Electricity.
I went down a rabbit hole looking for the “TSM of near-term electricity for data centers.”
There probably isn’t one.
And that’s actually the interesting part.
TSM became the obvious toll booth for AI because advanced chips have to pass through a relatively concentrated manufacturing bottleneck.
Electricity is different.
The bottleneck is scattered across gas turbines, transformers, generators, transmission, grid connections and companies that can provide power before the grid is ready.
The IEA says data center electricity use jumped 17% in 2025 and expects total data center power consumption to double by 2030.
AI-focused data centers could triple.
But here’s the part I think the market is starting to understand.
You can’t build a $10 billion data center and just plug it into the wall.
The grid may not be ready.
The IEA says supply chains for gas turbines and transformers have tightened, while grid connections and approvals are becoming major constraints.
That’s why companies are starting to look at behind-the-meter power and onsite generation.
And natural gas is suddenly looking a lot more interesting.
EIA says U.S. electricity demand has already accelerated after more than a decade of almost no growth.
From 2005 to 2019, electricity demand grew only about 0.1% per year.
From 2020 to 2025, it grew roughly 1.7% per year.
Now look at Texas.
EIA had previously expected Texas electricity load to grow 14% in 2027.
It just cut that forecast to 6% after Texas paused new data-center development and started reviewing projects.
And Reuters reported something even crazier this week.
Data-center power requests across the U.S. have exceeded 700 GW.
That’s more than 10 times the estimated power currently used by all U.S. data centers.
A lot of those requests may never become real projects.
Texas is now auditing them.
That’s the other side of this trade.
AI power demand is real.
But the amount of power being requested is getting ridiculous.
So I wouldn’t blindly chase every nuclear, utility, turbine or grid stock that has “AI electricity” attached to it.
Some of these stocks have already priced in years of demand.
I’d rather find the companies that can actually deliver power in the next 2–5 years.
Because if AI spending keeps exploding, electricity becomes a toll booth.
And if AI spending starts slowing?
That’s where it gets really interesting.
The AI bubble could cool down while electricity demand from projects already under construction keeps rising.
The hype can fade.
The power shortage doesn’t disappear overnight.
That’s a very different trade from simply buying another AI stock.
The question I’m trying to answer now isn’t:
“What’s the next Nvidia?”
It’s:
“Who gets paid when Nvidia needs electricity?”