AI data‑centres strain grids

OpenAI has paused part of its “Stargate UK” build after saying Britain’s energy prices and regulatory conditions do not support a long-term data-centre investment right now. The project had been presented in September 2025 as a major United Kingdom artificial intelligence push with local partner Nscale. (networkworld.com) The pause is not about one warehouse full of servers. Politico reported that OpenAI had been exploring leasing up to 8,000 Nvidia chips in the first quarter of 2026 across United Kingdom sites including Cobalt Park, an official “Artificial Intelligence Growth Zone” in northeast England. (politico.eu) A data centre is basically a factory for computing. Instead of assembly lines turning steel into cars, it runs rows of chips that turn electricity into answers, images, code, and predictions. (brookings.edu) Artificial intelligence makes that factory much hungrier for power than older internet services did. Brookings said on April 10, 2026 that advanced artificial intelligence models have become a new, fast-growing source of electricity consumption and that hyperscale operators are pushing for more generation and transmission as they build larger sites. (brookings.edu) The grid is the system of power plants, wires, and substations that moves electricity from where it is made to where it is used. When a single data-centre campus asks for hundreds of megawatts, it can force years of upgrades to transformers, transmission lines, and local substations before the first server turns on. (iea.org) That is why this fight is spreading beyond the technology sector. The International Energy Agency said in its February 2026 Electricity 2026 report that rising power demand is now being driven not just by transport and buildings but also by artificial intelligence and data centres, with grids and flexibility becoming central constraints. (iea.org) Once grid capacity gets tight, every new large user starts competing for the same headroom. Brookings says future constraints on power availability and grid capacity may place limits on artificial intelligence computing, which is another way of saying a chip campus and a new factory can end up arguing over the same wires. (brookings.edu) You can already see governments moving from welcome mats to gatekeeping. Ireland’s energy regulator now requires new data centres seeking grid connections to match their maximum import demand with generation or storage and to meet at least 80% of annual demand with additional renewable projects. (cru.ie) Virginia shows the same pressure in a different form. A 2025 Virginia legislative report said the state had 451 data centres in 2024, while later testimony cited by Virginia Business said Dominion Energy was handling 40,000 megawatts of data-centre-related capacity requests as of December 2024. (rga.lis.virginia.gov) (virginiabusiness.com) That changes how manufacturers think about location. A battery plant, chip plant, or chemicals expansion used to ask about land, labor, and tax breaks first; now it also has to ask whether an artificial intelligence cluster has already claimed the substation, the transmission upgrade budget, and the permitting bandwidth nearby. (brookings.edu) So the OpenAI pause is less a one-company stumble than a preview of a new siting map. Places with cheap power, faster transmission build-outs, and clearer rules will keep winning these projects, and places with expensive electricity and clogged approvals will watch both artificial intelligence campuses and ordinary factories slip elsewhere. (networkworld.com) (brookings.edu)