Broadcom's AI Revenue Keeps Accelerating. The Market Just Keeps Raising the Bar

Broadcom reported a record third quarter on Wednesday evening: revenue of $29.6 billion, up 86% from a year earlier, with non-GAAP EPS of $3.32. Both beat the street — consensus had looked for roughly $29.4 billion and about $3.22 a share.

And in the first minutes after the print, the stock still slipped a few percent.

That reaction is not a contradiction. It is the whole point of owning this stock right now. Broadcom keeps delivering exactly what it promised, and the market keeps moving the finish line. To understand whether that is a reason to buy, sell, or simply wait, you have to separate the record from the report the market was actually grading.

The June lesson: a record quarter was never enough

This is the second straight quarter the script has played out. Back in June, Broadcom's fiscal second quarter was a blowout — revenue up 48% to $22.2 billion, AI chip revenue more than doubled to $10.8 billion — and the stock still fell about 15% the next session.

The reason was guidance, not results. Management pointed AI chip revenue to about $16 billion for the quarter now behind us and re-stated, without raising, its target of more than $100 billion in AI semiconductor revenue for fiscal 2027. Investors who had already priced in even more acceleration sold what they thought was a pause. The market wasn't worried about whether Broadcom's AI business was growing; it was worried about whether it was growing fast enough against a hurdle that keeps resetting higher.

What changed this quarter: the numbers accelerated

Here is what makes this report different. The AI chip revenue that was guided to about $16 billion came in at $16.7 billion for the third quarter — up 221% year-over-year and 54% sequentially. More importantly, management did not hold the line this time. It guided fourth-quarter AI semiconductor revenue to about $21.7 billion, a 236% jump, and total revenue to roughly $34.8 billion, up 93%.

Read that for what it is. The fear baked into the June selloff was deceleration — that the AI surge would plateau. This quarter, the company told you the opposite: the next quarter's AI number is nearly a third bigger than the one it just printed. The bear case that the cycle is stalling did not show up here.

The economics that make this name different

What deserves attention is how Broadcom makes money on all this. Its free cash flow for the quarter was $13.7 billion — about 46% of revenue — against capital spending of only about $0.5 billion. Broadcom designs the custom accelerators and networking chips but does not own the factories, the power plants, or the data-center land the hyperscalers are buying.

That is the crucial contrast with the rest of the AI trade. Hyperscalers and foundries are still spending enormous sums on capacity that shows up as cash burn before it becomes profit. Broadcom is asset-light by design, so a surge in AI demand converts to cash almost immediately. The pressure that makes you nervous about other AI names — negative free cash flow during the buildout — is the absence of that problem here. This is AI revenue you can see turning into cash flow today, not a promise about 2030.

The multi-year claim, and the risk it carries

The bull case leans on the multi-year plan. CEO Hock Tan has said the company has line of sight to more than $100 billion in AI semiconductor revenue for fiscal 2027, backed by six core custom-chip customers — including examples like Anthropic, Google, Meta, and OpenAI — with order visibility stretching to 2028.

That is a real anchor, but it is also where the risk lives, and it deserves honest pressure. The concentration is extreme: a handful of customers, and a couple of giants within that handful, account for almost all of the AI pipeline. If one hyperscaler shoves its custom-silicon program out a year, the quarterly numbers wobble even though the long-term story survives. Broadcom has also said it will sell "chips only," stepping back from building complete AI systems, and management has noted bookings are not immediate deliveries — customers still need time to align power and infrastructure before taking chips.

So the forward-looking AI revenue is line of sight, not delivered revenue. It is a roadmap with a widening risk boundary — memory, packaging, power, and the timing of a few very large orders — even as it gives suppliers unusually clean visibility.

The so-what

None of this makes the stock cheap. It trades around $367, roughly a quarter below its record high, and has fallen about 12% over the past month — the market has already pulled forward enormous AI expectations, which is exactly why a record quarter can still produce a muted reaction. The earnings call later tonight is where the next signal sits: watch whether management firms up the fiscal 2027 AI revenue figure rather than just reaffirming "in excess of $100 billion."

The rerating condition is revenue conversion plus that multi-year number actually holding. The break condition is equally specific: a hyperscaler pushing custom-accelerator orders further out, or a single quarter where guided AI revenue goes flat or down after all this acceleration. As long as the guide keeps moving up and the cash keeps arriving, the deceleration story remains a fear, not a fact — and the stock's job is to keep making you wait for proof.