Broadcom's AI Revenue Is Growing at 221%. Here's Why Custom Chips Could Be a Bigger Business Than GPUs.

Broadcom (AVGO) just posted a 221% year-over-year surge in AI semiconductor revenue for Q3, and the driver is custom silicon, not GPUs. The growth comes from custom AI accelerators (XPUs) that AVGO designs with hyperscalers including Google, Meta, and OpenAI.

The pitch to those customers is practical. Custom chips deliver better power efficiency, faster inference, and higher returns on capital than GPUs. That last point matters as hyperscalers face capital constraints, which makes spending efficiency a priority.

Scale still favors NVDA. Its data center business is currently 4x larger than Broadcom's AI business, so this is a story about trajectory, not parity. The argument is that custom silicon is winning by volume among major customers and could eventually become a bigger business than GPUs.

Interpretation: a 221% growth rate off a smaller base, paired with the 4x size gap, suggests AVGO is closing ground fast. If hyperscaler budgets tighten, the efficiency case for XPUs strengthens relative to merchant GPUs. The named customers also tie AVGO's growth to the capex decisions of GOOGL, META, and OpenAI, which concentrates both the upside and the risk.