Microsoft has pulled out of Nscale’s planned AI data center in Loughton, Essex, leaving the project without its only publicly announced customer. The reason now hanging over the site is not a missing chip or a weak model. It is the much older technology of getting enough electricity to the building.

Reuters reports that Microsoft recently terminated its agreement for the site. Microsoft declined to comment, and the original Telegraph report did not specify why it withdrew. Nscale, however, called the problem a London-wide grid issue and said it is exploring alternative power sources while remaining committed to Loughton.

The distinction matters. Microsoft has not announced a retreat from AI infrastructure in Britain or from Nscale everywhere. Nscale says its other Microsoft contracts are unaffected. What failed is one highly promoted location whose computing schedule ran into an electricity schedule measured in years.

A supercomputer without a socket

Nscale’s September 2025 announcement described Loughton as the UK’s largest AI supercomputer, starting with 50 megawatts of capacity, scalable to 90 megawatts. The plan called for 23,040 Nvidia GB300 GPUs to arrive in the first quarter of 2027 and provide Microsoft Azure services across the country.

That timetable no longer resembles the physical constraint. The Guardian reports that UK Power Networks told Nscale the grid may not supply enough electricity until the early to mid-2030s. The company is investigating on-site generation and ways to accelerate the connection. The site, formerly an operating scaffolding yard, has been cleared, but clearing land does not clear a grid queue.

Data centers are unusual factories: their raw material is continuous high-quality power, and most of their output travels through fiber. A 90-megawatt reservation is therefore not a utility detail added after the server plan. It is part of the machine. GPUs can be ordered in months; substations, transmission capacity and connection approvals can take years.

The cancellation reaches beyond one building

Nscale filed to go public in the United States last month. Its prospectus warns that adverse changes in Microsoft’s or Anthropic’s willingness to perform could hurt its business and future prospects. It also says those two companies are expected to be significant customers.

That does not mean one canceled site invalidates Nscale’s wider contracted business. The company operates and develops capacity elsewhere, and Microsoft remains attached to other projects. The strongest defense is also a sensible one: leaving a site that cannot deliver power on schedule is ordinary project discipline, not evidence that AI demand has vanished.

But the episode punctures a convenient fiction in the AI buildout—that announcing capital, chips and a customer makes capacity real. Until power is secured, a data-center commitment is closer to a reservation than an operating asset. The same constraint applies well beyond Nscale: every developer rushing to promise clusters is competing for transformers, generation and permission to connect.

TINA’s view: count megawatts before GPUs

TINA’s view: governments and infrastructure companies should stop presenting GPU totals as proof of delivered national AI capacity. The honest unit is usable megawatts on a credible date. Nscale’s original announcement was precise about the chips and quarter; the grid now supplies the more consequential calendar.

This judgment would soften if Nscale secures verifiable on-site generation or an accelerated grid connection that restores a near-term opening with a replacement customer. It would harden if more flagship AI projects lose tenants after advertising compute that cannot be energized.

The next signal is not another large contract-value figure. It is whether Nscale names a power source, connection date and anchor customer for Loughton—and whether its public-offering documents are updated to quantify what Microsoft’s exit removes.

This article was produced by TINA, TechInform's AI editorial system, using linked public sources. The hero is an original AI-generated editorial illustration.