Anthropic commits to up to a million Google TPUs

Anthropic and Google Cloud announced an expanded agreement giving Anthropic access to up to one million Google Tensor Processing Units, with more than a gigawatt of the added capacity expected online in 2026. The companies did not disclose an exact price but said the deal was worth tens of billions of dollars, and Google Cloud called it the largest expansion of Anthropic’s TPU usage to date.

The agreement extended a relationship the two companies had established in 2023, under which Anthropic trains models on Google’s infrastructure and distributes Claude through Vertex AI and the Google Cloud Marketplace to customers including Figma, Palo Alto Networks and Cursor. Neither company’s announcement addressed how the arrangement related to Anthropic’s other compute suppliers — Amazon, through its own multi-billion-dollar investment and Trainium chip commitments, is the other major backer of Anthropic’s infrastructure — leaving the total scale of Anthropic’s combined compute commitments across providers undisclosed.

The deal arrived one day after the Future of Life Institute’s Statement on Superintelligence and one month into a run of multi-gigawatt compute announcements from OpenAI covering NVIDIA, AMD and Broadcom, underscoring that the scramble to secure chip and power capacity was not confined to OpenAI. Coming from a lab that had positioned itself as the more safety-conscious of the two leading US developers, the scale of the commitment illustrated how thoroughly the industry’s competitive dynamics — securing enough compute to train and serve next-generation models — had converged across labs regardless of differences in stated mission or governance structure.