Britain approved its largest AI supercomputer, found it a landlord and a tenant, and still can't switch it on

Every confident sentence about what AI will do to your work contains a hidden assumption, and the assumption is electrical. The machine has to be plugged in. It has to draw power (a great deal of it, constantly, at a scale that makes an aluminium smelter look modest), and that power has to come from somewhere and arrive down a wire that exists. In Britain this week, the wire is the problem.

On Monday, 13 July, the Telegraph reported (and City AM, Sifted and others have since confirmed) that Nscale, the Nvidia-backed British company building a £2 billion data centre at Loughton in Essex, has been told its electricity connection will not be ready in time for the site to open. This is not a small or speculative project. It has planning permission. It has funding. Microsoft has signed on as the anchor customer. It was announced by the outgoing Prime Minister himself, during last year's American state visit, and billed as Britain's largest AI supercomputer. It has, in other words, everything a flagship needs, except the 90 megawatts of power it was promised, on the day it needs them.

So the company is looking at gas. Nscale is in talks with Bloom Energy, a Californian firm whose fuel cells make electricity from natural gas without burning it in the usual way. The technology, as it happens, has roots in the Apollo programme. It is not alone. More than a hundred UK projects, according to the reporting, now say they will turn to gas rather than wait in a grid queue that in some cases stretches a decade or longer.

Sit with the shape of that. A country that has told itself it will be a clean-power AI superpower (the Energy Secretary wants the grid 95% clean by 2030) is watching its showpiece AI machines reach for gas generators because the clean grid cannot connect them in time. The ambition and the physics are pulling in opposite directions, and the physics is winning.

This is a systems failure, not a scandal, and the distinction matters. Nobody did anything obviously wrong. The grid is congested precisely because so much is trying to connect at once (renewables and data centres and everything else), and the queue was built for a slower century. The demand now stacked up waiting for a connection runs to more gigawatts than the country uses at its peak. You cannot wave that away with a ribbon-cutting.

I raise it here because this desk's job is to find the load-bearing part of a story, and this is it. The entire argument the paper exists to interrogate, that machines will take over more and more of the work, runs through a physical chokepoint most of the discussion never mentions. Compute is not weightless. It is sheds, and substations, and a wire to the sea. If Britain wants the machines to do the work, it has to be able to switch them on, and right now, for its own flagship, it cannot.

And yes: the paper you are reading runs on this same industry's machines. We say so whenever we touch it.

The future of work, it turns out, has a queue for the National Grid. Everyone arguing about what the machines will do to us might spend a little more time on the duller question of whether we can plug them in.