On Wednesday, Broadcom did something almost no company on Earth can do: it grew its AI chip revenue 143% in a single year, to $10.8 billion in one quarter, and posted record consolidated revenue of $22.2 billion, up 48%. Then the stock fell nearly 14% in after-hours trading and dragged the entire semiconductor complex down with it. Micron dropped 7%. AMD and Intel slid in sympathy. South Korea’s KOSPI opened down 1.8%, one of Asia’s worst performers of the day. This is what the AI trade has become: you can double your most important business and still get taken to the woodshed.
Read that again, because it’s the whole story. Broadcom beat on revenue. It beat on earnings. CEO Hock Tan guided Q3 AI semiconductor revenue to grow over 200% year-over-year to $16 billion, projected total Q3 revenue of $29.4 billion (up 84%), and reaffirmed that AI chip revenue will clear $100 billion. Those are not the numbers of a company in trouble. They’re the numbers of a company being punished for the unforgivable crime of merely being excellent.
The Bar Isn’t High Anymore. It’s Imaginary.
Here’s the math that broke the stock. Broadcom guided Q3 AI chip sales to $16 billion. Analysts wanted $17.2 billion. That’s a $1.2 billion “miss” — on a number that is itself growing more than 200% year-over-year. The company didn’t shrink. It didn’t stall. It grew at a pace that would make any other industry weep with envy, and the market decided that wasn’t enough.
This is the tell. When a stock can double its core business and lose a seventh of its value in an afternoon, the price was never about the business. It was about the story — the belief that AI demand is not just large but accelerating, infinitely, forever. The moment Tan handed Wall Street a number that implied AI growth might be decelerating from “insane” to merely “spectacular,” the story cracked. And when the story is the asset, a crack is a crash.
Follow the Two Sentences That Actually Did the Damage
It wasn’t really the $16 billion that did it. It was two things Tan said on the call that the headline numbers can’t paper over.
First: gross margins are getting diluted by the very thing everyone is cheering. Tan admitted that the explosive growth in AI semiconductors is dragging down Broadcom’s overall gross margin, because custom AI chips carry thinner margins than its legacy software and non-AI silicon. Sit with the irony. For two years, Wall Street has treated “AI revenue” as pure gold. Tan just confirmed that the more AI revenue Broadcom books, the worse its margin mix gets. The AI boom isn’t free money. It’s lower-quality money at higher volume — and the market only just noticed.
Second: Google is going to buy chips from other people. Tan acknowledged that Google — one of Broadcom’s six core custom-silicon customers, alongside Meta, Anthropic, and OpenAI — will likely draw on multiple chip suppliers. That single concession is a dagger. Broadcom’s entire premium rests on the idea that hyperscalers are locked into its custom accelerators. The moment the CEO concedes that his anchor customer is shopping around, the moat looks more like a puddle.
The Second-Order Effect: Everyone Else Pays for Broadcom’s Honesty
The most revealing part of Wednesday wasn’t what happened to Broadcom. It was what happened to everyone else. Micron, which sells the memory that goes around AI accelerators, fell 7% on a report that had nothing to do with Micron. AMD and Intel dropped. Korean chipmakers got hit before North American traders even woke up. This is contagion, and it tells you the entire sector is now trading as a single, correlated bet on one narrative.
That’s the part that should worry you if you own any of this. When dozens of distinct companies — logic, memory, foundry, networking — all move together on a single earnings call, you don’t own a diversified semiconductor portfolio. You own one trade, wearing a dozen tickers. And one-trade markets don’t correct politely. They gap down all at once, exactly like they did this week.
Who Gets Hurt
Not Hock Tan. Broadcom is still going to ship something close to $16 billion in AI chips next quarter and bank a 67% operating margin doing it. The business is fine. The people who get hurt are the late money — the retail investors and momentum funds who bought the narrative at the top, who were told that AI chip demand was a one-way escalator, and who just learned that “beat earnings, double AI revenue” can still mean “lose 14% overnight.”
And there’s a quieter casualty: every other AI hardware company about to report. Broadcom just reset the bar for what counts as a “good” quarter. After Wednesday, beating estimates isn’t enough. You have to obliterate them, raise guidance, and avoid saying a single honest word about margins or customer concentration. That’s not a sustainable game. It’s a setup for a string of disappointments where great companies post great numbers and watch their stocks bleed anyway.
The Verdict: This Was the Healthiest Bad Day the AI Trade Has Had
I’ll say the unpopular thing: a 14% drop on a doubled-AI-revenue quarter is the most rational thing the market has done in months. For two years, AI semiconductor stocks have been priced for perpetual acceleration — a physical impossibility. Broadcom’s Wednesday was the first time the market priced in the obvious truth that growth this fast eventually slows, that thin-margin AI silicon isn’t the same as fat-margin software, and that even the best customers eventually want a second supplier.
None of that means the AI boom is over. Broadcom will almost certainly clear $100 billion in AI revenue, and the data-center buildout is real. But “the boom is real” and “the stocks are priced correctly” are two completely different statements, and this week the gap between them finally showed up on the tape. The takeaway isn’t that Broadcom is broken. It’s that the AI trade has quietly stopped being about whether these companies can grow — and started being about whether they can grow fast enough to justify a price that already assumed a miracle. Wednesday was the day the miracle came up $1.2 billion short.