Daily
AI Infrastructure Daily: everything that happened across chips, memory, power, data centers and financing in a 3-minute read, every weekday. Free. Read today's →
NVDA$166.2−4.34%TSM$405.5−1.90%SK hynix₩1,254,000−1.80%Samsung Electronics₩208,000+0.97%MU$798.4−1.59%ASML$1,497.1−2.43%AVGO$364.4+5.54%LITE$624.6−3.61%COHR$232.1−1.04%ANET$206.6+5.06%GEV$818.5+1.27%ETN$359.0−0.44%VRT$209.6+1.52%CEG$282.4+2.79%VST$134.5+0.79%PWR$606.2−0.29%HD Hyundai Electric₩421,000+3.95%MP$37.45+1.96%FCX$60.45−2.69%CCJ$71.78−3.42%MSFT$523.0+0.36%META$623.6−0.82%CRWD$229.9+0.35%SMH$501.7−1.03%SPY$757.1−0.35%
Home/Blog/Power
Saturday 31 October 2026 · 2 min read · Power · Bottlenecks

Why power is the AI bottleneck: 8 facts on turbines, transformers and the grid

Chips can be made in a year or two. Power plants, transformers and transmission lines take much longer. That is why the slowest part of the AI build-out is electricity.

1Gas turbines are booked years out

Our tightest bottleneck. Large turbine makers have order slots stretching toward the end of the decade.

2Transformer prices have roughly doubled since 2020

US producer prices for power transformers show demand still outrunning supply.

3Grid connections can take three to seven years

Longer than building the data center itself.

4US electricity demand is rising again

After about 15 flat years, power output is climbing, largely because of data centers.

5Many campuses now build their own power

Fuel cells, gas engines and turbines on site, as a bridge until the grid catches up.

6Nuclear is back in demand

Hyperscalers are signing deals for existing plants and future small reactors, and uranium prices reflect it.

7Power construction spending keeps climbing

Spending on power plants and transmission is at record levels.

8Equipment makers have the pricing power

Companies with multi-year backlogs (turbines, transformers, switchgear, cooling) benefit most; developers that still need funding carry the most risk.

Where the data comes from

Our alternative data page charts US power construction spending, producer prices for transformers and electrical equipment, US power output and electricity prices, all from public US government data with full history. The data center map shows how much capacity is waiting for power.

What could ease the bottleneck

New turbine and transformer factories, faster grid permitting, and more efficient chips would all help, but each takes years. In the meantime, watch equipment makers' backlogs and prices: the day they stop rising is the day the shortage starts to ease.

Our Power & Cooling section scores every part of the power chain.

Explore Power & Cooling →

Frequently asked questions

Why is power a bottleneck for AI?

AI data centers need far more electricity than traditional ones, and the equipment to generate and deliver it has lead times of two to seven years.

Which stocks benefit from the power bottleneck?

Turbine, transformer, switchgear and cooling makers, grid contractors and power producers. Our Power & Cooling section lists them.

How long does it take to connect a data center to the grid?

Often three to seven years in busy markets, which is why many projects add on-site generation.

More guides

Figures quoted are as of the publish date; the linked pages show live data. Educational research by Singularity Research, not investment advice.