Anthropic’s big compute deal
Anthropic signed one of the largest AI‑compute deals recently, with substantial capacity coming from Google rather than Nvidia, signalling more competition in AI infrastructure choices. (timesofindia.indiatimes.com) The arrangement highlights that hardware and cloud partnerships are now strategic business decisions, not just technical details. (timesofindia.indiatimes.com)
Anthropic signed a new deal with Google and Broadcom for multiple gigawatts of artificial intelligence computing power, with capacity starting to come online in 2027. (anthropic.com) Anthropic said the agreement is its biggest compute commitment yet and will use Google’s Tensor Processing Units, or custom chips built for artificial intelligence workloads. Broadcom disclosed in a securities filing that Anthropic will access about 3.5 gigawatts through that arrangement beginning in 2027. (anthropic.com) (sec.gov) This expands a Google relationship Anthropic had already deepened on October 23, 2025, when it said it planned to use up to 1 million Google Tensor Processing Units in a deal worth tens of billions of dollars. Anthropic said that earlier expansion was expected to bring well over one gigawatt of capacity online in 2026. (anthropic.com) The chips matter because training and running large language models depends on vast clusters of processors, power, and networking, not just software. Google designs its own Tensor Processing Units, while Nvidia sells the graphics processing units that have dominated most commercial artificial intelligence deployments. (googlecloudpresscorner.com) (cnbc.com) Anthropic’s new deal shows that a top model maker is locking in supply with a cloud provider and a custom-chip partner, not only with Nvidia-based systems. Broadcom said separately that it will keep producing future versions of Google’s artificial intelligence chips under a supply agreement that runs through 2031. (cnbc.com) (sec.gov) Anthropic tied the expansion to surging demand for Claude. On April 6, 2026, the company said its annualized revenue run rate had reached $30 billion, up from $9 billion at the end of 2025, and that more than 1,000 business customers were spending over $1 million a year on an annualized basis. (techcrunch.com) (bloomberg.com) The arrangement also complicates Anthropic’s cloud picture. Amazon said in March 2024 that Anthropic was using Amazon Web Services as its primary cloud provider for mission-critical workloads and future foundation model development, even as Google continued investing and supplying chips. (aboutamazon.com) (cnbc.com) Google’s financial ties to Anthropic run alongside the infrastructure deal. Google committed up to $2 billion in October 2023, and later reporting said its total investment had risen to more than $3 billion through additional financing arrangements. (cnbc.com) (softwareunplugged.com) Google said thousands of customers already access Claude through Google Cloud, including Coinbase, Cursor, Palo Alto Networks, Replit, and Shopify. That gives the compute deal a second purpose: it supports both Anthropic’s model-building and Google Cloud’s effort to sell artificial intelligence infrastructure to outside companies. (googlecloudpresscorner.com) By 2027, the contest will not be only about who has the best model. It will also be about which companies secured enough chips, electricity, and cloud capacity early enough to keep those models running. (anthropic.com) (sec.gov)