As data center developers compete for a cut of the hundreds of billions of dollars flowing into the AI industry, the queue to join the UK's power grid has become jammed with projects that will likely never be built. The snarl is worsening already years-long wait times for viable projects and disrupting attempts to forecast energy demand and plan grid expansions.

The numbers

The queue began to balloon toward the end of 2024, around the time the government designated data centers "critical national infrastructure."

  • November 2024 — total energy demand of queued projects: 41 gigawatts.
  • June 2025 — the same figure: 125 gigawatts.
  • Of that, 73 gigawatts comes from new data centers: one and a half times the peak demand for the entire UK last year, and counting.

The government believes much of that demand is a mirage. Taco Engelaar, managing director at grid optimization company Neara, puts it bluntly: "It's absolutely insane. No one really understands what the real grid demand will be because of these phantom projects."

The problem is incentives

The crowded queue is largely the product of an incentives problem. Because developers face a years-long wait for grid access, and joining the queue has previously cost only a few thousand dollars, they might as well apply even without a watertight plan to develop a site. It is a no-downside bet that hedges for a future in which demand for compute stays sky-high.

Others pay the price. Grid operators have to account for the combined effect of large infrastructure proposals on network stability before granting individual connections. Olivier Darmouni, associate professor of finance at HEC Paris, explains the mechanism: "They have to treat every project as serious when they try to study whether the system can handle them. The more speculative projects are especially damaging because they make these studies more complex, more expensive, longer."

A lengthy queue also attracts middleman developers who apply for power intending to flip land later at a premium, packaged with grid access. Darmouni compares it to ticket scalping: "The congestion feeds on itself and gets out of hand."

The proposed fix and its dilemma

Ofgem, the energy regulator, laid out a proposal meant to force phantom data centers out of the swollen queue. Developers would be required to put down a steep, nonrefundable deposit that could balloon to hundreds of millions of dollars for the very largest data centers. They would also have to line up customers in advance and prove they have the funding to complete their builds.

Ofgem faces a balance problem: the reforms must be burdensome enough to deter speculators but not so burdensome that they drive legitimate projects to other shores. Otherwise the compute capacity the country's AI ambitions require simply goes elsewhere.

The industry is voicing that risk. Alex Burgoyne, head of data centers at real estate consultancy Knight Frank, warns that the buildout is bringing a huge amount of capital investment into the UK: "We don't want to shoot the golden goose."

Why it matters

This congestion is not unique to the UK; similar grid problems afflict the US and countries across Europe. But whereas the US is a highly desirable market, Ofgem's reforms risk making the UK — already unattractive for high energy costs and a dearth of land — one of the world's most expensive places to build a data center.

The real lesson concerns how numbers in the AI infrastructure debate should be read. "So many gigawatts of data centers are planned" sounds like a forecast, but it often reflects nothing more than a count of applications. As long as applying is cheap, that number measures not real demand but bets placed on the possibility of real demand.