Anthropic in Talks With Google for Major Cloud Compute Deal

Anthropic is in advanced talks with Google to secure a large-scale cloud computing agreement that would give the artificial intelligence startup access to Google’s custom tensor processing units and more than one gigawatt of compute capacity, people familiar with the discussions said. The arrangement, which has not been finalized, could be worth in the high tens of billions of dollars and is intended to support training of next-generation versions of Anthropic’s Claude chatbot.

Under the proposed deal, Anthropic could obtain as many as one million of Google’s TPUs and access more than one gigawatt of computing capacity that is expected to come online in 2026, according to the sources. Google would also provide additional cloud services to run and deploy the models. Anthropic has said it chose Google’s TPUs for their price-performance and efficiency and because of its existing experience training and serving Claude on those processors.

The talks follow a rapid expansion in demand for specialized AI compute. Anthropic has projected steep revenue growth tied to enterprise adoption of its products and is working toward an internal target of a $9 billion annualized revenue run rate...