Wednesday, October 14Global edition

Anthropic signs $45 billion compute deal with Nscale ahead of IPO

The six-year contract secures 460 megawatts of AI processing capacity at a West Virginia campus running Nvidia's next-generation chips. The facility isn't expected online until late 2027, but the commitment sends a signal to IPO investors that demand for Anthropic's products is real and growing fast.
Share

The deal is best read as an infrastructure lock-in timed to serve two converging pressures: surging demand for Anthropic's AI products and the company's preparations for a major initial public offering (IPO). By committing approximately $45 billion over six years to rent cloud computing capacity from Nscale, a London-headquartered AI infrastructure company, Anthropic is securing the processing power its business requires before capacity constraints become a competitive liability.

Bloomberg first reported the agreement on 26 August. Anthropic declined to comment.

The contract covers approximately 460 megawatts of power capacity, equivalent to roughly what 345,000 US homes draw at any one time. That capacity sits within the first of three planned buildings at Nscale's Monarch campus in West Virginia. The full campus is projected to reach around 1.35 gigawatts when complete, with total development costs, including an on-site power plant, estimated at about $71 billion, of which $47 billion would go toward AI chips. The facility is expected to come online in late 2027.

The hardware at the centre of the agreement is Nvidia's Vera Rubin chip system, a next-generation architecture that combines six different chips and represents the current frontier of AI chip design. Nscale itself is backed by Nvidia, which gives the supply chain a degree of vertical coherence that matters at this scale.

Nscale was founded in 2024 and, despite being only two years old, has moved quickly to position itself as an anchor provider for frontier AI companies. The Anthropic deal follows the collapse of an earlier relationship with Microsoft, which had a preliminary agreement with Nscale since March but subsequently withdrew, opening the Monarch campus's anchor tenant slot to Anthropic. The company is preparing for a US listing as soon as next month, and landing Anthropic as its primary customer materially strengthens that proposition.

For Anthropic, the strategic logic runs in the other direction. The company is expanding its compute infrastructure to meet growing demand for products including Claude Code, its AI coding tool. It is also moving toward annual revenues exceeding $65 billion, more than seven times its billing rate from the end of last year. A business scaling at that pace requires committed, large-volume compute that cannot be assembled incrementally.

The $45 billion figure also carries a signalling function. Anthropic is itself preparing for a major IPO, and a long-term infrastructure commitment of this size tells prospective investors that the company has both the demand to justify the expenditure and the confidence to lock in capacity years in advance. Compute, in the current AI market, is as much a strategic asset as it is an operational one. Firms that secure it early gain a structural advantage over those that must compete for it later at higher cost.

The exact payment structure of the six-year agreement has not been disclosed. The terms under which Microsoft withdrew from its earlier arrangement with Nscale are also unknown, as are the full details of both companies' IPO timelines.

Sources (5)

More from The Modern on this topic:

Devin RaoTech

Devin Rao writes on the business and strategy of technology — less about launch-day specs than who captures the value and who gets disrupted. Analytical, allergic to hype, happiest tracing a company’s incentives to their logical end.

Comments (0)