Anthropic will lease ~460 MW of compute at an Nscale data center in West Virginia running NVIDIA Vera Rubin systems — the latest in a spree that tops $100B in committed AI infrastructure.
Anthropic is buying compute the way nations stockpile oil.
Anthropic is buying compute the way nations stockpile oil. The AI lab has agreed to a roughly $45 billion, six-year cloud deal with Nscale — one of the largest AI-infrastructure agreements ever signed — and it is only the latest entry in a spending spree that has committed the company to well over $100 billion in future computing power. For a firm reportedly preparing to go public, the message to investors is blunt: capacity is the moat, and Anthropic intends to own it.
What the $45 billion buys
Under the six-year agreement, Anthropic will lease roughly 460 megawatts of compute capacity at a large Nscale data-center development in West Virginia — enough electricity to power about 345,000 US homes at any given moment. The facility is expected to come online by late 2027 and will run on NVIDIA’s Vera Rubin systems, which combine six different chips working in concert at the cutting edge of accelerator design. Nscale, a UK-based AI-infrastructure provider, becomes a central supplier to one of the world’s leading model developers.
A pattern, not a one-off
The Nscale deal lands on top of a remarkable run of commitments. In recent months Anthropic has reportedly signed compute agreements including around $50 billion with Fluidstack, $10 billion with Volta Infra Holdings and $45 billion with SpaceX. Stacked together, these deals lock in years of capacity — and years of obligations — ahead of a planned IPO. It is a bet that demand for frontier models will keep compounding, and that the scarcest resource in AI is not talent or data but raw, powered, ready-to-run silicon.
The risk hiding in the megawatts
There is a reason these numbers are measured in megawatts, not just dollars. Committing to 460 MW in West Virginia assumes the power, the grid connections and the physical build all arrive on schedule — and across the industry, roughly half of planned US data centers are now at risk of delay or cancellation as permitting, grid capacity and equipment shortages bite. Anthropic is pre-paying for capacity precisely because it may be scarce. Whether that looks visionary or over-extended depends on a variable no AI lab controls: whether the infrastructure gets built on time.
Written for Red Robot with AI assistance and human editing. Based on reporting by CNBC.
