For most of the past two decades, data centres were treated as a planning matter. They occupied land, generated few jobs relative to their footprint, and needed reliable power. Local authorities weighed them like any other large shed. That framing has collapsed. The scale of electricity demand created by AI training and inference has turned data centre construction into a question of national energy policy, and Britain’s grid was not designed with it in mind.
The numbers involved are genuinely large. A single hyperscale campus can require electricity comparable to a small city, and requests for grid connection now routinely exceed the capacity available in the regions where operators most want to build. This has produced a queue, and the queue has become the defining constraint on both digital infrastructure and industrial electrification.
Why the grid queue exists
Britain’s transmission connection process was built for a world of a few large generators connecting occasionally. It now handles thousands of applications from generators, storage projects, factories and data centres, many of which are speculative. Because there was historically little cost to holding a place, the queue filled with projects that will never be built, blocking those that will.
The Office of Gas and Electricity Markets and the system operator have moved to reform this, introducing tests that require projects to demonstrate readiness before holding capacity. The reform is sensible and slow, because reordering a queue that thousands of investment decisions depend upon cannot be done casually.
Geography is the real bottleneck
Britain generates a growing share of its electricity in Scotland and the North Sea, while demand concentrates in the South East. The transmission capacity to move power south is insufficient, which produces the perverse outcome of wind farms being paid to stop generating while gas plants run in the south.
Data centres cluster near London for latency and connectivity reasons, precisely where the grid is most constrained. Analysis published by the National Grid on network reinforcement sets out the transmission build required, and the timelines run into the 2030s. Planning consent for overhead lines is contested in almost every affected constituency.
The nuclear and behind-the-meter turn
Faced with grid delays, large operators have begun pursuing dedicated generation. That means power purchase agreements with specific plants, on-site gas turbines as a bridge, and serious interest in small modular reactors sited alongside compute capacity.
This shifts the policy question. If the largest new electricity consumers build their own supply, they stop competing for scarce grid capacity, but they also stop contributing to the shared network that everyone else depends on. How network costs are allocated when major users partially exit becomes a distributional question with consequences for household bills. The Department for Energy Security and Net Zero is working through exactly this tension.
Water, heat and the local case
Electricity is the headline constraint but not the only one. Large facilities need cooling, and depending on design that can mean substantial water consumption in a country whose water infrastructure is already under strain. Some operators have moved to closed-loop or air cooling partly for this reason, at a cost in efficiency.
There is a genuine opportunity in waste heat. Data centres reject enormous quantities of low-grade heat that could serve district heating networks, as happens in parts of Scandinavia. Doing so requires the network to exist and to be planned alongside the facility, which is a coordination problem British development has historically handled badly. Guidance from the Environment Agency on abstraction and discharge shapes what is feasible locally.
The economic argument cuts both ways
Advocates point to investment, construction employment and the strategic value of hosting compute domestically rather than depending on facilities abroad. Those arguments have force, particularly for sensitive public sector and health data.
Sceptics note that operational employment is modest, that the facilities consume scarce grid capacity that could serve manufacturing with higher employment intensity, and that much of the value accrues to overseas parent companies. Both positions are defensible. The honest conclusion is that data centres are neither the industrial saviour nor the parasite of competing rhetoric, and that their value depends heavily on whether they are additional to other electrified investment or crowding it out.
What a coherent policy would contain
Three elements stand out. First, a connection regime that prices capacity so speculative applications are filtered without excluding genuine projects. Second, transmission build treated as national infrastructure with a planning process that can actually deliver, since no amount of generation helps if the power cannot move. Third, explicit rules on how network costs are shared when large users self-supply.
Britain is not unusual in facing this. It is unusual in having a planning and connection system slow enough that the constraint bites harder than it needs to. Reporting from the National Audit Office on infrastructure delivery has made a version of this point about several sectors, and computing is now simply the most visible case.
