Anthropic expands compute partnership with Google and Broadcom

Anthropic announced an expanded compute agreement with Google and Broadcom, adding what it described as multiple gigawatts of next-generation TPU capacity, with the new supply expected to come online starting in 2027. The company said the expansion built on the $50 billion infrastructure commitment it had announced with the two firms in November 2025, and that the majority of the new capacity would be sited in the United States.

Anthropic tied the expansion directly to demand outpacing its existing capacity plans. It said run-rate revenue had passed $30 billion, up from roughly $9 billion at the end of 2025, and that the number of business customers spending more than $1 million a year with the company had passed 1,000 — a figure it said had doubled in under two months. Chief financial officer Krishna Rao said the company was “building the capacity necessary to serve the exponential growth we have seen in our customer base.”

The deal extends a compute strategy in which Anthropic spreads training and inference workloads across Google’s TPUs, Amazon’s Trainium chips and Nvidia GPUs rather than depending on a single supplier, a diversification the company has presented as both a resilience measure and a way to negotiate on price across providers. The pace of the revenue growth cited alongside the announcement — roughly tripling in a little over a quarter — underscored how quickly frontier labs’ compute commitments were scaling alongside claimed commercial traction in early 2026, a period in which several labs announced large infrastructure deals within days of each other.