Anthropic has agreed to rent approximately $45 billion worth of computing power from Nscale, a British infrastructure firm.
A source with knowledge of the arrangement told TechCrunch about the move.
The hardware behind the spend
Nscale was founded in 2024. The company has already struck agreements with Microsoft and will supply Anthropic using Nvidia’s Vera Rubin chips. This new system pairs six different processors to work together.
Capacity is expected to begin powering the AI lab’s services in late 2027, according to the source.
A pattern of rapid expansion
Bloomberg first reported the deal. The agreement spans six years and draws power from Nscale’s flagship data centre in West Virginia.
Anthropic has aggressively scaled its computing capacity over the past eight months in an effort to compete with rivals, most notably OpenAI.
Earlier this month, the company signed a $10 billion deal with Volta, an AI cloud startup founded in January. This secures a six-year supply of cloud computing from a data centre in Norway.
In July, Anthropic also signed a $5 billion compute-related deal with AMD.
May saw a large computing agreement with SpaceX. The firm is run by Elon Musk, whose rivalry with OpenAI CEO Sam Altman has made him an unlikely ally of Anthropic. The deal draws capacity from two different SpaceX data centres and reportedly provides $1.25 billion worth of capacity each month.
April brought a deal to significantly expand the partnership with Amazon, granting access to an additional 5 gigawatts of compute. That same month, the company expanded its relationship with Google and Broadcom, adding even more power capacity.
Anthropic is not alone in this pursuit. The race to acquire as much compute capacity as possible is ongoing, with other major players including Google, OpenAI, and Meta following a similar track.
What it means
For the people building these models, the practical result is a shift from renting cloud space to owning dedicated infrastructure. Anthropic is securing long-term, physical access to hardware rather than relying on public cloud providers. This reduces dependency on third-party pricing and ensures the specific chip architecture required for their latest models is available when needed. The strategy signals a move away from flexible, on-demand cloud usage toward a model where compute is treated as a permanent, capital-intensive asset.
