The real constraint on AI infrastructure isn't money, chips, or political will,it's a specialized steel produced by just two American factories in small towns most people have never heard of. Grain-oriented electrical steel (GOES) is the core material inside every large power transformer on the planet, and the United States depends entirely on Cleveland-Cliffs' mills in Butler, Pennsylvania, and Zanesville, Ohio, to supply it. With lead times for large power transformers stretching from 30 weeks before the pandemic to 128 weeks today, and some manufacturers quoting delivery dates in 2031, the AI infrastructure buildout faces a bottleneck that no amount of venture capital can solve.

Why Is Grain-Oriented Electrical Steel So Critical?

GOES is a highly specialized material whose magnetic grains are all aligned in the same direction, allowing electricity to flow through it with less than 1% energy loss. Regular steel would waste 10 to 20% of the energy passing through it as heat. Every large power transformer on the planet, those 400-ton machines that step grid voltage up and down between power plants and data centers, has a core made of GOES. There is no substitute, and you cannot build a transformer without it.

A single hyperscale data center, the massive facilities operated by Amazon Web Services, Microsoft Azure, Google Cloud, and Meta, needs between 200 and 400 megawatts of power. To put that in perspective, one megawatt is enough to power roughly 800 U.S. homes, so a single AI campus consumes as much electricity as a mid-sized town. Every megawatt requires approximately one ton of GOES steel embedded in its transformers, switchgear, and substations.

What Does the Math Actually Show About Supply and Demand?

The numbers reveal a crisis in the making. The United States plans to bring online 100 gigawatts of new data center capacity by 2030, which equals 100,000 megawatts. That translates to roughly 100,000 tons of GOES steel needed just for data centers. Cleveland-Cliffs' combined Butler-Zanesville output is under 200,000 tons per year, shared across transformers, motors, electric vehicles, grid replacement, and every other application. The math is stark: demand for data center infrastructure alone would consume half of the nation's total GOES production capacity.

Two data centers in Silicon Valley finished construction last year and still cannot process a single query, not because they lack chips or software, but because the transformers that would connect them to the grid have not been delivered yet. This is not a theoretical problem. It is happening now.

How to Understand the Broader Supply Chain Crisis

• Transformer Lead Times: Large power transformers now face 128-week wait times, up from 30 weeks before the pandemic, with some manufacturers quoting delivery in 2031 for new orders placed today.
• Generator Step-Up Units: High-voltage transformers installed at power plants face 144-week lead times, and in extreme cases, 4 to 5 years, creating bottlenecks at the source of power generation.
• Slot Reservation Deposits: When demand far exceeds production capacity, manufacturers ask customers to pay upfront deposits of 20 to 25% just to reserve a place in line for future production slots, signaling how tight supply really is.

The problem extends beyond transformers. Gas turbines, which power data centers and the broader grid, face similar constraints. GE Vernova, spun off from General Electric in April 2024, is sold out through 2029, with anyone ordering a new heavy-duty gas turbine today facing commissioning dates stretching to 2031. The company's slot reservation backlog jumped from 29 gigawatts to 83 gigawatts in a single year.

Why Did Siemens Energy's Stock Surge 10x in Two Years?

The clearest signal that the world has changed comes from Siemens Energy, the German industrial equipment manufacturer. In October 2023, CEO Christian Bruch had to publicly announce emergency talks with the German government to secure up to 15 billion euros in state guarantees to keep the business afloat. The stock fell 40% in a single day, with 3 billion euros in market cap vaporized by lunch. The company's wind business, Siemens Gamesa, had booked 4.4 billion euros in losses that year alone due to quality problems with onshore turbines and offshore ramp-up disasters.

By early 2024, the stock was trading around 14 euros. Today, Siemens Energy trades at 157 euros, more than a 10x increase from the lows. The company just reported a 146 billion euro order backlog, the largest in the history of European industrial equipment, with record quarterly revenue and grid technology revenue from hyperscalers more than doubling year-over-year, exceeding 2 billion euros in a single quarter. Siemens Energy sold 194 large gas turbines in Q1 FY2026, nearly double the prior year, with net income tripled.

"If the past five years have been about building the foundation, then fiscal year 2025 was the start of a growth journey with continuous margin expansion," stated Christian Bruch, CEO of Siemens Energy.

Christian Bruch, CEO at Siemens Energy

In plain English, Siemens Energy survived its near-death experience and now owns the world's grid bottleneck. A CEO begging for government survival in 2023, whose stock was left for dead, is now sitting on 146 billion euros of order backlog with visibility through 2030 and pricing power that no industrial company has enjoyed since the oil majors in 1974. The company repaid its government guarantees ahead of schedule and just authorized a 2 billion euro share buyback, part of a broader 6 billion euro capital return program running through 2028.

This transformation reveals something fundamental about market structure. Transformers are not an isolated case; they are a symptom of a much larger reality. The entire supply chain for AI infrastructure is constrained by physical manufacturing capacity, not by capital availability or political will. The 650 billion dollars in AI infrastructure capex announced for 2026 by Alphabet, Amazon, Meta, and Microsoft combined cannot be deployed faster than the factories can produce the equipment needed to connect those data centers to the grid. Until new manufacturing capacity comes online, and given that the lead time on a new transformer factory is four years, the AI buildout will be limited not by innovation or investment, but by the production capacity of a rolling mill in western Pennsylvania and the global supply chains that feed it.