Anthropic targets 5 gigawatts of compute by year end, New York Times reports
The figure covers an agreement with Amazon that includes nearly 1 gigawatt of Trainium2 and Trainium3 capacity due online by the end of 2026.
Published
Anthropic plans to have 5 gigawatts of compute capacity in place by the end of the year, the New York Times reported.
Anthropic has said it signed an agreement with Amazon covering up to 5 gigawatts of capacity for training and running its Claude models, with nearly 1 gigawatt of Trainium2 and Trainium3 chips coming online by the end of 2026, according to the company. Anthropic is committing more than $100 billion over the next ten years to AWS technologies as part of that deal, the company said. Separately, Anthropic has an agreement with Google and Broadcom for several gigawatts of next-generation TPU capacity, expected to start coming online in 2027, the company said.
The Amazon and Google deals follow a shift by Anthropic and OpenAI toward smaller data center agreements for AI computing capacity. The Information has reported that Anthropic has amassed up to $517 billion in compute commitments covering 14.8 gigawatts since October 2025, on top of the one to two gigawatts it already held, putting its total planned capacity at roughly 16 to 17 gigawatts, according to the report as cited by 24/7 Wall St. and Overcentral. That compares with around 12 gigawatts of Nvidia compute in OpenAI's existing and planned commitments, 24/7 Wall St. reported.
Anthropic confidentially submitted a draft registration statement to the Securities and Exchange Commission on June 1, 2026, ahead of a planned initial public offering, according to Measured AI. The 5-gigawatt target marks a step up from the smaller data center deals Anthropic had been pursuing, and signals confidence in near-term demand for Claude as the company prepares to go public and compete with OpenAI and Google for AI infrastructure. The scale of the commitment could strengthen its position with investors once it lists.