Sources: Anthropic has agreed to pay Nscale $45B over six years to rent about 460MW of power at a West Virginia data center using Nvidia's Vera Rubin chips
Anthropic PBC has agreed to spend $45 billion to rent AI cloud computing power from Nscale's flagship data center development in West Virginia …
Context & Ripple Effects
Anthropic had already committed to $30 billion of Azure capacity and, in August 2026, reportedly added a six-year $10 billion Norway capacity agreement. The reported Nscale arrangement extends that procurement pattern from hyperscaler capacity toward dedicated AI-cloud contracts.
For Nscale, the customer commitment is tied to the West Virginia campus it acquired in March, where it outlined plans for up to 8GW of compute. That makes a single 460MW reservation material evidence of demand for a development whose buildout is far larger than the initial allocation.
First-order effects
- On the reported terms, Anthropic secures about 460MW of Nscale AI-cloud capacity for six years, adding a West Virginia supply commitment alongside its Azure and Norway arrangements.
- Nscale gains a reported long-duration anchor customer for its West Virginia campus, while Nvidia’s Vera Rubin chips become the specified hardware for that allocation.
Second-order effects
- A long-term Anthropic commitment gives Nscale a clearer basis to sequence construction and financing around contracted demand rather than speculative capacity.
- AI-cloud providers pursuing Anthropic workloads face a buyer that is spreading large commitments across Microsoft, Volta Infra and Nscale, increasing the value of differentiated capacity, location and contract structure.
Third-order effects
- If comparable contracts continue, AI compute is moving toward a capacity market built on multi-year power-and-chip reservations, with model developers acting as anchor tenants for data-center projects.
- The resulting infrastructure model shifts execution risk toward operators such as Nscale: customer commitments may support buildouts, but delivery still depends on power, site development and the specified hardware arriving on schedule.
The trend: AI developers are turning compute procurement into long-duration infrastructure contracting, linking model demand directly to data-center power buildouts and accelerator supply.