Big promises, no plug: why Britain's AI ambitions are stuck in the queue

In January 2025 Keir Starmer launched the AI Opportunities Action Plan and set out an ambition for Britain to become an AI superpower. AI Growth Zones would speed up planning and give data centres better access to the grid, starting at Culham in Oxfordshire. That September came the Tech Prosperity Deal with the US, with £31bn of pledges from American tech firms and a second Growth Zone in the North East. Wales and Scotland followed. The October 2026 Budget added the Data Centre Covenant and a target of 6GW of AI data centre capacity by 2030.

The headlines have been big. The detail needed to deliver them has not kept up. Right now a data centre developer cannot apply for a new transmission connection, and nobody can say when that will change.

None of this is for lack of effort, and the ambition is genuine. But investors look at what they can actually build and when, and many are now looking at the Nordics, Iberia and France instead.

The connections mess

Start with the queue. Between November 2024 and June 2025, contracted demand went from 41GW to 125GW. Data centres make up about 73GW of that, across 315 projects. Peak demand for the whole UK in 2025 was roughly 45GW. Nobody thinks all of it gets built, and Ofgem is right that speculative applications clog the system.

The response has been slow and messy. NESO and the network companies have admitted “considerable uncertainty for customers” and had to reset their timeline in February 2026. Gate 2 offers have slipped, most recently to 31 October. The latest timeline published by NESO did not confirm when the next application window would open.

So the door stays shut. The working assumption is that Ofgem rules on its data centre commitment fee in Q1 2027 and the window follows. Some in the market now think it slips well into 2027. If so, a developer who found a good site in early 2025 may not even have applied for power until late 2027. Then the real wait starts.

The Covenant adds a Ministerial Control Centre, chaired by the AI minister, that can reserve and reallocate capacity and transfer pre-2030 connection contracts to “strategic” projects. Some of that could help. But “strategic” still has no definition, and meeting the Covenant’s tests gives you no right to a faster connection. Knight Frank’s Darren Mansfield put it bluntly: the Covenant “does not itself create new grid capacity”, and prioritising some projects “risks creating a two-tier market”.

How the neighbours do it

Look across the water and the contrast is stark. Ireland, which has its own grid crunch, now makes new data centres of 10MVA or more bring dispatchable generation or storage matching their import capacity, and source 80% of annual demand from new Irish renewables within six years. Tough, but at least developers know the rules.

France is actively selling. Its power is 95% low-carbon and it exported a record 89TWh net in 2024. It offers large data centres a potential tax rebate of €10.5/MWh. The state, RTE and Business France have identified 63 sites, including five fast-track locations with more than 700MW of connection potential within three to four years.

Ofgem’s package does none of that. It makes entry dearer and offers nothing back. That isn’t really Ofgem’s job, to be fair. But somebody in government has to give investors a reason to come, and on price the UK starts last: DESNZ figures put UK industrial electricity prices at the top of the IEA table for 2024. Business users paid about £259 per MWh in 2025, against roughly £130 in France and about £64 in Finland, the cheapest in the EU. With power passed through to customers, that gap goes straight onto their bills. Price alone rarely determines where tenants take space and capacity, but it is always in the mix.

A reprieve from America, for now

Here is the odd bit of luck. The US midterms on 3 November became a vote on data centres. Democrats took both chambers and flipped governorships in Texas, Georgia, Ohio, Iowa and Nevada, states holding about 37% of planned US capacity. Goldman Sachs reckons around 45% of planned US capacity could face moratoria or tighter rules.

Texas matters most, with 6.5GW under development. Greg Abbott had already paused approvals for a grid audit. His successor, Gina Hinojosa, says operations need to be paused until new protections are in place. In Ohio, at least 14 municipalities voted through bans or restrictions. Of 41 local ballot measures across 10 states, data centres lost almost all.

Politicians on this side of the Atlantic will have noticed. The backlash is already here. In Scotland it has paused development, and there have been protests and petitions across the UK, including in Northumberland. Ireland has had its own rows about data centres and their power. Data centres have become a proxy for AI itself, which understandably worries a lot of people. As the midterms showed, opposing them wins votes more easily than backing them, so most politicians would rather keep their distance.

That could mean a breathing space for Europe. Capital heading for Texas or Ohio will look elsewhere. It may not last: Eurasia Group, the political risk consultancy, thinks Hinojosa’s powers are limited. And a reprieve only helps if you can use it. If our window is shut, the UK’s paralysis just matches theirs, and the money goes to Paris, Madrid or Helsinki.

Enough power, wrong places

This question is often framed the wrong way. Britain regularly makes more clean power than it can use. The problem is moving it.

Most data centres sit around London and the South East. Most of the wind is in Scotland. The main border crossing carries about 3.4GW. When it is full, Scottish wind farms are paid to switch off and southern gas plants are paid to switch on. In 2025 more than 8.8TWh of northern Scottish wind was turned down, costing almost £300m, plus around £1bn to replace it.

NESO’s chief executive says AI data centres in Scotland could soak that up, and NESO reckons up to 20% of future data centre demand could go north with strong enough locational signals. Fibre, water and customer preference still pull south. So there is power. It is just stranded behind a network that is years behind. Government could do much more to make northern sites genuinely attractive.

Who can afford to play

This is the bit that worries me most. Ofgem’s proposed fee applies at 40MW and above and runs from £237,500 to £712,500 per MW. A 100MW scheme posts £23.8m to £71.3m when it accepts its offer. At 500MW it is roughly £118.8m to £356.3m. That sits on top of normal securities, with no relief as drafted for force majeure or change in law.

Then the milestones. Within six months you must pick a lane and show offtaker interest. By the planning milestone you must commit to long-lead kit like transformers. Later you need investment-grade credit and, for lease projects, a binding contract for 20% of the compute. Miss one and you can be thrown out of the queue.

This is not only about money. The milestones risk tilting the market towards the few players able to guarantee utilisation at the back end. In practice that means the hyperscalers. If not them, then only those with pockets deep enough to absorb the loss of their security deposit, however painful that may be. The milestones also overlook the role that speculators and development de-riskers play in the life cycle of a project. Not all capital can take planning and energy risk, but some has to. The queue needs cleaning up, but that cannot mean removing every speculator, or the result could be a bigger dry-up of projects in the long run, with no one willing to put capital at risk at the front end. The rules need to be very clear, weeding out speculative projects with low prospects of success without barring speculative consenting for genuine data centre projects.

The sequencing is backwards. Hyperscalers and neoclouds rarely sign until a developer has a firm power date. Lenders don’t fund until there is a customer and a planning consent. Ofgem wants the cash and the kit commitments first. Science Works, an independent science and technology policy group, found 42% of proposed UK capacity has no announced tenant, against 16% in the EU. That is a lot of exposed pipeline.

The Data Centre Alliance puts it well: “financial capacity is not the same as project viability”. A well-funded speculator can post the security. A credible mid-sized developer may not, especially when delays aren’t its fault. Stack the fee and the milestones on top of everything else, and the field narrows fast.

The industry is not looking for a free ride. The Covenant’s community and environmental commitments will add cost, but they are costs much of the industry is willing to take on, recognising that it has an image problem and wanting to show the positives it brings. The aim should be that the community investment arm of the Covenant directly benefits the communities in which data centres sit. That would help with public perception and show that the industry is willing to invest in these commitments, rather than treating them as a burden. The objection is to pointless cost, not to cost at all.

As the field narrows, that leaves the hyperscalers and the biggest private equity and infrastructure funds. Even some of their backers will baulk at staking nine figures on a site with no planning and no customer. UK colocation providers, which the DCA notes have carried our digital lives for decades, get squeezed out.

What this means for sovereign AI

Healey sold the Covenant as a way “to ensure our most sensitive data stays under British control”. The extra £60m for the Sovereign AI Fund is welcome.

But compute needs buildings and power. If the only people who can fund a large UK connection are US hyperscalers and a few global funds, the physical layer of British AI will be owned and controlled from elsewhere. You don’t grow domestic champions with a connection regime only the biggest foreign players can afford.

Ofgem has a real problem and doing nothing isn’t an option. The fix is calibration. Phase the deposit so it builds up with planning consent and customer commitment instead of landing in full on day one. Let credible developers without investment-grade ratings use other security, and line the offtaker milestone up with when offtakers actually sign. Give relief for delays developers can’t control. Most of all, publish a date for the next window and hit it.

America may have handed Britain a year or two while its biggest markets argue with themselves. It would be a shame to spend it with our own door shut.