OpenAI pauses UK data‑centre plans

OpenAI has put its planned United Kingdom data-centre project on hold after deciding the numbers did not work in Britain. The company said industrial power prices in the United Kingdom were about four times higher than in the United States, and it also pointed to unresolved copyright rules for training artificial intelligence systems. (thenextweb.com) The paused project was called Stargate UK, and it was part of OpenAI’s push to secure more of the giant computer warehouses that run tools like ChatGPT. Bloomberg reported the company is also reining in spending before a planned stock-market listing, which makes expensive overseas buildouts harder to justify. (bloomberg.com) A data centre is basically a factory for computation, and electricity is its biggest raw material after chips. If the same rack of servers costs far more to power in one country than another, the cheaper country starts winning before construction even begins. (cnbc.com) Britain has been trying to sell itself as a home for “sovereign” artificial intelligence capacity, meaning computing power located inside the country instead of rented abroad. OpenAI had announced the United Kingdom effort in September with Nvidia and Nscale as part of that pitch. (cnbc.com) The electricity problem is not a one-off complaint from one company. The International Energy Agency said in its 2026 electricity outlook that large differences in power costs for energy-intensive industry still separate Europe from the United States, even after the worst of the energy shock faded. (iea.org) The British government’s own energy data still tracks industrial electricity prices quarter by quarter, and the dataset was updated on March 31, 2026. That means OpenAI was judging the project against a live cost problem, not an old headline from the 2022 energy crisis. (data.gov.uk) The second problem was copyright, which sounds abstract until you remember how these systems are built. Training an artificial intelligence model means feeding it enormous libraries of text, images, audio, and code, so unclear rules over what can be used legally can turn into years of lawsuits or licensing bills. (gov.uk) Britain is still working through that exact question. The government’s copyright and artificial intelligence consultation ran from December 17, 2024, to February 25, 2025, and its March 2026 report reviewed options ranging from a broad text-and-data-mining exception to a stricter licensing model. (gov.uk) That leaves companies trying to plan billion-pound infrastructure with two moving targets at once: the price of every kilowatt-hour and the legal status of the data that fills the machines. OpenAI said it still sees “huge potential” in the United Kingdom, but for now it is keeping the public-services partnership and pausing the server buildout. (bloomberg.com) The pause is a reminder that artificial intelligence infrastructure is not just a software story. It is a map problem, where power markets, land, grid access, and copyright law decide where the next model gets built long before anyone types a prompt. (politico.eu)