Anthropic has agreed the Anthropic Nscale compute deal that could reshape how AI labs secure the processing power they need: a $45 billion, six-year contract giving Anthropic access to 460 megawatts of capacity at Nscale’s West Virginia data centre, powered by Nvidia’s Vera Rubin processors. Which leaves an obvious question: how does a startup founded in 2024 credibly back a deal of that size?
The Anthropic Nscale compute deal in context
The arrangement gives Anthropic a large slice of capacity at the West Virginia site, where Nscale is building out what its own materials describe as a Phase 1 target of 1.35 gigawatts of AI computing capacity, according to Forbes, with delivery targeted for early 2028. The 460 megawatts Anthropic is renting sits inside that broader build-out, underlining both the ambition of the site and how much of it is already spoken for.
For Nscale, the commercial logic is plain. According to daily.dev, the $45 billion figure is the single largest entry in Nscale’s $51 billion contracted revenue backlog, even as the company reported only about $100 million in revenue last quarter. The gap between backlog and current revenue is, to put it mildly, considerable. Whether Nscale can actually convert that pipeline into delivered infrastructure at this scale is the operative question.
Nscale describes itself as a “neocloud”: a provider of computing power oriented towards AI inference and training, competing for contracts that the large established cloud providers have historically dominated. The company is backed by Nokia, Nvidia, Dell and Blue Owl, and has raised over $5 billion in funding. It is building data centres across five countries, including the UK, and is targeting an initial public offering (IPO) this year.
A mixed track record so far
The West Virginia deal is not Nscale’s first major arrangement. The company has previously supplied compute capacity to Microsoft. But it has had rougher moments too. A partnership with OpenAI was paused owing to high energy costs, a reminder that the economics of large-scale compute supply are harder to manage than the headline contract figures suggest. And last month, reports emerged that Nscale’s UK data centre would be delayed because of grid connection problems, a constraint that has tripped up more than a few data centre developers in Britain.
Those grid difficulties are part of a broader pattern. AI infrastructure demand has grown faster than electricity networks were designed to accommodate, and the West Virginia site’s appeal partly reflects the relative availability of power in parts of the United States compared with more constrained markets in Europe.
Nscale has positioned itself as part of the answer to Europe’s dependence on overseas computing capacity. ‘We definitely want to support the UK AI ecosystem,’ Imran Shafi, UK managing director of Nscale, said earlier this month when the company announced it would provide compute to several British universities. ‘We’re a UK headquartered company that matters a lot to us.’ The UK data centre delay complicates that narrative somewhat, though it does not erase the underlying intent.
Nscale is not alone in trying to carve out a European alternative in AI infrastructure. Paris-based model maker Mistral, for instance, is pursuing its own local data centre build-out, backed by enterprises including ASML. The direction of travel across the continent is consistent: more AI capacity, closer to home, less reliant on a handful of American hyperscalers.
For Anthropic, the deal reflects a continuing scramble for compute at scale. AI model development remains constrained by access to sufficient processing power, and locking in 460 megawatts through a six-year agreement with a well-backed neocloud offers a degree of supply certainty. Whether Nscale’s Phase 1 delivery target of early 2028 holds is what Anthropic will be watching most closely.
