Anthropic signs $45 billion cloud computing deal with Nscale

Anthropic has signed a $45 billion, six-year cloud computing deal with Nscale, a British AI infrastructure company, to lease approximately 460 megawatts of computing capacity from a flagship data center Nscale is building in West Virginia. The facility is expected to come online at the end of 2027, and the arrangement will draw on Nvidia's next-generation Vera Rubin chip systems once they ship.

The deal is the latest and largest in an aggressive compute-buying spree Anthropic has pursued over the past eight months as it races to keep pace with rivals, most notably OpenAI, amid infrastructure strain that has affected Claude's reliability and performance during peak usage periods.

A pattern of escalating compute deals

The Nscale agreement follows a string of major capacity commitments: a $10 billion, six-year deal with Norwegian data center operator Volta in August 2026; a $5 billion compute arrangement with AMD in July; a SpaceX partnership announced in May providing roughly $1.25 billion worth of capacity per month across multiple data centers; expanded Amazon access adding 5 gigawatts of capacity in April; and a broadened Google/Broadcom partnership, also in April. Taken together, these deals represent tens of billions of dollars in committed spending inside a single year — a pace that reflects both genuine demand growth and the competitive pressure of an AI arms race where compute capacity has become the binding constraint on how fast any lab can ship and serve its models.

Why Nscale, and why West Virginia

Nscale, founded in 2024, has built partnerships with Microsoft and other major cloud and AI players in a short span, positioning itself as a fast-moving alternative to hyperscale cloud providers for AI-specific infrastructure. The West Virginia site represents a significant bet on U.S. data center capacity outside the traditional hubs of Virginia's 'Data Center Alley' and the Pacific Northwest, part of a broader industry trend of AI infrastructure spreading into new regions as power availability — not land or capital — becomes the primary bottleneck on where new facilities can be sited.

The choice of Nvidia's Vera Rubin systems, the successor to the current Blackwell generation, signals that this capacity is being built for training and serving Anthropic's next generation of models rather than simply expanding current inference capacity.

What this means for the broader AI infrastructure race

Anthropic's compute spending is part of an industry-wide pattern: Google, OpenAI, and Meta are each pursuing multi-billion-dollar capacity deals of their own, and the aggregate scale of commitments across the sector now runs into the hundreds of billions of dollars for infrastructure that in some cases won't be operational for another year or more. For Anthropic specifically, the reliability issues that have prompted this spending spree — service slowdowns and capacity constraints during high-demand periods — underscore how tightly coupled a frontier AI lab's product quality has become to its ability to secure raw compute, well ahead of any breakthrough in model efficiency that might reduce that dependency.