The AI chip trade just learned it has no brakes
Broadcom beat earnings and fell 14%. By Friday roughly a trillion dollars had left the chip complex. The lesson is about concentration, not about Broadcom.

Image: Peellden / Wikimedia Commons (CC BY-SA 3.0)
Broadcom beat. That is the part worth sitting with. The company reported earnings above expectations and its stock fell 14 percent the next day, and by Friday's close the selling had spread until roughly a trillion dollars had left the chip complex in two sessions. A good quarter helped erase a trillion dollars of market value. If that sounds irrational, it is not. It is what happens when a sector is priced not for success but for acceleration, and then merely succeeds.
The number that did the damage was not a loss. It was Broadcom's guidance for next-quarter AI chip sales: about $16 billion against the roughly $17.2 billion analysts had pencilled in, with no increase to the full-year AI forecast. In an ordinary market that is a rounding error attached to a very strong business. In this market it was a confession. When a stock is priced for a curve that only bends upward, the dangerous news is not bad news. It is adequate news. 'In line' is a downgrade when the price already assumes 'above.'
A good quarter helped erase a trillion dollars. That is not irrational. It is what 'priced for acceleration' means.
How a trillion dollars leaves in two days
The move did not stay with Broadcom. The SOXX, the semiconductor index a great many people own without quite realizing it, fell 10.4 percent on Friday. Arm dropped 12.8 percent, Intel 11.3, AMD 10.9. This is the mechanism worth understanding, because it is the same one every time. When a handful of names become a large share of an index, everyone who owns that index — including the pension fund that describes itself as conservative — is long the same trade, whether they chose it or not. On the way up that concentration is invisible and pleasant. On the way down it has no brakes, because the buyers and the believers are the same people, and they reach for the exit at the same moment.
It did not help that the May jobs report came in strong the same morning. A hot labor number pushes back the interest-rate cuts the whole AI trade has been leaning on, and a sector valued on cash flows years in the future is exquisitely sensitive to the rate used to discount them. Two unrelated facts — a chip company's cautious guide and a healthy labor market — rhymed into the same sentence: the cost of money is not falling as fast as these prices assumed.
The cycle hiding under the story
Underneath the AI headline is an older, more boring cycle the market keeps forgetting is still there. Memory. Producers like Micron have been steering output toward high-margin data-center customers, which is rational for them and brutal for everyone downstream: smartphones and PCs are being starved of parts just as global handset demand looks like it is rolling over. That is a classic memory squeeze, the kind that has humbled this industry every few years since the 1990s, and it is arriving at the exact moment investors had convinced themselves the cycle had been repealed by artificial intelligence. It has not been. AI demand is real. It is also sitting on top of a commodity cycle that still behaves like a commodity cycle.
None of this means Broadcom is a bad company, or that the data-center build-out is a mirage. It means something narrower and more useful: a great business and a crowded position are not the same object, and the past two days repriced the position, not the business. The question for anyone holding chips here is not 'is AI real' — it is — but 'how much of what I own depends on three or four names continuing to beat expectations that are already heroic, and what is my exposure if they merely meet them.'
The downside the tape has not finished pricing is the one where these two stories stop being separate: AI capex pauses to digest just as the memory cycle turns and phone demand keeps sliding, and the index that carried everyone up discovers it is the same small handful of stocks on the way down. That is not a forecast. It is the exposure. The lesson of the week is the oldest one on a trading desk and the easiest to forget in a bull market — the most expensive word in the market is 'only,' as in 'it was only in line.' Priced for perfection, a company can do everything right and still take a trillion dollars down with it.
References
- 24/7 Wall St. — Broadcom sinks 14% on soft AI chip outlook, dragging down AMD and Intel
- CNBC — Broadcom, Micron and Arm sink, leading chip stocks lower
- StartupHub — Arm and Intel lead AI chip rout as SOXX falls 10.4%
- Invezz — Broadcom rout hits Micron, AMD: AI demand and memory-chip cycle in focus
- Yahoo Finance — Intel, AMD, Micron shares sink as Broadcom results spark sector sell-off


