Anthropic Commits $45 Billion to Nscale for Six Years of AI Compute

News Summary

Anthropic has agreed to spend roughly $45 billion over six years to rent AI computing capacity from Nscale, a London-based AI infrastructure company, at a flagship data center campus the company is building in West Virginia. The agreement, first reported by Bloomberg and confirmed by multiple outlets on August 26, 2026 (Eastern Time), is the latest in a rapid-fire string of multibillion-dollar compute commitments Anthropic has made over the past several months as it races to secure processing power ahead of a planned stock market listing.

The Deal in Detail

Under the agreement, Anthropic will rent approximately 460 megawatts of computing capacity at Nscale's data center development in West Virginia. The facility will run on Nvidia's next-generation Vera Rubin chip system, a platform that combines six chips working together and sits, according to industry observers, at the cutting edge of current chip design. The Vera Rubin hardware is expected to begin coming online in late 2027, which is when the new capacity is expected to start powering Anthropic's Claude models and other services.

Nscale, founded in 2024, has moved quickly to establish itself as a major player in AI infrastructure, previously striking deals with Microsoft. Its West Virginia campus spans roughly 2,250 acres, and the company estimates the site could theoretically support more than eight gigawatts of computing capacity at full build-out. Notably, Microsoft had signed a letter of intent for the same West Virginia site in March 2026 but walked away from the arrangement during the summer, opening the door for Anthropic to step in as the anchor tenant. Nscale is also reportedly preparing a U.S. stock market listing as soon as next month, seeking to raise capital that industry watchers estimate at around $3 billion.

Part of a Broader Compute Spending Spree

The Nscale agreement extends an aggressive run of infrastructure deals Anthropic has signed over roughly the past eight months as it scales up capacity to compete with rivals such as OpenAI, Google, and Meta, all of which are similarly racing to lock in maximum compute capacity. Recent commitments include:

• An approximately $10 billion, six-year deal with Volta for a data center in Norway, announced in August 2026.
• A roughly $5 billion agreement with AMD, announced in July 2026.
• A deal with SpaceX allowing Anthropic to use available capacity at SpaceX's Colossus data center in Memphis, Tennessee, reported at about $1.25 billion per month running through May 2029.
• An expansion of Anthropic's partnership with Amazon representing an additional 5 gigawatts of capacity, announced in April 2026.
• An expanded partnership with Google and Broadcom, also announced in April 2026.

Taken together, these deals underscore just how much computing power modern large language models require to train and to serve growing numbers of users, and how aggressively leading AI labs are locking in long-term supply agreements with chipmakers and infrastructure providers alike.

Context: An IPO on the Horizon

The flurry of compute deals comes as Anthropic prepares for a widely anticipated initial public offering. The company confidentially filed a draft registration statement with the U.S. Securities and Exchange Commission in June 2026, with reports pointing to a potential Nasdaq listing as early as October 2026. Securing large, multiyear compute commitments now allows Anthropic to demonstrate to prospective public investors that it has the infrastructure runway needed to keep training and serving increasingly capable models well into the future, even as demand for AI services continues to climb worldwide.

Why It Matters

For an AI lab, compute capacity is as fundamental as an oil supply is to an energy company: it determines how large a model can be trained, how quickly it can be improved, and how many users it can serve at once. By diversifying its supply across chipmakers (Nvidia, AMD), cloud and infrastructure partners (Amazon, Google, Nscale, Volta), and even unconventional sources like SpaceX's data centers, Anthropic is working to reduce its dependence on any single provider while racing to keep pace with the enormous computing demands of next-generation AI systems used by people and businesses around the world.