AI infrastructure hits limits
OpenAI has paused part of a UK data‑centre project, citing high energy costs and regulatory hurdles, signalling that compute expansion is now constrained by local power economics. At the same time, major cloud builds continue elsewhere — CoreWeave signed a large cloud deal with Meta — showing that scale is still moving but with clear regional friction. (ghanamma.com) (networkworld.com) (enterpriseai.economictimes.indiatimes.com)
OpenAI spent months selling the idea that more artificial intelligence just meant more buildings, more chips, and more money. On April 9, 2026, it paused its main United Kingdom data-centre project instead, saying Britain’s energy costs and regulatory environment no longer worked for the plan. (networkworld.com) The paused site was part of “Stargate UK,” a project OpenAI announced in September 2025 with Nvidia and British developer Nscale. The build was pitched as a major boost for Britain’s attempt to become an artificial-intelligence hub. (cnbc.com) A data centre for artificial intelligence is basically a power station with computers attached. The expensive part is not just buying chips like Nvidia’s graphics processors, but feeding them electricity every hour and removing the heat they throw off. (networkworld.com) That is why this pause is different from an ordinary property delay. OpenAI did not say it lacked demand for computing power; it said the local math on power prices and rules stopped working. (reuters.com) Britain has been trying to speed up new data-centre projects, but operators still face grid-connection delays, planning fights, and some of Europe’s highest industrial electricity costs. Those problems hit hardest when a single facility may need the power draw of a small town. (capacityglobal.com) OpenAI has not abandoned the country completely. The company said it still sees “huge potential” in the United Kingdom and plans to keep working with the government on artificial-intelligence services for public agencies while the infrastructure piece is on hold. (bloomberg.com) At almost the same moment, the market got the opposite signal from somewhere else. On April 9, 2026, CoreWeave said Meta had expanded their cloud agreement to about $21 billion through December 2032 for artificial-intelligence capacity spread across multiple locations. (coreweave.com) CoreWeave’s deal shows the build-out is still happening, but it is moving to places and partners that can line up land, power, permits, and financing fast enough. The agreement also includes some early deployments of Nvidia’s Vera Rubin platform, which means the newest chips are being reserved for sites that can actually switch on. (coreweave.com) Meta’s extra spending also tells you who now absorbs the pain. Instead of every big technology company building every facility itself, some are paying specialist cloud operators like CoreWeave to secure the electricity, debt, and construction risk first. (cnbc.com) So the constraint on artificial intelligence in 2026 is not just chip supply anymore. It is whether a region can deliver cheap enough power, fast enough approvals, and enough grid capacity before the next model needs to train. (networkworld.com)
