Anthropic has reached an agreement to lease about $45 billion in AI compute capacity from British infrastructure firm Nscale, a person familiar with the deal told TechCrunch. The capacity will run on systems built around Nvidia’s Vera Rubin chips.
Founded in 2024, Nscale has already struck deals with major partners such as Microsoft. The Vera Rubin system combines six different chips working in concert and is regarded as a state‑of‑the‑art design. According to the source, the compute capacity is expected to begin powering Anthropic’s services in late 2027.
Deal specifics and Anthropic’s broader push
Bloomberg first reported the agreement; the deal reportedly covers six years and sources compute from Nscale’s flagship data center in West Virginia. Over the past eight months, Anthropic has been rapidly expanding its compute footprint to meet rising needs.
This agreement is one of several large, multiyear capacity deals the company has made recently. Earlier this month Anthropic signed a $10 billion contract with AI cloud startup Volta — founded this past January — securing a six‑year supply of cloud compute from a data center in Norway. In July, Anthropic also signed a $5 billion compute‑related deal with AMD.
In May, Anthropic disclosed a major computing agreement with SpaceX; that arrangement draws capacity from two different SpaceX data centers and is reported to provide roughly $1.25 billion worth of capacity each month. In April, the company expanded its partnership with Amazon, gaining access to an additional 5 gigawatts of compute, and in the same month extended relationships with Google and Broadcom.
Why it matters
Rapidly accumulating compute capacity is a strategic priority for AI developers because large-scale resources are central to training and running advanced models. Anthropic’s multibillion‑dollar, multiyear deals — including the $45 billion arrangement with Nscale — reflect an aggressive infrastructure build‑out aimed at keeping pace with competitors. Other major AI players, including Google, OpenAI, and Meta, are pursuing similar strategies to secure the compute needed for future model development and deployment.
