Britain’s AI Ambitions Are Running Ahead of Its Grid

Within a day of entering Downing Street, Andy Burnham put energy affordability at the centre of his government by removing VAT from domestic electricity bills from October. Days later, reports that ministers were considering rapid approval for the Rosebank and Jackdaw fields pushed North Sea production back into the same debate. Yet the most important energy decision facing Burnham’s new Energy Secretary, Miatta Fahnbulleh, may concern an industry with no rigs at all. Ofgem has identified about 140 proposed data centres seeking 50GW of grid capacity, more than Britain’s recent peak electricity demand of 45GW. Some applications will never be built, but 71 projects representing around 20GW, report having reached a final investment decision. Britain is encouraging an AI infrastructure boom before settling how its electricity system will accommodate it.

The scale becomes clearer when annual consumption, rather than connection requests, is considered. Data centres used an estimated 5TWh of electricity in 2023, about 2% of UK demand. Oxford Economics forecasts that this could rise to 26.2TWh, or 8.8%, by 2030, while NESO’s lower planning assumption is around 22TWh. Neither estimate is certain. Even the lower figure, however, implies that data-centre electricity use will more than quadruple within seven years. These facilities may serve a digital economy, but they are physical industrial sites, dependent on generation, storage, and networks that cannot be expanded through software.

Ed Miliband understood that collision and, as Energy Secretary, helped establish the AI Energy Council, backed AI Growth Zones, and supported work on reforming grid connections. Nevertheless, the problem remains the distance between recognition and delivery. At the Council’s April meeting, officials said that 6GW of data centre demand was broadly aligned with Clean Power 2030, and that planning was manageable. They described work on curtailment, non-firm connections, and flexibility incentives as at an “early stage”. By then, Ofgem had already found around 20GW of financially committed projects in the queue. Miliband left post with the machinery for coordination in place, but without the rules needed to control the pace of development. Once again, private investment moves faster than public infrastructure and regulation.

This imbalance is clearest when construction times are compared. Evidence submitted to Parliament by RenewableUK suggests that a data centre can become operational in two to three years. New onshore wind can take four to five years, offshore wind around ten, and grid infrastructure as long as fourteen years. A facility can therefore be ready to draw power years before the projects intended to supply it. In that gap, connections may be delayed, existing gas generation may run harder, or developers may rely on diesel backup. Higher household bills are not inevitable, but the cost of reinforcement does not vanish when a developer connects. The political question is who bears it. Britain is attempting to join two timetables that do not match.

However, data centres could also become useful participants in a cleaner grid. The AI Energy Council recorded that 8.3TWh of British wind generation was curtailed in 2024, largely because electricity could not be moved to where it was needed. Locating new facilities closer to areas where renewable energy is frequently constrained, shifting non-urgent computing away from peak periods, and installing on-site storage could absorb some of that wasted power. RenewableUK says sufficiently strong locational incentives might place up to 20% of future data-centre demand in Scotland, while large new users could spread fixed network costs across a broader customer base. Nevertheless, these benefits depend on where a facility is built and how it operates. Critical services requiring uninterrupted power cannot offer the same flexibility as deferrable AI training. As a result, the Government must distinguish between them before promising faster connections.

Burnham is right to make energy costs a political priority. His reported interest in approving Rosebank and Jackdaw, however, reaches for the wrong lever in this case. Domestic production can support employment, tax revenue, and resilience during a managed transition, but it will not build the transmission lines, storage, or generating capacity that data centres require. The government’s own North Sea consultation acknowledges that British oil and gas are traded internationally and that domestic output has only a minimal effect on prices. An Oxford Smith School analysis estimates annual household savings of £16 to £82 from maximising extraction, with even those savings dependent on tax revenues being transferred to households. Its renewable scenarios estimate savings of £105 to £441. Those figures are scenarios, not promises, but they expose the weakness in presenting more drilling as a solution to Britain’s electricity bottleneck.

Fahnbulleh now has enough evidence to move beyond another general review. NESO should publish a standing forecast of credible data-centre demand, with uncertainty ranges and regular updates. Accelerated connections should then depend on a project’s maturity, its location, and a credible plan for additional power or flexibility. Ofgem is already considering deposits, stronger readiness tests, flexible connections, and greater developer self-build. Fahnbulleh should turn those options into a clear settlement, including transparent rules for which network costs developers must meet. Corporate power purchase agreements can support new renewable generation, but only where they add capacity and match supply more closely to the site’s consumption. A financial contract with an existing wind farm elsewhere in Britain cannot supply power at a congested connection point.

Miliband left Fahnbulleh a diagnosis and a stack of unfinished consultations. She must now provide the decisions. Every data centre granted accelerated access to the grid should bring the clean power and network investment needed to sustain it, alongside verifiable flexibility where feasible. Otherwise, Burnham’s government will cut bills in October while storing up costs that could push them higher later. Britain’s claim to AI leadership will be credible only when its energy policy can move at the same speed.