Breaking

Transformer Delivery Delays Constrain US AI Data Center Buildout

July 9, 2026 at 00:24 EDT

Infra & Chips

The supply squeeze on electrical transformers, a device commercialized roughly a century ago, has become the biggest bottleneck for AI data center construction. With transformer lead times stretching into years, many planned facilities face the risk of delay or cancellation.

AI Infrastructure · The Grid Bottleneck

The AI Boom Runs on a 120-Year-Old Device Nobody Can Build Fast Enough

Chips flow freely from NVIDIA — but the transformers that connect data centers to the grid now carry lead times of years. The real constraint on AI isn't silicon. It's electrons.

143 wks

Avg. lead time — generation step-up transformers (Q2 2025)

128 wks

Avg. lead time — power transformers

+274%

Step-up transformer demand growth, 2019 → 2025

<10%

Electrical gear's share of total data-center cost

Filling the gap with imports

Chinese high-voltage transformer imports to the US, annual units

<1,500

2022

8,000+

by Oct 2025

A more than 5× jump in three years. Chinese-made units now supply roughly 30% of the US market for certain categories.

Under 10% of the cost. A 100% bottleneck on the schedule.

Of the 12 GW of US data-center projects planned for 2026, the numbers reveal how much hangs on gear that barely registers on the budget.

5 of 12 GW

Actually under construction

30–50%

Exposed to delay or cancellation

1 in 4

No disclosed power-procurement plan

Possible Fix — Solid-State Transformers

Semiconductor-based next-gen units could ease reliance on the legacy supply chain and expand alternative sourcing options.

But — The Catch

Costs run 2×–3.5× conventional types with a limited reliability track record — leaving critics to warn that many announced projects are "vapor capex" living only in press releases.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.