Anthropic agreed to spend $45 billion leasing AI cloud capacity from Nscale, a British infrastructure company barely two years old. The six-year deal ties roughly 460 megawatts of future capacity in West Virginia to Nvidia’s next-generation Vera Rubin systems, hardware that isn’t running yet and won’t start serving Anthropic’s workloads until late 2027.
Spread across six years, that works out to about $7.5 billion annually for cloud capacity Anthropic will never own outright. Nscale keeps the physical campus, the hardware, and the infrastructure risk, operating essentially as a specialist AI cloud provider. Anthropic buys the right to use that capacity. The arrangement pushes construction costs, procurement headaches, and equipment risk onto Nscale, while leaving Anthropic with a contractual claim on cloud compute it needs but can’t build fast enough on its own.
The obvious question is what 460 megawatts of cloud capacity will actually be worth by the time it arrives. Hardware ages quickly in this business. Rack designs shift, power density keeps rising, and model architectures change how efficiently teams extract value from that compute. Anthropic is essentially betting that locking in capacity now matters more than the risk that the economics look different in three years.
For Nscale, this deal changes everything. Founded in 2024, the company now anchors one of the largest AI cloud infrastructure commitments on record, and it’s building its Monarch campus in West Virginia at a scale well beyond a typical data center. A cloud customer the size of Anthropic, committed for six years, gives lenders and infrastructure investors exactly the kind of revenue certainty they look for before financing a project this large. Nscale still has to deliver, though, securing power, installing the hardware, and hitting deployment timelines that leave little room for delay once hundreds of megawatts are under contract.
This is not Anthropic’s only cloud bet, either. In July, the company struck a deal with AMD for up to 2 gigawatts of Instinct MI450 systems, and it separately signed a $10 billion, six-year capacity agreement with Volta tied to a Norwegian data center. Rather than leaning on one provider, Anthropic is drawing compute from multiple specialist clouds, hyperscalers, and chip architectures at once.
That approach reduces the risk of any single point of failure, but it adds a different kind of complexity: workloads now have to move across different cloud environments and hardware ecosystems, and engineering teams have to manage that on top of everything else already on their plate.
