Broadcom's AI revenue is up 143%, and the guide implies it triples again
$10.80bn in AI semiconductor revenue, guided to $16.0bn next quarter at over 200% growth, on custom accelerators and switching silicon. The hyperscalers are voting for parts they design themselves.
Broadcom reported AI semiconductor revenue of $10.80bn, up 143%, and guided the next quarter to $16.0bn — over 200% growth — on custom accelerators and Tomahawk/Jericho switching silicon sold to hyperscalers.
Custom accelerators are the part that matters strategically. A hyperscaler commissioning its own ASIC is making a multi-year commitment to a workload shape it believes it understands well enough to freeze into silicon. Google and Meta have both moved this way. The economics work only at volumes where a few percentage points of efficiency exceed the design cost, which is itself a statement about how large these workloads have become.
The switching half is less discussed and structurally similar. As training and inference clusters scale, the interconnect stops being plumbing and becomes a determinant of utilisation — an accelerator waiting on the network is an accelerator not amortising. Selling both the custom compute and the fabric around it is a strong position.
What this does not show is displacement. Nvidia's data-centre revenue exceeded $50bn in the quarter, which is several times Broadcom's entire AI line. Custom silicon is growing faster from a smaller base, in workloads the buyer has already characterised; the merchant GPU keeps the work that is still changing shape.
The number to watch is whether the guided figure lands. A 200% guide is a statement about design wins already booked, and a miss would say more about hyperscaler capex intentions than any survey.
Science-Technology News — Broadcom: The Architect of Custom AI ASICs → · Yahoo Finance — The 2 Chip Portfolio: Is AMD or Broadcom the Better Complement to Nvidia? →