Nscale signs $45B deal to deploy Nvidia’s Vera Rubin chips for Anthropic’s AI needs

A London-based AI infrastructure company most people haven’t heard of just locked in one of the largest cloud computing deals ever signed. Nscale has agreed to a $45 billion, six-year contract with Anthropic to provide computing services powered by Nvidia’s next-generation Vera Rubin chips at its Monarch Compute Campus in West Virginia.

To put that dollar figure in perspective, it’s roughly the GDP of Latvia. And the deal’s power footprint is equally staggering: approximately 460 megawatts, enough electricity to run about 345,000 American homes.

What Nscale is building

The deployment of Nvidia’s Vera Rubin chips under this agreement is slated to begin in late 2027. The Vera Rubin architecture represents Nvidia’s successor to the current Blackwell/Grace platform, designed as a cohesive supercomputer system rather than a collection of individual GPUs bolted together.

The platform features seven chips engineered to function as a unified AI supercomputer, specifically targeting the needs of frontier AI labs like Anthropic. Nvidia has pitched the architecture as delivering higher throughput per watt and lower costs per token compared to its predecessors.

Nscale isn’t putting all its eggs in the Anthropic basket, either. The company has a prior commitment with Microsoft for up to 1.35 gigawatts of Vera Rubin NVL72 systems at the same West Virginia site, announced in March 2026. Across its European operations, Nscale plans to integrate between 100,000 and 130,000 Vera Rubin GPUs.

From crypto mining to AI kingmaker

Nscale’s origin story adds an interesting wrinkle. The company evolved from a crypto-mining infrastructure spin-out into a specialized modular AI data center builder. The company has since attracted significant funding and notable board members, positioning itself as a credible partner for two of the most consequential technology companies on the planet.

Why Anthropic needs this much compute

Anthropic, the maker of the Claude family of AI models, is reportedly preparing to go public. Securing $45 billion worth of computing capacity ahead of an IPO isn’t just about having enough GPUs to train the next model. It’s a strategic signal to investors that the company has the infrastructure pipeline locked down to compete at the frontier for years to come.

The Vera Rubin platform’s focus on agentic AI workloads is particularly relevant here. Agentic AI, where models autonomously plan and execute multi-step tasks rather than just responding to prompts, is widely expected to be the next major capability frontier. These workloads are considerably more compute-intensive than standard inference, which partly explains why Anthropic is willing to commit to such an enormous power envelope.

The infrastructure arms race intensifies

Nscale’s combined commitments at the Monarch Compute Campus, between the Anthropic and Microsoft deals, suggest the site could eventually consume well over a gigawatt of power. That’s approaching the output of a nuclear power plant, dedicated primarily to training and running AI models.