Microsoft Abandons Flagship UK AI Data Center: The Grid Says Power May Not Arrive Until the 2030s
The AI race's bottleneck is no longer chips. It's the socket.
Three key facts
Microsoft walked away from a planned UK AI data center. According to CoinDesk citing the Guardian and Financial Times, the site in Loughton, Essex — about 6 miles northeast of London — was to be run by Nscale, a company spun out of bitcoin mining. The local grid operator, UK Power Networks, told Nscale there may not be enough capacity until the early-to-mid 2030s.
The original plan was no small thing. It was meant to house 23,040 Nvidia chips and supply 50 megawatts of computing to Microsoft's Azure cloud by early 2027. The project was announced last year by Nscale alongside Microsoft, Nvidia, and OpenAI; when the Guardian visited in February, the site was still a working scaffolding yard. Nscale says its other Microsoft contracts remain in place.
Behind it is a whole supply chain's predicament. The report notes grid-connection requests for UK data centers total about 73 gigawatts. Meanwhile, bitcoin miners pivoting to AI infrastructure face both money and power pressure: an AI-ready facility costs about $8 million to $15 million per megawatt, versus roughly $700,000 to $1 million for bitcoin mining, and AI customers expect servers to run around the clock.
Wang Dou's Take
Look at the time gap first: computing was to be delivered by early 2027; the grid says the 2030s. That's a five-to-six-year difference. This isn't "a delay" — it's "this project physically cannot get a place in line." Chips can be bought out of turn with extra money; power can't. The grid isn't an e-commerce site — place an order and it won't arrive next day.
What's most worth pondering is that line "UK data center grid-connection requests total about 73 gigawatts." Put it in perspective: the UK's national peak electricity demand is on the order of 40-something gigawatts, and the queued requests from data centers are already approaching twice that. Most of those requests will never be built, but every one of them holds a place in the queue. In other words, people who actually want to build are competing for power with a pile of people who may never build at all.
The third layer is harsher: bitcoin miners thought "we've got land, power, and construction experience" would let them pivot smoothly into AI, only to find neither holds up. Land, sure — but land isn't power. Construction experience, sure — but a mining facility and an AI facility aren't the same thing: ten times the cost, and higher reliability demands. Nscale's stumble here is a warning for the whole "miners pivot to AI" narrative: you can convert a mining site into a data center, but you can't convert your place in the substation queue. Problems capital can solve aren't problems — and power is precisely what capital cannot solve.
Source: CoinDesk · The Guardian