Broadcom to power Google’s next‑gen AI chips and deepen partnership with Anthropic

Broadcom has officially confirmed that it will manufacture future generations of Google’s custom AI chips, cementing its role at the heart of generative AI infrastructure. The deal covers Tensor Processing Units (TPUs), which power many of Google’s internal services as well as workloads for some of its cloud customers. At the same time, Broadcom announced an expanded partnership with AI startup Anthropic, which is set to gain access to around 3.5 GW of compute based on these chips.

Analysts at Mizuho estimate that the Anthropic partnership alone could generate up to $21 billion in AI‑related revenue for Broadcom in 2026 and as much as $42 billion in 2027, although the company itself has not disclosed specific figures. In parallel, Broadcom is also working on custom hardware for OpenAI, making it one of the central players in the emerging market for bespoke AI accelerators. In practice, this means that more and more large‑scale models deployed on Google Cloud will be able to run on alternatives to Nvidia’s ecosystem.

Experts note that the “Google + Broadcom + major AI startups” triangle is creating a new layer of vertical integration: from data centers and networking up to applied AI services. For the broader market, this is a signal that the “AI chip race” is moving beyond general‑purpose GPUs and increasingly into highly customized designs built for specific cloud platforms and their top‑tier customers.