All research

Energy · Deep dive · 2026-08-22

The transformer bottleneck: why an available megawatt can still wait

Power can exist on the system and still fail to reach a data-center rack. Between the grid and the load sits a custom chain of steel, copper, insulation, factory slots, testing, transport and site engineering.

, Founder and publisher, AI Bottlenecks

A data-center developer can secure land, announce a power agreement and order racks, yet still wait for the equipment that makes the electricity usable.

The transformer sits at that handoff. It changes voltage so power can move efficiently across the grid and then enter the campus at the levels required by switchgear, distribution systems and computing equipment. Without the right unit in the right substation, an available megawatt is an electrical fact, not a delivered service.

Calling this a transformer shortage is directionally correct and mechanically incomplete. A large power transformer is a custom manufacturing project made from constrained materials, wound and assembled in a scarce factory slot, tested on specialized equipment, moved under difficult logistics and integrated into a site-specific protection and cooling design.

The bottleneck is the chain.

What is inside the box?

At the center is a magnetic core, usually built from layers of grain-oriented electrical steel. Copper or aluminum windings wrap the core. Electrical insulation keeps conductors and voltage levels separated. Bushings carry current safely through the grounded tank. Tap changers adjust the voltage ratio. Oil and a cooling system move heat away. A steel tank holds the active assembly and protects it from the environment.

The U.S. Department of Energy's 2024 Large Power Transformer Resilience report maps the inputs in useful detail: grain-oriented electrical steel, continuously transposed copper conductor, insulating pressboard, mineral oil, tanks, tap changers, cooling systems and bushings all flow into winding, core and final assembly.

Each layer carries a different risk.

This is why an extra supply of one raw material does not automatically create an extra transformer.

Grain-oriented electrical steel is a real weak link

Transformer cores use grain-oriented electrical steel, or GOES, because its magnetic properties reduce core loss in the direction that flux travels. Higher-performance grades can lower losses and reduce the weight required for a given design.

DOE's report identifies domestic GOES production as a major weak link in the U.S. large-transformer supply chain. It also notes competition between GOES and non-oriented electrical steel for production capacity. The latter serves motors, including growing electric-vehicle demand.

The report cites a 2020 Department of Commerce survey in which GOES and continuously transposed copper conductor each represented roughly a quarter of final large-transformer production cost. Those are historical estimates, not a current bill of materials, but they show why steel and copper availability reach deep into factory economics.

Material is only the first gate. A manufacturer still has to turn thin steel laminations and insulated conductor into a unit that can pass a demanding electrical test.

The scarce asset is often the build slot

Large transformers are not produced like identical appliances. The voltage, capacity, impedance, cooling, noise limit, footprint, terminal arrangement and protection scheme can be specific to a project or utility standard.

A factory has a limited number of places where cores can be stacked, windings made, active assemblies dried, tanks filled and finished units tested. Some of the same equipment and people may serve large units, smaller transformers or refurbishment work.

DOE describes these as build slots and notes that capacity cannot be expanded quickly without new facilities. Its 2024 report said new factory construction can take one to three years. That statement predates the latest AI data-center demand wave, but the physical point is durable: a purchase order cannot create a winding hall, vapor-phase drying system, test bay and trained workforce overnight.

The important KPI is not factory square footage. It is qualified throughput from the constrained stations.

Testing is part of production

A transformer is valuable because it survives high voltage for decades, not because it looks complete.

Factory acceptance testing can include turns-ratio checks, winding resistance, insulation measurements, no-load and load losses, impedance, applied or induced voltage tests, temperature-rise work for a design, noise measurement and checks of controls and auxiliaries. Requirements depend on the unit and standard.

The test bay is therefore part of manufacturing capacity. A completed transformer waiting for a high-voltage slot is not shipped output. A failed test can send the unit back into diagnosis, drying, repair or rework.

Data-center projects can make this interface more demanding. The campus needs confidence about loss, harmonics, protection coordination, redundancy and behavior under rapidly changing load. The transformer does not operate alone. It sits inside a designed electrical system.

Transport turns hardware into a route-planning problem

Large transformers are heavy and difficult to move. Rail clearances, bridge limits, port access, heavy-haul trailers, road permits and crane availability can shape delivery. Some components may travel separately and be fitted on site.

This creates a geographical constraint that a global order-book number can hide. A factory may have capacity, but not for the required specification or delivery route. A transformer may be complete, but the site may not be ready to receive it. A damaged bushing or missed crane window can delay commissioning after the long manufacturing wait is over.

Delivered power is a schedule, not a nameplate.

Data centers compete with the rest of the grid

AI campuses are not the only source of demand. Utilities need transformers for grid expansion, replacement of aging equipment, renewable generation, storage, electrification and resilience.

DOE estimated that more than 80 percent of U.S. demand for new transformers above 60 MVA in 2019 was met by imports. It also described an installed fleet with long service lives and significant age. These figures are useful context, not a current market forecast. They show that data-center demand arrived on top of a supply system already balancing replacement and expansion.

Distribution transformers face pressure too. In a 2026 webinar transcript, DOE said distribution-transformer demand had risen 41 percent since 2019 and cited lead times of one to two years or longer in 2024, the latest year in that discussion. Large campus projects pull across voltage classes, so the constraint can appear in more than one box.

The public-company map

Transformer exposure appears across global grid-equipment manufacturers and more specialized suppliers. Hitachi Energy, Siemens Energy and GE Vernova sell major grid equipment and systems. Other public names sit farther downstream in campus distribution, protection and field execution, which the Energy system map separates from the transformer layer.

The cleanest stock pitch is rarely the cleanest system map.

Large diversified manufacturers can have strong grid backlogs while transformer revenue remains one piece of the group. Regional specialists can have tighter exposure but different liquidity and customer concentration. Campus equipment suppliers may benefit from the same delivered-power scarcity without building the large transformer itself.

The correct question is which constrained step the company owns, how much capacity it is adding, and whether price, mix and execution convert backlog into margin.

What would ease the bottleneck?

There is no single fix, which is encouraging and inconvenient.

Manufacturers can expand plants and test bays. Utilities and developers can standardize more specifications, reserve slots earlier and qualify additional suppliers. Material producers can add GOES and conductor capacity. Flexible designs and strategic spares can reduce replacement risk. Better project sequencing can ensure the site, route and crane are ready when the unit arrives.

Standardization has limits. Voltage, network studies, protection philosophy, transport restrictions and site geometry remain local. A standard core platform can reduce engineering work, but the last mile still has to fit the system.

The strongest evidence of easing would be shorter quoted lead times accompanied by stable quality, more available test capacity, rising accepted unit output and less customer prepayment or slot reservation. Price weakness alone could signal demand destruction rather than healthy supply.

What would prove the thesis?

Watch the receipts that connect factory capacity to delivered equipment.

The thesis weakens if announced capacity produces accepted units faster than demand grows, if project cancellations release meaningful slots, or if standardization materially shortens engineering and test cycles. It also weakens if data-center power plans move toward architectures that need less conventional grid equipment per delivered megawatt.

An available megawatt is not a delivered megawatt

The transformer bottleneck explains a recurring mistake in data-center power analysis. Generation capacity, grid connection and usable campus power are not interchangeable.

A megawatt becomes useful only after the system transforms its voltage, protects it, switches it, cools the equipment and commissions the site. Transformers sit near the center of that chain. Their scarcity is a compound problem across material, design, factory slots, test, transport and local engineering.

The electrical supply can be real. The rack can still wait.

Continue with AI's energy bottleneck: the race for delivered power for the full generation-to-rack stack.

Sources and receipts

This research is educational and is not investment advice.