAI for a data center — part of a SoftBank-built project expected to exceed $500 billion — plus a separate chip-procurement financing deal worth up to $350 billion. That's around $600 billion of exposure to a single customer, against Nvidia's own annual revenue of $216 billion. The money, it says, flows in a circle: Nvidia guarantees the financing, OpenAI leases the compute and uses that Nvidia-backed money to buy Nvidia chips, Nvidia books the revenue, and then backs the next project. This is the "circular financing loop," and it leaves Nvidia acting as supplier, investor and debt guarantor simultaneously.

Four Risks Beneath the Surface

Having laid out the loop, the video walks through four risks. The first is whether the buyer can actually pay: OpenAI has reportedly signed around $1.4 trillion in infrastructure commitments against roughly $25 billion in annual revenue, and Sam Altman has admitted the company loses money even on its $200 Pro subscribers. The video is fair here — if OpenAI drives costs down and turns profitable, it could become the greatest company on Earth.

The second risk is an accounting one that the video attributes to investor Michael Burry. How long a company claims its servers will last directly changes its reported costs. A $30 billion server fleet depreciated over three years books $10 billion of cost a year; stretched to six years, it books only $5 billion — instantly lifting reported profit by $5 billion for the same hardware. Per a table the video says Burry posted, Meta extended useful life from 3 to 5.5 years and Google from 3 to 6 years between 2020 and 2025, with Oracle and Microsoft following suit. Whether that's prudent or a trick, the video says, only time — around 2028 — will tell.

The third risk is hidden debt. With the AI build-out priced at over $3 trillion (or $5 trillion including power plants, per JP Morgan), and nobody holding that kind of cash, the money is being borrowed through shell companies. The video's example: a $29 billion Meta data center funded not by Meta but by a newly created shell entity that borrowed $27.3 billion from bond investors due in 2049 — with Meta as merely a 20% owner and rent-paying tenant, keeping the debt off its own balance sheet. This structure, the video says, is how Meta, Alphabet, Amazon, Microsoft and Oracle collectively carry $1.65 trillion in AI-related debt.

What the Insurance Market Is Whispering

The fourth risk is the systemic one: if the profits never come, Nvidia could lose its revenue, equity value and loans all at once. The video explains credit default swaps as insurance on a loan — and notes that on 27 July 2026, as OpenAI news broke, the price of insurance on Nvidia's debt recorded its biggest intraday jump. Its memorable line: the stock market tells you what people hope for, while the insurance price tells you what they fear — and on that day, the fear became evident.

The One Number That Decides It

The video ties everything together — Google's cash squeeze, Nvidia's circular financing, the depreciation math, the shell-company debt and high oil prices — as a combined paranoia that explains the fall. But it ends on a single deciding variable: the cost of AI tokens. If token costs keep dropping dramatically for the next three years as they did in the last three, enterprises will fold AI into every operation, OpenAI and Anthropic will turn profitable, and Nvidia's projections will come true — making it, in the creator's words, the greatest bet the world ever took. If they don't, you already know the story. The creator admits it's difficult, but bets that if any companies can pull it off, it's OpenAI, Anthropic and Google.

Conclusion

Think School's case isn't that the AI boom is doomed — it's that the market suddenly started pricing in the possibility, and that a web of vendor financing, stretched depreciation and off-balance-sheet debt made the fall sharper than record profits would suggest. Whether it was a wobble or a warning, the video says, depends almost entirely on one falling line: the cost of a token.

---

Originally published on Think School. Watch the full episode: https://www.youtube.com/watch?v=801JPiGAKOE