Anthropic Commits $45 Billion to Nscale Compute Deal for Nvidia Rubin Chips

Anthropic has agreed to rent roughly $45 billion in artificial intelligence computing capacity from British infrastructure firm Nscale, marking one of the largest hardware supply commitments ever made by an independent AI lab.

The six-year agreement will see Nscale supply processing power from its flagship data center in West Virginia, utilizing Nvidia's forthcoming Vera Rubin architecture. The Vera Rubin platform, which coordinates six distinct chips to handle frontier-model workloads, is slated to begin powering Anthropic's systems in late 2027.

Nscale, established in 2024, has rapidly positioned itself as an intermediary between specialized chipmakers and AI operators, having already inked supply agreements with hyperscalers like Microsoft. Securing Anthropic's business represents a major validation for the young infrastructure provider.

A Multi-Billion-Dollar Procurement Blitz

The Nscale transaction is the latest in an intensive, eight-month compute buying spree by Anthropic as it races to match the infrastructure scale of OpenAI and Google. Earlier this month, the company entered a six-year, $10 billion compute agreement with Volta to pull power from a data center in Norway, which followed a $5 billion capacity deal with AMD in July.

Anthropic's procurement push extends across multiple corporate ecosystems. In May, the firm signed a capacity pact with SpaceX worth approximately $1.25 billion per month across two data facilities. In April, the lab broadened its relationships with Amazon—gaining access to an incremental 5 gigawatts of power—as well as Broadcom and Google.

The Scale Report View

The staggering price tag reflects an era where frontier AI development is increasingly defined by capital structure rather than algorithmic novelty alone. By locking in long-term off-take contracts years ahead of physical hardware deployment, foundation model builders are taking on unprecedented forward balance-sheet commitments to avoid being starved of next-generation silicon when new architectures arrive.