OpenAI shelves UK datacentre
OpenAI paused its planned Stargate UK datacentre, saying high energy costs and regulatory uncertainty mean the project won’t proceed for now — though it left the door open to resume under the right conditions. The move coincides with OpenAI shifting toward enterprise customers (now >40% of revenue) and public projections of large ad revenue opportunities, a mix that reframes infrastructure decisions as both economic and policy tradeoffs. (cnbc.com) (decrypt.co) (axios.com)
OpenAI spent months talking about building a giant artificial intelligence data center in Britain, then stopped the plan before construction began. The company said the pause came down to two old-fashioned inputs: electricity prices and rules it could not model clearly enough to spend billions against. (cnbc.com) The project was called Stargate UK, and it was announced in September 2025 with Nvidia and Nscale as partners. The pitch was thousands of graphics processing units, which are the specialized chips used to train and run artificial intelligence systems, on British soil. (cnbc.com) A data center for artificial intelligence is basically a power plant customer wearing a software label. The chips do the math, but the real constraints are land, grid connections, cooling equipment, and a power bill that arrives every month whether users show up or not. (finance.yahoo.com) Britain wanted this project because it has been trying to sell itself as an “AI superpower” since 2023. A local cluster of high-end chips would have given the government something it has talked about for months: sovereign compute, meaning domestic capacity it does not have to rent from someone else abroad. (eandt.theiet.org) OpenAI did not slam the door shut. People involved in the talks told outlets that discussions with Nscale are still active, and OpenAI said it could restart if conditions improve. (cnbc.com) (techstartups.com) The timing matters because OpenAI’s business is changing fast. In a company post published this week, OpenAI said enterprise customers now make up more than 40% of revenue and are on track to reach parity with consumer revenue by the end of 2026. (openai.com) That changes how a data-center bet looks on a spreadsheet. If more revenue comes from banks, insurers, and large software contracts, OpenAI can buy capacity where it is cheapest and most predictable instead of building prestige infrastructure in every market that wants a flagship project. (openai.com) (cnbc.com) At the same time, OpenAI is telling investors a very different scale story on the consumer side. Axios reported that the company projects $2.5 billion in advertising revenue in 2026 and $100 billion by 2030, based on a forecast of 2.75 billion weekly users by 2030. (usnews.com) Put those two facts together and the UK pause looks less like a retreat from artificial intelligence and more like a filter. OpenAI is still chasing huge demand, but it is choosing locations where power, permits, and policy make that demand cheaper to serve. (cnbc.com) (openai.com) (usnews.com) For Britain, the awkward part is that this was exactly the kind of win it wanted to advertise. When a company building the most compute-hungry products in tech says the numbers do not work in your market, that is not a branding problem first; it is an energy and regulatory problem first. (cnbc.com) (eandt.theiet.org)
