I was reading The Week’s briefing on the latest data-centre wobble, and the detail that stuck with me wasn’t some complicated financial forecast. It was a proposed AI facility in Essex struggling to get enough electricity.
We’re endlessly told this technology will transform everything. Fair enough, maybe it will. But first somebody has to connect the building to the grid.
You can’t run a data centre on enthusiasm
The briefing describes a setback for Nscale, one of the companies trying to provide the infrastructure behind the AI boom. Microsoft had been lined up as the main customer for its Loughton site, but the electricity supply became a problem.
Reuters reported that Microsoft withdrew from the agreement after grid delays pushed the expected connection into the 2030s. Crucially, both companies said their wider relationship remained strong. This was a setback at one site, not Microsoft abandoning Nscale altogether.
Still, it exposes a fairly basic problem. You can announce contracts and ambitious expansion plans, but customers can’t use computing capacity that isn’t available.
I don’t think this proves AI is useless, or that every data centre is a terrible investment. It does suggest that the people selling the dream need to spend more time explaining how they’ll actually deliver it.
An impressive valuation won’t bring a grid connection forward by five years.
The debt is rather less futuristic
The more worrying part of the briefing is the borrowing. These businesses need expensive equipment and buildings before they can collect the revenues intended to pay for them. That leaves quite a lot riding on demand arriving at the right price and the right time.
This isn’t just newspaper doom-mongering. In its December 2025 Financial Stability Report, the Bank of England warned that debt financing in the AI sector was increasing rapidly. It also highlighted complex financing arrangements and the difficulty of tracing where some credit risks ultimately sit.
That matters because a disappointing investment becomes a different sort of problem when borrowed money is involved. Shareholders can watch the value of their shares fall. A lender still expects repayment, even if the promised customers haven’t arrived.
The Week also mentions chips being used as collateral. My immediate question is what those chips would fetch if several struggling companies needed to sell them at once. Equipment can be valuable without being a reliable safety net in every possible market.
And none of this requires AI to disappear. A technology can become widely used while some of the businesses built around it lose money. Being right about its importance doesn’t automatically mean being right about the price of every company involved.
