The report was, by any ordinary measure, exceptional. Broadcom posted revenue of $22.19 billion for its fiscal second quarter, up 48% from a year earlier. AI semiconductor revenue hit $10.8 billion, up 143%. Adjusted earnings came in at $2.44 a share, above Wall Street’s estimate. Then the stock fell roughly 14%, and took most of the semiconductor sector with it.
The reaction came down to two numbers that did not move. Broadcom guided third-quarter AI chip revenue to about $16 billion, below the $17.2 billion analysts had expected. And the company held its full-year AI guidance at $56 billion, when the consensus had drifted toward $57.6 billion. Chief Executive Hock Tan told analysts that demand for custom AI accelerators was running ahead of the company’s own forecast, and that AI bookings for the quarter topped $30 billion, with the order book now running at more than three times quarterly shipments. None of it was enough.
By the close on June 4, the Philadelphia Semiconductor Index had fallen more than 10%, its steepest drop in years, and the Nasdaq Composite lost about 4%. Analysts put the value erased from global chip stocks at more than $1 trillion. Broadcom itself fell about 13% on the day, its sharpest one-day decline in over a year, as investors who had bid the stock up roughly $270 billion in market value in the days before earnings reversed course.
“It’s not that AI demand is weak,” said Angelo Zino, a senior analyst at CFRA Research. “The bar was set extremely high before the release, and part of the stock’s reaction stems from that.”
The selloff exposed a new tension in the AI trade. For two years, investors treated every chipmaker earnings report as evidence that the buildout was accelerating. Broadcom’s quarter contained that evidence in abundance, and the market read it as a sign that growth was about to slow. Tan declined to raise a target that already implied 180% growth for the year. When a company that can book $30 billion of AI orders in a quarter stops raising guidance, the market asks whether the ceiling is visible.
NVIDIA, the sector’s biggest weight, fell about 6% during the selloff after briefly touching a $5 trillion market value earlier in the year. The company’s decision to raise its dividend drew an unusual amount of scrutiny. Dividend increases are usually welcomed; in this cycle, some analysts read the payout as a signal that management sees fewer high-return places to put cash — that capital returns are displacing reinvestment, and that AI chip procurement growth may have peaked.
Tan offered his own answer to the peak question. He said the company intends to focus on chips rather than the full servers and systems it had begun selling, a shift that narrows margins — the company guided third-quarter gross margin down to 74% from 77% — but sharpens its role as the silicon partner to the world’s AI labs. He named Anthropic, Google, Meta and OpenAI among six core custom-chip customers, all of which design their own accelerators through partners to control cost and supply. He also acknowledged that Google, Broadcom’s largest AI customer, will likely use more than one chip supplier going forward, raising the possibility that Broadcom’s share of a marquee account shrinks.
The market’s response to a record quarter is the clearest evidence yet that the AI infrastructure cycle has entered a new phase. In the previous phase, capacity was scarce and every forecast was a floor. Now, with hyperscalers planning to spend more than $700 billion on AI infrastructure this year, the question has become whether spending can keep compounding. When a company that sells the shovels declines to raise its forecast, investors start pricing the possibility that demand growth is decelerating from extraordinary to merely very strong.
Some analysts cautioned against reading too much into a single quarter. Broadcom’s book-to-bill ratio above three, they note, means customers are ordering more than three times as much as the company ships, a position that supports revenue well into next year. The company repeated its target of more than $100 billion in AI semiconductor revenue for fiscal 2027. A shortfall of $1.2 billion against a single quarter’s AI expectations — roughly 7% — does not look catastrophic in isolation.
But markets price narratives, and the narrative has shifted. The same week that Broadcom disappointed, ASML’s market value crossed $674 billion, making the Dutch lithography giant the most valuable company in European history, and Taiwan Semiconductor Manufacturing Co. told investors that its capacity could only support so much demand. Those two facts — record valuations for the equipment makers that gate the entire supply chain, and a foundry saying it cannot build fast enough — frame the debate that the selloff opened: AI chip demand is real, supply is constrained, and yet the stocks most exposed to the buildout are the ones being sold.
For investors, the quarter changed the rules of engagement. Growth alone no longer moves chip stocks; growth relative to an ever-higher bar does. Broadcom’s order book remains among the strongest in the industry, and the company’s position as the custom-silicon partner to the biggest AI spenders is intact. But the market has served notice that it will demand evidence, quarter by quarter, that the $56 billion target is a floor rather than a peak.
