AI Has a Transformer Problem, and Not the One You Think
I've been studying some pretty different industries lately. Hospitals in Australia, hyperscale data centers, residential construction in North America, water treatment and industrial automation, among others. Each has its own technologies, economics and problems to solve.
Something curious kept showing up as a barrier to progress. The same bottleneck across vast and seemingly unconnected domains. Transformers.
Not the "transformer models" that large language models are based on. Not the "Transformers" from the movies and '80s toys. Electrical transformers. The boring kind.
Transformers aren't exactly new technology. They've been around since the 1880s.¹ So why, more than a century later, are they showing up as bottlenecks across so many different industries?
A surprisingly long line
Consider what it takes to build a new data center. The chips, servers, power and cooling systems get most of the attention, particularly with all the excitement around AI. Yet somewhere between the power plant and the silicon, electricity needs to be transformed into voltages the equipment can actually use. That requires transformers, often several different kinds.
What's curious is that data centers are now competing for transformer manufacturing capacity with towns, hospitals, housing developments, water utilities and other customers we'd ordinarily never think of as competing with one another. They may need different kinds of transformers, but their equipment often depends on overlapping manufacturers, materials and production capacity.²
There's another issue. Sometimes having enough transformers to supply the electricity isn't enough. You also need enough to keep operating when something breaks. Some data center designs provide two transformer-fed power paths so either can carry the critical load.³ Hospitals and water treatment facilities can have similar requirements. Federal emergency power guidance describes redundant electrical services for critical facilities, while Virginia's regulations require separate transformers on independent electrical feeds for certain wastewater treatment facilities.⁴ ⁵
All that demand has been growing faster than much of the transformer industry can supply. According to the U.S. Department of Energy (DOE), delivery times for distribution transformers increased from three to six months in 2019 to 12 to 24 months or longer in 2024. DOE also reported that demand for distribution transformers had increased 41% since 2019.⁶ Its 2024 report on large power transformers documented extended procurement lead times, and DOE's 2026 briefing described some large transformer orders taking as long as 48 months.⁷
California’s Roseville Electric Utility is a good example. The utility, which used to buy equipment around a year ahead, is now buying large transformers for projects as many as five years in advance.⁸ That adds risk, not just delays. Utilities now have to forecast demand much further into the future, potentially tying up capital years earlier. A lot can change in five years. Projects can be delayed or canceled, demand can shift and utilities could end up with too much equipment, too little or the wrong kind.
You can build a data center, hospital or factory in less time than it takes to get the transformer needed to power it. That means a project can be otherwise ready to operate, with substantial capital already invested, and still be waiting for electricity.
How did we get here?
The obvious suspect is AI. Hyperscale data centers consume enormous amounts of electricity, and their expansion has increased demand for the equipment needed to connect them to the grid. AI is contributing to the shortage, but the problem predates the current data center boom.
Pandemic-related manufacturing disruptions coincided with utilities replacing aging equipment, expanding renewable energy capacity and increasing electrification.⁹ Transformer demand was already putting pressure on manufacturers before the hyperscalers began placing their enormous orders.
Still, transformers aren't experimental technology. Companies have been making them for generations. Why can't manufacturers simply produce more?
Why are there nearly 40,000 types of transformer?
As of March 2026, there were almost 40,000 different distribution transformer configurations nationwide.¹⁰ Some differences reflect genuine technical requirements, while others have accumulated through decades of utility purchasing practices. That much customization makes mass production difficult.
Even if manufacturers could standardize more designs, they’d still need enough transformer-grade electrical steel, specialized winding machinery, insulation systems and testing capacity.⁷ These are some of the upstream bottlenecks that repeatedly appeared in our value chain analyses.
Expanding a transformer factory doesn't necessarily solve the problem if the factory can't get the materials, components or specialized equipment it needs. An estimated 25% to 30% of U.S. distribution transformers are imported, and even domestically made transformers depend on imported components.¹¹ So the supply chain is global and complex, none of which makes manufacturing easier to spin up.
Why can't manufacturers simply make more?
Manufacturers are responding to the shortage, but expanding production takes years. Hitachi Energy is investing $457 million in additional manufacturing capacity, while Prolec GE is spending $140 million to nearly double production at one of its facilities.¹² ¹³ ¹⁴ Even that expansion is expected to take approximately 30 months. Demand can surge almost overnight, but the factories needed to meet it can't.
So, how much capacity should manufacturers add? Transformers typically last decades, so replacement demand has historically been spread over long periods. Today's surge includes substantial investment in AI infrastructure, but manufacturers have to decide how much of that demand will persist once their new factories are operating. Grid modernization, electrification and industrial expansion may provide more sustained demand, although their timing and scale also matter.
Prices suggest manufacturers have plenty of incentive to expand. According to the Congressional Research Service, inflation-adjusted transformer prices increased approximately 40% between 2020 and 2024.¹¹ The difficulty is deciding how much capacity to build against demand that may look quite different by the time those investments begin producing transformers.
The market is beginning to adjust. For example, the Congressional Research Service found that delivery times for some distribution transformers had fallen to approximately seven months by the second quarter of 2025.¹¹ That's good news, but this problem is deep and it isn't going away overnight. The implications are significant for data centers, property developments, new hospitals, cities and just about everyone with electrical power to activate.
When we first started noticing transformer bottlenecks across our industry analyses in early 2025, I wasn’t particularly interested in transformers. I was trying to understand what was holding each industry back. I certainly didn’t expect to keep finding the same bottleneck in industries that otherwise had so little in common. There’s something counterintuitive about the world’s most advanced AI being held up, at least in part, by a technology with its roots in the 1880s.
Endnotes
1. IEEE History Center, historical overview of early alternating-current transformer development. https://ethw.org/Transformers
2. Congressional Research Service, Electricity Distribution Transformers: Supply, Tariffs, and Policy Options, April 23, 2026, market participants and electrical steel sections. Supports overlapping materials and supply chains while distinguishing product markets. https://www.everycrsreport.com/reports/R48933.html
3. Caterpillar, Data Center Design Considerations, discussion of 2N independent power paths. This is an engineering configuration example, not a claim about the proportion of data centers with duplicate transformers. https://www.cat.com/en_US/by-industry/electric-power/Articles/ep-news/ep-news-data-center-design-considerations.html
4. Federal Emergency Management Agency, Emergency Power Systems for Critical Facilities: A Best Practices Approach to Improving Reliability, FEMA P-1019, Appendix B, 2015. https://www.fema.gov/sites/default/files/2020-07/fema_p-1019_final_02-06-2015.pdf
5. Commonwealth of Virginia, 9VAC25-790-490, Reliability Protection, subsections B through D. Requirements apply to specified Class I reliability facilities and have stated alternatives and exemptions. https://law.lis.virginia.gov/admincode/title9/agency25/chapter790/section490/
6. U.S. Department of Energy, Distribution Transformer Convening Webinar Transcript, March 5, 2026, remarks by Assistant Secretary Catherine Jereza at 00:03. https://www.energy.gov/oe/distribution-transformer-webinar-text-alternative
7. U.S. Department of Energy, Large Power Transformer Resilience Report to Congress, July 2024, section III.3; and Distribution Transformer Convening Webinar Transcript, March 5, 2026, 00:03. The latter reports lead times of up to four years for large transformers. https://www.energy.gov/sites/default/files/2024-10/EXEC-2022-001242%20-%20Large%20Power%20Transformer%20Resilience%20Report%207-10-24.pdf
8. Reuters, US power companies scramble to secure equipment as surging data center demand strains supplies, July 9, 2026. Reports Roseville Electric Utility’s five-year planning horizon. https://www.reuters.com/business/energy/us-power-companies-scramble-secure-equipment-surging-data-center-demand-strains-2026-07-09/
9. U.S. Department of Energy, Distribution Transformer Convening Webinar Transcript, March 5, 2026, 00:03, describing post-pandemic demand, aging infrastructure, workforce and raw-material constraints. https://www.energy.gov/oe/distribution-transformer-webinar-text-alternative
10. U.S. Department of Energy, Distribution Transformer Convening Webinar Transcript, March 5, 2026, 00:04:26. DOE says “almost 40,000” configurations. Its separate discussion of 80,000 possible attribute combinations should not be treated as the same measurement. https://www.energy.gov/oe/distribution-transformer-webinar-text-alternative
11. Congressional Research Service, Electricity Distribution Transformers: Supply, Tariffs, and Policy Options, Report R48933, April 23, 2026. Reports estimated 25% to 30% finished-unit imports, imported cores, inflation-adjusted price increases and 30-week delivery times by Q2 2025, approximately seven months. https://www.everycrsreport.com/reports/R48933.html
12. Hitachi Energy, Hitachi Energy breaks ground on the nation’s largest facility for the production of large power transformers in South Boston, Virginia, June 29, 2026. https://www.hitachienergy.com/us/en/news-and-events/press-releases/2026/06/hitachi-energy-breaks-ground-on-the-nation-s-largest-facility-for-the-production-of-large-power-transformers-in-south-boston-virginia
13. Prolec GE Waukesha, Prolec GE Waukesha Breaks Ground on Major Expansion in Wayne County, March 17, 2026. The $140 million investment was originally announced in May 2025; March 2026 marked groundbreaking. https://www.prolec.energy/news/prolec-ge-waukesha-breaks-ground-on-major-expansion-in-wayne-county/
14. GE Vernova, GE Vernova Completes Prolec GE Acquisition, February 5, 2026. Confirms acquisition of the remaining 50% stake. https://www.gevernova.com/news/articles/ge-vernova-completes-prolec-ge-acquisition-26b-senior-notes
Comments