Broadcom Didn’t Miss. It Just Stopped Promising More — and the AI Trade Cannot Survive on Less

Broadcom did not miss. That is the single most important thing to understand about Friday’s selloff, and the part almost no headline got right.

The company reported solid numbers. It simply guided next quarter to roughly $16 billion in AI chip revenue against the $17.2 billion Wall Street had penciled in, and — the real sin — declined to raise its full-year 2026 forecast for AI semiconductor sales. For an ordinary industrial company, this would be a footnote buried on page nine. For the AI trade, it was a verdict. Broadcom shares fell as much as 15% intraday, dragging Micron, AMD, Nvidia and Marvell down with them. The Nasdaq lost 1.13% and the S&P 500 slipped 0.63%, even as the broader economy sent a signal of rude good health.

Here is the thesis, stated plainly: the AI rally was never priced for growth. It was priced for acceleration. And acceleration, unlike growth, cannot be sustained indefinitely. It does not have to reverse to break the spell. It only has to slow down once.

What we have on our hands is a dead shark

In Annie Hall, Woody Allen’s character tells the woman he loves that a relationship is like a shark: it has to constantly move forward or it dies. “And I think what we’ve got on our hands,” he says, “is a dead shark.” The line is funny because it is true of more than relationships. It is true of any system that mistakes forward motion for proof of life.

The AI trade is that shark. It has never been valued on what these companies earn today; it has been valued on the assumption that next quarter’s spending will be bigger than this quarter’s, forever. Every three months the hyperscalers raised their capital-expenditure guidance, every three months the chipmakers raised theirs to match, and every three months the stocks rose — and the rise itself was offered as evidence the rise would continue. The circularity was the tell. The same handful of companies were investing in, buying from, and selling to one another, and calling the loop demand. It is the same logic that has Sam Altman chasing trillions to build his own chips: the buildout justifies the buildout.

Broadcom did not attack that logic on Friday. It did something quieter and more dangerous. It stopped swimming forward. And the whole school felt it.

Why a “good” quarter detonated the sector

A single soft guide from one supplier should not, in theory, reprice an entire industry. It did because of what Broadcom supplies and what it implies. As a maker of custom AI accelerators and a bellwether for hyperscaler appetite, its order book is read as a proxy for the whole buildout. When it declines to raise its number, the read-through lands instantly on the memory makers — Micron, whose high-bandwidth chips live or die on AI capital spending — and on every name leveraged to the same story.

That is why the people who spent two years learning how to buy Nvidia stock woke up to a market that no longer rewarded the reflex. It is why the AI names Morgan Stanley told investors to own moved together on the way down, as they had on the way up. Broadcom’s refusal to raise its forecast did not reveal a problem with Broadcom. It revealed that the feeding might not be infinite — and a market priced for infinity does not take that news calmly.

The economy is fine. For this trade, that is the trouble.

To be fair to the bulls, none of this is a crash, and the demand for computing is real. The hyperscalers are funding their buildout largely from profits, not debt, which makes this nothing like 2008. The Russell 2000 of smaller, domestic companies actually rose 1.45% on Friday — hardly the tape of a market in panic.

But the day delivered an awkward truth. The Bureau of Labor Statistics reported that the economy added 172,000 jobs in May, roughly double what economists expected, with unemployment steady at 4.3% and average hourly earnings up 3.4% over the year. “Total nonfarm payroll employment increased by 172,000 in May,” the agency wrote, “and the unemployment rate was unchanged at 4.3 percent.” A strong labor market is good news for workers and bad news for anyone holding a stock valued on the promise of cheap money. It pushes rate cuts further away and raises the rate at which all those distant AI profits get discounted into today’s price. Treasury yields rose; the most expensive stocks fell hardest. The economy is fine. For the AI trade, that is precisely the problem.

The most dangerous moment

We keep waiting for the AI bubble to burst, as if it owes us the drama of a single catastrophic morning. It may never grant us one. A bubble inflated by acceleration does not need a pin; it needs only a quarter in which nothing accelerates. Broadcom just delivered one, on a day the economy was busy proving how healthy it is.

The lesson is older than this cycle and will outlast it. When a market rests on the assumption that the line goes up because the line went up, the most dangerous moment is not the one that feels frightening. It is the one that feels safe — the morning the shark glides forward exactly as it always has, and no one notices it has already stopped breathing.