Anthropic’s latest infrastructure wager is not about a new chatbot feature or model release. It is a $45 billion reminder that, in the AI race, raw computing power is becoming the prize competitors cannot afford to miss.

The reported agreement would see Anthropic rent about $45 billion in AI compute from Nscale, the British infrastructure start-up founded in 2024. Nscale is expected to supply capacity through Nvidia’s Vera Rubin system, with service for Anthropic slated to begin in late 2027.¹

The scale matters as much as the timing. The arrangement reportedly runs for six years and draws power from Nscale’s flagship West Virginia data center, putting a young UK company at the center of an American-led scramble for the chips, power and facilities needed to train and run frontier AI systems.²

Anthropic’s move is the latest entry in a rapid-fire expansion that has unfolded over the past eight months. In April, it widened arrangements with Amazon, adding access to 5 gigawatts of compute, while also expanding relationships with Google and Broadcom. In May came a SpaceX arrangement reportedly worth $1.25 billion in capacity each month; July brought a $5 billion compute-related deal with AMD. Earlier this month, Anthropic signed a six-year, $10 billion cloud agreement with Volta for capacity in Norway.

Nscale’s offer points to the next phase of that campaign: newer infrastructure specialists are becoming strategic suppliers, not merely contractors. Anthropic’s view, implied by its dealmaking, is that abundant compute is essential to challenging OpenAI. Its rivals appear to agree. Google, OpenAI and Meta are pursuing the same finite pool of chips and data-center capacity, turning procurement into a frontline contest rather than a back-office task.

For now, the industry’s biggest AI ambitions are being measured less in model demos than in multibillion-dollar reservations for the machines behind them.