The UK Just Started Charging for Fake AI Demand — and Ireland Is Going Nuclear

The UK's Ofgem is charging data center developers up to £712,500 per megawatt to hold grid queue places, pricing out speculative projects, while Ireland reconsiders its nuclear ban as AI demand surges. A hosting founder on what the phantom-demand crackdown means for the real buildout.

Aug 27, 2026 - 14:11
0 6
The UK Just Started Charging for Fake AI Demand — and Ireland Is Going Nuclear

The UK Just Started Charging for Fake AI Demand — and Ireland Is Going Nuclear

Let me tell you something that's been rattling around my head since I read the Ofgem consultation last night. Two countries, one problem, two completely different answers — and both of them are telling you the same uncomfortable thing about the AI buildout that nobody wants to say out loud: nobody actually knows how much of the power demand is real.

I've been running hosting infrastructure for over a decade. I've watched power prices climb, chased grid connections, and sat through enough utility meetings to know that the electricity queue is where the AI trade either gets real or falls apart. So when the UK's energy regulator dropped a proposal in July to charge data center developers up to £712,500 per megawatt just to hold a place in the grid connection queue — and Ireland started talking about ending its 27-year nuclear ban in the same week — I paid attention. Because those two moves are the same story wearing different costumes. The phantom gigawatt problem has officially hit the policy level.

The Queue Nobody Can Trust

Here's the scale of the mess. Between November 2024 and June 2025 — under a year — the total demand of projects waiting in the UK's grid connection queue went from 41 gigawatts to 125 gigawatts. Data centers account for at least 80 gigawatts of that, and Ofgem's own consultation puts the data center share at around 73GW. Let me give you some perspective. Seventy-three gigawatts is one and a half times the peak electricity demand of the entire United Kingdom last year. The whole country runs on roughly 50GW at a winter peak. These proposed data centers, on paper, want one and a half times that.

And the queue has been dirt cheap to enter. Historically it cost a few thousand dollars to apply for grid access — peanuts compared to the option value of locking in capacity in case AI demand goes vertical. So developers applied for power they had no plan to use. Land flippers got in on the act too: middlemen grab a grid connection application, then flip the land-plus-queue package to a real data center company at a premium. Olivier Darmouni, a finance professor at HEC Paris who studies exactly this, called it "scalping for concert tickets." The congestion feeds on itself.

What Ofgem Is Actually Proposing

Ofgem's answer is a two-part gut check. First, a Data Centre Commitment Fee: £237,500 to £712,500 per megawatt, which works out to about 2.5% to 7.5% of average project costs. For a 100MW facility, that's £23.75 million to £71.25 million — before you've poured a single slab of concrete. The money is refunded when the project reaches energisation and forfeited if it bails out of the queue early. Second, queue management milestones: developers have to show financial capability, commercial maturity, and procurement activity to keep their place in line.

The consultation opened July 29 and runs until September 16. And Ofgem is not messing around with the rationale. Eleanor Warburton, the regulator's director for energy system design and development, put it bluntly: consumers should not bear the risks created by speculative projects taking up space in the system. She's right. When a phantom project sits in the queue, grid operators have to study the network as if it were real — which makes the studies slower, more expensive, and more complex for every viable project stuck behind it.

The Goldilocks Problem

But here's where it gets tricky, and this is the part that keeps founders up at night. Ofgem has to set the fee high enough to scare off the speculators and low enough that it doesn't drive legitimate projects to other shores. The UK is already an expensive place to build — high energy costs, a dearth of land, and now the threat of a seven-figure-per-megawatt deposit. Knight Frank's head of data centers, Alex Burgoyne, said exactly what I was thinking: "We don't want to shoot the golden goose."

And it's not just the hyperscalers who should worry. The fee structure could crush a whole generation of smaller, AI-specific data center operators — companies like Nscale, which has pledged billions to build in the UK. A small operator with a 20MW project is looking at a £4.75 million to £14.25 million commitment before energisation. That's a bet only well-funded players can take. Daniel Newton, a partner at the law firm Slaughter and May, nailed the balancing act: set it too low and you don't deter speculation; set it too high and you kill viable projects because nobody will put that kind of money at risk at the development stage.

Ireland's Answer — Open the Nuclear Door

Meanwhile, across the Irish Sea, a different government is answering the same question from the supply side. Ireland has had a legal ban on nuclear power since 1999. This week, the government was publicly keeping the nuclear door open as a long-term answer to data center demand, even while insisting the immediate strategy stays renewables-first.

You can see why they're rattled. Data centers consumed 23% of Ireland's total metered electricity in 2025 — almost as much as all the households in the country combined. Projections put that at 31% by 2032. The EU-wide number is just as wild: data centers across the bloc used roughly 70 terawatt-hours in 2024 and are projected to hit 115TWh by 2030 — enough to power almost 33 million homes for a year.

So you've got two governments, same problem, opposite instincts. The UK is trying to shrink demand on paper by pricing out the fakes. Ireland is trying to expand supply in reality by breaking a 27-year taboo. One is a demand-side fix. The other is a supply-side leap. Both are guesses.

The Two Readings

Here's how I read the Ofgem move — and honestly, it's both readings at once, which is why it's so interesting.

Reading one: this is the sane move. Cleaning the queue is the single most productive thing a grid regulator can do right now. Ofgem says its wider connections reforms have already accelerated about 7.8GW of projects by an average of six years. Six years. If the same logic applies to data center demand, then the fee isn't a tax on builders — it's a toll that makes the whole queue move faster for the people who actually intend to build. Every hour a phantom project sits in the queue is an hour a real one can't get its network studies done.

Reading two: this is a tax on uncertainty wearing a policy costume. The reason Ofgem has to charge £712,500 a megawatt is that it cannot tell a real project from a fake one, and neither can anyone else. The queue data — the very information grid planners use to decide where to build new transmission — is so polluted that the regulator's only tool is to make the bet expensive enough that only true believers pay. That is not a sign of a healthy, confident AI buildout. That is a regulator saying: I have no idea how much of this demand is real, so I'm going to make you put your money where your mouth is.

Ireland's nuclear flirtation is the same confession from the other direction. When a country that banned atomic power for 27 years starts leaving the door open, it's not because nuclear suddenly got cheap. It's because the demand forecast is so uncertain that the government is willing to overturn a bedrock policy to hedge against it.

What This Means for Independent Hosting Providers

First: treat queue cleanup as a demand signal, not a crash signal. When the phantom gigawatts get priced out, the real buildout gets faster and clearer. Ofgem's 7.8GW accelerated by six years is the proof — clean queues mean real projects move. The AI buildout isn't collapsing because a regulator started charging for queue spots. It's getting honest.

Second: budget for the new cost layer. Commitment fees are coming to every jammed market, not just the UK. When your local grid operator copies Ofgem — and it will — the cost of holding grid capacity becomes a line item that hits every project, including yours. Model £237,500 to £712,500 per megawatt into your colo pricing now, while you can still explain it to customers as a forward-looking cost instead of a surprise.

Third: grid access just became a tradable, expensive asset. Land-plus-queue packages are already being flipped in the UK, and the fee regime is going to flush the weakest hands. Before you sign any colo, lease, or partnership deal, find out who actually holds the grid position — and whether they can afford to keep holding it through energisation. A counterparty that can't cover the commitment fee is a counterparty that's about to sell you a problem.

Fourth: don't chase hyperscale leases in expensive-energy markets unless power is locked. The UK and Ireland are about to become case studies in how energy scarcity reshapes the data center map. Secondary markets with cleaner queues, cheaper power, and fewer regulatory surprises are going to win the next wave of real projects. That's where independent operators have always had the edge — and now the economics are swinging back our way.

Fifth: watch which direction your regulator jumps. Fee (UK), moratorium (US), nuclear door (Ireland) — every market is picking a response to the same phantom-demand problem. The hosting providers who read the direction early and position before the rules land are going to eat the lunch of the ones who wait for the fine print.

The Bottom Line

Here's the truth nobody in the AI hype cycle wants to hear: the queue numbers were always a mirage, and the regulators just figured it out. The UK is charging for the mirage. Ireland is building an escape hatch from it. Both moves are expensive, both are risky, and both are healthier than pretending 125GW of connection requests was ever going to get built.

I've said it before and I'll say it again: this isn't the AI buildout ending. It's the AI buildout growing up. When the fake demand gets priced out and the real demand gets faster connections, the people who actually own servers, power contracts, and grid positions are the ones who win. Ent? The next five years are going to be shaped by who could tell the difference — and who was willing to pay for it.

— Allan Ali, Founder

This article was produced with AI-assisted research and editorial support. Sources: WIRED, Ofgem, Euronews, DatacenterDynamics.

What's Your Reaction?

Like Like 0
Dislike Dislike 0
Love Love 0
Funny Funny 0
Wow Wow 0
Sad Sad 0
Angry Angry 0
Allan Ali

Publisher of Global1.News. Automation architect, systems builder, and the guy making sure the truth gets published.

Comments (0)

User