AI data-centre bottleneck: transformers
- Reports say many 2026 U.S. AI data-centres are delayed due to power transformer lead times and price spikes. - Transformer lead times are now two to four years, and prices have roughly tripled, creating capacity constraints. - That hardware and power bottleneck turns physical grid and supplier concentration into a core risk for AI deployments (x.com).
A basic AI data center needs more than chips and land: it needs giant transformers, and those are now taking years to get. (bloomberg.com) Bloomberg reported on April 1 that more than half of the U.S. data centers planned for 2026 are expected to be delayed because key electrical gear is hard to source. CBRE separately said U.S. capacity under construction fell to 5.99 gigawatts at the end of 2025, down from 6.35 gigawatts a year earlier, as developers ran into permitting, zoning, and power-procurement delays. (bloomberg.com 1) (bloomberg.com 2) A transformer is the grid’s voltage gearbox: it steps electricity up for long-distance transmission and down so a site can use it safely. For the largest units, procurement times have stretched to 120 to 210 weeks, according to the National Infrastructure Advisory Council and industry reports. (cisa.gov) (powermag.com) The backlog did not start with AI. The National Infrastructure Advisory Council said average transformer lead times rose to 120 weeks or more, while the largest units can take as long as 210 weeks, after years of underinvestment, labor shortages, and tight supplies of grain-oriented electrical steel and copper. (cisa.gov) (utilitydive.com) AI is adding a new load shock on top of that older shortage. The Electric Power Research Institute now projects U.S. data centers could consume 9% to 17% of national electricity by 2030, up from a much smaller share today. (powering-intelligence.epri.com) The Financial Times reported that new data centers will need an estimated 44 gigawatts of additional capacity by 2028, while grid constraints leave only about 25 gigawatts available in the next three years. That leaves a gap of roughly 19 gigawatts. (ft.com) The equipment pinch is broader than one box in a substation. POWER Magazine said switchgear averaged 44 weeks in the second quarter of 2025, and Utility Dive reported high-voltage transformers were taking about three years on average even as manufacturers announced reshoring plans. (powermag.com) (utilitydive.com) Developers have responded by hunting for imported gear and on-site generation. Bloomberg reported that shortages in transformers, switchgear, and batteries have pushed some U.S. projects to rely on Chinese electrical equipment imports, while some large campuses have turned to gas turbines to get computing capacity online faster. (bloomberg.com 1) (bloomberg.com 2) Utilities and manufacturers have pressed Washington for help. The Department of Energy’s July 2024 report to Congress said extended replacement times for large power transformers are a grid-reliability problem, and industry groups have backed ideas including a transformer reserve, standardized designs, and incentives for new domestic production. (energy.gov) (makeitelectric.org) For AI builders in 2026, the constraint is no longer only how many graphics processors they can buy. It is also whether a transformer ordered this year arrives before the data center it is supposed to power. (bloomberg.com)
