A utility in Virginia recently told a data-center developer that the high-power transformer needed to energize its planned campus would not arrive for five years. The developer had budgeted for roughly 30 months. That gap, now showing up in procurement timelines across the country, is quietly reshaping where and how fast the next generation of AI infrastructure can be built.
The problem is straightforward in concept and brutal in practice. Large data centers drawing hundreds of megawatts need dedicated power transformers rated at 100 megavolt-amperes (MVA) or more to connect to the high-voltage transmission grid. Only a handful of factories worldwide produce equipment at that scale, and every one of them is backlogged. According to Bloomberg reporting, lead times for these units have stretched to as long as five years, roughly double the timeline utilities quoted as recently as 2022.
A shortage years in the making
The squeeze did not start with AI. The Department of Energy’s Office of Electricity has tracked a steady deterioration in transformer supply chains since before the pandemic. Distribution transformers, the smaller units that step voltage down for neighborhoods and commercial buildings, went from a lead time of three to six months in 2019 to 12 to 30 months by 2023, according to the office’s supply-chain analysis. Those are not the same class of equipment that a 200-megawatt data center needs, but the trend illustrates how broadly the market has tightened at every tier.
A 2023 Government Accountability Office investigation, GAO-23-106180, laid out the structural forces behind the delays: limited global manufacturing capacity, a constrained supply of grain-oriented electrical steel (the specialized material used in transformer cores), and a domestic workforce too small to support rapid production increases. The GAO also flagged a reliability consequence that extends well beyond data centers. When utilities cannot replace aging or storm-damaged transformers quickly, the entire grid becomes more vulnerable to prolonged outages.
Then came the AI construction wave. Announcements from Microsoft, Amazon, Google, and Meta over the past two years have added tens of gigawatts of projected electricity demand to utility planning forecasts. Each large campus needs not just power but the physical hardware to receive it, and that hardware is now the binding constraint.
The interconnection queue is already overwhelmed
Even before transformer delivery times doubled, the process of connecting new projects to the grid was strained. The Federal Energy Regulatory Commission’s interconnection queue, the formal pipeline through which generators and large loads secure grid access, hit record volumes in 2023. FERC’s Order No. 2023 attempted to clear the backlog by streamlining study and approval procedures. But the rule was designed primarily for generator interconnection, not for demand-side loads like data centers, and it does nothing to solve the physical shortage of equipment waiting at the end of the process.
The result is a two-stage bottleneck. A data-center project can spend years working through interconnection studies and permitting, only to discover upon approval that the transformer it needs will not ship for several more years. No federal agency currently publishes a regularly updated index tracking lead times specifically for high-power units rated at 100 MVA and above, so developers often do not learn the true timeline until they place an order.
What manufacturers are up against
Building a large power transformer is not like assembling commodity electronics. A single unit can weigh more than 400 tons, require thousands of pounds of grain-oriented electrical steel, and take months of specialized labor to wind, insulate, and test. The global market for these machines is dominated by a small number of manufacturers, including Hitachi Energy, Siemens Energy, and ABB, and their order books were already full before data-center demand surged.
Expanding production is slow. A new transformer factory takes three to five years to build and staff, meaning that capacity additions announced today will not ease the current backlog. The supply of grain-oriented electrical steel, produced by only a few mills worldwide, adds another constraint. Even if a manufacturer secures factory space and workers, it may not be able to source enough core material to increase output significantly.
How developers are adapting
Faced with multi-year waits, data-center developers are changing how they plan and build. Some are phasing their campuses so that initial buildings operate on whatever grid capacity already exists at a site, deferring full buildout until new transformers arrive. Others are installing on-site natural gas generation or exploring small modular nuclear reactors as a way to bypass the grid connection entirely, at least for a portion of their load.
Site selection has shifted as well. Locations near substations with spare high-power transformer capacity, or in utility territories where equipment was pre-ordered years ago, now carry a premium that rivals cheap electricity or abundant fiber. A site in central Ohio with an available 345-kilovolt transformer connection may beat a cheaper site in Texas where the nearest suitable equipment is three years from delivery.
None of these workarounds fully replaces a dedicated grid connection through a properly rated transformer. On-site generation adds fuel cost and emissions. Phased buildouts delay revenue. And small modular reactors remain, as of mid-2026, a technology with more announcements than operating units.
Policy levers that could help, eventually
At the federal level, several ideas are circulating. Standardizing transformer designs could allow manufacturers to shift from one-off custom builds to something closer to serial production, cutting months from each order. Targeted incentives for domestic transformer factories could reduce dependence on a small number of overseas suppliers. And better coordination between FERC’s interconnection process and utility procurement schedules could prevent the scenario where a project clears every regulatory hurdle only to stall on equipment delivery.
The DOE has signaled interest in all of these approaches, but none can eliminate five-year waits in the near term. Building a new factory takes years. Changing procurement practices across dozens of independent utilities takes longer. And the demand side of the equation shows no sign of easing: beyond AI data centers, the electrification of transportation and heavy industry is adding its own pressure on transformer supply.
A few pieces of steel shaping the digital economy
What makes this bottleneck unusual is its specificity. The constraint is not land, or capital, or even electricity generation. It is a single category of industrial equipment, produced by a small number of factories, from a specialized material, on timelines that no software update can compress. The DOE and GAO have documented the broad supply-chain deterioration. Bloomberg’s reporting pins the sharpest delays to the largest, most complex units. FERC’s data confirms that the queue of projects waiting for grid access far exceeds the system’s capacity to process them.
For data-center developers, grid planners, and the communities that depend on reliable power, the practical reality as of mid-2026 is this: the availability of a few hundred high-power transformers will do more to determine the pace of AI infrastructure buildout than any amount of investment capital or computing innovation. Until manufacturing catches up, the digital economy’s expansion runs on the schedule of a steel-and-copper machine that takes years to build and cannot be rushed.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
