Anthropic has agreed to rent roughly $45 billion in AI compute from Nscale, a British infrastructure company founded in 2024, in a six-year deal that will run on Nvidia's new Vera Rubin chip system. Capacity will come from Nscale's flagship data center in West Virginia and is expected to begin powering Anthropic's services in late 2027. Bloomberg first reported the agreement.
Vera Rubin is Nvidia's next-generation system, combining six different chips working in concert and marketed as the current cutting edge of chip design. Nscale, despite being two years old, has already signed compute deals with Microsoft, and the Anthropic contract vaults it into the top tier of AI infrastructure suppliers. The $45 billion headline number puts the arrangement on par with the largest disclosed AI compute commitments to date.
It is also Anthropic's fifth major compute deal in eight months. Earlier in August, the company signed a $10 billion, six-year agreement with Volta, an AI cloud startup founded in January, for capacity out of a Norwegian data center. In July, Anthropic committed $5 billion to a compute-related deal with AMD.
Key facts
- 01Anthropic will rent roughly $45B in AI compute from Nscale over six years, powered by Nvidia's Vera Rubin chip system.
- 02Capacity comes from Nscale's flagship West Virginia data center and is expected to come online in late 2027.
- 03The deal is Anthropic's fifth major compute agreement in eight months, following a $10B Volta pact earlier in August.
- 04Anthropic's May deal with SpaceX delivers $1.25B of capacity per month across two data centers.
- 05An April expansion with Amazon added 5 gigawatts of compute on top of existing Google and Broadcom arrangements.
In May, Anthropic entered a large computing arrangement with SpaceX that reportedly provides $1.25 billion of capacity per month, drawn from two separate SpaceX data centers. That deal alone works out to roughly $15 billion a year — a striking figure given the well-known rivalry between SpaceX chief Elon Musk and OpenAI CEO Sam Altman, which has effectively turned Musk into a compute supplier for one of OpenAI's closest competitors.
April brought two more expansions. Anthropic significantly enlarged its Amazon partnership, gaining an additional 5 gigawatts of compute, and it deepened its relationship with Google and Broadcom to add still more power capacity. Taken together with the Nscale agreement, Anthropic has now stacked disclosed compute commitments worth more than $60 billion, on top of the SpaceX monthly run-rate.
The pace tells its own story about where frontier AI economics have landed. Training and serving models at Anthropic's scale now requires locking in multi-year, multi-gigawatt supply years in advance, because the physical buildout — power, cooling, chip allocations from Nvidia — cannot be conjured on demand. The Nscale capacity does not come online until late 2027, meaning Anthropic is buying compute today that will only be usable more than a year from now.
Anthropic's motivation is straightforward: keep pace with OpenAI, which has been on its own compute-securing spree with Microsoft, Oracle, and others, and stay in reach of Google and Meta, both of which are building out custom silicon and hyperscale data centers on their own balance sheets. Anthropic does not own its own hyperscaler, so it is stitching together supply from Amazon, Google, SpaceX, AMD, Volta, and now Nscale — a diversified compute stack that reduces single-vendor risk but adds coordination overhead.
The Nscale arrangement is notable for its choice of hardware. Vera Rubin will be one of the earliest large-scale customer deployments of Nvidia's new architecture, and the late-2027 timing suggests Nscale is committing to a specific Nvidia roadmap slot well ahead of general availability. For Nvidia, locking a marquee frontier-lab tenant onto Vera Rubin capacity at a two-year-old British startup is a useful proof point for the broader Rubin cycle.
The West Virginia siting is also worth noting. The state has been courting hyperscale data center investment on the strength of available land and grid interconnects, and a facility large enough to host $45 billion of compute represents a substantial local power and construction footprint. Nscale has not publicly detailed the facility's power draw or cooling design.
The risk in all of this is timing. Anthropic is signing six-year deals in a market where model architectures, chip generations, and unit economics are all moving quickly. If test-time compute demand grows as projected, the company will look prescient. If efficiency gains from smaller models or algorithmic improvements compress inference costs faster than expected, some of this capacity will land as expensive overhang. Neither Anthropic nor Nscale has disclosed pricing terms or off-ramps.
For the AI market, the Anthropic-Nscale deal is another data point that the frontier is now a capital-allocation contest as much as a research contest. The labs that can credibly finance and secure late-decade compute — through equity raises, cloud partnerships, or rental agreements like this one — are the ones that get to stay at the frontier. Anthropic has spent eight months answering that question in public, one multibillion-dollar contract at a time, and the Nscale agreement makes clear the answer is not close to finished.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
