AI power crunch
Data‑centre builds for AI are colliding with electricity capacity and permitting, forcing utilities to plan roughly $1.4 trillion in grid investment to handle rising demand. Delays and cancellations of many U.S. data‑centre projects are already happening, and a Stanford estimate put global AI compute growth at roughly 3.3× per year with AI energy use hitting about 29.6 GW in 2025. (cbsnews.com) (fool.com) (businesstoday.in)
Artificial intelligence is running into a physical limit: many new data centers cannot get enough electricity fast enough. (cbsnews.com) Investor-owned utilities now expect to spend at least $1.4 trillion through 2030, up from about $1.1 trillion a year earlier, according to a PowerLines report cited by CBS News and Yahoo Finance. Utilities told the survey that data centers are now a top driver of capital spending. (finance.yahoo.com) The strain is already hitting construction plans. More than half of United States data-center builds have been delayed or canceled, The Motley Fool reported on April 14, and other industry coverage has put the share of troubled projects in a roughly 30% to 50% range. (fool.com) A data center is a warehouse of servers, plus cooling gear and backup equipment, that turns electricity into computing power. The International Energy Agency said artificial intelligence training and deployment happen mainly in these facilities, making power supply a hard cap on how much new capacity can come online. (iea.org) The demand curve is steepening. Stanford’s Artificial Intelligence Index said global artificial intelligence computing capacity grew about 3.3 times a year and reached 17.1 million graphics processing units by early 2026, while Business Today reported the same research estimated artificial intelligence energy use at 29.6 gigawatts in 2025. (hai.stanford.edu) (businesstoday.in) That growth is colliding with a grid that was built for steadier demand. Deloitte said United States power demand could rise 10% to 17% by 2030 from 2024 levels, driven by data centers, manufacturing reshoring, and wider electrification. (deloitte.com) The power problem is not just generation. Utilities also need new transmission lines, substations, transformers, and local distribution upgrades, and each piece can face multi-year permitting, equipment, and interconnection delays. (deloitte.com) The International Energy Agency estimated global electricity use by data centers will rise from 460 terawatt-hours in 2024 to more than 1,000 terawatt-hours in 2030 in its base case. That is why utilities, regulators, and developers are now arguing over who pays for the build-out and how quickly projects should get approved. (iea.org) Utilities and grid planners say the spending surge is also about aging equipment, storm resilience, and renewable integration, not only artificial intelligence. Consumer advocates and some policymakers warn that if regulators approve the full build-out, household and business electric bills could absorb part of the cost. (cbsnews.com) The next phase of the artificial intelligence boom will depend less on model demos and more on whether developers can secure megawatts, permits, and grid connections before the end of the decade. (iea.org)
