AI Needed a Power Grid. So It Built One.

Somewhere in Mississippi, nearly fifty gas turbines are running. Not at a utility. Not at an energy company. At a data center — xAI's Colossus 2 facility, which, according to reporting from TechCrunch, has drawn a lawsuit over its use of so-called "mobile" gas turbines as what amount to permanent power plants. The word "mobile" is doing a lot of work in that sentence. It implies temporary. Provisional. A stopgap while something better gets sorted. What it actually describes, apparently, is the permanent infrastructure of one of the most talked-about AI projects on earth.

Around the same time that story broke, a geothermal startup called Fervo Energy went public and popped 33% on its debut, per TechCrunch. Investors had asked, before the IPO, why Fervo wasn't raising more money. The IPO was upsized multiple times. The demand driver everyone cited: AI data centers.

Two stories. One about a company burning gas unchecked to keep its AI running. One about a company that went public because AI needs so much electricity that geothermal — geothermal, a technology that has spent decades being described as promising — suddenly looks like a growth stock. Read them together and you stop thinking about energy policy. You start thinking about something stranger: the tech industry has quietly become the energy industry, and it got there fast enough that nobody had time to file the right permits.

The Infrastructure Was Always Coming

This is how it goes with technology at scale. The thing that seems like software turns out to require hardware. The hardware turns out to require real estate. The real estate turns out to require power. And the power — this is where it always gets interesting — turns out to require choices that can't be abstracted away into a press release about sustainability goals.

For years, the major cloud providers treated their carbon commitments like a brand attribute. Renewable this, net-zero that. The language was ambient, inoffensive, designed to signal virtue without requiring specifics. What the xAI situation in Mississippi represents isn't a scandal so much as a candid photograph. This is what building at speed actually looks like when the grid can't keep up. You find turbines. You call them mobile. You plug them in and you run them, and you deal with the legal consequences later — if there are any.

There's no villain required for this story to be troubling. The economic logic writes itself. If your competitive position depends on having more compute than the next company, and the next company is moving fast, then waiting for permitted infrastructure is not a neutral choice. It's a decision to lose. The turbines are just the physical record of that calculation.

What Fervo Means

The Fervo IPO is the other side of the same coin, and in some ways it's more interesting. Enhanced geothermal has always been a compelling technology in theory — tapping into heat that exists miles underground, reliably, without the intermittency problems that plague solar and wind. The reason it never fully arrived wasn't technical skepticism. It was that nobody wanted to pay for it at scale.

Now someone does. Multiple someones, actually, and they are apparently so eager that they asked a geothermal company why it wasn't raising more money. That is a sentence I would not have expected to read about geothermal five years ago. The fact that it happened because AI data centers need reliable baseload power is either a hopeful sign or a deeply cynical one, depending on your mood. Probably both.

What it tells you is that the grid is now the bottleneck. Not the models. Not the chips. Not the talent. The electrons. The companies building the most powerful AI systems in the world have run into the oldest constraint in industrial history — you cannot run machines without energy — and they are responding the way industrial companies have always responded: by building their own supply chains. By becoming, in some functional sense, power companies.

A Cycle You Recognize If You've Watched Long Enough

I've watched enough technology hype cycles to know that the part everyone ignores is always the infrastructure. The apps get the coverage. The founders get the profiles. The thing underneath — the fiber, the server farms, the power contracts — gets treated as a detail until it becomes a crisis or a lawsuit or, occasionally, an IPO.

We're at the moment where AI's infrastructure is becoming visible. Not because anyone chose to reveal it, but because it got too large to hide. Fifty gas turbines in Mississippi cannot be soft-launched. A geothermal company going public cannot be kept quiet. The physical footprint of this technology is now legible in ways that lines of code and model weights are not.

What the two stories share, underneath the surface details, is an admission. AI at scale requires energy at scale — more than the existing grid was built to deliver, more than existing permitting frameworks were designed to accommodate, more than any renewable commitment was sized to cover. The companies building it know this. They've been operating around it for a while. The rest of us are just now getting the receipts.

The question worth sitting with isn't whether any of this was handled well. It clearly wasn't, in at least one case. The question is what you think it means that the most sophisticated software ever written turns out to need, at its foundation, something as old and blunt as fire.