The AI Boom Has a Power Problem

Most people who got rich in the Gold Rush never mined a day in their lives.

The prospectors chased the shiny stuff, and most of them went broke.

The ones who built real fortunes sold the picks, the shovels, the pans, and the blue jeans.

They didn’t bet on which prospector got lucky. They sold to every prospector who showed up, lucky or not.

I think about that every time a new mania takes over the market, and right now the mania is artificial intelligence.

Everybody is piling into the chip stocks, betting on which company builds the smartest model.

That’s the gold. I’d rather sell the shovels.

So what is the shovel this time? It is plain old electricity.

Almost nobody chasing AI has stopped to ask the obvious question. What runs all of it?

These AI models don’t float in the cloud. They live in enormous data centers packed with chips, and those chips are pigs for power.

The demand is staggering. Government estimates say data center power use could nearly triple by 2028.

Essentially, the whole AI story is running headfirst into a wall the market has barely priced in: we don’t have enough power to feed it.

For years the easy answer was wind and solar.

Green doesn’t work here, and the reason is simple. A data center runs 24 hours a day, 365 days a year, and it can never blink.

The wind stops and the sun sets. Renewables tap out at around 80% of what one of these facilities needs around the clock, and a single hour of downtime at a hyperscale center can cost millions.

You simply can’t run the most valuable machines on earth on power that quits when a cloud rolls in.

But there is one source that checks every box: Nuclear.

A nuclear plant runs at full power more than 93% of the time, goes 18 to 24 months without refueling, produces zero operating emissions, and packs enormous output into a small footprint.

It’s the only clean source that never sleeps. And the smartest, richest companies in the world have already figured this out.

Look at where their money is going, because this is the part that matters. Microsoft (MSFT) cut a deal to restart a reactor at Three Mile Island, the most infamous nuclear site in America.

They’re so excited about it, they’re calling it a “once-in-a-lifetime opportunity.”

Amazon (AMZN) signed a 17-year deal to pull nearly 2 gigawatts straight from a Pennsylvania nuclear plant.

Google, Meta, and Amazon are all funding a new generation of smaller reactors built in a factory and dropped on site.

These are not press releases and pledges.

These are binding, decades-long contracts signed by the most sophisticated buyers on the planet, and they are spending billions.

When companies this smart all sprint in the same direction at the same time, it pays to notice what they’re running toward.

The government sees it too. The Department of Energy put up a $17.5 billion loan program to speed up construction on ten new large reactors, aimed squarely at the shortage the tech giants are staring down.

None of this is a straight line, and I won’t pretend otherwise. Licensing a reactor through the regulators still takes years, and the smaller reactor startups have to raise serious money before the federal loans kick in.

This is a multi-year story, not a next-week trade.

But here’s what the doubters keep missing. The demand is already locked in with signatures, the buyers have the deepest pockets in the market, and the government is subsidizing the buildout.

That’s about as clear a runway as this business ever gives you.

Logo

YOUR ACTION PLAN

The AI race will mint winners and losers among the chipmakers, and picking the right one is close to a coin flip. The one thing every single one of them needs, no matter who wins, is power.

That’s the shovel, and that’s where I want to be.

The big winners in this space aren’t the obvious mega-cap names everyone already owns.

They’re further down the chain, in the companies that build and run these reactors, and one of them sits at the top of my list right now.

Popular posts