Microsoft has walked away from the data center campus that ministers once called Britain's largest AI project, after revised grid forecasts pushed its connection date from 2027 to the early 2030s.
Power Queues Stall Britain's Flagship AI Build
The withdrawal turns a spotlight on a bottleneck that has quietly reshaped the UK's infrastructure pipeline. Grid operators simply cannot energise new hyperscale loads fast enough to match the pace at which AI developers want to build.
Loughton was meant to be different. Announced in 2025 and backed by Nscale, the Essex campus carried the weight of the government's "sovereign AI" rhetoric, promising domestic compute capacity rather than reliance on overseas facilities.
By March 2026, that promise looked fragile. Reporting revealed the site was still functioning as a scaffolding yard, with construction nowhere near the stage its public messaging implied.
What You Need to Know About the Withdrawal
Microsoft's exit removes the anchor tenant from a project that had few other named backers. Without a committed buyer for the compute, financing the build becomes considerably harder.
Nscale, however, has pushed back against the suggestion that the project is dead. The company framed the problem as regional rather than structural, pointing to constraints across the London grid area.
- Grid timeline: Connection slipped from 2027 to the early 2030s, a delay of roughly five years.
- Demand pressure: More than 300 data centers are waiting to connect across the UK.
- Capacity gap: Data center electricity demand exceeds current capacity by 60%.
- Customer base: Microsoft was the only publicly announced client for the site.
"We remain committed to both the Loughton project and the UK and we are exploring alternative power sources for the site," Nscale said in a statement, adding that its relationship with Microsoft "remains strong and long-term."
Why This Matters
The immediate casualty is Nscale's credibility as Britain's self-styled sovereign compute provider. A flagship site without a flagship customer invites sharper questions from investors ahead of its planned Wall Street listing at an estimated $35 billion valuation.
More broadly, the episode exposes a mismatch at the heart of the AI buildout. Neoclouds such as Nscale, many of them heavily backed by Nvidia, are borrowing aggressively to construct infrastructure on the assumption that demand will arrive on schedule.
That assumption depends on electricity that may not exist. Analysts have suggested the AI sector needs to generate roughly $2.5 trillion in annual revenue to justify the infrastructure spending already committed, while current estimates place yearly AI revenue between $150 billion and $250 billion.
Previous technology cycles offer a cautionary pattern. The companies that built railways, fibre networks and early cloud capacity often failed to recoup their outlays, while the businesses layered on top captured the gains.
The Wider Signal for UK Tech Policy
For Westminster, the message is uncomfortable. Ministers have repeatedly sold AI data centers as evidence of post-Brexit industrial strategy, yet the binding constraint is now mundane: wires, substations and planning timelines.
Unless grid reinforcement accelerates, similar withdrawals become likely. Compute follows power, and at present the power is not there.



