Here is a number worth sitting with: 50 gigawatts. That is the combined electricity demand of roughly 140 datacenters currently queued up for connection to Britain's national power grid. To put it in terms that require no engineering degree, Britain's energy regulator Ofgem has confirmed that this figure exceeds the entire peak electricity consumption of the country, which stood at 45 GW on 11 February this year. The UK government wants to be an AI superpower. The grid, at present, has other ideas.
The figures, reported by The Register and drawn from a recently published Ofgem call for input on Demand Connections Reform, reveal a system under serious strain. Since November 2024, total contracted offers in the demand queue tripled from 41 GW to 125 GW by June 2025. A significant portion of that surge is attributable to AI-related datacenter projects, many of which are at early or speculative stages of development. The National Energy System Operator (NESO), which manages Britain's electricity and gas systems, identified approximately 140 datacenter facilities in the queue, the majority of which are likely to receive a "ready-to-connect" offer, known as a Gate 2 agreement.
The fundamental question is not whether this demand is real. Some of it clearly is. Nvidia, Microsoft, and Google have all committed billions of dollars to AI infrastructure in Britain, and the government's AI Opportunities Action Plan, launched in January 2025, has designated five AI Growth Zones across England, Scotland, and Wales, reportedly generating £28.2 billion in investment and more than 15,000 jobs. The question is whether the system for allocating grid access is remotely fit for purpose when confronted with this scale of demand.
Ofgem's own document is admirably candid about the structural failings. The queue, it says, contains a significant number of projects that are "likely non-viable." Viable projects are being held up by the sheer volume of speculative applications clogging the system. And critically, there is currently no mechanism to prioritise strategically important projects over ones that exist mainly on a spreadsheet. For a regulator whose job is to protect the interests of consumers and manage national infrastructure, that is a significant admission.
Omdia principal analyst Alan Howard, speaking to The Register, explained the dynamic clearly.
"The strategy for many datacenter operators is to secure multiple land parcel rights, request a grid load connection for each (often requiring a costly load study), and see what gets approved so they can build."The result, as Howard notes, is that the capital investment required to treat all those applications as genuine is "clearly untenable" and represents a major financial risk for the energy sector if the demand does not materialise. A similar pattern was documented in a report by the Uptime Institute, which found that developers routinely reserve power for projects that may never be built, while operators apply for more than they need to accommodate future uncertainty.
This is not a uniquely British problem. Howard told The Register that the grid interconnection queue issue is replicated in the US and other global markets. In the UK's case, it is complicated by some of the world's highest household energy prices and by net-zero commitments that constrain what new generation capacity can look like. The government has pointed to its AI Growth Zones, including one in North Lanarkshire and one at the Culham Science Centre in Oxfordshire, as areas where new clean power generation will serve future datacenter demand. A spokesperson for the Department for Energy Security and Net Zero told The Register that "datacenters will only be connected where the grid can support them."
The counter-argument deserves serious consideration. Proponents of the UK's AI strategy are right to point out that grid reform and digital infrastructure investment are not mutually exclusive. The government has assembled an AI Energy Council, comprising Ofgem, NESO, energy companies, and major datacenter operators including Google, AWS, Microsoft, and Equinix, specifically to address the energy challenge. Ofgem is proposing a two-phase reform of the demand connections process: an immediate phase targeting datacenters with stronger financial commitments and readiness requirements to flush out non-viable projects, followed by a broader strategic plan for data facilities across all demand sectors. Phase one decisions are expected in spring 2026, with responses to the consultation due by 13 March 2026.
The government has also launched an Advanced Nuclear Framework to attract private investment into next-generation nuclear technology. A small modular reactor has been announced for Wylfa on Anglesey in northwest Wales, to be sited near one of the designated AI Growth Zones in North Wales. However, that reactor will not generate power until the mid-2030s, which is not particularly reassuring for datacenter operators hoping to connect in the next few years.
Strip away the talking points and what remains is a regulatory architecture that was never designed for this volume or velocity of demand. The Ofgem reform process is a genuine step forward, and the proposal to use evidence such as a Final Investment Decision or secured planning permission to distinguish real projects from speculative ones is sensible. But the fact that this common-sense mechanism does not already exist, and that the queue was allowed to nearly triple in seven months, raises fair questions about how seriously the energy system infrastructure was stress-tested before the AI buildout ambitions were publicly announced.
Reasonable people can disagree about the pace and shape of Britain's AI strategy. The economic case for building world-class digital infrastructure is real, and the risk of falling behind the United States or Europe is not trivial. But economic ambition without grid capacity is not a strategy; it is a wish list. The Ofgem consultation, and the two-phase reform programme it sets in motion, represents the kind of pragmatic, evidence-based course correction that should have preceded the fanfare rather than followed it. Getting this right matters not just for the datacenters, but for every household and business already connected to a grid being asked to do far more than it was built for.