Mentatcurated
Energy & Climate high · independent

The bottleneck is a transformer

About half of the AI data-center capacity meant to come online in the US this year is stalled — not for want of chips or money, but because the power transformers now take two and a half years to arrive.

The shortage everyone watched for was supposed to be silicon. Instead, of the roughly twelve gigawatts of US data-center capacity slated to switch on in 2026, only about a third is actually under construction; the rest is waiting on grid hardware. A large power transformer that took two years to order before 2020 can now take two and a half, and up to four or five for the biggest high-voltage units — longer than the entire data center is meant to take to build.

In the United States, exactly one company still makes the grain-oriented electrical steel that transformer cores are wound from: Cleveland-Cliffs, at a single mill in Pennsylvania.

Follow the wait down to its root and it ends at a single material. Transformer cores are wound from grain-oriented electrical steel, a fussy iron-silicon alloy tuned for low magnetic loss, and in the United States exactly one company still makes it: Cleveland-Cliffs, at one mill in Pennsylvania. The trillion-dollar buildout narrows to one specialty-steel line.

The four largest spenders plan to put more than $650 billion into AI capacity this year, and that money can buy GPUs faster than the grid can deliver the iron to power them. Capital and compute turn out to be the easy parts; the slow part is century-old electrical gear that no amount of cash can fast-forward, and which the rest of the country's electrification is queued behind too.

The lenses

Novelty 2
Impact · breadth 3
Impact · depth 4
Actionable 1
Substance 3
Hype 4

The facts

Planned 2026 capacity delayedAbout half; only ~1/3 under construction
Transformer lead time now~2.5 years, up to 4-5 for the largest units
US producers of the core steelOne (Cleveland-Cliffs)
Open bloomberg.com →

How this connects

Tap a node to open it