Broadcom Raised Its AI Revenue Target to $58 Billion. The Bar, Not the Business, Is the Story

Broadcom just did the thing that, three months ago, cost its shareholders $280 billion. On September 2, after reporting its fiscal third quarter, CEO Hock Tan raised fiscal 2026 AI revenue guidance to $58 billion from $56 billion — roughly 186% above last year. The growth behind the number is hard to argue with: AI semiconductor revenue more than tripled to $16.7 billion in the quarter, up 221% year over year, and fourth-quarter guidance points to an acceleration to about $21.7 billion, a 236% jump. Total revenue came in at $29.6 billion, up 86%.

The instructive part is not that Broadcom raised its number. It is how the market treated the same company the last time it stood exactly in this spot. In early June, Broadcom beat earnings and guided fiscal 2026 AI revenue at $56 billion — and the stock collapsed roughly 13% in a single day, erasing about $280 billion of market value in one of the largest single-stock wipeouts of the megacap era, because investors expected a bigger jump and a raised long-term target, and Tan held the line instead. The shares, which peaked near $495, now trade around $367, about 25% below that high.

That context reframes what this raise actually proves. Broadcom's AI business is not the question the stock is arguing about. It is the biggest winner — by a wide margin — of AI demand that does not run through Nvidia.

Broadcom builds the custom AI accelerators the hyperscalers design themselves rather than buying Nvidia's chips off the shelf: Google's TPUs chief among them, plus custom-silicon commitments from Meta, OpenAI, and Anthropic. These chips win on cost per watt and efficiency for a workload the customer controls — the dimensions that matter more as AI shifts from the giant training runs that built Nvidia's monopoly into running inference at scale, where hyperscalers want cheap, specialized compute of their own. That is the specific stage where Broadcom's design wins and its networking hardware give it the clearest "who is better at which stage" case against Nvidia.

The commitments behind that story are enormous, and that is exactly why the raise reads as two things at once. more than 10 gigawatts of 2027 AI chip shipments, a multi-generation TPU agreement with Google, an additional 5 gigawatts for Anthropic beginning in 2027, 1.3 gigawatts for OpenAI next year, and 3 gigawatts from Meta through 2028, while bookings for the quarter exceeded $30 billion. The fiscal 2027 target stands at AI revenue in excess of $100 billion. That is genuine, documented, multi-year visibility.

It is also concentrated visibility. That backlog belongs to a handful of hyperscalers, and their capital-spending decisions — not Broadcom's engineering — are the swing factor. A single pause in hyperscaler spending discipline can turn "visibility" into "leverage" in a hurry, which is the dual reading these supply commitments demand: record demand and rising delivery risk in the same number.

Where does that leave an investor deciding whether any of it matters to them? The raise confirms delivery; it does not build a new bull case. The growth is real, multi-year, and now on the books. But at a market value near $1.75 trillion, the stock trades well below its high precisely because the market is already paying for a $100 billion AI year in 2027 — roughly 19 times those expected fiscal-2027 earnings, against about 25 times at the peak. The near-term return no longer depends on whether Broadcom grows; it depends on whether Broadcom keeps raising a bar the market has already set very high.

June proved the market can erase $280 billion in a day on a quarter where the business was plainly fine. That is the opportunity cost of owning a stock priced close to perfection: the execution is excellent, but the multiple leaves little room for a quarter that merely meets the climb rather than exceeds it. Broadcom has one of the strongest multi-year demand stories in this cycle — the question is whether the current price has already spent the best of it, and whether each guide-up from here can beat a bar that only gets higher.