The Data Center Bottleneck Moved to the Electrical Room

Most of the commentary on the data center buildout is still written as if the binding constraint were silicon. It isn't. For anyone actually sourcing to these projects, the constraint moved to the electrical room some time ago, and it has stayed there.

Three numbers frame the problem.

The power math

GE Vernova closed its second quarter of 2026 with a gas turbine backlog of 116 GW, up from 100 GW a quarter earlier, and is now taking reservations for 2031 delivery slots. Like Siemens Energy and Mitsubishi, it books four to five years out. Data center customers are roughly 20% of its customer base and are bidding against utilities for the same slots.

Interconnection is worse. More than 2.2 terawatts of generation and storage sat in U.S. interconnection queues as of March 2026. Average time from initial request to commercial operation had risen to nearly five years by 2024, against under two years in 2008. And the attrition is severe: of the projects that requested interconnection between 2000 and 2019, only 19% had reached commercial operation by the end of 2024.

Read those together and the implication is uncomfortable. A queue position is not a power supply. Four out of five projects that get in line never energize.

The equipment math

Power transformers are running about 128 weeks in 2026. Generator step-up units are closer to 144 weeks. Specialty units are being quoted at four years.

One distinction matters here and gets flattened constantly in the trade press: distribution transformers are not power transformers. Distribution units came back down to roughly 30 weeks by the second quarter of 2025. The crisis narrative you still see quoted is often two-year-old distribution data applied to large power equipment, where the situation is genuinely worse and has not recovered.

If your transformer is 128 weeks and your turbine slot is 2031, those two dates, not construction and not permitting, are your schedule.

What it looks like on a purchase order

I have been buying medium-voltage equipment into heavy industrial packages for years, and the published numbers match what actually shows up on a quote.

On large drive packages, motors in the 1,500 to 10,000 HP range, the transformer is reliably the long pole. At 4160 volts, oil-filled or dry-type, 52 weeks has been a normal quotation. The balance of that package, motors and drives included, comes in around 20 to 25 weeks.

That gap is the entire problem in miniature. The transformer is not marginally longer than the equipment around it. It is more than double, and it sets the date the line can be energized no matter how well everything else is managed. You can expedite the drives. You cannot expedite the transformer, and if you release it late you have bought yourself a six-month delay that no amount of attention downstream will recover.

Note what this is not. It is not a data center problem. It is a medium-voltage equipment problem that data centers are making worse by absorbing the available capacity. Anyone building anything that draws serious power, whether process plants, mining, recycling or steel, is standing in the same queue, competing against buyers with better funding and less price sensitivity.

The layer most people are missing

Here is where this stops being general infrastructure commentary and becomes a sourcing problem.

Proclamation 11021, effective 12:01 a.m. Eastern on April 6, 2026, changed the assessment basis for Section 232. Duty on covered metal articles and derivatives now applies, in the proclamation's own words, to "the full customs value of the imported product, regardless of metal content."

That is a different calculation, not a different rate, and the distinction is the whole point. A transformer or switchgear lineup that previously carried duty on its declared steel and copper content now carries duty on the entire entered value. Depending on the metal fraction, absolute duty can rise even where the headline percentage falls.

The structure now runs roughly: 50% on articles made entirely or almost entirely of steel, aluminum or copper; 25% on the broader derivative list; 10% on certain articles made from U.S.-melted and poured or smelted and cast metal; and 0% for articles outside Chapters 72 to 76 with less than 15% metal content by weight.

There is also a transitional tier, and it is worth understanding precisely because it is easy to misread. Annex III of the proclamation covers what the administration describes as metal-intensive industrial equipment and electrical grid equipment. For those articles, through December 31, 2027, the combined Column 1 MFN rate plus the Section 232 rate is capped at 15%. If the Column 1 rate already meets or exceeds 15%, no additional Section 232 duty applies. After that date, Annex III articles move to the 25% derivative rate.

Two cautions on this, both of which I would want confirmed before pricing anything. First, "electrical grid equipment" is the category description, not the legal scope. Annex III is a finite list of HTS codes, and published counts of that list vary. Whether your specific transformer, switchgear or e-house classification sits inside it is a question you answer by reading the annex against your code, not by reading the category name. Second, the 15% is a ceiling on a combination, not a Section 232 adder. Modeling it as a flat 15% on top of your landed cost will overstate your duty on some parts and understate it badly on others.

This sits on top of what was already in force: the August 2025 derivative expansion that pulled dry-type and control transformers into scope, and the 50% copper action effective August 1, 2025, which lands directly on windings.

Classification and valuation on this equipment stopped being clerical work some time ago. It is a material line in the project cost, and on a nine-figure electrical package the gap between a defensible position and a lazy one is real money.

The bottleneck nobody can procure around

The part I find most interesting is the part capital cannot fix.

About 95% of U.S. transformer cores use grain-oriented electrical steel, and there is a single domestic producer. Before the February 2025 tariff changes, roughly 85% of imported GOES came from South Korea. That concentration is a real exposure and it is the one most people name.

But steel is not actually the binding constraint. The U.S. transformer workforce is somewhere around 15,000 people, and meeting projected demand would require something close to tripling it. Coil winding, hand-placing insulated conductor, has resisted automation and takes years to train. The workforce is aging out of a vocational pipeline that stopped refilling a generation ago.

New GOES capacity is being built. It will not resolve this. You cannot issue a purchase order for coil winders.

What I would do if I were sourcing to these projects

Three things, and none of them are exotic.

Release the transformer first, before the design is finished. If the transformer is double the lead time of everything around it, it cannot wait for a complete package. Reserve the slot on preliminary ratings and absorb the change-order risk on the details later. That trade is almost always cheaper than the delay, and the projects that land on time are the ones that made it.

Run the classification against the annex before the PO, not at entry. The April basis change means duty exposure on electrical equipment has to be modeled at quote stage, and it has to be modeled off the actual HTS code rather than the category label. Origin, metal content by weight, and melt-and-pour or smelt-and-cast documentation are pricing inputs now. A supplier who cannot produce mill certs traceable to melt location is a supplier who will surprise you at the border, and the surprise arrives after the equipment is built.

Qualify fabrication capacity outside the transformer bottleneck. Not everything in the electrical scope is transformer-constrained. E-houses, power modules, skids, structural steel, enclosures and piping assemblies are fabrication work, and there is qualified capacity in North America and offshore that is not competing for coil winders. Moving that scope to available capacity protects the schedule items that genuinely have no substitute.

The buildout is real and the capital is committed. What is scarce is not money or intent. It is transformer slots, turbine slots, queue positions, and trained hands, and increasingly the ability to land equipment at a landed cost you predicted correctly.

Sources: Proclamation 11021, "Strengthening Actions Taken To Adjust Imports of Aluminum, Steel, and Copper Into the United States," Federal Register, April 9, 2026, and CBP guidance CSMS #68253075; GE Vernova Q2 2026 results as reported by Utility Dive; RMI analysis of FERC interconnection queue data, March 2026; Congressional Research Service report R48933 on electricity distribution transformers; Lawrence Berkeley National Laboratory interconnection queue completion data.

Mike Dates is the founder of Supply Chain Works LLC, an independent consulting practice focused on global sourcing, tariff analysis, HTS classification and import/export compliance.