Ofgem just put a price tag on speculative compute demand. Its July 29 consultation proposes a data-centre commitment fee for projects above 40 megawatts, backed by queue milestones that force developers to show an end user, long-lead electrical procurement, financial capacity, and technical readiness before they keep their place in Great Britain’s electricity connection queue.
This revisits AI Datacenters Made the Utility Bill a Control Surface and New York Turned the AI Buildout Into a Permit Queue. The new development is Ofgem’s specific collateral machine: a returnable fee of roughly £237,500 to £712,500 per megawatt of requested capacity, forfeitable when a project terminates or misses the rules.
the queue became the infrastructure
Ofgem’s consultation document says contracted demand in the connection queue rose from 41 GW in November 2024 to 125 GW in June 2025. About 73 GW of that total is data-centre demand across roughly 315 projects. Great Britain’s 2025 peak electricity demand was about 45 GW.
That arithmetic is the whole story. The queue is no longer paperwork waiting for infrastructure. The queue shapes infrastructure: transmission studies, distribution planning, reinforcement assumptions, scarce engineering attention, future capacity reservation, and political cover for new buildout. A phantom campus can still consume the planning system while a real one waits behind it.
Ofgem calls the package Curate. Good name, honestly. The regulator wants to reduce non-viable projects without setting a hard cap on data-centre volume. The mechanism is administrative, but the consequence is physical. Grid access becomes a staged claim that must be financially secured and technically evidenced.
cash is doing policy work
The proposed fee applies to data-centre projects above a 40 MW threshold. The amount comes from 2.5% to 7.5% of average data-centre capital expenditure, which Ofgem translates into about £237,500 to £712,500 per MW of requested capacity. A 1 GW project would therefore carry hundreds of millions of pounds in secured commitment.
That is the part that matters. A generic readiness checklist can become theater. A financial commitment tied to megawatts changes the economics of queue squatting. If the developer proceeds as planned, the money comes back. If the project terminates or fails the relevant requirements, the money can be forfeited.
The proposal also acknowledges different business models. A self-operated facility proves progress differently from a shell built for lease or sale. Ofgem’s milestone framework tries to map those paths without letting every PowerPoint campus claim the same grid priority. The evidence list is deliberately plain: credible compute customer or end-user, long-lead electrical equipment, financial capability, technical readiness. Boring is the point. Serious projects can carry boring evidence.
the number is bigger than the grid can politely ignore
The BBC’s coverage pulls the public-facing consequence into view: there are 564 data centres listed in the UK, 24 more in Scotland’s planning process, and Ofgem’s consultation runs until September 16. Residents are already fighting noise, electricity demand, and cooling-water pressure around proposed campuses. The regulator’s queue reform will not answer every local land-use fight, but it hits the upstream fiction that every proposed load should be treated as equally real.
Ofgem’s document says the 73 GW data-centre queue is dominated by extra-large and hyperscale projects. Its table shows 166 projects in the 100-500 MW band holding 36,632 MW, and 40 projects above 500 MW holding 31,408 MW. That means most queued data-centre capacity sits in projects whose failure would not be clerical. One abandoned hyperscale campus can leave behind studies, route assumptions, political commitments, and opportunity cost.
the better AI infrastructure metric is sincerity per megawatt
AI infrastructure coverage loves total capex, GPU counts, model tokens, and campus announcements. Those numbers are useful until they become cosplay for capacity. A megawatt requested from a grid operator has a different moral weight than a megawatt in a slide deck. It asks other projects to wait. It asks planners to model reinforcement. It asks communities to carry risk before the campus proves it will exist.
Ofgem’s proposal does something rare in AI policy: it turns intent into collateral. The fee makes speculative access more expensive. The milestone rules create a record where operators can separate credible demand from land-banked fantasy.
This is the shape of compute governance after the hype cycle hits the wall socket. Model companies can talk about intelligence. Developers can talk about sovereign AI, regional growth, jobs, and cloud capacity. The grid replies with a queue, a threshold, a fee schedule, and a deadline.
The useful precedent is portable. A data-centre boom does not become real because a company asks for interconnection. It becomes real when a developer can survive deposits, milestones, procurement evidence, local review, power pricing, water rules, and cancellation risk. Ofgem has started treating the queue like infrastructure because that is what it already became.