OpenAI pauses UK 'Stargate' data‑centre plans

OpenAI has stopped work on its Stargate project in Britain less than seven months after unveiling it with Nvidia and Nscale, and the company said on April 9 that it will wait until regulation and electricity costs support “long-term infrastructure investment.” (cnbc.com) (openai.com) The plan announced on September 16, 2025 was to give OpenAI local computing power inside the United Kingdom for jobs where the data cannot easily leave the country, including finance, public services, research, and national security work. (openai.com) (datacenterdynamics.com) That local computing power was supposed to start with up to 8,000 graphics processing units, which are the chips used to train and run large artificial intelligence models, and then grow toward 31,000. (datacenterdynamics.com) The site most often linked to the project was Cobalt Business Park on Tyneside in northeast England, with an initial target of launching in the first quarter of 2026, and that deadline passed without construction beginning. (telegraph.co.uk) (bmmagazine.co.uk) This was not just a company expansion plan. Prime Minister Keir Starmer’s government had made onshore artificial intelligence infrastructure a national goal in its January 13, 2025 AI Opportunities Action Plan, which promised to expand sovereign compute capacity by at least 20 times by 2030. (gov.uk 1) (gov.uk 2) “Sovereign compute” is government language for keeping enough chips and data centres at home that the country can run important artificial intelligence systems without depending entirely on servers in another jurisdiction. (gov.uk) Britain has been trying to fix the two bottlenecks that make those projects hard: power and permits. In November 2025, the government published an AI Growth Zones plan built around faster planning approvals and better access to electricity for large data centres. (gov.uk) OpenAI’s pause suggests those fixes were not enough for this project, at least not yet. The company pointed directly to the cost of energy and the regulatory environment, which are the two inputs that decide whether a data centre can run like a factory or sit like an empty warehouse. (cnbc.com) (bloomberg.com) The energy problem is not abstract. United Kingdom government statistics have shown industrial electricity prices were exceptionally high by international standards, and industry groups said in September 2025 that British manufacturers were still paying materially more than European rivals. (gov.uk) (uksteel.org) That leaves Britain in an awkward spot. It wants local control over powerful artificial intelligence systems, but the economics of running tens of thousands of power-hungry chips can still push those systems toward whichever country offers cheaper electricity, quicker grid connections, and fewer planning delays. (gov.uk) (cnbc.com) So the likely near-term shape of “local artificial intelligence” is less a giant national supercomputer in every country and more a split system: smaller tasks on devices or regional servers, with the heaviest model training and inference concentrated where power and permits are easiest to secure. That is an inference from the economics behind this pause, not a new OpenAI announcement. (openai.com) (gov.uk) (cnbc.com)