Britain's AI Data Centers Requested 50 Gigawatts of Grid Power. The Entire Country Peaks at 45.
That comparison sounds like Britain is about to run out of electricity. It isn't — most of those projects will never get built. The real story is that Europe's grid-connection queues, not chip supply or construction budgets, have become the binding constraint on where the AI boom actually lands, and the fix Britain tried this year offers a preview of the fight every other country is about to have.
The grid queue, two ways
A simplified model, not real UK project data: 12 projects in a queue, 4 of them missing land, permits or funding ("zombie" projects) — roughly the ~30% share NESO's own reform found unviable. One connection can be processed per simulated year. Toggle the queueing rule, then press Run.
Initial queue — first-come, first-served. Whoever is first in line goes first, ready or not. Press Run.
The plain version
Picture a line at a government office, except the line is for plugging into the electrical grid, and reaching the counter can take a decade. That's roughly what's happening across Europe's biggest tech hubs — London, Frankfurt, Amsterdam, Paris, Dublin — where companies wanting to build AI data centers face connection queues running seven to ten years, even though building the data center itself takes only 18 to 24 months. The bottleneck isn't the building. It's the wire.
In February, Britain's grid operator found roughly 140 proposed data centers asking for a combined 50 gigawatts of power — more than the entire country delivers at its single busiest moment of the year, 45 gigawatts. That doesn't mean Britain is about to go dark; most of those projects will never actually get built. That's exactly the problem: for years, the queue worked strictly first-come-first-served, so speculative projects with no land, no permits and no real financing could hold a spot for a decade, blocking real, fundable projects stuck behind them. The industry calls these "zombie" projects.
This June, Britain rewrote the rule. Instead of first-come-first-served, projects now have to prove they're ready — land secured, permits in hand, financing lined up — to keep their place in line. The effect was immediate: the national queue shrank from over 700 gigawatts of requests to 238 gigawatts of genuinely viable ones, mostly by deleting projects that were never getting built in the first place.
Ireland shows what happens without that fix. Dublin's grid has been effectively closed to new data-center connections since around 2021, and roughly €5.8 billion worth of fully permitted, land-bought data centers now sit idle, waiting for power that hasn't come.
The expert version
The mismatch driving Europe's data-center power crunch is structural, not cyclical: in the FLAP-D markets (Frankfurt, London, Amsterdam, Paris, Dublin), grid interconnection queues average 7–10 years, against a hyperscale data-center construction timeline of 18–24 months. A developer can secure land, planning consent and a tenant faster than a transmission operator can issue a firm connection offer, so financeable projects sit stalled behind speculative ones with no real prospect of completion.
In Great Britain, National Energy System Operator (NESO) data show the demand-side queue climbing from roughly 41 GW (November 2024) to 125 GW (June 2025) to about 140 identified data-center projects totaling ~50 GW by February 2026 — a figure that alone exceeds Great Britain's own peak system demand of 45 GW, recorded February 11, 2026. The generation-side queue told the same story in reverse: more than 700 GW of contracted capacity sat in the pipeline under a strict "first come, first connected" allocation model, much of it non-viable.
Ofgem approved NESO's TMO4+ reform package in April 2025; it went live June 10, 2025, replacing first-come-first-served with "first ready and needed, first connected." Projects lacking land rights, consents or demonstrated progress were placed in a non-priority "Gate 1" tranche — roughly 217 GW of the queue, informally termed "zombie" projects — cutting the retained development pipeline to 238 GW. Battery storage, at 83 GW, was the largest surviving category; roughly 36 GW of solar was removed. By June 2026, more than 700 projects representing 37 GW had received connection offers under the new regime.
Ireland illustrates the cost of reforming too slowly: the Commission for Regulation of Utilities' effective 2021–22 moratorium on new Dublin-area data-center connections, driven by transmission-capacity limits, has left an estimated €5.8 billion in fully permitted, land-secured data-center investment unable to reach commercial operation, pending EirGrid's 2026–2030 grid-backbone upgrade and an April 2026 Grid Code Modification proposal to the CRU.
Why it matters for tech + supply chain: AI's site-selection constraint is shifting from chips and capital to which country can turn a grid-connection application into actual power the fastest.
Why it matters for tech + supply chain: queue-allocation mechanism design, not raw generation capacity, is now the marginal lever governments have over where AI compute physically gets sited — and Britain's readiness-gated model is the template other grid operators are now racing to copy.