How to Feed a Data Centre 24/7: Drill Into 460-Degree Rock

The AI boom can buy chips, land and engineers by the truckload. The one thing it cannot buy enough of is clean power that never switches off. So a geothermal company drilled its hardest well yet, into rock hot enough to cook it, and named Utah as the home of the first data centres built to run on heat mined from deep underground.

Solar stops at dusk. Wind drops when the air goes still. A hall of AI chips does neither. It runs flat out at three in the morning on a windless night, and it wants power that will do the same, for years.

The answer that is finally starting to work is not in a laboratory. It is a few miles under the Utah desert, in rock hot enough to boil away a careless drill. And the company chasing it just set a record that says the real barrier was never the heat. It was the digging.

Ordinary geothermal power is simple and rare. In a few lucky spots, like Iceland, nature left hot water in cracked rock near the surface. You drill in, steam rises, it spins a turbine. The trouble is that these ready-made underground kettles sit almost nowhere useful.

Enhanced geothermal, often shortened to EGS, drops the need for a natural kettle. Go deep enough and the rock is hot everywhere, not just in Iceland. So you drill down into hot, dry stone, pump in water to open a web of thin cracks, then send water down one well and pull it back up another. It soaks up the heat and returns as steam. In plain terms, you build your own boiler where nature never left one.

That last part is the whole prize. It turns geothermal from a geographic fluke into something you could do almost anywhere. The reason we have not been is money, and the money problem lives at the drill bit.

Here is the number that matters. In July, Fervo Energy finished a well it calls Sawtooth 7 at its Cape Station site in Utah. It reached a measured depth of 19,448 feet, more than three and a half miles of hole, into rock at about 460 degrees Fahrenheit. It did it in 21 days.

Speed is the point. That well took roughly 70 percent less time to drill than the company's earlier design, and its drilling rate has climbed 143 percent since its first well at the site. Deeper, hotter and longer, dug faster. When a rig costs a fortune for every day it turns, time in the ground is the cost.

None of this came from a physics breakthrough. It came from the oil and gas fields. Fervo borrowed the tricks that cracked shale: drilling sideways through a hot layer, not only straight down, and threading fibre-optic cable down the hole to feel what the rock is doing. Those tools spent twenty years getting cheap while chasing oil. Point them at hot rock, and the learning curve that made shale gas cheap starts making deep heat cheap.

A clean source that runs day and night has always struggled to find a buyer willing to pay the premium up front. AI changed that overnight.

A big data centre is a brutal customer in the best way: it wants a lot of power, in one place, running every second. That is a near-perfect match for geothermal, happiest running flat out around the clock. So Fervo has aimed its Utah project straight at that demand, naming the state as the site of the first cluster of data centres designed to run on enhanced geothermal, and it has signed a framework with Google to line up as much as 3 gigawatts, or 3,000 megawatts, over time.

The near-term piece is concrete. Fervo says the first 100 megawatts from Cape Station should start feeding the grid around October 2026, which it bills as the first commercial-scale geothermal power to reach that mark. Another 400 megawatts is slated for 2028. That first 100 is enough to run a serious data centre, not the whole boom, but it is real electrons on a real date.

It is fair to ask why this matters when solar panels keep getting cheaper. The answer is one piece of jargon worth knowing: capacity factor, the share of time a plant actually delivers its full rated output.

A solar farm rated at 100 megawatts might average about a quarter of that across a year, because the sun keeps banker's hours. A geothermal plant can run near 90 percent of the time, because the heat three miles down ignores clouds, seasons and nightfall. Engineers call that firm power: the kind you can promise will be there. Solar and wind are cheaper, but part-time. Geothermal turns up for every shift, and a data centre loses money every minute it sits dark.

Now the cold water, because there is some.

This is a drilling record and a power plan, not a discovery. Fervo did not invent new physics in July; it dug a hole faster and lined up a customer. Both matter; neither is a magic wand.

Still, this story is the rare hopeful kind. The heat was always down there. What was missing was a cheap way to reach it and a customer rich enough to want it. This year, both showed up at once.

You drill several miles down into hot, dry rock, pump in water to open a network of cracks, then circulate water through those cracks so it carries the heat back up as steam to spin a turbine. Ordinary geothermal needs a natural pocket of underground hot water, which exists in very few places. Enhanced geothermal makes its own underground boiler, so it can work in far more locations, as long as the deep rock is hot enough.

Data centres full of AI chips run every hour of every day and cannot afford to go dark, so they need firm power that is always on. Solar and wind are cheaper but part-time. Geothermal runs around the clock at very high availability, which fits a data centre almost perfectly. Fervo has named Utah as the site of the first cluster of data centres built to run on enhanced geothermal and signed a framework with Google for up to 3 gigawatts over time.

Fervo drilled the well at its Cape Station project in Utah to a measured depth of 19,448 feet, into rock at about 460 degrees Fahrenheit, in 21 days. That is roughly 70 percent faster than its earlier well design and continues a steep drop in drilling time and cost. The speed matters because the biggest barrier to enhanced geothermal has been the price of drilling, not the heat. The first 100 megawatts from the site is due to start reaching the grid around October 2026.

Related from Frontier Signal: our running coverage of the AI power crunch and the race for firm, always-on electricity. Frontier Signal explains frontier technology in plain English. This is general information, not investment advice.