Ten years ago, Aberdeen was told it would die with the oil. Instead it went back to sea. How the Granite City became the place where Britain trains its machines.

There is a joke in Aberdeen, told mostly by people who work offshore, that the city's economy has always run on two things: what lies under the North Sea, and the willingness of people to fly out over it in the dark. For fifty years, what lay under the sea was oil. Now, mostly, it's computers.

It is hard to overstate how unlikely this looked in the late 2020s. North Sea production was in steep decline. Decommissioning, the slow and fabulously expensive business of dismantling platforms, was the only growth industry left, and it was growth in the way a funeral is an event. The city's population was falling. Its young engineers were leaving for Norway, the Gulf or Australia. "Managed decline" was the polite phrase. People in Aberdeen used less polite ones.

What changed was not one thing but several, arriving at once.

Too much wind, not enough wires

The first was a waste problem. By the late 2020s Scotland was generating more wind power than its grid could carry south to where people used it, and on windy days wind farms were paid to switch off. "We were paying hundreds of millions a year to not use free electricity," says one former grid engineer. "Everybody knew it was mad. Nobody had a use for it on the spot."

The second was the AI boom's appetite for power. Training a frontier model needed more electricity than a small town, and a great deal of cooling. The big American firms were building in deserts and talking, with a straight face, about orbiting data centres powered by sunlight, apparently untroubled by the difficulty of getting rid of heat in a vacuum, or of sending an engineer to replace a failed server.

The third, oddly, was an old experiment. In 2018 Microsoft had sunk a sealed data centre on the seabed off Orkney. When it was brought up two years later, its servers had failed at about one eighth the rate of an equivalent setup on land. The project was quietly shelved. Nobody in Britain built on it, although companies in China did.

Fiona Gauld, who had spent her career in subsea engineering, read the old reports during the long Covid winters and never quite stopped thinking about them. "Everyone looked at that experiment and saw a curiosity," she says. "I looked at it and saw a job description for Aberdeen."

Oil rigs to training rigs

The idea Gauld took to investors in 2029 had a slogan before it had a business plan: oil rigs to training rigs. Take a platform due for decommissioning. Keep the steel, the cranes, the helideck and the accommodation. Ring the seabed around it with sealed compute pods, cooled by the sea. Power them from the wind farms going up all around. Store the surplus in batteries, in the Highlands' pumped hydro reservoirs, and as hydrogen in the empty gas fields underneath. And staff all of it with people who already knew how to work offshore.

The first reaction, she admits, was laughter. The second was a list of objections, most of them correct.

Chips go out of date in three years, and you can't upgrade them at the bottom of the sea. (Answer: design the pods to come up. That's what the cranes are for.) The wind doesn't always blow, and idle chips are ruinously expensive. (Answer: keep a steady base load and add flexible capacity on top. Expensive chips are what make the batteries worth buying.) The old platforms are falling apart. (Answer: many are. Survey them and take only the good ones. Of the first twenty assessed, four were suitable.) The oil companies will never hand over the liability. (This one took two years and some very expensive lawyers, and was settled with ring-fenced decommissioning bonds that stayed with the original operators.)

And then there was politics. The first proposal was described in one Scottish newspaper as "Westminster's AI farms in Scottish waters," and for a while it looked fatal. What rescued it was a lease condition that now seems obvious: a guaranteed share of compute for British research and public models, and a slice of revenue paid into a Scottish community fund. "It stopped being their project," says one MSP who opposed it at first. "It became ours."

Two weeks on, three weeks off

Today there are four training rigs in the central North Sea, with two more in conversion. Around three thousand people work on them or for them. The helicopters still leave Dyce at ten to seven. The ROV pilots who used to inspect pipelines now inspect pods. The crane operators who lifted drill pipe lift compute modules. A good number of the control-room staff have no offshore background at all, including several blind technicians, recruited after one of them redesigned Skerry Alpha's monitoring system so that the pods could be heard rather than watched.

The rigs have trained, among other things, part of Will, Britain's national open-weights model, and Avalon, the video model built on it. Its makers like to say it learned to see while sitting at the bottom of the sea.

The project has its critics. Environmental groups point out, correctly, that the rigs aren't green in every hour. During the "Long Calm" of January 2034, Skerry Alpha ran for three days on stored hydrogen, and the hydrogen round trip throws away two thirds of the energy. Haar Compute's answer is that it never claimed otherwise. Its target is ninety per cent wind and stored wind across a year, and it publishes the number. Others worry that too much of the capacity is still rented to foreign firms, and that Britain risks becoming the North Sea's landlord rather than its builder. They may be right.

But stand on the helideck of Skerry Alpha at dusk, with the turbines turning to the horizon and the knowledge that sixty-four pods are humming in the cold water beneath your feet, and it's hard not to feel that something has been recovered, not just a platform or a few thousand jobs but an idea about what this country is for.

"For fifty years the North Sea paid for Britain," Gauld says. "Then everyone assumed it was finished. It wasn't finished. We just needed a different reason to go to sea."

Next: Dyson Shoal.