Microsoft pulls out of a 50 MW AI data center as grid capacity isn't available until the 2030s

Nscale, a company that grew out of bitcoin mining and is backed by Nvidia, had announced last year, alongside Microsoft, Nvidia and OpenAI, a 50 MW site in Loughton, Essex, for Microsoft's Azure cloud, built around 23,040 Nvidia chips. The site was due online in early 2027. UK Power Networks, the local grid operator, told Nscale it could not guarantee sufficient power capacity before the early-to-mid 2030s — a delay incompatible with the project's timeline. Microsoft walked away; Nscale has cleared the site and is exploring other ways to secure power, while stating that its other Microsoft contracts remain in place.

UK regulator Ofgem counts about 73 GW of grid-connection requests from data center projects — a volume the grid cannot absorb within the timelines project developers have announced. Ofgem is weighing tougher queue rules, noting that some applications show no real intention of ever being built, which delays the projects that are genuinely ready to go.

The cost gap between the two uses of the same electrical asset illustrates part of the problem: building AI-ready capacity costs an estimated $8–15 million per megawatt, versus $0.7–1 million per megawatt for bitcoin-mining capacity, according to CoinShares. Public bitcoin miners have announced more than $70 billion in AI and HPC contracts since March 2026, but CoinShares tracks only about 550 MW actually billing customers out of more than 4 GW under contract — a sign that reserving a slot in the connection queue is far cheaper than building the capacity behind it.

For power-intensive industrial sectors — energy, O&G, aerospace — this UK case is a warning sign: grid availability, not chip or capital availability, is becoming the real bottleneck for large-scale AI deployment, with connection delays now measured in years, not months.