AI Needs More Than GPUs. Why Are Transformers Taking Years to Deliver?
The AI boom has an oddly old-fashioned problem.
Companies are spending billions of dollars on some of the most advanced computers ever built. Yet one of the hardest pieces of equipment to obtain is based on technology developed more than a century ago: the transformer.
In the United States, a typical large power transformer can take roughly 2.5 to three years to procure. Some extra-high-voltage units can take as long as five years.
That matters because a data center full of GPUs is useless without the electrical infrastructure needed to power it.
The machine behind the electricity
A transformer does a simple but essential job: it changes the voltage of electricity.
Power needs to travel efficiently across long distances at high voltage, then be converted to appropriate voltage levels before it can be used by buildings, factories and data centers.
AI is adding enormous new demand to that system.
JLL expects nearly 100 GW of new data-center capacity to be added globally between 2026 and 2030, roughly doubling existing capacity. Total spending associated with that expansion, including the GPUs and networking equipment installed by tenants, could approach $3 trillion.
All those computers eventually need electricity.
And that means they need transformers.
The waiting list is already affecting construction
The shortage isn't theoretical.
JLL says the average lead time for data-center equipment in the United States is now about 42 weeks, 83% longer than in 2019.
For transformers specifically, its 2026 estimates vary sharply by region:
Americas — 43 weeks
Asia-Pacific — 20 weeks
Europe, Middle East and Africa — 100 weeks
These are data-center equipment averages; the largest grid-scale transformers can take much longer.
Developers are responding by ordering critical equipment as much as 24 months in advance. Even then, 57% of data-center projects surveyed by JLL experienced construction delays of at least three months in 2025.
The AI industry is discovering that computing capacity can grow faster than industrial capacity.
Why not simply make more transformers?
Large transformers are difficult to manufacture quickly.
They require specialized electrical steel, copper, insulation and other components. Many are engineered for particular voltage and grid requirements. The largest units are so heavy that transporting them can require specialized vehicles and logistics.
Only a limited number of factories can manufacture extra-high-voltage models.
The United States is particularly exposed. Government-supported research says more than 80% of U.S. demand for large power transformers is met by imports.
The domestic industry also depends heavily on imported materials such as grain-oriented electrical steel and refined copper.
And AI isn't the only customer.
New factories need transformers. Renewable-energy projects need them. Expanding power grids need them. Aging equipment needs replacement.
AI arrived in a market that was already under pressure.
$457 million says more than a forecast
The clearest evidence of a shortage is often not a market report. It is what manufacturers are willing to spend.
Hitachi Energy is building what it says will be the largest U.S. manufacturing facility for large power transformers in South Boston, Virginia.
The investment:
$457 million.
Expected new jobs:
about 825.
The factory is part of more than $1 billion Hitachi Energy is investing in U.S. grid-equipment manufacturing. The company specifically identifies data centers alongside power generation, transmission and industrial customers as sources of demand.
Factories of this scale also explain why the shortage cannot disappear quickly.
Demand can surge in months.
Industrial capacity takes years to build.
A former Tesla executive sees another opportunity
Established manufacturers aren't the only ones paying attention.
Drew Baglino spent roughly 18 years at Tesla before leaving the company. His new company, Heron Power, is trying to rethink part of the electrical infrastructure itself.
Heron raised $140 million in Series B financing and is planning a $100 million factory in California.
Its first major product, Heron Link, is a 5-megawatt, semiconductor-based power-conversion system aimed partly at data centers. The company hopes to begin mass production in late 2027.
Baglino told Reuters that conventional transformers can take as long as four years to obtain.
Heron is still a startup. Its technology has not proved that it can displace conventional transformers at scale.
But its existence tells us something about the market.
When customers are willing to wait years for an essential product, entrepreneurs start looking for another way to make it.
The opportunity — and the risk
For established transformer manufacturers, today's environment is attractive.
Long waiting lists can support pricing power. Large order backlogs provide visibility. AI, grid modernization and electrification are creating additional demand.
Reuters reported in July that lead times for some U.S. power equipment had exceeded 160 weeks, while transformer prices were expected to rise another 4% to 10% over the following year. U.S. data-center power demand could reach about 110 GW by 2030.
But shortages have a habit of creating their own solution.
High prices attract investment.
Factories expand. New competitors arrive. Customers find substitutes.
Hitachi Energy is adding capacity. Other industrial companies are expanding. Startups such as Heron are trying new technologies.
At the same time, manufacturers themselves are aware that today's data-center boom may not continue indefinitely. Reuters recently reported that some suppliers are structuring long-term customer agreements partly to protect new factory investments from the risk of a future downturn.
Today's shortage could become tomorrow's excess capacity.
If AI infrastructure spending remains strong, transformer manufacturers may enjoy years of demand.
If billions of dollars of new manufacturing capacity arrive just as AI investment slows, today's pricing power could disappear surprisingly quickly.
Both possibilities matter.
AI is becoming an infrastructure story
The first phase of the AI investment boom was easy to see.
Nvidia sold the GPUs.
But GPUs need servers. Servers need cooling. Data centers need electricity. Electricity requires generators, cables, switchgear, substations and transformers.
The further we follow the money, the less AI looks like a purely technology story.
It starts to look like an industrial infrastructure boom.
That doesn't mean transformer manufacturers are automatically good investments. A growing industry and a good investment are not the same thing.
It does mean investors and business owners may need to look beyond the companies with “AI” in their descriptions.
Sometimes the most important company in a technological revolution is not making the revolutionary product.
It is making the ordinary thing the revolution cannot operate without.
Today, one of those ordinary things is a transformer.
BEYOND THE OBVIOUS.
Sources
National Laboratory of the Rockies — Large Power Transformer Supply Chain
U.S. government-supported research on transformer manufacturing, import dependence and procurement times. Typical large power transformers can take roughly 2.5–3 years to procure, with extra-high-voltage units taking up to five years.
NLR — Large Power Transformer Supply Chain
JLL — 2026 Global Data Center Outlook
Data on equipment lead times, construction delays, infrastructure investment and projected global data-center growth.
JLL — 2026 Global Data Center Outlook
Hitachi Energy — U.S. Transformer Manufacturing Expansion
Hitachi Energy's $457 million Virginia large-transformer factory and wider U.S. manufacturing investment.
Hitachi Energy — Virginia Transformer Factory
Reuters — U.S. Power Equipment Shortage
Reporting on transformer and grid-equipment shortages as data-center electricity demand increases.
Reuters — U.S. power companies scramble for equipment
Reuters — Heron Power
Reporting on Drew Baglino's Heron Power, its $140 million funding round and planned $100 million manufacturing facility.