AI Data Center Transformer Shortage: The Grid Equipment Bottleneck Beneath the Boom

September 16, 2026

Altsets

Research by Altsets Research

Share

Why AI data-center growth is straining transformers, electrical steel, switchgear, cable, grid hardware, and cooling infrastructure at the same time.

Data used:Altsets Supply Chain Intelligence: 90k+ entities, 400k+ relationships, 20+ years of history.

Key findings

  • The AI infrastructure constraint extends beyond large power transformers into electrical steel, components, switchgear, wire, cable, grid hardware, and cooling.
  • Hitachi Energy's announced U.S. manufacturing commitments illustrate the scale and long lead times involved in expanding transformer capacity.
  • The relevant investment exposures differ by layer: long-cycle equipment, specialized upstream inputs, and higher-volume campus infrastructure should not be treated as interchangeable.

Research brief

Summary: The AI data center buildout is creating a broader electrical equipment constraint than the phrase "transformer shortage" suggests. Large power transformers remain difficult to source, but the same investment cycle is now pulling on electrical steel, transformer components, switchgear, medium voltage equipment, wire and cable, grid hardware, and cooling capacity.

The transformer bottleneck beneath the AI data center boom is becoming an electrical infrastructure bottleneck. Hitachi Energy's latest US expansion makes the scale visible. On September 15, 2026, the company announced a $528 million transformer factory in Gallman, Mississippi that is expected to more than double production capacity in the area and create more than 700 jobs. That follows the $457 million large power transformer expansion now under construction in South Boston, Virginia, plus a $106 million expansion of transformer component production in Tennessee. Hitachi now describes its US manufacturing expansion commitment as roughly $1.5 billion.

This matters to investors because adding servers does not create a usable data center by itself. New computing load has to be connected to generation and transmission, transformed to lower voltages, switched and protected, distributed through the campus, and cooled. A shortage at any sufficiently critical point can delay the revenue producing asset even if GPUs and servers are available.

The transformer shortage is increasingly a layered electrical infrastructure constraint, not a single-product shortage.

$528M
Hitachi Energy Mississippi transformer factory
Announced September 15, 2026
~$1.5B
Hitachi Energy US manufacturing expansion commitment
11.8%
Central estimate of US electricity consumption from data centers in 2030
LBNL 2025 update
12 to 30 months
Distribution transformer lead times cited for 2023
US Department of Energy
80,000+
Distribution transformer varieties used across the United States
US Department of Energy
>$5B
GE Vernova data center orders in Electrification in H1 2026

The result is a different way to read AI infrastructure spending. Semiconductor demand is concentrated inside the computing layer. Electrical infrastructure demand spreads outward into Hitachi, GE Vernova, Eaton, Siemens Energy, Hammond Power Solutions, Cleveland-Cliffs, Prysmian, Hubbell, ABB, Schneider Electric, Vertiv, Modine, and other companies that occupy different parts of the power path. They should not be treated as interchangeable AI beneficiaries. Some are exposed to genuinely difficult manufacturing bottlenecks, while others are benefiting from higher unit volumes in markets where additional capacity can arrive more quickly.

Why data centers are colliding with the grid equipment supply chain

The underlying demand increase is no longer theoretical. Lawrence Berkeley National Laboratory's 2025 update estimates that US data centers could account for 11.8% of national electricity consumption in 2030 in its central estimate, with scenarios ranging from 9.5% to 15.3%. The Energy Information Administration has also found that US electricity demand grew about 1.7% annually from 2020 through 2025, compared with roughly 0.1% annually from 2005 through 2019, with data centers driving much of the recent increase.

That load arrives on an electrical system whose equipment supply chain was already strained. The Department of Energy says distribution transformer lead times increased from three to six months in 2019 to 12 to 30 months in 2023, the latest comparable data it cites. DOE was still describing distribution transformers as supply constrained in 2026 and highlighted shortages in materials and components rather than treating the problem as an isolated factory scheduling issue.

There is also a specification problem. DOE has identified more than 80,000 distribution transformer varieties used across the United States. That fragmentation makes it harder to convert nominal factory capacity into interchangeable units that any utility can deploy. The transformer shortage therefore cannot be solved simply by adding one generic production line. Designs, voltages, ratings, testing requirements, utility standards, skilled labor, cores, windings, insulation, bushings, and other specialized components all constrain throughput.

AI data centers intensify this problem because they add large, concentrated loads rather than evenly distributed incremental electricity demand. The utility may need transmission upgrades and large power transformers before the site can be energized. The campus then needs its own medium voltage and dry-type transformers, switchgear, breakers, backup power architecture, distribution equipment, and cooling systems. The useful investment question is therefore not simply who makes transformers. It is which layer of this electrical stack remains difficult to expand.

Exhibit 1

The electrical path from grid connection to usable compute

A data center can be delayed by constraints at several different equipment layers.

Embed this
  1. 01
    Grid connection
    Transmission and utility interconnection
  2. 02
    Transformation
    Large, medium voltage, and dry-type transformers
  3. 03
    Switching and protection
    Switchgear, breakers, and related equipment
  4. 04
    Campus distribution
    Cable, busway, connectors, and distribution hardware
  5. 05
    Compute and cooling
    Power delivery and thermal infrastructure
Structural diagram based on the equipment sequence described in this article. It does not imply that every project uses an identical configuration. Source: Altsets

The bottleneck is a stack

Exhibit 2

The electrical infrastructure stack exposed to the data center buildout

The listed companies occupy different equipment layers and should not be treated as interchangeable exposures.

Embed this
LayerCompanies with relevant exposureWhy the layer matters
Large and medium power transformersHitachi Energy, GE Vernova through Prolec GE, Siemens Energy, EatonChanges voltage across transmission, utility interconnection and campus power systems. Long manufacturing cycles and specialized designs make this one of the hardest capacity layers to expand.
Dry-type and three-phase transformersEaton, Hitachi Energy, Hammond Power SolutionsImportant inside commercial and industrial facilities, including data centers. Capacity is also being expanded rapidly in the US.
Transformer cores and electrical steelCleveland-Cliffs and foreign electrical steel producersGrain oriented electrical steel is a specialized core material for power and distribution transformers. US production is unusually concentrated.
Transformer componentsHitachi Energy, Hubbell and specialized private suppliersBushings, insulation, conductors, tap changers, connectors and related components can constrain finished equipment even when assembly capacity is available.
Copper and electrical cablePrysmian through Encore Wire, Southwire and other wire manufacturersCopper conductors are required throughout the power path, including transformer windings, cables and electrical distribution. Fabrication capacity matters in addition to raw metal supply.
Switchgear and breakersEaton, GE Vernova, Siemens Energy, ABB, Schneider ElectricControls, isolates and protects increasingly large electrical loads. Manufacturers are adding capacity alongside transformer expansions.
Grid connection hardwareHubbell, Hitachi Energy, GE Vernova and other utility suppliersConnectors, bushings, arresters, insulators and substation equipment connect new load to the surrounding transmission and distribution network.
Data center coolingVertiv, Modine, Schneider Electric and othersNot part of the transformer itself, but a separate infrastructure requirement that grows as rack power density increases.
Displayed companies are examples discussed in the article, not a complete supply chain. Source: Altsets

The most important upstream material distinction is electrical steel. Large power transformers require high performance magnetic cores, and DOE has specifically identified additional grain oriented electrical steel, or GOES, capacity as a supply chain priority. Cleveland-Cliffs says its Butler Works in Pennsylvania is the only US producer of regular GOES and the only domestic producer of its high permeability TRAN-COR electrical steel, both of which are used in power and distribution transformers.

That concentration makes GOES analytically different from ordinary steel demand. A transformer manufacturing boom can increase demand for a product made on specialized equipment to specific magnetic performance requirements. It also means that expanding domestic transformer assembly does not automatically eliminate upstream supply risk. DOE's June 2026 review again described electrical core steel and distribution transformer manufacturing capacity as national supply chain constraints.

Copper is different. It is economically important to transformers and the wider data center electrical system, but the exposure spreads across mining, refining, winding conductors, cable and fabricated electrical products. DOE has identified expanded access to copper and greater capacity to turn it into transformer windings and continuously transposed conductor as an investment need for the large power transformer supply chain. That does not establish a direct supplier relationship between a particular copper company and a particular transformer OEM, but it shows why transformer capacity cannot be analyzed independently of conductor manufacturing.

Prysmian offers one public example on the fabrication side. It acquired Encore Wire in 2024 and is investing $500 million over five years at Encore's McKinney, Texas campus, including additional medium voltage cable capacity. Prysmian explicitly ties the expansion to data centers, industrial electrification and grid investment. The exposure is broader than transformer windings, but that is precisely the point: a data center draws conductor demand throughout the electrical system rather than at one piece of equipment.

Components can form narrower constraints. Hitachi Energy's $106 million Alamo, Tennessee project is increasing production of transformer bushings, including products for high voltage AC and DC applications. The company expects the site to become the largest bushing manufacturing facility in North America once the expansion is complete. Hubbell also manufactures utility bushings, connectors, insulators and arresters used throughout transmission and distribution systems. These businesses sit several layers away from a GPU purchase, but their products are necessary to connect and protect high voltage equipment.

Transformer capacity is expanding, but so is the rest of the electrical stack

Hitachi is not expanding in isolation. Eaton announced a $340 million investment in a new South Carolina plant for three-phase transformers, with production expected to begin in 2027. In April 2026, Eaton also announced a new Nebraska factory for medium voltage switchgear, explicitly citing AI data center demand. Its second quarter results showed Electrical sector backlog 43% above the prior year, while Electrical Americas organic sales increased 18%.

GE Vernova has moved more aggressively into transformers through acquisition. In February 2026 it completed the $5.275 billion purchase of the remaining 50% of Prolec GE, bringing a major North American transformer and transformer component manufacturer fully inside its Electrification segment. Prolec has seven manufacturing sites across the Americas, including five in the United States. GE Vernova reported more than $5 billion of data center orders in Electrification during the first half of 2026, more than double its total for all of 2025.

The significance of that acquisition is greater than adding another AI label to GE Vernova. Transformers now sit alongside high voltage switchgear, capacitors, substations and other grid products within a broader electrification portfolio. GE Vernova is also expanding its Charleroi, Pennsylvania operation, which produces high voltage circuit breakers, switchgear and instrument transformers. The company is effectively building exposure across several pieces of the same grid connection problem.

Siemens Energy is following a similar pattern. It is expanding large power transformer production in Charlotte, North Carolina and announced a $1 billion US manufacturing program in 2026 that includes additional transformer output and a new high voltage switchgear factory in Mississippi. The expansion shows why transformer demand and switchgear demand should be analyzed together. Increasing the number of transformers without increasing the equipment needed to control and protect those circuits only moves the constraint.

The smaller Hammond Power Solutions provides a more concentrated view of one transformer segment. Hammond manufactures dry-type transformers rather than the full utility transmission stack. In the second quarter of 2026, it reported sales growth of 44.7% from the prior year and backlog growth of 96.9%, while specifically identifying US data center investment, electrification and power infrastructure spending as demand drivers. The distinction matters because dry-type transformer exposure is closer to industrial and facility-level power distribution than a large transmission transformer installed by a utility.

Switchgear may be the clearest evidence that the bottleneck is spreading. ABB has been adding US production of busway, circuit breakers and other low voltage electrical products for data center and utility customers. Schneider Electric committed more than $700 million to US operations through 2027 after earlier building additional Tennessee capacity for custom switchgear and medium voltage distribution equipment. The industry's response is becoming broad enough that transformer factories should be viewed as one part of a synchronized electrical manufacturing expansion.

Cooling sits beside this power chain rather than inside it, but it creates another investable capacity cycle. Vertiv announced multiple Americas manufacturing expansions in March 2026 and a roughly $50 million Ohio investment to increase liquid cooling and chilled water system capacity. Modine has been expanding its Airedale data center cooling manufacturing footprint and in May 2026 disclosed a capacity agreement covering more than $4 billion of cooling products for one strategic data center customer between 2027 and 2029. Schneider Electric has also signed large supply agreements that combine electrical distribution equipment and cooling.

Cooling therefore belongs in the data center infrastructure map, but not because it is another transformer component. Its demand mechanism is different. Higher compute density increases the amount of heat that must be removed from a given amount of floor space, while transformer and switchgear demand is driven by the power that must be delivered and controlled. Both can benefit from the same data center project, but their capacity constraints, competitive structures and product cycles are different.

The investment read-through is about scarcity, not simply exposure

The electrical infrastructure buildout can be separated into three types of exposure.

Exposure type 01

Long-cycle equipment scarcity

Large power transformers, some medium voltage equipment and high voltage switchgear require specialized plants, engineering, testing and skilled labor. The simultaneous capital spending by Hitachi Energy, GE Vernova, Eaton and Siemens Energy suggests manufacturers expect elevated demand to persist long enough to justify substantial fixed investment. This is the portion of the theme where order backlogs and new factory announcements provide the strongest evidence of a structural capacity response rather than a temporary increase in component purchases.

Exposure type 02

Specialized upstream capacity

Cleveland-Cliffs' position in domestic GOES and Hitachi's investment in transformer bushings show that finished transformer output depends on narrower manufacturing processes upstream. These inputs can matter disproportionately even though their dollar value is only part of a completed transformer. Investors analyzing new transformer capacity should therefore distinguish between announced assembly capacity and the supply chain needed to run that capacity at high utilization.

Exposure type 03

Volume exposure across the wider campus

Wire and cable, busway, breakers, power distribution systems, connectors and cooling all scale with data center construction. These markets can produce strong growth without necessarily carrying the same scarcity economics as large power transformers. Prysmian, Hubbell, ABB, Schneider Electric, Vertiv and Modine belong in the AI infrastructure discussion for this reason, but the mechanism is increased electrical and thermal equipment volume rather than the same transformer shortage.

That distinction also changes how backlog should be interpreted. A transformer order can be placed years before equipment enters service because procurement has to occur early in a project's schedule. A cooling order is generally closer to the construction and fit-out of the actual data hall. Strong orders in both categories support the same broad data center capital cycle, but they reveal different points in that cycle.

There is also an important diversification effect. Grid equipment manufacturers are not dependent only on AI. Utilities are replacing aging assets, transmission is expanding, industrial electrification is increasing load, and generation projects require many of the same transformers, substations and switchgear. AI is accelerating a market that already had other demand drivers. That can make the electrical equipment cycle less dependent on any single generation of servers than the most direct semiconductor exposures.

What could weaken the bottleneck thesis

Capacity catches up

The biggest risk to the scarcity argument is the industry's own investment response. Hitachi Energy, Eaton, Siemens Energy, GE Vernova, Schneider Electric, ABB, Vertiv, Modine and Prysmian are all adding manufacturing capacity. Much of that output begins ramping between 2026 and 2029. If demand growth slows while those plants come online, lead times could fall and the pricing power associated with scarce capacity could normalize even if absolute shipment volumes remain high.

New capacity clears different constraints at different speeds

Not all capacity additions are interchangeable, however. A new dry-type transformer plant cannot automatically substitute for a large transmission transformer factory. More cable output does not solve a shortage of transformer cores. More transformer assembly space does not solve every shortage of bushings, windings or qualified electrical steel. The relevant question over the next several years is therefore which individual constraints clear first.

Project schedules move

Project timing is another risk. A proposed data center can create equipment reservations before the campus is ultimately completed. Generation availability, transmission interconnection, permitting, financing and customer demand can all alter construction schedules. Order books should therefore be analyzed alongside cancellations, deposits, delivery schedules and manufacturing utilization rather than treated as guaranteed future economics.

Corporate exposure can be diluted

Finally, exposure can be diluted inside diversified companies. Hitachi Energy is part of Hitachi rather than a separately traded pure play. Cleveland-Cliffs has substantial steel operations outside electrical steel. Eaton, Schneider Electric, Siemens Energy, ABB and GE Vernova serve many markets beyond data centers. A company can occupy an important physical position in the supply chain without having the same position in its consolidated income statement.

Conclusion

The AI data center transformer shortage is best understood as a layered electrical infrastructure constraint.

Hitachi Energy's latest US investments show that large transformer capacity remains valuable enough to justify factories costing hundreds of millions of dollars. But the more important conclusion is what sits around those factories. Transformer production requires specialized electrical steel, windings, bushings, insulation and other components. Connecting the resulting equipment requires switchgear, breakers, cable, substations, connectors and grid hardware. Once electricity reaches the data center, another capital cycle begins in power distribution and cooling.

That creates a broader investment universe than the transformer manufacturers alone, but it also argues against treating every electrical supplier as equivalent. Hitachi Energy, GE Vernova, Eaton, Siemens Energy and Hammond Power Solutions have direct transformer manufacturing exposure. Cleveland-Cliffs occupies a concentrated domestic position in transformer core steel. Prysmian participates in the conductor and cable buildout. Hubbell supplies utility connection hardware. ABB and Schneider Electric add switchgear and power distribution exposure. Vertiv and Modine sit on the thermal side of the same data center construction cycle.

The central finding is that AI infrastructure is increasingly constrained by the equipment required to move electricity, not merely by the equipment that consumes it. As new US factories begin production, the investment question will shift from whether electrical equipment is scarce to which layer of the stack remains scarce after the current capacity expansion is absorbed.

How to Cite This

According to Altsets Supply Chain Intelligence (altsets.com), AI infrastructure is increasingly constrained by the equipment required to move electricity, not merely by the equipment that consumes it.

For research inquiries or data access: press@altsets.com

Go Deeper

Sources

Methodology

Read the methodology for this research.