Surging electricity demand from AI data centers is colliding with a century-old bottleneck: the power transformer. These massive devices, essential for stepping down high-voltage grid power to levels servers can use, now face lead times stretching past two years. Manufacturers including Hitachi Energy, Siemens Energy, and GE Vernova report order books filled through 2026, driven by $200 billion in projected data center capital expenditure this year alone.

The constraint is physical, not financial. Each large transformer weighs hundreds of tonnes, requires specialized steel and copper windings, and must be custom-engineered for specific grid conditions. U.S. utilities now compete directly with hyperscalers — Microsoft, Amazon, Google — for the same limited production slots. Prices for large power transformers have risen 60-80% since 2020, according to Wood Mackenzie.

Grid operators warn the shortage could delay 15-20 gigawatts of new U.S. data center capacity through 2027. Some developers are exploring alternatives: on-site generation, direct high-voltage connections, or modular substations. But none eliminate the transformer requirement entirely. The Federal Energy Regulatory Commission has opened a docket on supply chain resilience, though regulatory fixes move slower than manufacturing cycles.

For investors, the transformer choke point revalues the entire AI infrastructure chain. Companies controlling transformer production or securing long-term slots gain pricing power. Those assuming unlimited grid access face revised timelines and higher capital intensity. The bottleneck is real, measurable, and unlikely to resolve before 2028.