The US has a transformer problem. The metal-and-oil kind that sits on utility poles and converts voltage so electricity can actually reach your data center. Lead times for electrical transformers now stretch past two years, and prices have quadrupled since the 2000s. This matters because AI data centers need enormous amounts of power, and every connection between different voltage levels on the grid requires a specific transformer. No transformer means no power. No power means no AI.
The bottleneck starts with grain-oriented electrical steel, or GOES, the specialized material used in transformer cores. About 80% of global GOES production sits in China, Japan, and South Korea. The US has exactly one primary domestic producer: Cleveland-Cliffs' AK Steel, operating with limited capacity. The Department of Energy has formally identified GOES as a critical material for energy infrastructure. Section 301 tariffs on Chinese steel were supposed to help domestic manufacturing, but they've actually increased costs for US transformer makers, even with existing exemptions for GOES.
The numbers are stark. The National Renewable Energy Lab estimates the US has somewhere between 60 and 80 million distribution transformers, compared to roughly 5,000 large power transformers. NREL projects distribution transformer capacity needs to grow 160% to 260% by 2050 just to meet baseline demand from residential, commercial, industrial, and transportation sectors. That's before you factor in the accelerating power hunger of AI training and inference clusters. Each transformer is a single point of failure in the network. Lose one to a wildfire or a car accident, and that part of the grid goes dark for years.
Investment bank Harris Williams identified six drivers of transformer spending in 2023: aging infrastructure, grid resiliency, legislation driving new demand, renewable energy generation, high-growth end markets like AI and EVs, and reshoring initiatives. All arrows point up. Yet as of 2026, no new US supply has come online. Rob L'Heureux, who wrote the original analysis, puts it plainly: this is a country that let its industrial base decay and hasn't figured out how to rebuild it. Building GPT-5 turns out to be the easy part.