Anthropic strikes Google TPU deal

Anthropic said on May 19 that it had struck a multibillion-dollar deal with Google for TPU computing capacity, putting fresh attention on how frontier AI companies are locking up long-term access to chips and power. The company did not disclose a precise dollar amount in the social-media announcement, but described compute as a strategic priority and tied the agreement to model training and expansion. The post followed a broader April 6 announcement in which Anthropic said it had signed a new agreement with Google and Broadcom for multiple gigawatts of next-generation TPU capacity expected to come online starting in 2027. ### What exactly did Anthropic say on May 19? Anthropic said on May 19 that it had reached a multibillion-dollar agreement with Google for TPU capacity, according to the company’s announcement on X referenced in the source briefings. The company did not publish a contract value in that post, and the available public language described the arrangement in broad terms rather than as a line-by-line commercial disclosure. (anthropic.com) Google’s TPU, or Tensor Processing Unit, line is the company’s in-house AI accelerator family. Anthropic’s post framed the agreement around securing computing capacity for model workloads, a signal that the company is treating access to hardware as a core operating constraint rather than a routine cloud purchase. ### How does this fit with Anthropic’s earlier announcement? (usnews.com) Anthropic said on April 6 that it had “signed a new agreement with Google and Broadcom for multiple gigawatts of next-generation TPU capacity” and that the capacity was expected to come online starting in 2027. In the same announcement, CFO Krishna Rao called it the company’s “most significant compute commitment to date.” (anthropic.com) That April statement is the clearest official description now in public. Anthropic said the expanded infrastructure would power its frontier Claude models and help it meet customer demand, while adding that the vast majority of the new compute would be sited in the United States. ### Is there a number attached to the deal? Reuters reported on May 5, citing The Information, that Anthropic had committed to spend $200 billion with Google Cloud over five years as part of a recent agreement. (anthropic.com) Reuters said it could not independently verify that figure. Anthropic’s own April 6 announcement did not give a total contract value. (anthropic.com) It described the agreement in capacity terms — multiple gigawatts of next-generation TPU infrastructure — rather than as a fixed spending figure, and the May 19 social post described the arrangement as multibillion-dollar without a more precise number in the materials reviewed here. ### Why Google TPUs, and not just one hardware stack? (usnews.com) Anthropic said on April 6 that it trains and runs Claude on AWS Trainium, Google TPUs and Nvidia GPUs. The company said that mix lets it match workloads to the chips best suited for them and provides resilience for customers. Amazon remains Anthropic’s primary cloud provider and training partner, Anthropic said, adding that it continues to work with AWS on Project Rainier. (anthropic.com) That means the Google TPU agreement expands Anthropic’s hardware base rather than replacing its existing cloud relationships. ### What does this say about Anthropic’s scale right now? Anthropic said on April 6 that its run-rate revenue had surpassed $30 billion, up from about $9 billion at the end of 2025. (anthropic.com) The company also said the number of business customers spending more than $1 million on an annualized basis had risen to more than 1,000 from over 500 in less than two months. Those figures help explain why Anthropic is talking about compute in gigawatts rather than in ordinary cloud terms. (anthropic.com) The company linked the infrastructure expansion directly to demand for Claude and to serving customers worldwide. ### What comes next, and when? Anthropic said on April 6 that the next-generation TPU capacity covered by its agreement with Google and Broadcom is expected to come online starting in 2027. (anthropic.com) The company also said most of that new compute will be located in the United States. Future updates are likely to come through Anthropic’s announcements page, Google Cloud disclosures or additional company statements, with Google, Broadcom and Anthropic the named participants in the next phase of deployment. (anthropic.com)