Transformer shortage delays AI centers
- Around half of US AI data-center projects in 2026 are delayed due to a global power transformer shortage. - Prices reportedly tripled and lead times stretched to two-to-four years, with Cleveland‑Cliffs cited as a key supplier. - These supply constraints could meaningfully slow AI deployments and raise build costs for hyperscalers and enterprise clouds (x.com).
A transformer is the grid’s gearbox: it changes voltage so electricity can move from substations into a data center. In 2026, shortages of those machines are delaying or canceling roughly a third to half of U.S. data-center projects planned to open this year. (bloomberg.com) (techspot.com) Bloomberg reported that 12 to 16 gigawatts of U.S. data-center capacity had been slated to come online by the end of 2026, but only about 5 gigawatts is under construction. Typical build times run 12 to 18 months, which leaves little room for equipment that arrives years late. (techspot.com) (tech.yahoo.com) The bottleneck is not just land or financing. Developers also need switchgear, batteries and large transformers, and the International Energy Agency said procurement for large power transformers has stretched to as long as four years, about double 2021 timelines. (bloomberg.com) (iea.org) That squeeze lands as Alphabet, Amazon, Meta and Microsoft are forecast to spend about $650 billion on capital expenditures in 2026, much of it tied to artificial intelligence infrastructure. Money can buy land, chips and contractors, but it cannot instantly expand transformer factories. (bloomberg.com) (globalbankingandfinance.com) The shortage did not start with AI. The Department of Energy and the National Renewable Energy Laboratory have tied transformer shortages to post-pandemic demand, grid modernization, renewable-energy build-outs, electrification, extreme weather replacements and shortages of steel, copper and labor. (energy.gov) (docs.nrel.gov) In a 2024 report, NREL said utilities were already seeing transformer lead times of up to two years and price increases of five to six times over two years for some distribution units. A 2024 National Infrastructure Advisory Council report said average transformer lead times had climbed to about 120 weeks, with some large units taking 80 to 210 weeks. (osti.gov) (cisa.gov) One chokepoint is grain-oriented electrical steel, the specialized metal used in transformer cores to reduce energy loss. Cleveland-Cliffs said in 2024 that it was the exclusive U.S. producer of that steel and tied a new $150 million transformer plant in Weirton, West Virginia, to rising demand. (clevelandcliffs.com) (manufacturingdive.com) Manufacturers are adding capacity, but slowly. Cleveland-Cliffs said the Weirton plant was scheduled to start production in the first half of 2026, while Eaton said in 2025 that a new South Carolina transformer facility would come online by 2027. (manufacturingdive.com) (eepower.com) For cloud companies, the result is a mismatch between software timelines and power-equipment timelines. Data centers for AI are often planned around 12-to-18-month construction cycles, while key electrical gear now carries waits measured in years. (techspot.com) (iea.org) The immediate question is not whether demand for AI computing exists. It is whether the grid equipment behind that demand can be built and delivered fast enough for 2026 projects to open on schedule. (bloomberg.com) (cisa.gov)
