Broadcom’s AI Chip Revenue Jumped 221% Last Quarter. Here’s What’s Driving It.

Broadcom reported its fiscal third-quarter earnings yesterday — and the AI numbers are genuinely hard to absorb. AI semiconductor revenue reached $16.7 billion in the quarter, up 221% year over year and 54% from the previous quarter. Total consolidated revenue hit $29.6 billion, up 86%.

Then came the guidance. Broadcom expects AI semiconductor revenue of $21.7 billion in the fourth quarter alone — a 236% year-over-year increase. Full-year fiscal 2026 AI semiconductor revenue is now guided to $58 billion, up 186% from last year. And beyond that: management said it has line of sight to $115 billion in AI semiconductor revenue in fiscal 2027 and $230 billion in fiscal 2028.

Those numbers are so large they require context to make sense. Here’s the context.

What Broadcom Actually Makes

Most people in the tech industry know Nvidia. Far fewer know Broadcom — despite the fact that it is just as central to the AI infrastructure boom, in a completely different way.

Broadcom doesn’t make the GPUs that Nvidia is famous for. Instead, it designs custom AI accelerators — called XPUs — that are purpose-built for specific customers’ workloads. It also makes the networking chips that connect thousands of GPUs together inside AI data centres. Both categories are in extraordinary demand.

The custom accelerator business is particularly important to understand. Google’s TPUs — the chips powering Google’s AI training and the ones being supplied to Anthropic — are designed by Google and manufactured by TSMC, but the custom silicon design work runs through Broadcom. So do Meta’s MTIA chips for recommendation and inference at scale, and OpenAI’s “Jalapeno” custom chips. CEO Hock Tan confirmed yesterday that Broadcom expects Anthropic to deploy 5 gigawatts of TPU 8i chips in 2027, with line of sight to deliver another 10 gigawatts.

This is the business model that makes Broadcom’s numbers possible. Rather than selling off-the-shelf silicon at competitive prices, it builds bespoke silicon for the largest AI spenders in the world — companies that can commit to enormous multi-year volumes in exchange for chips optimised specifically for their workloads.

Why the Stock Fell After the Results

Broadcom’s share price dropped roughly 5% after hours despite the extraordinary numbers. This sounds paradoxical. The explanation is in the expectations.

Wall Street had modelled Q4 AI semiconductor guidance of approximately $22 to $23 billion. Broadcom guided to $21.7 billion. That miss — of roughly $300 million to $1.3 billion on a $21.7 billion figure, meaning less than a 6% shortfall — was enough to trigger a selloff.

The same pattern played out after Broadcom’s Q2 results in June, when the stock fell 13% despite reporting $10.8 billion in AI revenue up 143% year over year. The market had priced in an even higher number. Extraordinary growth, extraordinary guidance, and a market that consistently prices in something more extraordinary still.

The Long-Range Forecast Is the Headline

The quarterly numbers are striking. The multi-year forecast is what makes this story significant for anyone thinking about where the AI industry is heading.

Broadcom is guiding to $115 billion in AI semiconductor revenue in fiscal 2027 and $230 billion in fiscal 2028. The company says it has supply secured for those targets — and that demand is currently running ahead of even those projections.

To put $230 billion in perspective: Nvidia’s total revenue for all of fiscal 2025 was approximately $130 billion. Broadcom is projecting that its AI chip business alone — not counting its infrastructure software division, which contributes another $8-9 billion per quarter — will be nearly double that in two years.

These projections rest on commitments from a small number of hyperscale customers. Broadcom has six named XPU customers — Google, Meta, OpenAI, and others not publicly named — all of which are making multi-year commitments to custom silicon as a deliberate strategy to reduce reliance on Nvidia GPUs. The April 2026 long-term agreement with Google to supply custom AI chips through 2031, extending a partnership that has been in place for about a decade, is the clearest example of how durable these relationships are.

What This Means for the AI Chip Landscape

Broadcom’s results, read alongside Nvidia’s acquisition of Hugging Face this morning, paint a picture of an AI chip market that is simultaneously consolidating and diversifying.

Nvidia is the dominant GPU supplier and is now reinforcing its position by acquiring the open-source platform most developers depend on. Broadcom is the dominant custom chip designer and is watching its revenue scale at triple-digit rates as hyperscalers invest billions in bespoke silicon. Both are winning — but they’re winning in different parts of the market.

The companies that are losing are the ones that failed to establish deep relationships with hyperscalers early: smaller GPU makers, startups promising alternative architectures, and anyone trying to compete with Nvidia on GPU performance alone. The market is not a monolith — but the leaders are entrenching fast. For more on the AI chip race, see our earlier coverage of AMD’s $5 billion investment in Anthropic and the Qualcomm-Tenstorrent acquisition talks.

Read more tech related articles here.