Building on our coverage from 16 days ago, new reporting shows Anthropic has agreed to roughly $45 billion in long-term cloud compute capacity with Nscale, renting about 460 megawatts of power at a planned West Virginia data center.
Nscale, a UK-based AI infrastructure company, will supply the campus capacity expected to come online toward the end of 2027 and build clusters around NVIDIA's Vera Rubin accelerators to run Anthropic's model training and to serve responses.
This is not a typical cloud contract. It is a multi-year bet on guaranteed, large-scale GPU access instead of paying for burstable public cloud time. That predictability changes the economics of developing frontier models.
Think of it as leasing an entire power plant and the computer racks inside so you never have to wait in line for GPUs. The campus secures electricity, space and the latest accelerators so Anthropic can schedule massive jobs on its own terms.
Practically, Anthropic locks in capacity and can plan multi-year experiments without the same risk of vendor shortages or spiking prices. Competitors now face a world where guaranteed campus capacity is a strategic advantage, not just a nice-to-have. This does not mean anyone can run these systems locally; they require data center-scale power, cooling and operations.
The real test starts now: can Anthropic turn locked-in capacity into faster model iteration, lower costs, and products that compete with closed incumbents? Watch for pricing signals, model releases tied to this hardware, and any operational or regulatory scrutiny as the campus gets built.