Announced against financed: Britain's data centre grid queue

Britain's data centre connection queue measures applications, not investment. The distinction is now doing real work in planning and regulation.

Author: NM AI Research (ORCID: 0009-0003-4213-7769) · Version: 0.1 (draft) · Date: 2026-08-01 · Licence: CC BY 4.0

Method lineage: applies the announced-versus-financed frame from Contingent vs Robust AI Power Demand (concept DOI 10.5281/zenodo.20559430) to a UK denominator.

1. The figure in circulation

Great Britain's electricity demand connection queue grew from 41 GW in November 2024 to 125 GW in June 2025. Ofgem publishes the split: 17 GW transmission and 24 GW distribution at the start, 97 GW and 29 GW at the end. For scale, peak electricity demand in GB on 11 February 2026 was 45 GW.

A queue approaching three times national peak demand is a striking number, and it has carried a good deal of commentary about whether Britain can power the AI build-out. The number itself is not in dispute. What it measures is.

2. What the queue counts

A connection queue records requests for capacity. It does not record capacity that anyone has committed to build. Developers apply at multiple points, hold options, and revise. This is ordinary, and a queue containing speculative entries is not by itself evidence of anything improper.

Two features of the current queue are less ordinary.

The first is composition. Ofgem's own analysis, published in its July 2026 consultation, puts approximately 73 GW of the demand queue as data centres, across roughly 315 projects ranging from 1 MW to 1,500 MW. The concentration is heavy: 166 extra-large projects (100 to 500 MW) hold 36,632 MW and 40 hyperscale projects (above 500 MW) hold 31,408 MW. Those 206 projects, under two thirds of the count, account for 93.2% of the data centre capacity in the queue.

The second is movement between categories. Ofgem records that between May 2024 and August 2025, at least 9 GW in the transmission queue changed its connection request from battery technology to data centre. A queue position is an asset, and its stated purpose can be revised.

3. The financed slice, and why it has two denominators

The relevant question is how much of the queue carries a Final Investment Decision. Ofgem's February 2026 call for input reports the answer, drawing on NESO's Demand Queue Call for Input. The call for input identified around 140 data centres representing 50 GW. Of those, 71 projects representing around 20 GW reported that they had achieved financial commitment with FID.

Two features of that sentence bear on how the ratio should be read.

The FID figures are self-reported. Developers told the system operator they had reached a final investment decision. The number is an assertion by the party with the strongest interest in holding its place in the queue, and it was not independently audited. If it is wrong in either direction, the more likely direction is upward bias.

Ofgem reads the same figure as a sign of maturity, not of speculation. The call for input describes 71 of 140 as "indicating a significant volume of mature data centre projects in the demand queue". Anyone citing the figure as evidence of a soft queue is departing from the regulator's reading of it.

With that stated, the arithmetic. Against the call for input's own population, 20 GW of 50 GW is 40% financed, and 71 of 140 is 50.7% by project count. Against Ofgem's later and larger figure for the whole data centre queue, the same 20 GW is 27% of 73 GW, and 71 of 315 is 22.5% by count.

Both are defensible, and they are five months and two populations apart. The 140 and 50 GW come from NESO's call for input, which covered transmission-level demand connections and counted respondents. The 73 GW and 315 projects come from Ofgem's own analysis of the full queue across transmission and distribution. Nobody has published an updated FID count against the larger denominator.

The financed share is therefore a band rather than a point. Somewhere between roughly a quarter and two fifths of queued data centre capacity carried a claimed final investment decision when it was last counted, and the more recent the denominator, the smaller that share looks. No published source narrows it further.

4. The regulator is already pricing the gap

Ofgem has published its own assessment of what the queue implies, and it is more direct than anything a third party would need to add.

Its July 2026 consultation converts the pipeline into money. At an assumed £9.5m per MW, the 73 GW of queued data centres implies approximately £693 billion of total capital expenditure, which Ofgem puts at around 23% of UK GDP in 2025. It sets this beside Bloomberg's February 2026 estimate of roughly $650 billion of global AI infrastructure capital expenditure in 2026 by Alphabet, Amazon, Meta and Microsoft combined.

The comparison is Ofgem's own, and it needs stating precisely: the two figures sit in different currencies, and Ofgem does not convert them. Taken at face value, the implied capital cost of Britain's data centre connection queue exceeds the combined worldwide annual infrastructure spending of the four largest buyers of that infrastructure, and the currency difference widens the gap rather than narrowing it. Ofgem introduces the comparison explicitly as a test of "the feasibility of this pipeline of data centre development materialising and the volume of possible speculative, non-viable projects".

Ofgem is acting on that reading. On 29 July 2026 it opened a consultation proposing a data centre commitment fee of between £237,500 and £712,500 per MW, equivalent to around 2.5% to 7.5% of average project costs, alongside data centre specific queue management milestones. Developers would pay on accepting an offer and recover the fee once energised, forfeiting it if they leave early. The consultation closes on 16 September 2026.

Ofgem is also explicit about the limits of its own case. Its consultation accepts there is "no evidence to suggest" that data centres carry a lower rate of non-viable applications than other sectors, and rests the argument on the scale of the sector's demand rather than on data centres being unusually speculative.

5. Land is being allocated against the announced figure

While the financed share remains uncounted against the current denominator, the announced figure is being used to make decisions that are hard to reverse.

The draft London Plan, published on 16 July 2026 and open for consultation until 15 October 2026, gives data centres a dedicated policy for the first time, GLE3. Data centres are removed from the Plan's definition of industrial land. More consequentially, the Green Belt policy gives them their own sustainable location test: under Box PV7A, data centre proposals qualify where the site uses grey belt land or land at the edge of a residential-led growth location offering low-carbon heat opportunities.

Press coverage has described this as making data centres the only exception to Green Belt controls. That overstates it, and the accurate version is narrower and still notable. Data centres are not exempted from Green Belt policy; they are given a test written for them, distinct from the tests applied to housing and to industry. The Plan is explicit that this "differs from the approach to industrial uses, such as storage and distribution", where heavier traffic generation justifies a more targeted approach.

This may well be justified on its merits, and the Plan attaches substantive conditions on heat recovery, water and renewable supply. What is notable is that the need being met is evidenced substantially by demand figures whose financed component has not been re-measured.

6. The first legal test has already landed

A hyperscale data centre of 90 MW and up to 72,000 square metres was proposed on the former Woodlands Park landfill site at Iver, in the Buckinghamshire Green Belt. Buckinghamshire Council refused it in June 2024. The appeal was recovered, and permission was granted on 9 July 2025 by the Minister of State on behalf of the Secretary of State.

Foxglove and Global Action Plan brought a statutory review under section 288 of the Town and Country Planning Act 1990. On 19 January 2026 the Government Legal Department wrote accepting a "serious logical error" and that the decision should be quashed: the screening decision had relied on mitigation measures to conclude the scheme was not EIA development, without those measures being secured by condition or obligation. Permission to bring the claim was granted on all grounds on 22 January 2026. In April 2026 the developer accepted that the environmental commitments would be made binding through a contract with the council.

The government's response was to concede. Whatever the eventual planning outcome at Iver, the mechanism by which announced mitigation becomes enforceable mitigation has now been tested once, and it failed the first time it was examined.

7. The pattern is not domestic

Britain is not unusual in this. On 30 July 2026 the European Commission opened bidding for up to seven AI gigafactories, with a headline of around €10bn in public funding against roughly €30bn in total investment. The figure Brussels can commit under the current budget is reported as approximately €1bn, with the balance dependent on the next Multiannual Financial Framework, which member states are still negotiating. The structural limit is firmer than the headline either way: Council Regulation (EU) 2026/150 provides that the Union contribution "shall cover up to 17 % of the capital expenditure (CAPEX) investments in the overall computing infrastructure of the AI gigafactory", with participating states required to at least match it. A headline figure that is capped by regulation at a minority share, and committable today at a tenth of its announced value, is the same distinction, drawn in euros by the announcing body rather than in gigawatts by a regulator. The gap between announced and financed is not a British artefact of how the connection queue is built.

8. What would change this read

The queue measures applications, the financed share is smaller and less precisely known than the headline implies, and decisions are being taken against the headline. Three developments would falsify that.

If the FID share rises sharply once the commitment fee applies, the queue was carrying better information than this piece credits, and the deflation was wrong.
If an updated FID count against the 73 GW denominator lands near 40% or above, the band above is too pessimistic and should be withdrawn.
If the queue converts, that is, if connected and energised data centre capacity begins tracking the queue rather than the financed slice, the announced figure will have been the better forecast.

The first of these resolves on a known date. Ofgem's consultation closes on 16 September 2026, and a further consultation on self-build and ownership of high voltage transmission assets is expected in the autumn.

Verification: the 17% ceiling on the Union contribution is quoted from Council Regulation (EU) 2026/150 of 16 January 2026 (OJ L, 19.1.2026), amending Regulation (EU) 2021/1173. The €10bn headline, the €30bn total and the approximately €1bn committable under the current budget are from press coverage of the Commission's 30 July 2026 announcement and an unnamed Commission official's briefing, not from a Commission document, and should be treated as reported rather than primary. Queue and FID figures traced to Ofgem, Demand Connections Reform call for input, published 13 February 2026, paragraphs 2.5, 2.7 and 2.8. Data centre queue composition, capital expenditure arithmetic and the battery-to-data-centre reclassification traced to Ofgem, Proposed data centre connection reforms (Curate), 29 July 2026, paragraphs 2.6, 2.7, 2.8, 2.10 and Table 1. Commitment fee range traced to the same consultation and Ofgem's accompanying press release. Green Belt policy traced to the draft London Plan 2026, policy GLE3, paragraphs 4.36 and 4.37 and Box PV7A. The 27% and 22.5% ratios are derived, not reported, and combine a February 2026 FID count with a July 2026 denominator drawn from a different population; no published source states them. The Iver chronology is from the claimants' and their solicitors' public statements and contemporaneous coverage, not from a court judgment; no final order quashing the permission was located, and the status should be checked before relying on it.

Disclosure [1]: the author holds no position in, and receives no compensation from, any entity named. No third party reviewed, funded or directed this work. This piece was produced with the assistance of an AI model made by Anthropic, a company operating in the sector whose projected electricity demand the piece examines.