OpenAI pauses UK build

OpenAI has put its Stargate UK data-centre plan on hold less than seven months after launching it with the British government, Nvidia and Nscale in September 2025. Politico reported the pause on April 8, 2026, and said OpenAI tied the decision to British copyright rules and power prices. (politico.eu) Stargate is OpenAI’s name for the giant buildings full of chips that train and run artificial intelligence systems. In the United States, OpenAI said in January 2025 that Stargate aimed to invest $500 billion over four years, starting with Texas and other future campuses. (openai.com) Britain was supposed to get its own version in Northumberland, on the site of a former coal power station near Blyth. When the project was announced on September 16, 2025, Politico said the first phase was part of a broader £31 billion wave of United States tech investment into Britain timed around a new U.S.-U.K. technology pact. (politico.eu) OpenAI’s own announcement said Stargate UK would start with up to 8,000 graphics processing units, which are the specialized chips used for artificial intelligence work, and could scale to 31,000 over time. OpenAI said those chips would let its models run on local British computing power for finance, public services, research and national security work where location matters. (openai.com) That local angle is why this was never just another warehouse project. OpenAI said the point of Stargate UK was “sovereign compute,” meaning British organizations could keep sensitive work on machines inside the country instead of sending it abroad. (openai.com) The British government had built policy around exactly that idea. Its January 2025 Artificial Intelligence Opportunities Action Plan called for “AI Growth Zones” to speed up land, power and planning for data centres, and the North East site was later designated as one of those zones. (gov.uk 1) (gov.uk 2) The snag is that an artificial intelligence data centre does not run on hype. A single campus needs huge amounts of electricity, and the British government is still consulting in April 2026 on how to speed grid connections for “strategic demand including data centres,” which is official language for the fact that projects can be stuck waiting for power. (gov.uk) At the same time, Britain is still arguing over how artificial intelligence companies can use copyrighted material for training. The government published a fresh report on March 18, 2026, saying it was weighing four policy options after a consultation that drew more than 11,000 responses, which leaves a company planning a multibillion-pound build without a settled rulebook. (gov.uk) (politico.eu) While Britain works through those bottlenecks, OpenAI has been moving faster in the United States. In September 2025, OpenAI, Oracle and SoftBank said Stargate had nearly 7 gigawatts of planned United States capacity and more than $400 billion in investment lined up over the next three years. (openai.com) So the pause is a reminder that artificial intelligence strategy is now partly a map problem. If one country offers cheaper power, faster permits and clearer rules, the chips go there first, and the products built on top of those chips tend to follow. (openai.com) (gov.uk) Britain is still trying to fix that. A government update in January 2026 said it had designated five Artificial Intelligence Growth Zones and committed £2 billion to expand national compute capacity twentyfold by 2030, but OpenAI’s pause shows that policy promises and construction economics are not the same thing. (gov.uk)