Anthropic extends its Google and Broadcom deal to multiple gigawatts of TPUs from 2027

2026-04-06 · Publisher : Anthropic · Domain : Cloud & DevOps

Anthropic signs with Google and Broadcom for multiple gigawatts of next-generation TPU capacity from 2027, mostly in the United States, after announcing run-rate revenue above USD 30 billion.

### Background On 6 April 2026, Anthropic announces a new agreement with Google and Broadcom for **multiple gigawatts of next-generation TPU capacity**, expected online starting in 2027. Anthropic calls it its most significant compute commitment to date. It says its run-rate revenue now exceeds USD 30 billion, up from about USD 9 billion at the end of 2025, and that more than 1,000 customers each spend over USD 1 million a year, up from more than 500 in February. Most of the new capacity will be sited in the United States. ### Key facts - Volume: multiple gigawatts of next-generation TPU capacity, from 2027. - Figures in context: run-rate revenue above USD 30 billion in April 2026, up from about USD 9 billion at end 2025, according to Anthropic. - Customers: more than 1,000 customers above USD 1 million a year, versus more than 500 in February 2026. - Location: most of the capacity in the United States, extending the USD 50 billion commitment of November 2025. - Multi-platform: Claude runs on AWS Trainium, Google TPUs and NVIDIA GPUs; Amazon remains the primary cloud provider and training partner. ### Why it matters Anthropic confirms a model where one lab reserves multiple gigawatts from several providers, with horizons of two to three years. For IT departments, it is an indicator of the pressure on AI capacity and of how model availability depends on these reservations. The announced ramp-up is a capacity intention: the revenue figures are Anthropic's. ### Takeaway Anthropic reserves multiple gigawatts of TPUs for 2027, extending its October 2025 agreement.

Tags : Puces, Anthropic, Google Cloud, Multi-cloud, Datacenter