Microsoft Just Abandoned Its Big UK AI Datacentre Bet

Microsoft has walked away from its agreement to anchor a planned UK AI Datacentre in Loughton, Essex, dealing a major blow to the project British ministers had billed as the country's largest sovereign AI datacentre. The decision, reported by The Guardian on October 18, came down to electricity: the local grid could not promise enough power to run the facility until the early-to-mid 2030s.

The timeline tells the story. The site was announced in September 2025 as a flagship of Britain's AI strategy, a fifty-megawatt AI campus scalable to ninety megawatts that would initially house twenty-three thousand forty of Nvidia's newest graphics processors, delivered in early 2027 to power Microsoft's Azure cloud services across the UK. By March this year, the Guardian found the location still operating as a scaffolding yard. In September, grid operator UK Power Networks told developer Nscale that sufficient electricity would not arrive until the 2030s. Last weekend, Microsoft informed Nscale it was ending the agreement.

Power, not chips, is the binding constraint

That last step matters because it flips the AI industry's usual story. For years the bottleneck was supposed to be semiconductors — who could buy enough advanced chips to train frontier models. Microsoft's exit shows the real choke point is moving down the stack: transformers, substations, and grid connections. You can win planning consent and a ministerial launch, then wait a decade for electrons.

Microsoft was the anchor tenant, and according to the newspaper's reporting it was Nscale's only publicly announced customer for the Essex scheme. The pullout therefore does double damage: it removes the contracted demand that justified the build, and it embarrasses the political branding around the project. Nscale, which is backed by Nvidia, told reporters it remains committed to the site and to the United Kingdom, that its relationship with Microsoft "remains strong and long-term," and that the London-wide grid problem does not affect its other contracts. The company says it is exploring on-site generation and ways to accelerate its grid connection.

What happens to Britain's sovereign AI plans

The stakes go beyond one site. The project was planned as a two-billion-pound build creating roughly two hundred fifty permanent jobs, according to the BBC. Nscale filed for a United States stock market listing last month, with reports suggesting a potential valuation of thirty-five billion dollars — an ambition that looks shakier without an anchor customer on its flagship British site. The irony is that Microsoft has not abandoned Nscale entirely: it remains involved in the company's separate project in Norway, where power apparently is not the problem.

The same weekend brought a mirror image from across the Atlantic. Meta announced the completed first phase of a gigawatt-scale training campus in Ohio, now operating hundreds of thousands of chips — proving that the engineering of massive AI clusters works when the power shows up. The contrast is the story: America can feed its superclusters; parts of Britain's grid apparently cannot, at least not on any schedule an anchor tenant would sign.

For young people in Britain, the fallout is concrete rather than abstract. Two hundred fifty permanent jobs were supposed to anchor a local tech ecosystem in Essex, and the campus was meant to host the AI tools that British startups build on. When a flagship UK AI Datacentre project cannot get a grid connection, the servers powering the next wave of apps get built somewhere else — and the tax revenue, jobs, and technical know-how follow them. As one account of the episode put it, the pullout is casting doubt on the country's sovereign AI ambitions just as rival governments are racing to secure their own capacity.

For anyone watching where AI infrastructure money goes, the lesson is practical. Data centres are the factories of the chatbot era, and factory locations are now decided by grid capacity as much as by tax breaks or land prices. Britain still wants to be a sovereign AI hub, but sovereign ambitions need sovereign megawatts. Until the grid catches up, the country's largest AI datacentre exists mostly on paper — and Microsoft, having waited through a scaffolding yard and a power delay measured in decades, decided it would rather build where the lights can stay on. For more on the buildout, see Meta's million-chip training campus and how Broadcom is lining up over fifty billion dollars for custom AI chips while Tencent leases advanced chips through Oracle.