Energy
Data centres and the grid: the quiet fight over who gets the electrons
AI's real constraint in the UK isn't chips or talent. It's a queue for grid connections that at its worst stretched into the late 2030s — and the scramble to fix it is reshaping both industries.
Ask anyone in the data-centre business what keeps them up at night and, as of early 2026, the answer is rarely chips, planning permission or even money. It's a substation. Or rather, the lack of one — and the multi-year queue standing between a shovel-ready site and the electrons it needs. The most important fight in British technology is happening not in labs but in the unglamorous paperwork of grid connection offers.
How the queue got so long
Britain's grid connection process was designed for a sedate era: a handful of large power stations, demand growing slowly, first come, first served. Then two revolutions arrived at once. Thousands of renewable generation and battery projects applied to connect — and a wave of new demand, led by data centres, applied at the same time. The queue swelled to hundreds of gigawatts of proposed projects, several times the country's roughly 60-gigawatt peak demand. Some applicants were quoted connection dates in the late 2030s.
The rot in the system was the "zombie project": speculative applications that grabbed a place in the queue with no realistic prospect of being built, blocking serious schemes behind them. When holding a queue position costs little, hoarding queue positions becomes a business model.
Enter the data centres
Data centres transformed this from an energy-sector problem into a national economic one. A large AI-oriented campus can demand more than 100 megawatts — small-town quantities of electricity on a single site — and the growth of inference workloads means the demand curve keeps steepening (we explain why in our piece on inference chips). The industry's traditional heartland in west London hit the wall first: local grid capacity around the Slough–Hayes corridor became so constrained that new housing developments in some boroughs were warned of connection delays, an early collision between homes and hyperscale that made national headlines.
Government responded by embracing the sector. Data centres were designated critical national infrastructure in 2024, and the AI Opportunities Action Plan of early 2025 created "AI Growth Zones" — areas promised accelerated grid connections and friendlier planning, with the first anchored at Culham in Oxfordshire. The signal was unambiguous: compute is now treated as strategic infrastructure, alongside ports and power stations.
When holding a queue position costs little, hoarding queue positions becomes a business model.
The reform: ready beats early
The structural fix arrived through connections reform led by the National Energy System Operator (NESO), approved by Ofgem in 2025. The principle changed from "first come, first served" to "first ready, first connected": projects must now show land rights, planning progress and financial credibility to keep their place, and alignment with the country's strategic energy plans determines priority. Hundreds of gigawatts of speculative applications faced removal or resequencing, pulling connection dates forward for projects that are actually real.
It is genuine progress, with an uncomfortable corollary: connection capacity is now explicitly rationed by policy. Somebody in an office decides whether a battery farm, a housing estate or a data centre better serves the national plan. The fight over electrons has not ended; it has moved indoors.
What the data centres are doing meanwhile
Operators aren't waiting politely. The pragmatic play is siting flexibility: following available grid capacity to Scotland, the north-east and Wales rather than paying the west London premium — quietly redistributing tech investment around the country in the process. The bolder play is self-supply: on-site gas turbines as bridging power, co-location deals with wind and solar farms, and — the endgame several hyperscalers are pursuing globally — small modular nuclear reactors, for which the UK picked Rolls-Royce SMR as preferred bidder in 2025. Batteries help smooth the peaks too, though grid-scale storage has its own supply story — see our solid-state batteries reality check.
There's also a subtler shift: making the computing itself more grid-friendly. Training runs can pause when the grid is stressed; inference can be routed between regions; and the growing share of AI running on end-user devices (the small-model shift) never touches a data centre at all.
Why this matters beyond the industry
Three reasons to care. Your bills: grid upgrades are ultimately socialised across consumers, and how much of the cost data centres bear is a live regulatory argument. Your region: power availability is becoming the decisive factor in where tech investment lands, which means grid maps are quietly becoming economic development maps. And the national bet: Britain wants AI-driven growth and clean-power targets simultaneously, and both draw from the same constrained grid. As of early 2026 the honest summary is that the UK has diagnosed the problem correctly, started the right reforms, and remains years away from the boring abundance of power that its digital ambitions assume.