Anthropic signs multi-gigawatt TPU deal with Google and Broadcom

Anthropic said on April 6, 2026 that it signed a new agreement with Google and Broadcom for multiple gigawatts of next-generation TPU capacity beginning in 2027. The company described it as its biggest compute commitment so far, designed to power frontier Claude models and keep up with what it called extraordinary customer demand. In practical terms, this is a capacity reservation story: Anthropic is locking in future training and inference supply before competition for advanced AI hardware becomes even tighter.

The announcement also gave a rare look at demand. Anthropic said run-rate revenue has surpassed $30 billion, up from about $9 billion at the end of 2025. It also said the number of business customers spending more than $1 million on an annualized basis has risen above 1,000, double the level it cited when it announced its Series G fundraising in February. Those numbers help explain why the company is willing to commit so aggressively to long-dated infrastructure: it is not only chasing model leadership, but also trying to preserve service reliability as enterprise adoption accelerates.

Anthropic positioned the TPU deal as part of a broader multi-platform strategy rather than a move away from other partners. The company said it trains and runs Claude across AWS Trainium, Google TPUs, and NVIDIA GPUs, while Amazon remains its primary cloud provider and training partner through Project Rainier. Anthropic also emphasized that Claude is available across the three largest cloud platforms through Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Azure Foundry. The message is that platform diversity is now part of the product promise.

Why this deal stands out

Multiple gigawatts of future TPU capacity is not a routine procurement announcement. It signals that compute planning has become a strategic moat on par with model quality and distribution. By tying together Google, Broadcom, and its existing cloud relationships, Anthropic is trying to reduce supply risk while keeping leverage across vendors. For the broader market, the deal is another reminder that frontier AI competition is increasingly shaped by who can secure power, silicon, and data-center scale years ahead of actual model launches.