AI data‑centre builds face scrutiny
A wave of reporting and commentary suggests AI data‑centre projects are being delayed or cancelled as economics, power constraints and permitting bite. One YouTube piece highlights 48 delayed/cancelled AI data‑centre projects, and OpenAI says it has paused its main UK data‑centre plan citing regulation and costs—signals that some large builds may be getting reappraised. That shift matters because it could change GPU demand, cloud pricing leverage and how buyers structure long-term infrastructure commitments. (youtube.com) (finance.yahoo.com)
A data centre is basically a warehouse full of computers, and the surprise in April 2026 is that some of the biggest artificial intelligence warehouse plans are no longer moving in a straight line. OpenAI told Reuters on April 9 that it is pausing its main Britain project because energy costs are high and the regulatory environment is unfavourable. (finance.yahoo.com) That Britain project was not a small pilot. Reuters said OpenAI’s planned Stargate project with Nvidia and Nscale was meant to deploy thousands of graphics processing units, which are the chips that do the heavy lifting for training and running artificial intelligence models. (finance.yahoo.com) The timing matters because the whole artificial intelligence boom has been built on the assumption that more demand would always justify more buildings. A YouTube video published in April 2026 claims 48 artificial intelligence data-centre projects were delayed or cancelled in 2025, which shows how quickly the conversation has shifted from “build everything” to “which sites can actually pencil out.” (youtube.com) The bottleneck is not just money. The International Energy Agency said data centres used about 460 terawatt-hours of electricity in 2024, and its base case projects demand rising to more than 1,000 terawatt-hours by 2030, which means every new site is competing for power lines, substations and generation capacity. (iea.org) That creates a very ordinary problem inside a very futuristic industry: you can order servers faster than you can build the electrical plumbing around them. The International Energy Agency said artificial intelligence workloads happen mainly in data centres, but those facilities also need cooling systems, backup equipment and grid connections before a single chip can be switched on. (iea.org) Permits are the other slow part. Reuters said OpenAI paused its Britain plan over regulation as well as energy costs, and that lands squarely in the part of the project where local approvals, land use rules and utility negotiations can stretch for months or years before concrete is poured. (finance.yahoo.com) Britain is awkward here because the government has been pitching the country as an artificial intelligence hub. Reuters said the pause is a setback for Prime Minister Keir Starmer’s effort to attract global artificial intelligence investment, which means the issue is no longer just corporate budgeting but national industrial policy. (finance.yahoo.com) The bigger shift is that companies may stop treating giant campuses as the default answer. If power is scarce and permits are slow, buyers start favouring shorter contracts, phased rollouts and existing cloud capacity over betting billions on one site that might sit half-finished. (iea.org) That does not mean artificial intelligence demand disappears. It means the winners change from whoever announced the biggest campus in 2025 to whoever can secure electricity, permits and financing in 2026 without paying so much that the economics break before the servers arrive. (iea.org)
