Anthropic’s Reported $45B Nscale Deal Shows AI Compute Is Now A Land Grab

Anthropic is reportedly preparing to spend $45 billion on cloud computing capacity from Nscale, and the deal captures the new reality of AI; the model race is now a power, chip and financing race.

The Financial Times reports that Anthropic has agreed to a six-year arrangement with UK cloud startup Nscale for capacity from a planned data-centre campus in Mason County, West Virginia. The reported deal would give Anthropic access to about 460MW of computing power from late next year, using Nvidia’s next-generation Vera Rubin processors.

The numbers are large even by today’s AI standards. Nscale’s wider West Virginia campus is expected to be built around 1.35GW of data-centre capacity and supported by a 2GW natural-gas power plant. Reports say the overall project could cost about $69 billion, with Nscale backed by investors and partners including Aker, Nvidia, Dell and Nokia.

For Anthropic, the logic is straightforward. Claude is no longer simply a chatbot brand. It is now an enterprise platform used for coding, research, workplace tasks, security reviews and agentic workflows. Those workloads need reliable compute, and the companies that wait until demand arrives may find that the chips, power and data-centre capacity have already been reserved by someone else.

This is why the AI race now looks more like an infrastructure arms race. The biggest labs are not only hiring researchers and releasing models. They are locking up power contracts, buying or leasing future GPU capacity, signing cloud commitments and spreading bets across multiple infrastructure partners.

Anthropic already has deep cloud relationships with Amazon and Google, so a separate Nscale deal would make strategic sense. It gives the company another source of capacity and reduces dependence on one infrastructure partner. It also gives Nscale a flagship AI customer before its own expected public-market moment.

The West Virginia location matters too. AI infrastructure is moving into regions where land, energy and political support can be assembled at huge scale. That creates jobs and local investment, but it also raises questions about power demand, water use, grid pressure, emissions and whether communities are being asked to host infrastructure built mainly for a handful of AI giants.

Nvidia sits in the middle of all this. Its Vera Rubin platform is becoming the next prize for AI labs, cloud providers and sovereign AI projects. After Nvidia’s $96.2 billion Q2 FY27 quarter, the message is clear: demand for AI systems is still running ahead of what the supply chain can comfortably deliver.

The risk for Anthropic is that compute commitments this large only make sense if AI revenue keeps scaling quickly. If enterprise adoption slows, model costs fall faster than expected, or rivals find cheaper ways to serve the same workloads, these long-term contracts could become expensive obligations.

For now, the reported deal shows where the industry is heading. The most powerful AI companies will not only be the ones with the best models. They will be the ones that can secure electricity, chips, data centres and capital before everyone else does.