Microsoft Ends Data Center Deal with Nscale in Essex

What Was Reported

Microsoft has ended the agreement to occupy the data center that Nscale plans in Loughton, in the English county of Essex, according to a report by the Telegraph published on Sunday (18) and picked up by Reuters. According to the newspaper, the operators of the British power grid have informed Nscale that they do not expect to provide power to the site before the 2030s. Microsoft had not publicly commented until the closure of this article.

Nscale responded with a statement that neither confirms nor denies the departure. The company stated it remains "committed" to Loughton and the UK, which is exploring alternative energy sources for the site and that its relationship with Microsoft "remains strong and long-term." It characterized the episode as a power grid issue "affecting all of London" and asserted that it "does not impact our other contracts."

The Symbolic Weight of Loughton

The site was not a peripheral project. Microsoft and Nscale presented Loughton as the site for what would be the largest supercomputer in the UK, with about 23,000 Nvidia Grace Blackwell Ultra generation GPUs, according to announcements from both companies. The site was central to the UK’s discourse on sovereignty in AI, and Nscale, founded in London, was a local showcase for this discourse.

The problem described is not about demand or funding. It is about connection. A client of Microsoft’s caliber purchases capacity with a delivery date, and a substation promised for after 2030 removes the project from any planning window for a hyperscaler. Microsoft, according to the Telegraph, did not explain its decision, but the timing of the power grid suffices to make it understandable.

The Countdown to the IPO

The news arrives at the worst time for Nscale, which is seeking to raise up to $3 billion in an initial public offering in the United States, according to reports from Bloomberg covered by the industry press, with Goldman Sachs and JPMorgan at the forefront of the process. It is a company selling capacity still under construction, and the stock price depends on the credibility of the delivery schedule. The prospectus submitted to the SEC shows a contracted portfolio of about $103 billion, of which Microsoft and Anthropic account for approximately 85%, according to outlets that reviewed the document.

The largest individual slice comes from Anthropic: agreements of up to about $44.6 billion for dedicated GPUs at the Monarch Compute Campus in West Virginia, according to the cited S-1. Loughton is expected to be a small portion of that total, and Nscale itself insists on this. The effect on the roadshow is different. The question this case raises for investors is how much of the remaining portfolio depends on electrical connections that do not yet exist, and whether other contracts in the portfolio might be terminated for the same reason.

Insights Across Three Markets

In the UK, this episode exposes a bottleneck that no investment announcement can resolve in the short term. If Nscale is correct in saying that the issue affects the entire London area, any AI project relying on the southeast England grid faces the same horizon, and capacity is likely to migrate to where stable energy is available. For a country that promoted the partnership with Microsoft and Nvidia as proof of its attractiveness for AI investment, losing the tenant of its flagship project due to a lack of power supply is a political embarrassment, not just a commercial one.

Norway presents the most obvious counterpoint within Nscale itself. The company has capacity agreements with Microsoft in Narvik, in the north of the country, where hydroelectric supply is abundant, and the Loughton case reinforces the Nordic argument for loads that do not require low latency compared to London. In the United States, Nscale's bet in West Virginia with Anthropic follows the same logic: to prioritize land and energy over proximity to the end customer.

For CIOs of banks, insurance companies, and British public agencies planning to run regulated AI loads on physically located capacity in the UK, the consequences are concrete. Data residency requirements that necessitate servers on British soil will now compete against a supply that grows more slowly than announced, while new capacity settles in Scandinavia and the American interior. The question these executives will carry to the next renewals with hyperscalers shifts from price per GPU to the energization date of the contracted region.