On paper, it was a clean sweep. Broadcom reported quarterly results on Wednesday that beat Wall Street’s expectations on nearly every line, raised its target for next year’s artificial intelligence revenue by $15 billion and described an AI business growing faster than almost anything else in the semiconductor industry. The stock fell anyway. The stumble came one sentence into the outlook: fourth-quarter guidance that came in just shy of the number analysts had built into their models.
The numbers were large by any standard. Revenue for the third fiscal quarter rose 86 percent from a year earlier to $29.59 billion. Net income was $13.09 billion, up 216 percent, or $2.68 a share on a GAAP basis. Stripping out one-time items, operating profit rose 92 percent to $20.10 billion and earnings were $3.32 a share, ahead of the consensus estimate.
The engine was artificial intelligence. AI semiconductor revenue grew 221 percent from a year earlier to $16.7 billion in the quarter, up 54 percent from the prior three months, and now accounts for 56 percent of the company’s total revenue. Run at that rate for a year, the AI business alone would be larger than the annual revenue of almost every other chip company in the world.
The composition of that business matters. Broadcom’s AI revenue is not built on the merchant graphics processors that made Nvidia famous; it comes from designing custom accelerators for a small group of the world’s largest computing companies, along with the networking equipment that ties those chips together in giant clusters. Executives have said six custom-chip customers are driving the growth. The appeal of the model to those customers is control: a chip designed around their own software, their own power budgets and their own cost targets, rather than a product shaped for everyone.
The forecast is where the trouble started. Broadcom said it expects AI semiconductor revenue of $21.7 billion in the current quarter, up 236 percent from a year earlier, and total revenue of about $34.8 billion, up 93 percent, with a non-GAAP operating margin of roughly 66 percent. Wall Street had been looking for about $35 billion of total revenue. The gap was small, roughly $200 million on a number that large, but the market reacted as if it were enormous: shares fell as much as 6 percent in after-hours trading before recovering to about 3.5 percent below the closing price.
The sensitivity has an explanation. Broadcom has spent more than a year beating expectations and raising its outlook, and investors in the AI trade have come to price in continuous upside; a guide that merely meets the consensus can disappoint even when the quarter behind it was exceptional. The after-hours reaction said less about the quality of the business than about the expectations wrapped around it.
On the earnings call, Chief Executive Hock Tan pushed the story forward. He raised the company’s AI revenue target for fiscal 2027 to $115 billion from $100 billion, and sketched a fiscal 2028 figure of $230 billion, saying the six custom-chip customers are accelerating their deployments rather than deferring them. For a sense of scale: the whole company generated $29.59 billion of revenue in the quarter just reported, and Mr. Tan is describing an AI business that would produce nearly twice that amount in a single quarter by fiscal 2028.
The raised targets are a bet that the custom-silicon wave has years left to run. The customers commissioning these chips are spending on the assumption that demand for AI computing will keep outstripping supply, and that owning the design of the chips they buy is cheaper than buying a merchant product at whatever price the market will bear. Broadcom has become the design partner of record for that strategy, selling not just silicon but the architecture, the networking and the engineering that surrounds it.
The other side of the story is concentration. Nearly all of Broadcom’s AI growth rests on six customers’ budgets and product roadmaps, and a delay at any one of them would show up immediately in the numbers. The same customers are among the most demanding buyers in technology, and several maintain design teams capable of taking work in-house; Broadcom must keep winning programs its own clients could, in principle, build themselves. Analysts who follow the company say the margin guidance, at roughly 66 percent on a non-GAAP basis, shows how much of Broadcom’s AI revenue is design and intellectual property rather than commodity hardware, but also that the mix depends on programs staying on schedule.
The quarter leaves Broadcom in an unusual position: reporting AI growth that nearly every chip company would envy, yet judged on a forecast that fell a fraction of a percent short of what investors had already decided they wanted. The punishment was swift and then partly withdrawn, as investors digested the raised targets and concluded the long-term story had not changed. For a company whose AI revenue grew 221 percent, the surprise of the day was that the market wanted more still, and that the penalty for slightly less arrived within the hour.
