OpenAI pauses UK Stargate project
OpenAI has put part of its Stargate infrastructure project on hold in the UK, citing runaway power costs and an unfavourable regulatory environment. The pause highlights that AI capacity roadmaps are increasingly constrained by energy, grid and policy realities rather than purely by compute economics. For large enterprises, the move is a reminder that assuming limitless on‑demand model capacity is risky when physical infrastructure becomes the bottleneck. (networkworld.com) (thehindu.com) (computerweekly.com)
OpenAI spent months talking about building more artificial intelligence capacity, then stopped one of its biggest British data-centre plans before it really started. On April 9, Reuters and CNBC reported that OpenAI had paused its United Kingdom Stargate project because British power prices were too high and the regulatory climate was too uncertain. (reuters.com) (cnbc.com) This was not a small pilot in a forgotten corner of the company. Politico reported that OpenAI had said it would explore leasing up to 8,000 advanced Nvidia chips from London-based Nscale at sites across Britain, including Cobalt Park in North Tyneside. (politico.eu) A data centre for artificial intelligence is basically a warehouse full of chips that turn electricity into answers, images, and code. The harder the model works, the more power it draws, so the real input is not just silicon from Nvidia but steady electricity from the grid. (networkworld.com) That is why this story starts with energy bills, not software. Network World reported in February that OpenAI had already shifted toward a “power-first” strategy and was looking at funding generation and transmission because electricity had become one of the fastest-growing limits on large-scale artificial intelligence deployment. (networkworld.com) Britain has wanted these projects badly. The Financial Times reported in September 2025 that OpenAI, Nvidia and Nscale had unveiled “Stargate UK” as part of the British government’s push to turn the country into an artificial intelligence hub, with ministers highlighting sites in the North East of England. (ft.com) The problem is that a minister can promise faster growth, but a grid connection still arrives on engineering time. Industry coverage this week said OpenAI’s British plan ran into industrial electricity prices that are higher than in some rival markets, plus uncertainty around the rules companies would face as they build and train models there. (datacentrereview.com) (capacityglobal.com) This also lands after a year of questions about Stargate itself. Network World reported in July and August 2025 that the broader Stargate buildout had already been slowed by land, energy, financing and partner coordination, even after earlier splashy announcements around hundreds of billions of dollars in planned spending. (networkworld.com 1) (networkworld.com 2) So the British pause is less a sudden reversal than a very visible example of the new math. In older cloud computing, you could often assume more servers would show up if you paid enough; in artificial intelligence, you now need chips, transformers, substations, permits and long-term power contracts to line up at the same time. (networkworld.com) (cnbc.com) OpenAI did not walk away from Britain entirely. Bloomberg reported that the company said it would keep working with the British government on a separate agreement to provide ChatGPT and other services for public services, even while the data-centre buildout was paused. (bloomberg.com) The immediate lesson is simple: artificial intelligence capacity is starting to look like electricity infrastructure, not just internet software. When a project backed by OpenAI, Nvidia and government support can still stall over power and policy, the bottleneck is no longer demand for models but the physical systems needed to run them. (politico.eu) (networkworld.com)
