Microsoft Walked Away From Britain's Biggest AI Datacenter: the Grid Had No Power to Spare

Microsoft just walked away from the anchor-tenant deal for what was supposed to be Britain's flagship AI datacenter, and the reason was not cost or demand. It was electricity. The Microsoft Nscale datacenter project in Loughton, Essex, a 50-megawatt facility planned to house 23,040 Nvidia GB300 GPUs for Azure, fell apart after the local grid operator said it could not deliver enough power until the 2030s. Microsoft terminated the single-site agreement, stripping the anchor customer from a project the UK government had hailed as a cornerstone of its 2025 AI strategy.

Why the Microsoft Nscale datacenter deal collapsed

The Loughton project was announced in September 2025 as part of Britain's AI infrastructure push, with Microsoft signed on as the only publicly announced tenant. Nscale bought the Essex site and planned to install roughly 23,000 Nvidia chips, with the first equipment arriving in the first quarter of 2027 and capacity expandable from 50 to 90 megawatts. The plan looked solid on paper and shaky on the ground: earlier this year, reporting showed the site was still a working scaffolding yard. Nscale has since cleared it, but the schedule was already in doubt before the grid news landed.

The grid news is what killed the deal. UK Power Networks told Nscale it could not guarantee sufficient power to the site until the early to mid-2030s, according to reports first published by The Telegraph. Microsoft walked, with a spokesperson stressing the decision concerned a single site and that the company remains committed to its broader partnership with Nscale, including a separate project in Norway. Nscale says the issue is a London-wide grid problem that does not affect its other contracts, and that it is exploring alternative power sources for Loughton alongside other UK expansion sites.

The timing is awkward for Nscale, which is preparing a US stock-market listing that reports suggest could value the company at up to $35 billion. It has also raised $3.36 billion in convertible loan notes led by Third Point, and counts Nvidia among its backers. Losing your only named anchor tenant is never a good look in the weeks before an IPO.

Power is now the binding constraint on cloud capacity

This is not a one-off embarrassment; it is the market catching up with physics. Britain's datacenter grid connection requests now total about 73 gigawatts, per local reporting, far beyond what the network can deliver on any reasonable timeline. Across the industry, the bottleneck has moved from GPUs to power, chillers and permits. NextDC lost 44% of its value in a single day when Melbourne's grid could not keep up with its expansion. The Nscale collapse is the same story with a bigger name attached: the clouds can raise unlimited capital, but they cannot conjure a substation.

For cloud operators, the lesson is that siting is now a first-order engineering decision. The cheapest land in the right metro area is worthless if the connection queue runs into the 2030s, and "exploring alternative power sources" is what a developer says when the grid says no.

What this means for your cloud footprint

If you run production workloads in the UK or anywhere with a constrained grid, treat power availability as a capacity risk, not a vendor problem. Three moves worth making now:

1. Ask your providers for committed, not planned, capacity numbers with dates tied to grid connection, not construction milestones.
2. Build multi-region redundancy across grids, not just across availability zones in the same metro.
3. Pad capacity timelines by quarters, not weeks, and keep a fallback region warm if you cannot afford a Loughton-style surprise.

Microsoft absorbed this one with a press statement. Most companies would feel it as an outage that never gets fixed.