Story
Oct 20, 2026
Anthropic’s $45 Billion Nscale Expansion Faces a Power-Grid Reality Check
Anthropic’s race for more AI computing capacity could deepen its reliance on Nscale, while Nscale argues that a power problem near London will not derail its wider ambitions. The prospective deal highlights both the scale of AI demand and the fragility of the infrastructure being built to meet it.
Anthropic’s push to feed demand for its Claude models began with a first Nscale agreement reported in August, part of a broader run of infrastructure partnerships with AMD, SpaceX, Google and Broadcom. The company has said surging use of Claude has put “inevitable strain” on its infrastructure.1
Now, Anthropic is reportedly in talks to lease capacity at a second 460-megawatt data centre at the same West Virginia site. The prospective agreement could be worth about $45 billion, according to a person familiar with the negotiations; Anthropic declined to comment, while Nscale was approached for comment.1
The timing makes the proposed expansion more than a simple vote of confidence in a fast-growing supplier. Nscale is among the so-called neoclouds building or leasing data centres and selling computing power to AI labs and hyperscalers — a capital-intensive model that investors have begun to scrutinise. In the past month, rivals CoreWeave and Nebius saw their shares fall 28% and 27%, respectively, amid concern over heavy borrowing.1
Over the weekend, the pressure became more tangible. The Telegraph reported that Microsoft had withdrawn from an agreement to anchor an Nscale data-centre project outside London after delays connecting the site to the UK power grid. Nscale’s response was defiant rather than dismissive: it said it remained “committed” to the project and was “exploring alternative power sources for the site.”1
The company also maintained that its relationship with Microsoft was “strong and long-term,” and said the London-wide grid problem had “no impact on our other contracts.”1 If the West Virginia talks produce a deal, that claim will face a far larger test: delivering hundreds of megawatts of reliable power for one of AI’s hungriest customers.