On June 3, Broadcom did everything an investor is told to want a company to do. It beat earnings — $2.44 per share against the $2.40 analysts expected — and posted record revenue of $22.2 billion. “Q2 semiconductor revenue from AI of $10.8 billion grew 143% year-over-year, above our forecast,” CEO Hock Tan told investors. He then guided that number italichigher/italic, telling the market to expect AI chip sales to grow more than 200 percent in the current quarter, to $16 billion. The next day the stock fell roughly 14 percent, and it dragged Nvidia, AMD and Intel down with it.
Why would a company that beat, raised, and reported 143 percent AI growth lose a seventh of its value overnight? Because Tan declined to lift the italiclonger/italic horizon — the roughly $100 billion AI prize Wall Street had already penciled in for the years ahead. He beat, he raised, and he refused only to promise that the acceleration would itself accelerate. A single ounce of restraint was enough to erase tens of billions of dollars of market value in an afternoon.
That reaction is the whole story, and it is worth stopping on. A market that punishes a company for beating expectations is no longer pricing earnings. It is pricing a italicnarrative/italic — and narratives, unlike earnings, do not have a floor. When the only thing holding a price up is the promise of acceleration, the first day acceleration merely italiccontinues/italic instead of italicincreasing/italic becomes the day the spell breaks. We have been here before. The Greeks had a name for the thing Wall Street has built: the italicouroboros/italic, the serpent that eats its own tail. They drew it as a symbol of eternity, of a system that sustains itself forever. The modern version is made of silicon and venture capital, and it has persuaded itself of exactly the same immortality.
Where is the demand actually coming from?
Consider how the money now moves. In the fall of 2025, Nvidia agreed to invest up to $100 billion in italicOpenAI/italic. OpenAI, in turn, is using that capital to build data centers stuffed with Nvidia’s chips. Weeks later, OpenAI signed a separate agreement to buy roughly $300 billion of cloud computing from Oracle over five years. The chipmaker funds the customer; the customer buys the chips; the cloud provider books the revenue; everyone’s stock goes up. By 2026, analysts have counted more than $800 billion of these interlocking arrangements across the AI supply chain.
This is not, in itself, fraud. Vendor financing is as old as commerce. But it has a specific historical odor, and anyone who lived through 1999 will recognize it. In the telecom bubble, Lucent and Nortel lent money to the very companies that then bought their equipment, booking the loans as sales. The demand looked real until the financing stopped, and then it vanished all at once — because much of it had never been demand at all. It was a company buying its own products with its own money and calling the round trip “growth.”
The tell, today, is OpenAI itself. The company is reportedly on track to lose around $14 billion in 2026 — nearly triple its 2025 losses — while projecting $100 billion in revenue by 2029. The entire edifice rests on a demand curve that has not yet arrived, financed by the suppliers who most need it to exist. A ruminant, chewing and re-chewing the same cud, can look busy for a very long time. It is not, for all that, eating anything new.
The numbers nobody wants to frame as a warning
The macro picture rhymes with the micro one. The Shiller price-to-earnings ratio for the U.S. market has climbed above 40 — a level last seen at the peak of the dot-com mania. The five largest companies now account for roughly 30 percent of the S&P 500, the greatest concentration in half a century. At its November high, Nvidia traded above 30 times sales; Palantir began 2026 above 100 times sales. Historically, a price-to-sales ratio north of 30 has been less a valuation than a confession.
None of this means the technology is fake. This is the part the bears get wrong, and honesty requires saying it: Nvidia’s fiscal 2026 revenue reached $215.9 billion, up 65 percent, with $68.1 billion in a single quarter. That is not a Pets.com selling dog food at a loss. The chips are real, the data centers are real, and the productivity gains, eventually, will be real too. The AI stocks that have outrun the S&P 500 did so on genuine sales, not vapor.
But — and this is the distinction the market is refusing to make — italica real revolution and a real bubble are not mutually exclusive/italic. The railroads were real and the 1873 railway crash was real. The internet was real and so was the Nasdaq’s 78 percent collapse between 2000 and 2002. The mistake is never believing in the technology. The mistake is believing that because the technology is real, the italicprice/italic cannot be wrong.
What Broadcom was really telling you
Hock Tan refused to raise his number. Read that as the most honest sentence spoken on Wall Street this month. He was not signaling weakness; he was declining to participate in the ouroboros — declining to promise an acceleration he could not see. The market, addicted to the next upward revision, treated candor as catastrophe.
That is the condition we are in. The boom has reached the stage where it must keep manufacturing its own demand to justify its own price, and a market that has to finance its own customers has, by definition, run short of the organic kind. The serpent is eating well. It is also getting shorter.
For the investor, the lesson is older than any chip. Before you decide that this time the snake really can swallow itself forever, understand what you actually own: not a story about the future, but a claim on cash a company has not yet earned. When the financing loop tightens — and loops always tighten — the price will discover, very suddenly, the difference between demand and its own reflection.
The echo is loud right now. It is still only an echo.
