Last week we covered ABF substrates. The week before, power semiconductors. This week the bottleneck moved beneath the transformer.
Most AI supply-chain bottlenecks can eventually be attacked with capital. Capacity gets announced, tools get ordered, suppliers expand, and the constraint either eases or rotates somewhere else.
Power does not move on that timeline.
Part of the reason is that expanding grid capacity itself takes years with new generation, new transmission, land deals, permitting, regulation. None of that compresses no matter how much capital you throw at it.
The second problem is less obvious. Even when the power exists, moving it from the grid into a data center requires physical equipment. That equipment is constrained too.
Power transformers: the gate between the grid and the load
The equipment in question is the power transformer and within that, specifically the large power transformer (LPT).
The smaller grey and green cylinders that you see on the side of the road are distribution transformers, which step voltage down for residential and small-commercial use. They are important, and they're constrained too, but they aren't the main piece that gates a data center power transmission.
It is usually the LPT, which is a custom-built, 100-plus-ton machine that steps voltage up at generation plants and back down at substations. Each one is engineered for the specific site it's going to. None are off the shelf, and none ship from inventory.
Without one connecting a substation to a campus, the megawatts a hyperscaler signed up for don't actually arrive. That makes the LPT the gate.
The next question is what's happening at the gate right now.
A Large Power Transformer (Source: tteusa.com)
What the order books are saying
The transformer manufacturers are sitting on the strongest order books in decades, and they're falling behind anyway. The recent earnings prints show both sides:
Demand at the OEMs is at a generational high: GE Vernova’s backlog hit $163B in Q1 2026, and its Electrification segment booked $2.4B of data center equipment orders in the recent quarter, which is more than it booked from data centers in all of last year.
Pricing on new orders is repricing higher before it converts to revenue: Transformer buyers are pre-buying production slots, accepting longer lead times and paying premiums to secure delivery. Prices have risen about 80% over five years, while some developers are willing to pay premiums or buy future slots before a specific project is finalized.
Capacity is being added, but slowly and far out: GE Vernova, Hitachi, Siemens, Eaton, Hyundai and others are all expanding, but most of the meaningful new transformer capacity only starts showing up in 2027 and 2028. That is too late for projects trying to get energized now.
Lead times keep stretching anyway: Average transformer lead times have gone from ~50 weeks in 2021 to well over 140 weeks today for substation-class LPTs.
The shortage is now project-defining: Roughly half of US data center builds planned for 2026 are expected to slip or get cancelled because of energisation challenges.
From GE Vernova’s Earnings Analysis on Tessara
The 2nd order transformer constraint: GOES
With capex flowing and order books at record highs, the question is what's actually keeping the makers from catching up with the demand.
One of the hardest-to-replicate upstream constraints is the steel they have to buy.
That steel is grain-oriented electrical steel, or GOES. It's the material every LPT's magnetic core is built from.
GOES is a high-silicon steel whose crystal grains have been aligned in one direction during rolling, so it loses very little energy to heat when alternating current cycles through it. That property is what makes a transformer efficient and at AI data center-scale loads, efficiency translates directly into how much steel you need per megawatt, how hot the unit runs, and how long it lasts.
There is no easy substitute. Amorphous metal works in some distribution transformers, but it is not a drop-in replacement for high-voltage LPTs. For the largest units, transformer makers still build around qualified GOES.
The problem is that GOES is not normal steel. Making it requires specialized rolling, heat treatment, coating, tight process control, and years of production know-how. A new line is a multi-hundred-million-dollar project, and qualifying the material for high-voltage transformer use takes years. You cannot buy that capability off the shelf.
That is why the supplier base is so narrow. The qualified high-grade market is dominated by a small group of steelmakers, creating an oligopoly:
Baowu in China
Nippon Steel in Japan
POSCO in Korea
Thyssenkrupp in Germany
Cleveland-Cliffs in the US.
This global supply chain worked when transformer demand was steadier. The US could rely on imported LPTs, and those imported LPTs carried foreign GOES inside them. But the US is now one of the largest and fastest-tightening LPT markets in the world, while still importing more than 80% of the large power transformers it uses.
That is the tension. The demand shock is increasingly American. The manufacturing base is still mostly foreign. And the most important steel inside the transformer is controlled by a small group of qualified suppliers.
Which brings the story to the only domestic GOES producer in the US.
Cleveland-Cliffs and the US monopoly
Inside the US, the picture narrows to one name: Cleveland-Cliffs (CLF).
Cliffs is the only domestic producer of GOES, with its electrical steel position centered around Butler Works in Pennsylvania. There is no real number two. For years, that was a niche fact inside a much larger steel company. Now it is becoming strategically relevant.
The reason is simple: the US is trying to localize the equipment needed to power data centers, factories, renewables, and the grid itself. That makes the steel inside that equipment matter more.
Three things are moving in Cliffs’ direction.
First, transformer capacity is moving closer to home: The AI data center buildout is heavily US-weighted, and several transformer OEMs are adding US plants on 2027–2028 timelines. Those plants will need qualified electrical steel. Imported GOES will still be part of the market, but the only domestic supplier already sitting in position is Cliffs.
Second, tariffs now protect the domestic GOES position: Imported steel faces a 50% tariff, and in 2025 that protection was extended to electrical steel cores and laminations, which are the transformer-core parts made from GOES. The US still needs imports, but the tariff wall now sits directly around the material Cliffs makes, making domestic GOES more strategically valuable.
Third, the government is treating the supply chain as critical infrastructure: The Defense Production Act gives the federal government tools like direct purchases, loans, and loan guarantees to support industries deemed critical to national defense. In April, the White House added transformers and their supply chain to that list, explicitly naming electrical core steel alongside them. For Cliffs, that puts federal funding and procurement support behind the only US source of GOES.
Cliffs will certainly be using this window to improve its electrical steel position.
But this is not a current earnings story yet. Electrical steel is still a small slice of the company today. In Q1 2026, stainless and electrical combined were just 3% of product sales volumes, while the rest of the business was still mostly flat-rolled steel tied to autos and broader industrial demand.
That is why the market still prices Cliffs like a steel company, not like a potential AI power bottleneck company. And based on today’s revenue mix, that makes sense.
CLF is not a pure-play AI power stock. It is a steel stock with a mispriced domestic electrical-steel option
Who captures the rent
Pays rent: the AI hyperscalers and the utilities and developers building for them. Microsoft, Google, Amazon, Meta, OpenAI, plus the merchant power developers and colos sitting between them, all of whom need transformers to actually deliver the megawatts they've contracted.
Captures rent: the transformer OEMs (GE Vernova, Hitachi Energy, Siemens Energy, Eaton) at the equipment layer, and increasingly the GOES producers beneath them. Baowu, Nippon Steel, POSCO, Thyssenkrupp, and Cleveland-Cliffs.
What to watch
CLF Q2 and Q3 earnings. Watch for any uptick in the stainless and electrical steel volume share above 3%, management commentary on new customer interest or order pipeline for the GOES line.
Transformer OEM earnings. GE Vernova, Hitachi Energy, Siemens Energy, Eaton. Track US capacity buildout progress, any input cost or steel-sourcing commentary, and lead-time language.
The Week Ahead
A few important earnings this week. Read our post and pre-call briefs here and stay prepared.
This week’s Earnings Calendar from Tessara
Wednesday, June 3
AVGO (Broadcom) — Networking. Watch: If AI semiconductor revenue exceeds $10.7B guided level.
Thursday, June 4
CIEN (Ciena) — Optical Components. Watch: Cloud provider order mix and pluggable ramp
Want the live data behind The Chokepoint? Tessara is the live supply-chain map of the AI build, for investors. We track what's binding in the supply chain and what it means for what you own. 300+ companies across compute, memory, foundry, networking, and power.
See you next week,
Teng & Arvind
This article is for informational and research purposes only. It is not financial advice, investment advice, or a recommendation to buy or sell any security. Tessara Research does not publish price targets. The views expressed here reflect our analysis at the time of publication and may change as new evidence arrives. Readers should do their own research and consult a qualified financial adviser before making investment decisions.