Broadcom CEO Hock Tan just handed investors a pretty clear signal about where the AI infrastructure wars are heading. During the company’s Q3 2026 earnings call, Tan revealed that Anthropic, the Claude AI maker, is on track to become Broadcom’s single largest customer for custom AI accelerators, known as XPUs, in 2027. And it’s expected to hold that top spot into 2028.
The numbers behind Anthropic’s compute build-out
Anthropic’s hardware ambitions are staggering even by the standards of the current AI arms race. The company placed a $10 billion order for Google TPU Ironwood racks back in December 2025, followed by an additional order valued at $11 billion.
The deployment timeline paints an equally aggressive picture. Anthropic is planning 1 gigawatt of TPU capacity in 2026, scaling to 5 GW of TPU v8i capacity in 2027, with the potential to reach 10 GW in 2028.
Broadcom has already begun production shipments of the next-generation TPU v8i, which will power a significant chunk of that planned capacity. The company is also supplying custom accelerators for OpenAI’s Jalapeño chips, meaning Broadcom is quietly becoming the foundry of choice for the two most prominent frontier AI labs on the planet.
Why custom silicon is eating the GPU market
Custom accelerators, or XPUs, are designed for specific workloads, which means they can deliver better performance per watt and per dollar for the companies that can afford to design them. Broadcom’s AI semiconductor revenue is expected to double in the coming years, driven almost entirely by this demand for custom silicon.
What this means for the AI infrastructure landscape
Broadcom is now confirmed as a custom silicon supplier for at least three of the most important AI organizations in the world: Google, Anthropic, and OpenAI.
The power consumption figures also deserve attention from anyone tracking the real-world constraints on AI scaling. Going from 1 GW to 10 GW of deployed compute in two years would represent an unprecedented ramp in energy demand from a single company.