OpenAI stalls Stargate build

OpenAI just hit pause on part of its Stargate buildout in the United Kingdom, less than seven months after unveiling “Stargate UK” with Nvidia and Nscale in September 2025. The company said the project no longer made sense under current power prices and regulation. (openai.com) (cnbc.com) That sounds like a company problem, but it is really a physics problem first. A modern artificial intelligence data center is a warehouse full of chips that draw huge amounts of electricity every hour, so cheap power is as important as the servers themselves. (openai.com) (networkworld.com) Stargate was announced on January 21, 2025 as a giant infrastructure venture led by OpenAI and SoftBank, with Oracle and MGX as initial equity funders and Arm, Microsoft, Nvidia, and Oracle listed as technology partners. The plan called for up to $500 billion of investment and 10 gigawatts of capacity, which is utility-scale, not startup-scale. (openai.com) (group.softbank) By September 24, 2025, SoftBank said Stargate had nearly 7 gigawatts of planned capacity and more than $400 billion in investment lined up over three years, with Abilene, Texas as the flagship site and five new United States sites added. That made the United Kingdom project look like one spoke in a much bigger machine. (group.softbank) The United Kingdom spoke was pitched as “sovereign compute,” which means keeping computing power inside the country for jobs where location and legal jurisdiction matter. OpenAI said on September 16, 2025 that Stargate UK would run its models on local infrastructure with Nvidia and Nscale. (openai.com) (nscale.com) Nscale’s announcement put numbers on that promise. It said the broader UK commitment involved up to 58,640 Nvidia graphics processing units, while early Stargate UK deployments were reported at up to 8,000 chips with room to scale to 31,000. (nscale.com) (datacenterdynamics.com) Then the economics broke the story. CNBC reported on April 9, 2026 that OpenAI would revisit Stargate UK only when regulation and energy costs support “long-term infrastructure investment,” and multiple reports tied the pause to electricity prices that are about four times higher than in the United States for industrial users. (cnbc.com) (ons.gov.uk) The regulatory piece was not just generic red tape. Reporting around the pause pointed to unresolved United Kingdom rules on artificial intelligence and copyright, which matters because companies building local data centers want to know what models they can train, what data they can use, and what liabilities come with that. (thenextweb.com) (networkworld.com) At the same time, some of the people who helped map OpenAI’s compute expansion are heading to Meta. Bloomberg reported on April 11, 2026 that Peter Hoeschele, Shamez Hemani, and Anuj Saharan, all tied to OpenAI’s computing and Stargate work, are joining Meta. (bloomberg.com) (economictimes.indiatimes.com) That link matters because data centers are not built by money alone. The people who know how to secure land, power contracts, chip supply, cloud partners, and construction timelines are a bottleneck of their own, and Meta is trying to buy its way through that bottleneck as it ramps spending for its artificial intelligence buildout. (bloomberg.com) (outlookbusiness.com) So the pause in Britain is not a story about OpenAI suddenly losing interest in infrastructure. It is a story about where the next wave of artificial intelligence gets built: in places with cheaper electricity, clearer rules, and enough experienced operators to turn a promise on paper into a working grid-connected machine. (openai.com) (cnbc.com)