Mark Moore


Articles

Departures, part 2: is any train in Europe leaving on time?

Published 6 November 2026 · Read the post · Read part 1

In part 1 the board was Euston’s, and London’s data centre train was showing DELAYED. If you want the one genuinely fast service out of London, though, you don’t go to Euston at all. You walk round the corner to St Pancras and catch the Eurostar. Read the full piece

That’s roughly where this edition ends up. The board is FLAP-D plus two: Frankfurt, London, Amsterdam, Paris and Dublin, the five markets the industry groups as FLAP-D, with Oslo added for the Nordics and Scotland added because Holyrood has put its own service on hold. Same board, same columns: the date that was scheduled or targeted, the date the latest public evidence says to expect, and a remark. The carriages show how full each market already is, using CBRE’s colocation vacancy for the first quarter of 2026. Where CBRE doesn’t publish a figure, the carriages are left blank.

Guy Massey’s US arrivals board started this series, and it’s still the best format I’ve found for a queue.

The board

The departures board for FLAP-D plus Oslo and Scotland: scheduled or targeted date, expected date, remark and CBRE colocation vacancy for each market

London: DELAYED

I covered London in part 1, so briefly: NESO’s reformed pipeline holds more than 100 GW of large demand, most of it data centres, and around 13 GW of that can connect before 2030. A further 86 GW is slated for 2030 to 2035. Part 1: https://lnkd.in/p/egqJGnD8

The Caledonian Sleeper: HELD

The Caledonian Sleeper leaves from Euston too, and Scotland’s data centre service is currently held at the platform. This hold is in the planning system. On 16 September the Scottish Parliament rejected a Green motion for a moratorium, but agreed an amended motion, 80 votes to 26, which says no decisions on hyperscale applications should be made until national planning guidance is complete. The Scottish Government doesn’t accept that this is a moratorium. Its Minister for Public Finance has since told Holyrood’s Climate Action Committee that the guidance will be published before Christmas, and that the government doesn’t expect any applications to be decided before then.

Two directions now apply. Since August, planning authorities must notify Ministers of any data centre application above 50 MW, and since 17 September those applications have required a full environmental impact assessment. Around two dozen hyperscale applications are in the system. Scotland’s installed data centre capacity was estimated at 10 to 30 MW in 2024.

Frankfurt: DELAYED to the mid-2030s

Frankfurt’s distribution operator, NRM Netzdienste Rhein-Main, no longer works on first come, first served for large requests. Anything of 3.5 MW or more goes into a once-a-year pro-rata allocation, and at the West grid node the threshold drops to 1 MW until a new substation at Eschborn is commissioned. NRM’s own words are that demand clearly exceeds what the upstream transmission grid can supply, and it names the boom in data centre requests as one of the reasons. It expects the restrictions to last until the mid-2030s, when new TenneT supply points at Eschborn and Ostend come in. Mainova, NRM’s parent, says it and TenneT plan around 1,000 MW of additional supply capacity in the next few years, without a date.

One operator has had the answer in writing. Firstcolo was reportedly told that power for its planned Frankfurt expansion would not be available before 2035.

Frankfurt also adds conditions on top of the wait. New data centres coming into operation from July 2026 must meet a design PUE of 1.2 and source at least half their electricity from renewables under Germany’s Energy Efficiency Act, and the city restricts where data centres can be built through its own data centre concept. A bill to relax parts of that Act, giving new sites four years rather than two to meet the efficiency standard and moving the full renewables requirement from 2027 to 2030, has had its first reading in the Bundestag but has not yet passed.

Amsterdam: DELAYED from 2031 to 2036

In January, Liander published the results of TenneT’s recalibrated congestion study for Noord-Holland. For the Vijfhuizen part of the network, which serves Amsterdam Nieuw-West, Zuid and the Zuidas, the expected connection date moved from 2031 to 2036. Diemen moved from the end of 2036 to the end of 2037. Around 230 Amsterdam customers are waiting across those two areas.

It isn’t all one way. Oostzaan moved a year earlier, to 2029 to 2030, and Liander expects to grant around a fifth of the requests waiting there. In April a court in Arnhem upheld TenneT’s decision to put a 70 MW data centre request near Schiphol on the waiting list.

Planning is tight as well. Amsterdam’s stance is “no, unless”: new or expanded data centres over 2,000 m² or 5 MVA are not permitted under its zoning, and nationally, hyperscale sites of 10 hectares and 70 MW or more have been confined to two locations, Eemshaven and Middenmeer, since January 2024.

Amsterdam has the most spare space on the board at 9.4%, which can look odd for a market this constrained. It isn’t a contradiction. Vacancy measures halls already built and available to rent; the grid governs what gets built next.

Dublin: SUSPENDED

EirGrid’s May 2026 overview says it plainly: the greater Dublin region is considered fully constrained. Ireland has reopened applications under the regulator’s new large energy user policy, decided in December 2025, but reopening applications isn’t the same as reopening capacity. The new policy also changes what a data centre has to bring. Sites of 10 MVA or more must provide dispatchable generation or storage, on site or nearby, equal to their full import capacity, and match 80% of their annual demand with new Irish renewables over six years. The policy is now the subject of a judicial review.

The scheduled date on the board, 2028, is the date reported in 2022 for the end of EirGrid’s Dublin pause. There is no expected date to put against it.

Paris: the Eurostar, EXPRESS for fast-track sites only

Paris is the one service on the board running early, and it’s worth being precise about why. In May 2025 France’s regulator, CRE, approved a fast-track procedure under which the State identifies large sites, 400 MW to over 1 GW, whose connection to the 400 kV network isn’t limited by grid capacity, with a 2028 to 2029 target. RTE lists five such sites, two in Île-de-France and three in Hauts-de-France, and four had been allocated by Choose France in May 2026. The first fast-track contract, signed with RTE in January 2026 for the Campus IA site at Fouju, provides 240 MW by the end of 2027, ahead of the target, and 700 MW before the end of 2029.

Off the fast track, France looks much more like everyone else. RTE had reserved nearly 18 GW for around 80 data centre projects by May 2026, against about 5 GW at the end of 2024, with around 7 GW in Île-de-France. Operators report ordinary connection waits of two to seven years.

There’s also a detail in RTE’s figures that every grid planner will recognise. Recently connected data centres draw on average about 20% of the power they asked for, and RTE expects it to take 10 to 15 years for them to reach around 80%.

So the express runs because the State chose the route and the stations before the passengers arrived.

Oslo: a cheaper ticket, but the platform is filling

I started this edition expecting the Nordics to be the train running on time. The evidence doesn’t support that. Statnett’s own median times from project start to operation are four to six years for a transformer upgrade, five to ten for a new substation and seven to twelve for a new line. Mature applications for consumption rose by around 2,000 MW year on year in the first half of 2026, and Statnett describes the queue as growing sharply, while 600 MW of new consumption was connected over twelve months. Stricter maturity rules for new applications came in from July.

Elsewhere in the region, Denmark’s Energinet paused new connection agreements in March with around 60 GW of consumption queued against peak demand of around 7 GW, and Finland is drafting rules to replace first come, first served. The Nordic offer is cleaner and cheaper power. It isn’t a faster connection.

What the board says about the UK

My reading: every market on this board that queues by order of arrival shows DELAYED or worse. The one service that runs early is the one where the State picked the sites first and built the line to them.

The UK has the pieces of that. AI Growth Zones designate sites, and Gate 2 reorders the queue by readiness and need. Whether they add up to a Eurostar, or just a revised timetable for the same trains, is the question for the next twelve months. While the answer comes, each month a data centre waits costs its investors between £65,000 and £105,000 per MW of IT load, the floor I set out in part 3 of “Who pays for the wires?”. Part 3: https://lnkd.in/p/eFCyB6vb

Should the UK pick its fast-track sites the way France has, or keep letting the queue decide?

Sources

NESO results of connections reform, via DCD, 9 December 2025

NRM Netzdienste Rhein-Main, Pro-Rata-Verfahren Strom (2026/27 allocation round)

Mainova, Stromnetz (grid investment and 1,000 MW plan with TenneT)

Energieeffizienzgesetz (EnEfG), section 11

Deutscher Bundestag, first reading of the Energy Efficiency Directive implementation bill (21/8027), 24 September 2026

Stadtplanungsamt Frankfurt, Rechenzentrenkonzept (2022)

Liander, Langere wachttijden in Noord-Holland maar ook eerder ruimte voor deel Amsterdamse klanten, 15 January 2026

Gemeente Amsterdam, Vestigingsbeleid datacenters 2020–2030, eerste herziening (2026)

Tweede Kamer, Kamerstuk 26643-819 (hyperscale data centre locations)

EirGrid, Data Centre Constrained Area Overview, May 2026

CRU, Large Energy Users Connection Policy decision CRU2025236, 12 December 2025

ClientEarth, press release on judicial review, 27 April 2026

CRE, fast-track procedure for large consumers, 7 May 2025

RTE, Les data centers en chiffres clés (updated 1 June 2026)

RTE, accelerated connection contract with Campus IA (Fouju), press release, 26 January 2026

Statnett, How to get capacity for direct customers

Statnett, half-year report 2026

Energinet, temporary pause in new grid connections, 2 March 2026

CBRE, Global Data Center Trends 2026 and European Data Centres Figures Q1 2026

Mark D. Moore, “Who pays for the wires?” Part 3, LinkedIn, 9 October 2026

Mark D. Moore, “Departures: where the UK’s data centre trains actually stand”, LinkedIn, 30 October 2026

Scottish Parliament, motion S7M-01177 as amended, 16 September 2026

Scottish Government, Chief Planner letter and EIA Direction, 16 September 2026; Notification Direction, 17 August 2026

Minister for Public Finance to the Climate Action Committee, 8 October 2026

Reported, not primary: Firstcolo (dpa via Energie & Management, 18 March 2026); Arnhem court ruling (NL Times, 29 April 2026); Paris ordinary waits (operator statements, 2026); Finland (Borenius, 26 August 2026); Dublin 2028 (DCD, 11 January 2022); Scotland autumn 2026 decision timing (DCD, 16 September 2026); two dozen Scottish applications (CIBSE Journal, 1 October 2026); Scotland 10 to 30 MW installed (DSIT and Scottish Enterprise, via APRS, December 2025).

Provenance: Synthesis, with original comparison across markets.

Earlier posts and articles are archived at markdmoore.co.uk

Graphic accompanying the article: Departures, part 2: is any train in Europe leaving on time?

Wholesale CfD’s timing trap: Give notice first, find out later

Published 5 November 2026 · Read the post · Read the paper

Last Wednesday, DESNZ and the Scottish Government published a consultation on Wholesale Contracts for Difference. It came up at LCCC Connects in Birmingham on Tuesday, and having read the consultation and its analytical annex since, I think it deserves more attention than it has had so far. It closes on 9 December. Read the full piece

What’s on offer

The offer is for generators that already exist. A wind farm, solar farm or run-of-river hydro scheme that isn’t on a CfD could swap the wholesale part of its revenue for a fixed strike price. When the day-ahead reference price is below the strike price the generator is topped up, and when it’s above, the generator pays the difference back. Under DESNZ’s lead option, ROC income carries on unchanged and the RO term isn’t extended. Contracts of 5 or 10 years are proposed, allocated through a one-off sealed-bid, pay-as-clear auction and funded through the supplier levy, the same route as the existing CfDs. DESNZ assumes payments start in 2028.

The pool is large. At a 5 MW minimum, up to 29.8 GW across 1,063 sites could bid, of which 868 sites are on the RO and 195 are merchant. A 10 MW threshold would cut that to 26.9 GW across 666 sites. DESNZ is clear that these are ceilings on eligibility, not forecasts of take-up. Biomass, anaerobic digestion, energy from waste and storage are proposed to be left out, and the existing nuclear fleet may be offered a bilateral deal of its own.

Why the government wants it

On DESNZ’s estimate, gas set the marginal power price around 90% of the time in the early 2020s and could still do so about half the time in the early 2030s. Every megawatt-hour the existing fleet sells at market is paid that gas-linked price. In 2022/23 the average intermittent reference price reached about £190/MWh. Under a two-way contract, the money above the strike price would have gone back to consumers.

There’s a stick alongside the carrot. The Electricity Generator Levy rises from 45% to 55% and will now run for as long as the longest WCfD, while a generator holding a WCfD is exempt for the life of its contract.

What an existing asset bids

A new wind farm bidding for a CfD has to recover its construction cost through the strike price. For a farm built ten years ago that capital is already spent, so its bid should come down to the wholesale revenue it expects over the contract term, less whatever it’s prepared to give up for certainty. That discount is the consumer’s gain, and only the auction will show how big it is. The analytical annex doesn’t try to estimate it. It’s entirely qualitative, with no monetised savings, no gas price scenarios and no assumption on take-up, and it says the numbers will follow.

Four details that decide whether it works

1. The Capacity Market timetable. A generator can’t hold a Capacity Agreement and an allocation-round CfD at the same time. To switch, it has to give a transfer notice at least 16 months before the delivery year starts. DESNZ’s own example: for the delivery year beginning 1 October 2028, the notice is due by 1 June 2027. The auction is expected towards the end of 2027. So the generator gives up its capacity payments first and finds out whether it has won a WCfD several months later. DESNZ says it’s looking at a shorter notice period for the WCfD, and it needs to, or I’d expect most Capacity Agreement holders to sit the round out. The consultation isn’t consistent on the date either: one section expects the round towards the end of 2027, another has contracts awarded in 2028.

2. Start dates. DESNZ is wary of flexible start dates, because a generator could wait out a spell of high prices before starting and then miss the low prices at the far end. That risk falls away if the end date is fixed. Every 10-year contract would end on the same day, and a generator that starts later simply gets a shorter contract at the same strike price. A window of up to twelve months would leave room for existing PPAs and Capacity Agreements to run off, and waiting would cost the generator something.

3. Two contract lengths, one clearing price. In a pay-as-clear auction every winner is paid the price of the last bid accepted. A 5-year bid and a 10-year bid are pricing different periods, 2028 to 2033 against 2028 to 2038, so ranking them together sets one price for two products. I’d clear each length separately.

4. A price set after the bids are in. DESNZ is considering not publishing the administrative strike price or the budget. The Secretary of State would look at the anonymised bid stack and decide how much of it to buy, and in pay-as-clear that also sets the strike price for every winner. I can see why government wants the flexibility. In return, the test it will apply should be published before the round, and the anonymised bid stack after it. Without the bid stack, the discount generators accepted can’t be measured, even by DESNZ’s own evaluation, and that discount is the whole value-for-money case.

Who else should be watching

DESNZ notes that a WCfD may put generators off long-term PPAs, because its payments are referenced to the day-ahead price. That tends to be discussed as a problem for suppliers’ hedging. It matters to buyers too. An operating wind farm with years of output data and no construction risk is about the best counterparty a large user can find for a fixed-price contract. Data centres, battery projects that need revenue floors, and industry that is electrifying all draw on that same pool, and their alternative is a new-build project with a connection date that may move. If much of the existing fleet moves onto WCfDs, that supply thins out just as demand for it grows. Letting generators bid only part of a site’s output, which DESNZ is also considering, would keep some of it available.

Where I’ve landed

I’m in favour, and I sent my response to DESNZ on 4 November. It makes the four points above, supports the 5 MW threshold, and asks for the monetised analysis to be published before the auction framework. The full response is on the White Papers page at markdmoore.co.uk.

If you hold a Capacity Agreement on an RO wind farm, would you give it up in June 2027 without knowing the auction result?

Sources

1. DESNZ and Scottish Government, Wholesale Contracts for Difference: consultation, 28 October 2026, gov.uk (Sections 1 to 5; Tables 1 to 3)

2. DESNZ, Wholesale Contracts for Difference: analytical annex, 28 October 2026, gov.uk

3. Electricity Capacity Regulations 2014, regulation 34, as described in source 1, Section 4

Provenance: Analysis. The fixed end date and the separate clearing of contract lengths are my own proposals.

Earlier posts and articles are archived at markdmoore.co.uk

Graphic accompanying the post: Wholesale CfD’s timing trap: Give notice first, find out later

Not all data centres are the same, Part 1: what they pay

Published 3 November 2026 · View on LinkedIn · Read the post

I’m at LCCC Connects in Birmingham today, so this seems the right morning to look at three of the levies that sit on every business electricity bill: the Renewables Obligation (RO), Feed-in Tariffs (FiT) and the Capacity Market (CM). Read the full piece

From April 2027, around 10,000 manufacturers stop paying the RO and FiT costs, and from October 2027 the CM cost, under the British Industrial Competitiveness Scheme (BICS). Data centres aren’t in it. The scheme’s eligible codes are manufacturing, and data processing and hosting (SIC 6311) isn’t on the list.

So what do the three levies cost a data centre that keeps paying them? It depends less on how big the site is than on how much of its booked capacity it actually uses, and when.

This year’s rates

The RO costs £32.73 for every megawatt-hour a supplier sells this year: 0.472 certificates per MWh at the £69.34 buy-out price. FiT adds about £7.16 per MWh, which is £1.73bn of scheme costs spread over 241 TWh of supply in the last year Ofgem has reported.

The CM works differently. It is charged only on electricity drawn between 16:00 and 19:00 on working days from November to February, 246 hours this winter. This year’s £3.47bn of capacity payments, spread over 11.3 TWh drawn in those hours, comes to £307.71 for each MWh drawn in the window. Averaged over a flat load across the whole year, that’s £8.64 per MWh.

Together, a site with a flat load pays £48.53 per MWh. The government puts what BICS is worth to eligible firms at around £35 to £40 per MWh. The two figures measure different things: the government’s is an average across the scheme from 2027, including firms on half relief and later years when the RO and FiT costs fall.

What 100 MW pays depends on what it draws

The levies follow the energy drawn, not the capacity booked. For 100 MW of booked capacity, each percentage point of load adds about £0.42m a year.

Table: the three levies a data centre pays per year for 100 MW of booked capacity, by share of the booking it draws

Published analyses of UK Power Networks’ half-hourly data put the average data centre at about a fifth of its booked capacity. At that level, 100 MW booked pays about £8.5m a year in the three levies. National Grid’s distribution study and NESO’s long-term scenarios plan on 70%, which would be £29.8m. Flat at the full booking it would be £42.5m. Part 3 of this series, on 30 November, looks at what data centres book and what they draw.

So today’s levy bills are well below what a full site would pay, and they rise as halls fill.

When the training run pauses

The CM is the levy that varies by type of data centre, because it depends on when power is used rather than how much. Lawrence Berkeley National Laboratory’s report on US data centres assumes AI training servers run about 80% of the time. A training site drawing 80% steadily, including in the winter peak hours, would pay about £6.1m a year in CM per 100 MW booked. If it stepped down for those 246 hours, its CM charge would fall to nothing, and its total from £34.0m to £28.0m a year, on a supply contract that passes the charge through at cost.

Training jobs can be paused and restarted. Sites that serve users as they arrive, such as inference and enterprise sites, have less room to move, so the same saving is harder for them to capture.

What it adds up to

From 2027 a manufacturer drawing the same power with full BICS relief would pay little or none of this, which is the policy’s intention. A data centre keeps paying, and what it pays is set more by how full it is and whether it can step down at winter peaks than by the size of its connection. Part 2 of this series, on 23 November, looks at how fast different types of data centre change their load.

If a training site can cut its levy bill by a sixth by stepping down at winter peaks, should that flexibility be something it is paid for, or only something that saves it money?

How this was counted

Rates are for 2026/27 as published. FiT uses the last full reported scheme year (April 2024 to March 2025). CM is charged per MWh drawn in Periods of High Demand, which the regulations define as 16:00 to 19:00 on working days from November to February, excluding England and Wales bank holidays. Load cases are illustrative constant shares of a 100 MW booking; the stepped-down case assumes no draw at all in the peak window. The figures leave out network charges, balancing costs and the Contracts for Difference levy, which BICS does not cover either, and assume suppliers pass the levies through at cost. BICS eligibility is from the scheme’s final code list.

Sources

GOV.UK, British Industrial Competitiveness Scheme: guidance for applicants, 10 August 2026, updated to 12 October 2026

DESNZ and DBT, Government response to the BICS consultation, 16 April 2026

BICS Annex A, eligible SIC and HS codes, final list (8 July 2026)

House of Commons Library, CBP-10315, 27 May 2026

DESNZ, Renewables Obligation level 2026 to 2027, 29 September 2025

Ofgem, RO buy-out price and mutualisation ceilings 2026 to 2027, 31 March 2026

Ofgem, Feed-in Tariff Annual Report, scheme year 15, December 2025

EMR Settlement, Key Figures for Payments 2026/27 (capacity figures 31 July 2026), checked 2 November 2026

Electricity Capacity (Supplier Payment etc.) Regulations 2014 and Electricity Capacity Regulations 2014, regulation 2(1)

Computer Weekly, analysis of UK Power Networks data centre demand data, 13 May 2026

TwentyForty, Power In, August 2026

National Grid DSO, Data Centre Impact Study, November 2025

NESO, Future Energy Scenarios 2025

Lawrence Berkeley National Laboratory, 2024 United States Data Center Energy Usage Report

Scenarica, data centre levy costs, 9 September 2026

Provenance: Analysis, my own arithmetic on published 2026/27 rates. It builds on Scenarica’s estimate of the levy stack (September 2026) and adds the levy-by-levy split by load and the effect of Capacity Market timing.

Earlier posts and articles are archived at markdmoore.co.uk

Graphic accompanying the article: Not all data centres are the same, Part 1: what they pay

Three lines to power, and the main one is suspended

Published 2 November 2026 · Read the post

As of November 2026, a developer who wants to power a new data centre from the transmission network can’t apply. NESO closed the transmission queue to new demand in March 2025 while it reissued offers under connections reform, and it still hasn’t published a date for the next application window. Everything else on the map has to be read with that in mind. Read the full piece

I’ve drawn the routes as a metro map because most of the questions I’m asked about data centre power are really route questions. Which line am I on, where does it cross planning, and which station is the slow one? There are three power lines (transmission, distribution and on-site gas) and two consent lines (local planning, and the national infrastructure regime). They cross in more places than most site plans allow for.

Each station carries a time where a source supports one, marked S for a statutory or licence period and O for an observed figure or a stated aim. There’s deliberately no total for any line. The long legs (transmission reinforcement, distribution build, most gas network builds) have no reliable published durations, and a total built on them would look far more certain than it is.

Where every line starts: land

All three power lines leave from the same station. For a transmission demand project, secured rights over all the land, through an option of at least three years leading to a lease of 20 years or more or a purchase, meet Gate 2 readiness on their own. Council land sits at this station too. A council can grant permission for development on its own land, but that permission benefits only the council and any named joint developer, which shapes how any deal with a private operator has to be written.

Transmission: service suspended

When the line does reopen, the timetable is short at the front and open-ended at the back. NESO issues an offer within three months of a complete application, and a Gate 2 offer is then open for 90 days. After acceptance come the queue milestones, counted back from the contracted connection date. For a date more than five years out, the planning application has to go in 48 months before it and consent has to be in hand 30 months before it. Then the transmission owner builds whatever reinforcement the connection needs. I’ve found no published typical duration for that leg on demand projects, which is why the station shows none.

The government’s response of 28 October to its strategic demand consultation sets out what has to happen before the next train. Ofgem’s data centre queue reforms come first, and the response puts Ofgem’s decision in the first quarter of 2027. As consulted on, they include a Data Centre Commitment Fee of £237,500 to £712,500 per MW, held from acceptance until energisation, which the government expects to apply to projects above 40 MW that connect to or affect the transmission system.

The same response confirms a priority lane, though not one a developer can join on its own. A project needs a sponsor (a government department, a devolved government or a mayoral strategic authority), goes through the Connections Accelerator Service, and ministers make the final decision. Both measures are meant to be in place before the window reopens. Later, at a window not yet named, data centre connections after 2030 are to be held to regional targets set by grid capacity, water and land.

Distribution: open, with a dotted line to NESO

DNOs kept taking applications through the pause and, as far as I can find, still are. A DNO has to make an offer within three months of a complete application, and under the Guaranteed Standards an EHV demand quotation is due within 65 working days. Since June 2026, demand applications of 5 MVA or more have had to come with a site plan, a preliminary timeline and, where the applicant doesn’t own or lease the site, heads of terms. A single-line diagram is needed where the equipment is likely to cause disturbance.

The dotted line on the map runs from the Transmission Entry Assessment back up to NESO. That’s the ENA’s working name for any case where NESO is asked to assess a distribution project’s effect on the transmission network; the formal processes in the industry code have other names, and the government’s response uses a third. A distribution connection large enough to affect the transmission network needs one, and there’s no national threshold for demand; each DNO decides case by case. For larger demand connections the ENA’s queue guide puts the NESO interface milestone 12 months after acceptance, with the planning and land milestones running from it. The government’s response is now explicit that embedded demand sits outside the gated process whatever its effect on transmission, and that it is looking at regional thresholds above which a distribution project would need an assessment and join the gated queue. No date is given. If your project depends on a large distribution connection, how that threshold is set is the question to put to your DNO.

The response also commits to flexible connections for demand: non-firm or phased agreements, and a last-resort power for NESO to direct large users to cut demand. No terms have been published, so neither appears on the map yet.

On-site gas: the gas main is the long station

Generating on site swaps one queue for others. Gas transporters must quote for a large load within 21 working days, and the build is where the time goes. Cadent aims to deliver its largest new connections within 12 months, though a new pressure-reducing governor on fresh land can take up to two years. National Gas’s indicative timeline for a bespoke connection to the national transmission system comes to about 36 months from the offer clock starting (my sum of its published stages). The gas networks’ own trade body puts new demand connections at 6 to 24 months.

The environmental permit is three months by statute for a standard permit, or four for a bespoke permit with public participation. One recent data centre permit, for standby plant, took about a year from being duly made to being issued. Above 50 MW of generation in England the plant is a nationally significant infrastructure project in its own right, so the gas line joins the DCO line at that interchange. And any grid connection for standby or export puts the site back on a grid line.

A branch line: the bridge

There’s a shorter route the main lines don’t show. A data centre that can’t wait for its full grid connection can start on a smaller on-site supply (gas engines, batteries or both), run at part load, and hand over to the grid when the grid arrives. On the map it’s a dashed branch. It crosses local planning, passes the DCO line without stopping, and joins the grid at energisation.

Size decides which stations it has to call at. Under 50 MW of generation, a plant in England stays in local planning rather than becoming a nationally significant project, and in Scotland it stays outside section 36. The environmental permit turns on a different number: the Environment Agency adds together the thermal input of every combustion unit permanently on the site, standby generators included, and at 50 MWth or more the site needs the large-installation permit. For gas engines, 50 MWth of fuel input is roughly 20 MW of electrical output (my conversion, at about 40% efficiency). On a site that already carries large standby sets, the bridge may be over that line before it starts.

The bridge doesn’t skip the queue for gas or the permit, so it isn’t instant, and it still needs a grid line in the end. What it changes is the date the first halls can run.

Where the consent lines really cross

Planning isn’t a condition of Gate 2 for demand. It becomes a queue milestone after the offer, which is why the consent lines on the map cross the grid lines after acceptance rather than before the application. I started this map assuming the opposite.

Local planning in England has a statutory determination period of 13 weeks for a major application, or 16 with an environmental impact assessment. Hoare Lea’s analysis of 33 data centre applications put the observed average at about 490 days, and I’ve not found anything more recent. The NPPF published in August asks plans to provide for data centres together with their associated generating capacity and the network infrastructure for grid connections (policy E1). It gives substantial weight to proposals that reflect AI Growth Zones, and it names the co-location of large generators with large users of power as a reason a location can matter (policy E2).

The national route has been open to data centres since January 2026 through a section 35 direction. The three directions I found took between 52 and 61 days from request. On 28 October the government laid a draft National Policy Statement for data centres before Parliament. It sets an indicative minimum of 100 MW of rated IT load for a direction, expects detailed reasons for any request below that, and gives applications that meet its requirements a presumption of consent. It is out to consultation until 8 January 2027, Parliament’s scrutiny period runs to 17 February, and it has to be laid again before it takes effect, so for now it is a draft.

Statutory pre-application consultation was removed from 24 July, which MHCLG estimates could save up to a year, though no project has yet been through the new process end to end. After submission, the statutory steps are 28 days to accept, six months to examine, three to recommend and three to decide; on this year’s Planning Inspectorate averages that’s about 16 months.

Scotland and Wales

The DCO line stops at the English border, because a section 35 direction can only be given for development wholly in England, and the draft policy statement covers England only. In Scotland, data centres go through local planning. Since 17 August councils have had to notify Ministers of any application over 50 MW, since 17 September those applications have been EIA development, and gas generation over 50 MW needs a section 36 consent from Ministers. Wales replaced its Developments of National Significance with Infrastructure Consent Orders in December 2025. Gas generation between 50 and 350 MW is consented that way, larger plant stays in the DCO regime, and the data centre itself goes through local planning. The grid lines run the same way in both, with the same NESO window and the same suspension.

Reading the map

The construction and reinforcement stations are where the map runs out of evidence, and they’re also where most of the months go. I’ll come back to them on 12 November, with the gap between ordering the major kit and having it delivered. In the meantime, if the transmission window reopened tomorrow, would your project be ready for the first station, or still negotiating the land?

Sources

DESNZ, Accelerating electricity network connections for strategic demand (consultation), 11 March 2026, and government response, 28 October 2026, gov.uk

Ofgem, Decision on a modification to transition to connections reform, January 2025

NESO, connections reform pages, checked 31 October 2026

NESO, Gate 2 Criteria Methodology, 2026 version

Ofgem, Proposed data centre connection reforms (consultation), 29 July 2026; consultation page checked 31 October 2026

Ofgem, agreement to revise the Gate 2 to Whole Queue timetable, 2 October 2026

ENA, Queue Management User Guide v6.0, September 2025 (sections 4.1.2, 5.3 and 6.3.1; glossary)

Distribution licence, Standard Licence Condition 12; Electricity North West, Notice of Rights, April 2025

SP Energy Networks, demand connections: minimum information for applications of 5 MVA and above, citing ENA demand queue entry requirements, checked 31 October 2026

Cadent, New Large Load Connections re-opener submission, January 2024

National Gas, Applications Guide 2025

Wales & West Utilities, Notice of Rights 2025/26

Future Energy Networks, via IGEM, April 2026

Environment Agency, permit decision for the Linmere data centre standby plant, 29 January 2026

Planning Practice Guidance, determination periods; Hoare Lea analysis via DCD

Ministry of Housing, Communities and Local Government, National Planning Policy Framework, 17 August 2026, chapter 7, policies E1 and E2

Planning Act 2008, sections 35, 55, 98 and 107; SI 2026/641; MHCLG section 35 directions for Wapseys Wood, Ampthill Road and New Barn Road, 2026

Initial Proposal for a National Policy Statement for Data Centres, laid before Parliament 28 October 2026; consultation closes 8 January 2027

Planning Inspectorate, NSIP management information, July 2026

Scottish Government, notification direction (August 2026) and Chief Planner letter on EIA (September 2026); Electricity Act 1989, section 36

Infrastructure (Wales) Act 2024, section 2

Town and Country Planning General Regulations 1992, regulation 3

Planning Act 2008, section 15 (generating stations in England over 50 MW)

Environment Agency, RGN 2 Understanding the meaning of regulated facility, Appendix 1, Section 1.1 notes, v4.0, April 2019

Environment Agency, SR2018 No 7 standard rules for low-risk medium combustion plant (site total under 50 MWth; excludes electricity generation other than excluded back-up generators)

BCLP, The energy nexus: UK data centres and power strategy, June 2026 (three-route framing)

Provenance: Synthesis, with an original schematic. As far as I’ve found, this is the first single map of the UK routes to data centre power with times per station.

Earlier posts and articles are archived at markdmoore.co.uk

Graphic accompanying the article: Three lines to power, and the main one is suspended

Departures: where the UK’s data centre trains actually stand

Published 30 October 2026 · Read the post

Anyone who has travelled north from Euston on the West Coast Main Line will know the old concourse. No seating, thousands of people milling about under the split-flap board, everyone watching for their train to flip from a time to DELAYED. The wait was bad enough. The real problem came when the platform was finally announced, and the whole concourse moved at once. Read the full piece

That’s roughly how the UK’s data centre pipeline looks from the outside in October 2026, so I’ve drawn it as a departures board. Nine services, each with the time it was meant to leave, the time the latest public evidence says it will, and a remark. Every figure on the board is sourced below, and where a number is my own arithmetic I say so.

Guy Massey’s arrivals board for US data centre kit prompted this, and the UK version swaps his supply chain rows for grid, planning and policy.

The grid: two lines, one queue

The top two rows are the ones that matter most. NESO’s reformed connections pipeline, published in December 2025, holds more than 100 GW of large demand projects, most of them data centres. Around 13 GW of that can connect before 2030. A further 86 GW is slated for 2030 to 2035. If your project was planned around a 2030 energisation, the expected column now reads 2035 for most of the queue.

The second row is the queue itself. In written evidence to Parliament, NESO put data centre connection requests at around 72.8 GW through to 2039, with about 59 GW of that waiting for a transmission connection across 173 projects. Forty of those projects are between 500 MW and 1,500 MW each. For scale, Ofgem’s own figures show the whole demand queue going from 41 GW in November 2024 to 125 GW by June 2025, and almost all of the increase was at transmission level. These are big trains, and they all want the same platform.

Planning: a statutory 13 weeks, an observed 490 days

A major planning application in England has a statutory determination period of 13 weeks, or 16 where an environmental impact assessment applies. Hoare Lea’s analysis of 33 data centre applications put the average time to secure consent at about 490 days. The two figures don’t measure exactly the same thing (one starts at a validated application, the other covers the whole journey to a decision), so read the board as the timetable set against the experience. In Scotland, data centres over 50 MW have been EIA development since 17 September 2026, which adds work before an application is made.

The fourth row is the large-scale route. Since 8 January 2026, data centres can opt into the nationally significant infrastructure regime through a section 35 direction. The SDC M40 Campus in Buckinghamshire (around 300 MW of IT load, with an on-site gas turbine energy centre) has its direction and expects to submit in January 2027. On current Planning Inspectorate timings, that’s about 16 months from submission to a decision, so mid 2028. That date comes from my own arithmetic on published timings, and it assumes nothing slips.

On-site power: a train that leaves on time, if the gas arrives

Plenty of developers are now looking at generating on site while they wait. The constraint then becomes the gas main. Cadent aims to deliver large new gas connections within 12 months, but where a pressure-reducing governor needs new land, installation can take up to two years. The remark on the board says AT RISK: 12 months is achievable on the right site. I set out what that route can and can’t do in Part 1 of my UK Colossus article (12 October), and Part 2 (29 October) put a value on the time it saves.

The policy services

Four rows on the board are about the machinery itself.

Gate 2 offers are the reformed connection offers that decide where each project sits in the queue, and they are the platform announcement. Phase 1 offers, already extended once to 30 September, can now be issued by 31 October, after Ofgem agreed a second extension on 2 October following late offers from the transmission owner. Phase 2 offers remain due by 31 January 2027, and distribution offers complete by 30 November 2026 (Phase 1) and 31 March 2027 (Phase 2). Each holder then has 90 days to accept, with extra time only in narrow cases, such as four more weeks where an offer has had to be corrected. When the offers land together, everyone who accepts will be chasing the same designers, contractors and equipment slots at the same time. That’s the Euston rush, and on present evidence it starts in November.

The Strategic Spatial Energy Plan was commissioned in October 2024, when the government said the first iteration would be published in 2026. NESO now expects to deliver the final plan in Autumn 2027, and on 23 October the government formally moved its statutory deadline from the end of 2026 to 31 December 2027. The network plans that depend on it, the Centralised Strategic Network Plan and the Regional Energy Strategic Plans, follow by the end of 2028. A data centre choosing a site in 2026 is choosing it before the plan that says where the network will be built.

AI Growth Zones were first announced in January 2025, with Culham designated first. There are now at least five. Each has to show it can support at least 500 MW of AI infrastructure, but a freedom of information analysis by Search Acumen found just one relevant planning permission granted in South Oxfordshire in 2025. The board says AWAITING until planning permissions start to follow the designations.

The newest row went up on 28 October. The Budget’s Data Centre Covenant offers “strategic” projects priority connections, the transfer of pre-2030 connection contracts and temporary on-site power while they wait, in return for a local benefit plan, a design PUE of 1.3 or better and annual energy and water reporting. There is no MW threshold, no criteria and no application route yet; those will follow “in due course”. The expected column says TBC and the remark says AWAITING until the definitions arrive.

What the wait costs

The band along the bottom of the board is my own figure. Each month a data centre waits for power costs its investors somewhere between £65,000 and £105,000 per MW of IT load, which I published in Part 3 of “Who pays for the wires?” on 9 October. It’s a floor, built from the low end of every sourced range. On a 100 MW site, the low end is £6.5 million a month.

The train that isn’t on the board yet

On 28 September the government announced GB Grid, a state-owned company within Great British Energy that could co-invest in new connections. I wrote about it on 1 October. It isn’t on the board because it has no timetable yet: no published date for its first connection, and no detail on how it will sit alongside NESO’s queue or Ofgem’s rules. When it gets one, it goes on the next edition.

I’ll update the board as the dates move, and next Friday’s edition will set London against Europe’s other FLAP-D markets: Frankfurt, Amsterdam, Paris and Dublin.

What’s the next destination, and will you be ready to rush when you hear the clatter of the board?

Sources

1. NESO results of connections reform, via DCD, 9 December 2025: datacenterdynamics.com/en/news/neso-reveals-results-of-uks-grid-connection-reforms

2. NESO written evidence to Parliament, DCU0081: committees.parliament.uk/writtenevidence/166090/html

3. Ofgem, demand connections update, November 2025 (41 GW to 125 GW)

4. Hoare Lea, analysis of 33 data centre applications (about 490 days to consent), reported by DCD, 27 May 2026: datacenterdynamics.com/en/news/england-data-center-planning-applications-doubled-in-2025-foi-requests-reveal

5. Planning Practice Guidance, determination periods (13 and 16 weeks)

6. Scottish Government, Chief Planner letter and EIA Direction on data centres, September 2026: gov.scot/publications/chief-planner-letter-eia-direction-on-data-centres

7. Forsters, data centres brought into the NSIP regime (8 January 2026)

8. Planning Inspectorate, SDC M40 Campus project page (EN0110030) and NSIP management information, July 2026

9. Cadent, New Large Load Connections re-opener submission, January 2024

10. Ofgem, agreement on alternative Gate 2 to Whole Queue timetable, 2 October 2026: ofgem.gov.uk/transparency-document/ofgem-agreement-alternative-gate-2-whole-queue-g2twq-timetable

11. DESNZ press release, Strategic plan for long-term energy infrastructure, 22 October 2024

12. NESO, new timelines for strategic energy plans, 5 December 2025

13. DESNZ, Strategic Spatial Energy Plan deadline extension, 23 October 2026: gov.uk/government/publications/strategic-spatial-energy-plan-deadline-extension

14. Data Centre Review, AI Growth Zones yet to accelerate planning activity, 20 August 2026

15. DSIT, AI Growth Zones collection: gov.uk/government/collections/ai-growth-zones

16. HM Treasury, The Data Centre Covenant, policy paper, 28 October 2026: gov.uk/government/publications/data-centre-covenant/the-data-centre-covenant

17. Mark D. Moore, “Who pays for the wires?” Part 3, LinkedIn, 9 October 2026

18. Osborne Clarke, NESO allows extensions to Gate 2 offer signing deadlines, 16 June 2026: osborneclarke.com/insights/energy-transition-neso-allows-extensions-gate-2-offer-signing-deadlines

Provenance: Synthesis, with original analysis on cost of delay.

Earlier posts and articles are archived at markdmoore.co.uk

Departures, UK data centres: where the grid, planning and policy stand, October 2026. Data centre demand: scheduled 2030, expected 2035, delayed. Connection queue: 72.8 GW of data centre requests to 2039, 59 GW waiting at transmission across 173 projects, queued. Planning: statutory 13 weeks for a major application, average time to consent about 490 days, delayed. SDC M40 Campus, 300 MW: pre-application, expected 2028. Gas main on site: 12 months, up to 24, at risk. Gate 2 offers: 30 Sep 26, now 31 Oct 26, delayed. Spatial Energy Plan: 2026, now Autumn 2027, delayed. AI Growth Zones: awaiting. DC Covenant, 28 Oct 26: TBC, awaiting. Cost of the wait: £65,000 to £105,000 per MW of IT load, for every month a data centre waits for power.

If the UK had its own Elon Musk, what would the time be worth? (Part 2 of 2)

Published 29 October 2026 · Read the post

Part 1 asked how fast a UK Musk could build a data centre, and the answer was quickly, within limits. Keep the site under the consenting thresholds, power it on site, and Project Flowers reaches first power in about eight months with fuel cells on a council-owned site, or about twelve with gas engines. On an average site, with ordinary planning, either route takes about twenty. Memphis took four. Read the full piece

Developers count in pounds, so this part puts a price on the months.

The rate

In “What is the grid worth to a data centre?” (9 October) I published a cost of delay of £65,000 to £105,000 per MW of IT load per month. It rises with the length of the wait, and it’s a floor: development overhead is left out, and every range behind it is taken at its low end.

Everything below uses the lowest published rate, £65,000 to £69,000, which was modelled for longer waits. The savings here run to months rather than years, so read the results as indicative.

What a month is worth

Part 1 set the speed limit at roughly 31 to 40 MW of IT load with fuel cells and 11 to 14 MW with gas engines. At those sizes a month of delay costs a fuel cell site about £2.0m to £2.8m, and an engine site about £0.7m to £1.0m.

How the months become money: months saved x MW of IT load x the published cost of delay. Fuel cells, council against average site, 8 against 20 months: 12 months x 31 to 40 MW IT x £65k to £69k, £24m to £33m. Gas engines, council against average site, 12 against 20 months: 8 months x 11 to 14 MW IT x £65k to £69k, £6m to £8m. UK best case against Memphis, fuel cells, 8 against 4 months: 4 (slower), £8m to £11m. Gas engines, 12 against 4 months: 8 (slower), £6m to £8m. These are gross figures. The net value has to be worked out site by site.

The chart sets out the arithmetic for each case: months saved, times MW of IT load, times the rate.

The council dividend

The biggest saving in Part 1 came from the planning route. A council deciding an application on land it already owns works to a statutory clock of 13 weeks, or 16 with an environmental impact assessment. The average UK data centre takes about 490 days to secure consent.

With fuel cells, the council route brings time to power down from about twenty months to eight. Twelve months at the rate above is worth roughly £24m to £33m. Gas engines gain less, because engine delivery sets their pace: eight months saved, about £6m to £8m.

That is what a quick planning decision, on land a council already holds, is worth to the developer.

The price of the rules

Memphis had three things the UK can’t offer: a ready factory shell, a gas main to hand and a 364-day window for temporary turbines. Even in its best case, Project Flowers finishes four months behind with fuel cells and eight behind with engines. Four months at fuel cell scale comes to about £8m to £11m, and eight months at engine scale to about £6m to £8m.

Some of that buys protection. Memphis’s turbines ran under a local exemption while residents objected, and the cost of the speed fell on the neighbours. Whether the UK’s safeguards are worth £8m to £11m a site is a judgement for policy, and these numbers don’t settle it.

There is one smaller lever. Ready-to-run used gas engines (50 Hz, low hours, within the emissions limit) could bring the engine case on a council site from about twelve months to eight, which adds roughly £3m to £4m.

What these figures leave out

Every figure here is gross. It is the value of reaching power sooner, before the cost of the plant, the fuel, the financing, and whatever becomes of the plant once the grid arrives. The net position depends on the site, the fuel route and what the plant does after connection, and it can sit well below the gross. That calculation has to be done site by site, and I’m happy to talk it through one to one.

The figures are also a private value to the investor. Some of the margin a delayed site forgoes is earned by another site, here or abroad.

Who keeps the value

Part 1 noted the catch in the council route. Permission granted on a council’s own land benefits the council and any named joint developer, so a private builder has to be named in the application. On my reading, that makes the time saved something the council and the developer would negotiate over, through land terms or a share in the project. I haven’t yet seen a council put a price on it.

The wider picture

These are single-site numbers. “What does the wait add up to” (16 October) sets out what grid delay is costing UK data centre investors in total.

Tommy Flowers’ team had the first Mark 2 Colossus in service four months after the first machine started work, in time for D-Day. On a council site, the months Project Flowers saves are worth tens of millions.

If a council could hand a developer £24m to £33m of time on land it already owns, what should it ask for in return?

Earlier posts and articles are archived at markdmoore.co.uk

What the months are worth: Project Flowers. Gross value of time to power, at £65,000 to £69,000 per MW of IT load per month; fuel cells 31 to 40 MW IT, gas engines 11 to 14 MW IT. Council site against average site: fuel cells, 12 months saved, £24m to £33m; gas engines, 8 months saved, £6m to £8m. UK best case against Memphis: fuel cells, 4 months slower, £8m to £11m; gas engines, 8 months slower, £6m to £8m. Gross, before plant, fuel and financing.

Who owns the battery queue?

Published 22 October 2026 · Read the post

Ofgem counts about 83 GW of battery storage holding Gate 2 positions, against a need for 24 to 29 GW by 2035, and its CMP470 commitment fee is built on that gap. The TEC Register names the project company behind each entry. That tells you very little about who controls the capacity, so I traced the largest of them through Companies House. Read the full piece

I took the NESO TEC Register of 5 October 2026 and kept battery-only entries; sites pairing storage with solar or wind are left out. Fifty-eight companies hold battery capacity labelled Gate 2 and not yet built, 11.5 GW in total. The label only appears once an agreement is countersigned, so this is the countersigned part of the queue, a fraction of Ofgem’s figure. I ranked the companies by capacity, counting each project stage once, and followed the largest 25 (9.5 GW, 82% of the labelled total) up through the persons-with-significant-control filings to the last company or person the register names. Every check is dated 5 or 6 October.

Twenty-five names, seventeen owners

The 25 project companies resolve to 17 owner groups.

Five of them control 52% of the capacity: Fidra Energy (1.4 GW), Copenhagen Infrastructure Partners (1.35 GW), Statera Energy (1.19 GW), Matrix Renewables (1.0 GW) and Revera (0.8 GW). Behind those five sit EIG, CIP’s fourth flagship fund, EQT, TPG and Carlyle.

Who controls the Gate 2 battery queue: the largest 25 battery companies holding countersigned Gate 2 capacity (9,507 MW) by type of ultimate owner and owner group. Infrastructure or private equity funds 6,540 MW (68.8%), led by Fidra Energy (EIG; National Wealth Fund 25 to 50%) 1,400 MW, Copenhagen Infrastructure Partners 1,350 MW, Statera Energy (EQT) 1,190 MW, Matrix Renewables (TPG) 1,000 MW and Revera (Carlyle) 800 MW. Privately held developers 1,000 MW (10.5%), state-owned 828 MW (8.7%), listed storage funds 440 MW (4.6%), utility 350 MW (3.7%), joint venture 349 MW (3.7%).

Sorted by type of owner, the picture is starker. Infrastructure and private equity funds control 68.8% of the ranked queue. Privately held developers hold 10.5%, state-owned companies 8.7%, the two listed storage funds 4.6%, and a single utility, RWE, 3.7%.

Set against what’s already built

The same tracing on the 57 companies with built battery capacity on the register (4.2 GW) gives a different mix. Funds control 38%. Utilities, listed funds, joint ventures and smaller private owners share most of the rest, and two insurers, Generali and Tokio Marine, each control a battery. The register only carries transmission-connected and larger embedded storage, a little over half of the roughly 7 GW in operation. For the whole operating fleet, two September analyses traced ownership through the same filings: The Renewable Guy & Cities (9 September) found Gresham House and Zenobe holding 26% of everything running, and Vitreous Labs, writing for Voice of Renewables (12 September), showed that most operating batteries are traded by a company outside the owner’s group.

Read together, the batteries waiting to connect are much more likely to be owned by a fund than the batteries already running. Fund ownership brings capital and a defined exit, and a sale keeps capacity in the queue under a new name, which is the stay, leave or sell choice I described when the CMP470 proposal was published.

The register can’t tell us which way these owners will go.

The names on the register mislead

Twelve companies on the register carry the Harmony or HEIT name and hold 521 MW of built batteries. They answer to three unrelated owners: Foresight, which took the HEIT fund private in 2025; TagEnergy; and a private investor. Three of them were renamed at Companies House some time ago and still appear on the register under their old Harmony names.

On 8 July 2026 a newly formed company, Bess Holdco, replaced Harmony Energy as the minority holder in five of those battery companies, and from that date Bess Holdco has been held by two individuals. The next day Companies House stopped naming a controller of Harmony Energy, following Alpiq’s purchase of 90% of it. The filings show the stakes moving out the day before. They don’t say why.

Public money in the queue

The largest single entry, Thorpe Marsh (1.4 GW, under construction), belongs to Fidra Energy. EIG holds 50 to 75% of Fidra’s holding company, and since 5 September 2025 the National Wealth Fund, owned by HM Treasury, has held 25 to 50%. Statkraft (owned by the Norwegian state) and Masdar (Abu Dhabi) add a further 828 MW. Add Equinor’s shareholding in Blackhillock (349 MW) and the entries with a state-controlled shareholder holding 25% or more come to 2.6 GW, 27% of the ranked queue.

Where control is shared, or the trail stops

Zenobe Energy, which sits above two of the ranked companies and four built ones, has no majority holder: KKR and a vehicle split between Infracapital and a private fund partnership each hold 25 to 50%. At Blackhillock (349 MW) Equinor holds the shares and Noriker Power holds the votes. Carlyle and BlackRock appear on the register as having significant influence or control as members of a firm, which is how a manager is recorded when its fund holds the shares, so I describe them as controlling rather than owning.

Some trails go dark. TagEnergy’s UK holding company has filed a statement of no registrable controller since 3 August 2026, and the Gore Street Energy Storage Fund did the same on 28 August. Where the register stops and the public record names an owner, I’ve used the public record and said so; EQT behind Statera is one example. Field’s backer, DIF, is confirmed from the parent company’s own accounts, which put its holding at 42%, against the 50 to 75% shown on the register.

What this doesn’t show

Ofgem’s Initial Impact Assessment puts 26% of battery capacity at Gate 2 or energised with no securities or liabilities posted. It publishes that figure in aggregate only, and nothing here links it to any owner. The ranking covers the countersigned queue on one date. Ownership is as recorded at Companies House on the dates checked, and the register shows who controls what, not why.

If two-thirds of the batteries Ofgem’s fee is aimed at belong to funds with their own exit timetables, will the fee change who builds them, or only who holds them while they wait?

Original analysis. I haven’t found a published trace of who ultimately controls the countersigned Gate 2 battery queue (trade press, research providers and newsletters searched on 6 October 2026). The built comparison applies the same method to TEC-registered storage and credits the September analyses named above. Sources: NESO TEC Register, 5 October 2026; Companies House persons-with-significant-control records, 5 and 6 October 2026; Ofgem, CMP470 minded-to consultation (para 2.5) and Initial Impact Assessment (Figure 2).

Earlier posts and articles are archived at markdmoore.co.uk

Funds dominate the battery queue: share of capacity by type of ultimate owner, TEC Register 5 October 2026. Funds hold 68.8% of the Gate 2 queue (largest 25 companies, 9,507 MW) against 38.0% of built capacity (57 companies, 4,172 MW).

Andy Burnham can make the UK the fastest time to power. He just needs to want to do it

Published 19 October 2026 · Read the post

Microsoft has ended its agreement to anchor Nscale’s data centre at Loughton in Essex. The Telegraph reported it on Sunday and Nscale has confirmed it. Neither company has given a reason. The background is not hard to read, though. Grid officials reportedly told Nscale that power would not arrive until the 2030s, and UK Power Networks is reported to have put it at the early to mid 2030s. Read the full piece

None of this should surprise anyone who works on connections.

The site was named in the 2025 AI strategy as the UK’s largest sovereign AI data centre. A minister wrote to the district council in support. It had planning permission and a 90 MW connection agreement.

It still could not get power on time.

Demand does not wait for a connection date. A customer that needs compute in 2027 will not sign for a site that energises in 2032. It goes where the power is, and that may well be outside the UK. Nscale says it is looking at alternative power for Loughton and at other UK sites, which is the right instinct.

The question is whether policy helps that search or gets in its way.

AI is likely to be the biggest productivity story of the next decade. Estimates range from modest to transformative, but even the middle of that range would matter to a country whose productivity has lagged its main competitors for years. Using AI does not require the compute to be here. Owning the capacity does, though. It decides where the investment lands, where data that has to stay in the UK can be processed, and whether the businesses that build on compute grow here or somewhere else.

The UK has the most to gain and, on current grid dates, the most to lose.

To be fair to the Prime Minister, he has made grid delays a priority. Last month the government announced GB Grid, a publicly owned company inside Great British Energy that will invest in networks and can co-invest in connections that businesses build themselves. That is a real step. It mainly helps the large sites, though, and large sites still wait years for reinforcement. The government’s own Delivering AI Growth Zones paper calls timely grid connection the single biggest blocker, and the zones were built around scale, with applicants asked to show they could deliver at least 500 MW by 2028.

The regulator’s numbers point somewhere else. In its Curate consultation this July, Ofgem counted around 315 data centre projects in the demand queue, roughly 73 GW, and found that 99% of that capacity sits in projects above 40 MW. It has proposed a commitment fee for data centres of 40 MW and above, and new milestones from 10 MW in the consultation text, though the drafted instruments go as low as 1 MW at distribution. None of it is final; a decision is due later this year. Read plainly, projects under 40 MW are not what is blocking the queue.

Why 30 MW? Because it can be bridged. On-site power at that scale is ordinary engineering: gas engines, fuel cells, batteries. I’ve built a bridge strategy modelling process for exactly this question: which bridge options, in what sequence, and at what point waiting for the grid is the better call. Each site comes out differently. One rule holds everywhere, though: whatever is built must not become obsolete when the grid arrives.

Owners are already trying this. The gas networks logged 113 data centre connection enquiries across 2024 and 2025, roughly tripling in a year, and say a new gas connection typically takes 6 to 24 months. ACRE’s planned data centre at Pinewood Nurseries in Stoke Poges, near Slough, has applied to run on gas engines because its own filing warns of at least ten years before a firm grid connection. The Thames Valley Park scheme in Wokingham proposes gas fuel cells for its first data halls until a larger grid connection in 2037.

Each one runs into a wall of conditions. Slough has drawn a formal parish council objection on air quality, carbon and noise. Thames Valley Park’s own energy strategy concedes that keeping the fuel cells would lock in a high emissions profile, which is exactly the argument objectors will use. Then there is permitting. Gas engines at 44% electrical efficiency (a manufacturer’s figure) burn about 68 MW of fuel to make 30 MW of power. Environment Agency guidance treats new combustion units sharing a stack as one plant, and above 50 MW thermal the large combustion plant rules apply, not the lighter medium plant regime. A 30 MW bridge can therefore land in the heavier permitting category by default. Each of these is reasonable on its own. Together, they mean a bridge that should take 18 months takes much longer, and much of the point of bridging is lost.

The obvious objection is carbon, and it needs answering head on. A bridge should be temporary by design. A fixed end date sounds tidy, but nobody can set one honestly: I track grid connection dates every week and they keep moving, and no one can publish a delivery date for the network upgrades that people would trust. So tie the consent to the event, not the calendar. The bridge runs until the grid connection is energised, then has a short, fixed window to move to standby or convert, with a route to biomethane where it is available.

The queue problem is a big-site problem: 99 per cent of data centre capacity in the demand queue sits in projects above 40 MW (Ofgem Curate, July 2026). The proposed commitment fee applies from 40 MW; a bridged 30 MW site falls below it and Loughton, at 90 MW, above.

Three policy changes that would make the difference

1. A bridged data centre category in the Curate decision. Projects under 40 MW that will run on on-site power should be able to meet the new milestones by energising on that power, and keep their queue position while they wait. Ofgem and NESO are still deciding, so the timing is right.

2. One consent route, tied to energisation, with a bridge-to-grid condition. A model Local Development Order, or a national prior-approval route, that consents the data hall and its on-site generation together until the grid connection is energised, with permitting handled in the same process. After energisation, generation has a short, fixed window to drop to standby or convert, which keeps the bridge consistent with Clean Power 2030 without anyone having to guess a grid date.

3. A register of ready public plots. A common rule of thumb is 3 to 4 MW per acre, so a 30 MW site needs around ten acres including its power. Councils and other public bodies hold plots of five to fifteen acres on employment land near gas and fibre. List them, with the planning and power position checked in advance, and let the AI Growth Zone delivery unit run it.

None of these needs new primary legislation to start. GB Grid shows the government is willing to change how connections work for the big sites. The same willingness applied to small, bridged sites would put compute into the UK, potentially in less than two years rather than ten.

GB Grid’s operating rules have not been published yet. That makes this the right moment to ask what it is for. If part of its job were to co-invest in the connections that bridged sites will eventually need, so they move from bridge to grid the moment the network arrives, the UK would already have the vehicle to cut time to power. Could GB Grid be that vehicle?

Provenance: Synthesis. Public reporting (The Telegraph, Reuters, The Guardian, Bloomberg, City A.M.), Ofgem’s Curate consultation (July 2026), the Delivering AI Growth Zones policy paper (November 2025), Environment Agency guidance and Future Energy Networks. Views are my own.

Earlier posts and articles are archived at markdmoore.co.uk

Graphic accompanying the article: two data centre sites, one waiting for the grid and one bridged with on-site power

What does the wait add up to?

Published 16 October 2026 · Read the post

Thoughts for the weekend. Who pays for the wires? A postscript, by popular demand (mine).
Last week’s piece put a price on waiting. For a data centre that’s ready to run and can’t get power, each month of delay costs between £65,000 and £105,000 per megawatt of IT load, and the figure rises the longer the wait goes on. Nobody asked what that adds up to across the country. I wanted to know anyway. Read the full piece

The obvious sum gives the wrong answer

Ofgem’s July consultation counts about 73 GW of data centres in the demand connection queue, spread across roughly 315 projects. That’s more than one and a half times Britain’s peak electricity demand. Multiply all of it by last week’s figure and a plausible range of delays, and the total comes to something over £300 billion.

The number means nothing, as Ofgem’s own evidence demonstrates. Built out in full, the queue implies £693 billion of capital spending, about 23 per cent of UK GDP in 2025. It isn’t going to happen. Much of the queue is speculative, and a queue position with nothing behind it loses very little by waiting. The cost lands on the sites that will actually be built.

Four inputs, two of them unpublished: queue about 73 GW across about 315 projects (Ofgem); share built 7 to 30 per cent, most likely 15 (NESO); wait per site 1 to 10 years, most likely 4 (Loughton, Pinewood, JLL); cost per month £65k to £105k per MW of IT load (Part 3).

A realistic pipeline

So the sum becomes two questions. How much of the queue gets built, and how long does each of those sites wait?

The first has no published answer, but the evidence brackets it. NESO expects around 5.2 GW of data centres to be connected by 2030, about 7 per cent of what’s been requested. At the other end, 71 data centre projects, around 20 GW, told NESO last winter that they had reached a final investment decision. That’s about 30 per cent of the queue, self-reported, and it sits awkwardly beside the same survey’s finding that only 32 per cent of data centre projects had a customer signed. I’ve used 7 and 30 per cent as the ends of the range, with 15 per cent as the most likely value.

The second is harder still. There’s no public register of demand connection dates, only what developers say about their own sites. Nscale’s 90 MW site at Loughton, planned for 2027, is reported to be waiting until the early to mid 2030s for enough grid capacity. A data centre application at Pinewood Nurseries in Buckinghamshire, submitted in August, expects its firm connection about ten years after acceptance. JLL puts the wait for a new 50 MW facility in London at around eight years. From those I’ve taken delays of one to ten years, with four the most likely.

Then I ran it 20,000 times (no really, Monte Carlo simulations): drawing the share built and each site’s wait from those ranges, sizing the projects from Ofgem’s own breakdown of the queue, and costing each wait at last week’s rate for a wait of that length.

The answer

Counting only the data centres likely to be built, grid delay is costing their investors somewhere between £30 billion and £80 billion, with a central estimate of about £55 billion. That’s before anyone else waiting for a connection.

Behind the central estimate sits about 9.5 GW of IT capacity that gets built and has to wait. Set against what that capacity costs to build, at Ofgem’s average of £9.5 million per megawatt, the delay comes to a little under half. Most of it is margin never earned, which is what a finished building with no power does to a balance sheet.

What moves the answer

What moves the answer: share of queue built moves the central estimate from £22bn to £95bn; most likely wait of 2 to 6 years moves it from £45bn to £62bn; long waits causing exits take it to £49bn. Base case about £55bn.

The share of the queue that gets built matters far more than anything else in the model. If 7 per cent of the queue gets built, the central estimate falls to about £22 billion. At 30 per cent it rises to about £95 billion. Moving the most likely delay from four years to two or six shifts the answer by about 16 per cent either way. Letting the longest waits push projects out of the queue altogether, which is what one developer says would happen, takes about 8 per cent off.

That last point has a witness. Pinewood’s energy consultant writes that a ten-year wait for power “would render the data centre undeliverable” and that it “would likely not be built”. The same application seeks consent for either a data centre or up to 16,000 square metres of industrial units. Even a named site in the queue may never become a data centre, which is the realisation problem in miniature.

One site

At last week’s rate, Nscale’s Loughton site, around 69 MW of IT load, loses about £290 million if its connection arrives four years late and about £690 million at eight. Those figures rest on press reports of the date, not a published offer.

The small print

This is what investors lose. The country loses less, because some of the margin a delayed site never earns is earned by another site, in Britain or abroad.

Each site’s cost is measured at the date it would otherwise have connected, and the total adds those together. It isn’t a single present-day figure. Delay is counted from the date a site would otherwise have been ready, and the rate underneath is a floor: it leaves out development overhead and takes the low end of every sourced range.

The unpublished dates

The whole estimate turns on two things that could be public and aren’t: which data centre projects hold connection offers, and what dates they’ve been given. Generation and storage have public registers. Demand has nothing equivalent. Ofgem’s current consultation on connections data proposes a demand capacity register, and it closes on 6 November.

A £50 billion spread is an honest statement of what can be known today. With a register, most of it would narrow to a number.

If a delay costing tens of billions can only be estimated from press reports and planning files, who should be required to publish the dates?

Sources

Ofgem (29 July 2026). Proposed data centre connection reforms: Curate consultation. Paras 2.6 to 2.7, Table 1, and Chapter 4.

Ofgem (13 February 2026). Demand Connections Reform: call for input. Paras 2.7 and 5.6.

NESO (March 2026). Demand Call for Input: high level summary.

NESO. Written evidence to Parliament, DCU0081.

Buckinghamshire Council, application PL/26/06414/OA, Land at Pinewood Nurseries, Stoke Poges: description of development, and Energy Centre Report v1.0 (Electric Land Digital, August 2026), section 3.2.

Sifted (13 July 2026), reporting The Telegraph, on the Nscale Loughton opening date; Superpower Daily (24 September 2026) on the reported UK Power Networks estimate.

JLL Research, via SymTerra, as reported by Archdesk (July 2026).

Ofgem (25 September 2026). Transforming connections data: consultation. Closes 6 November 2026.

Cost of waiting: M D Moore, “What is the grid worth to a data centre?”, 9 October 2026, and forthcoming paper. Real 2026 pounds; 8 per cent nominal cost of capital at 2 per cent inflation; 70 per cent load factor.

Aggregate estimate: own simulation, 20,000 runs, on the inputs stated above.

Provenance: Original analysis. I haven’t found a published estimate of what grid delay is costing UK data centre investors in total. Queue data from Ofgem and NESO; delays from developers’ statements and press reports; the calculation is my own.

Earlier posts and articles are archived at markdmoore.co.uk

Graphic accompanying the article: What does the wait add up to?

Fourteen UK data centres, one test, and why the map is smaller than you’d expect

Published 13 October 2026 · Read the post

EnergyTag recently published a map rating 21 US data centre projects on how they buy their power: good, bad or ugly, sized by planned demand. The obvious question was whether the same thing could be done for the UK. It can, with one condition that shapes everything that follows: only schemes that can be verified go on it. Read the full piece

This article sets out the rules. The map itself will be updated on the second Tuesday of every month, and the rules will not move to make the numbers look better.

Why the connection register doesn’t answer the question

The first place to look was the NESO TEC Register. On the 25 August 2026 file, 106 rows carry a Demand plant type, totalling 64.7 GW. Every one of them is a hybrid, demand sitting alongside generation or storage (Drax BECCS, Pembroke, a run of solar-and-battery Green Energy Centres). Not one row names a data centre, and no customer name contains “data” or “compute”. That isn’t an error. TEC is a generation register, and demand customers connect under agreements that are not published scheme by scheme.

So the evidence has to be assembled from what developers, councils and government actually publish: planning committee reports, section 35 requests to the Secretary of State, Written Ministerial Statements, procurement notices and developers’ own statements.

The inclusion test

A scheme reaches the map only if all four of these are true, and each rests on a developer, planning authority or government source:

1. It is a named scheme at a specific location.

2. It has an MW figure stated by the developer, the planning authority or government.

3. Its supply arrangement is publicly stated.

4. Its status carries a date.

Trade press is useful for finding schemes. A figure that only appears in a news report stays off the map until it appears in a primary document.

The ratings

Green means new clean supply contracted to the site, or co-located with it, with the counterparty or asset named, and no on-site fossil prime power.

Amber means grid supply only, or a clean-power claim that isn’t tied to a named new asset. Renewable certificates, “100% renewable” tariffs with no supplier named, a private wire that is still being explored, and a memorandum of understanding all sit here.

Red means on-site or dedicated gas (or other fossil) generation planned as the primary or bridging supply.

Standby plant is recorded but doesn’t move the rating. Almost every site has it. Elsham Tech Park in North Lincolnshire is the clearest case: the committee report describes up to 650 diesel generators of 2,480 kW and gas engines of just under 50 MW, all for emergency use and testing, with the site supplied from the grid. That is Amber, and the generator count sits in the register for anyone who wants it.

The scale differs from EnergyTag’s for a reason. In much of the US, new load on the grid means new gas. Great Britain’s grid is a different mix. Gas supplied 26.8% of Britain’s electricity in 2025, according to NESO, against about 41% of US utility-scale generation on the EIA’s figures for the same year. Drawing from the GB grid isn’t treated as “bad” here for that reason.

What happened when the test was applied

Thirty-eight candidate schemes were researched across England, Scotland, Wales and Northern Ireland. After a final re-check on 9 October, fourteen pass. Twenty-four are on a watch list, each with the specific reason it failed.

Most of the failures come down to the MW figure. QTS at Cambois in Northumberland, consented in March 2025, is reported at 720 MW in one place and 1.1 GW in another, and neither figure comes from a primary document I could read. Humber Tech Park, Google’s North Weald scheme and Teesworks all have figures that trace back only to trade press or a law firm’s briefing note. Others fail on supply. Two appeal-approved schemes in Buckinghamshire and Hertfordshire (90 MW and 96 MW) say nothing public about where their power comes from.

For scale, several further schemes would qualify if trade press reporting were accepted in place of a primary document. They stay on the watch list.

Google Waltham Cross is the case that tests the rule hardest. It has a named clean-energy arrangement, with Shell as 24/7 carbon-free energy manager across a portfolio that includes ENGIE’s Moray West offtake, but Google describes that arrangement for its UK operations as a whole, and I could find no MW figure for Waltham Cross from Google or the council. It stays on the watch list until one is published.

Of the fourteen that passed, one is Green, ten are Amber and three are Red.

The Green is Kao Data’s Harlow campus, 71 MW of IT load once its current phases are complete. Downing Renewable Developments has agreed to develop, build and operate a solar farm of up to 40 MW for it, north of Foster Street in Harlow, with the output sold to Kao Data under a power purchase agreement and delivered by private wire. Both companies announced it in November 2025. The farm covers only part of the load, the grid supplies the rest, and on Downing’s own timeline it won’t be generating until about 2029. The test asks for new clean supply contracted to the site with the asset and counterparty named, and this has both. The register records the timing and what remains open, including the farm’s planning status.

Two of the three Reds came through the section 35 route, which is where the detail is most complete. SDC M40 Campus in Buckinghamshire is consulting on two options for about 300 MW of IT load. One is an on-site energy centre of about 200 MW running on natural gas from an existing pipeline, powering up to two of its three buildings until grid capacity arrives, which the developer expects by 2040. The other is a direct 400 kV connection with no energy centre. Only one would be built, and while the gas option is on the table it stays Red. Quest Park at Stewartby plans gas-fuelled generation at each building for 720 MW of IT load, running as the prime supply until a 200 MW grid connection arrives in October 2034.

The third Red arrived during the final check. Thames Valley Park, near Reading, was on the watch list until its outline application was validated by Wokingham Borough Council on 28 September. The applicant’s own documents describe 72 MW of IT capacity, with gas-fed fuel cells as the principal energy source until grid connections are completed in 2037.

How to read the donut

The monthly graphic carries a small donut showing the total MW by rating. It sums each scheme’s headline figure as the source states it. Some of those figures are IT load, some are campus or site power, one is “compute” and several don’t say. The donut indicates where the stated capacity sits and shouldn’t be quoted as a measure of load. The register records what each figure measures, row by row.

What changes each month

On the second Tuesday of each month every row is re-checked. Schemes move from the watch list to the map when a primary source appears, and back off it if a source is withdrawn or contradicted. Ratings change when supply arrangements change. Thames Valley Park has already made that move once, between the first draft of this article and publication. Every change goes in a change log.

Corrections are welcome, provided they come with the primary document.

A scheme can ask a planning committee for consent and a network operator for a connection without ever stating, in one public place, how much load it will draw and where that power will come from. Should that be a condition of either?

Analysis

Earlier posts and articles are archived at markdmoore.co.uk

Graphic accompanying the post: Fourteen UK data centres, one test, and why the map is smaller than you’d expect

If the UK had its own Elon Musk, how fast could they build a data centre? (Part 1 of 2)

Published 12 October 2026 · Read the post

Britain has done fast before. In 1943 Tommy Flowers and his team at the Post Office Research Station at Dollis Hill built the first Colossus in eleven months. It started work at Bletchley Park in early February 1944, and the first Mark 2 was in service on 1 June, four months later and in time for D-Day. Read the full piece

Eighty years on, xAI borrowed the name. Its Colossus in Memphis went from start to finish in 122 days, with 100,000 Nvidia GPUs. So it’s a fair question: could someone with the same appetite for speed do that here?

I’ve worked it through against the UK rules. The short answer is yes, up to a certain size, and in about twice the time.

Call the British version Project Flowers, after the man who built the first one. The chart at the top sets the ways it could be powered against Memphis, from decision to first power.

What Memphis actually had

Three things were already in place in Memphis. The building was a disused Electrolux factory, so there was a shell to fit out rather than one to design and consent. There was a gas main to hand. And the turbines powering the site ran under a local exemption that lets temporary generators operate for up to 364 days without an air permit. A permit for 15 turbines was approved in July 2025, after the cluster was already running.

Residents and environmental groups have challenged those turbines ever since. That’s a cost of the speed, and it was carried by the neighbours rather than by xAI.

So the recipe was a ready site, a ready fuel supply and a regulatory window. The engineering was quick as well, but those three did most of the work.

The UK rulebook comes in three sizes

Bring your own power in England and the size of the plant decides which rules apply.

Bring your own power: the rulebook in three sizes, England. Under 50 MWth of fuel, a medium combustion plant or specified generator permit (statutory 3 months) plus local planning; 50 MWth to 50 MWe, a bespoke installation permit plus local planning; over 50 MWe, a development consent order, about 16 months from submission. Mobile generators have no 364-day window.

Below about 50 MW of fuel burned (roughly 17 to 23 MW of electricity, depending on the kit), generators need a medium combustion plant or specified generator permit, with a statutory determination period of three months for a standard permit, plus local planning.

Between 50 MW of fuel and 50 MW of electrical output, the whole site counts as one installation and needs a bespoke permit. In the year to March 2026 the Environment Agency decided its two most complex permit categories on time between 55 and 65 per cent of the time.

Above 50 MW of electrical output, the generating station is a nationally significant infrastructure project and needs a development consent order. On the Planning Inspectorate’s 2026 averages that takes about 16 months from submission, with pre-application on top.

The 364-day window has no equivalent here either. The Environment Agency can treat a mobile generator as static if it does a job a static one could do, and will if it sits in one place for six months or more.

Colossus scale, the British version

There’s a live example. The SDC M40 Campus in Buckinghamshire plans up to about 300 MW of IT load across three buildings, with an on-site gas turbine energy centre of roughly 250 to 350 MW. It has been directed into the nationally significant regime, and the application is expected in January 2027. At Colossus scale, the British answer is measured in years.

So how big can a fast one be?

Here’s the arithmetic, one step at a time.

Start with turbines. One of the Solar Titan 130 units used in Memphis burns about 47 MW of fuel on its own. Under the 50 MW line you’d get exactly one, with no spare, which rules turbines out at this size.

Gas engines fit better. Six 3.3 MW engines come to about 44 MW of fuel. Keep one as a spare and you have about 17 MW of firm power. After cooling and losses that supports roughly 11 to 14 MW of IT load, or about 8,500 to 11,000 H100-class GPUs. About a tenth of Colossus.

Fuel cells move the line. They turn gas into electricity without burning it, so on my reading they sit outside combustion permitting altogether, and I’ve asked the Environment Agency to confirm. The limit then becomes the 50 MW planning threshold, which allows roughly 31 to 40 MW of IT load and 24,000 to 31,000 GPUs. A quarter to a third of Colossus.

Those figures assume no diesel standby on site and count server power only. Treat them as a ceiling.

The council shortcut

Inside that limit, planning is the slow step. The average UK data centre takes about 490 days to secure consent, and this is where a council with land it already owns could change the outcome.

A council can decide a planning application for its own land, and the statutory clock is 13 weeks, or 16 where an environmental impact assessment applies. There are two catches. A permission granted that way belongs to the council and any named joint developer, so a private builder would need to be named as a joint developer. And the clock only starts once the application is in, so the assessment work comes first.

The other route is to grant permission in advance with a Local Development Order. Slough Trading Estate already works on a similar basis through its Simplified Planning Zone. Havering’s exploration of an order for a data centre campus is a reminder that willingness alone doesn’t make it quick.

Time to power

Put it together. Assume our UK Musk builds as fast as Musk did, 122 days, and orders the kit on day one. That’s the chart at the top.

On a council site with fuel cells, about eight months from decision to power. Suppliers claim delivery in as little as 90 days, though that’s a US claim. With new gas engines it’s about twelve, because that’s roughly how long the engines take to arrive; ready-to-run used engines could bring it down to about eight if the gas is trucked in. On an average site, either way, about twenty. At Colossus scale, twenty months plus pre-application.

Fuel is the other constraint. A large new gas connection takes about a year (Cadent aims to deliver them within 12 months, longer where a new governor needs land). Until it arrives, a virtual pipeline of trucked compressed gas can carry the load. At full output that’s roughly 7 to 23 trailer deliveries a day depending on the kit, and the major-hazard regime applies from 50 tonnes of gas on site, which is around a day’s supply or less.

So the best UK case is about twice as long as Memphis, and it’s measured in months.

Then what?

A site built this way runs on its own power until the grid arrives, and the first question any investor will ask is what happens to the plant then. Whatever gets built to beat the queue shouldn’t become a stranded asset the day the connection comes through. There’s a way to plan the plant so it keeps its value after the grid arrives, and I’ve worked it through for UK sites. It needs doing site by site, and I’m happy to talk it through one to one.

Part 2 puts a value on that time: what getting to power in eight months, against twenty on an average site, would be worth to a developer.

The only remaining question is whether we have a UK Musk single-minded enough to channel the pioneering spirit of Tommy Flowers and make this happen.

Original analysis: converting the UK consenting thresholds into GPU capacity and time to power is my own work, from public sources (Environment Agency, Planning Inspectorate, Cadent and manufacturers’ data). I haven’t found it published elsewhere (checked 10 October 2026).

Graphic accompanying the article: If the UK had its own Elon Musk, how fast could they build a data centre? (Part 1 of 2) How big can a fast UK data centre be? From the consenting limit to GPUs: gas turbines, no spare possible; gas engines, 16.8 MWe firm, 11 to 14 MW of IT load, 8,500 to 11,000 GPUs, about a tenth of Colossus; fuel cells, about 48 MWe, 31 to 40 MW of IT load, 24,000 to 31,000 GPUs, a quarter to a third

What is the grid worth to a data centre?

Published 9 October 2026 · Read the post

Thoughts for the weekend. Who pays for the wires? Part 3 of 3.
Last week finished at Packet Bridge, with a canal company that had leased its canal to a railway. The canal itself didn’t go anywhere when the trains arrived. It was still there, still able to carry goods, and it’s the canal I want to start from this week. Read the full piece

In 1964 Robert Fogel asked a question that sounds simple. How much poorer would the United States have been in 1890 without its railroads? He answered it by working out what it would have cost to move the same goods by the next best route: rivers, canals and wagons. The difference between the two is what economic historians now call the social saving. For farm freight, Fogel put it at no more than 2.7 per cent of GNP. That was far smaller than anyone had expected, and the reason was the alternative. The waterways could have done most of the work. Fogel went on to share the 1993 Nobel prize in economics.

The British numbers came out larger. Gary Hawke put the social saving of the railways in England and Wales at around 4 per cent of national income in 1865 on fares and freight charges alone, and about 7.5 per cent once the comfort of passenger travel was counted. Tim Leunig went back to the timetables and valued the time passengers saved, and found it grew to over 10 per cent of national income by 1912. Canals were no competition for a passenger in a hurry, so the alternative was weaker and the saving bigger.

The idea underneath all three is the same. A network is worth what it saves over the best thing you could do without it.

The worst alternative is waiting

For a data centre waiting on a grid connection, the simplest alternative is no power at all. The building may be finished and the customers may be signed. Nothing earns until the connection arrives.

That alternative has a price, and it’s one I can put a number on. Using the hold-up-cost decomposition on sourced inputs, the cost of waiting comes to between £65,000 and £105,000 per megawatt of IT load for every month of delay. For a 50 MW site that’s roughly £80 million over a two-year wait, £285 million over five and £500 million over eight. The figures are in real 2026 pounds, valued at the date the site would otherwise have connected, at a 70 per cent load factor, for a site in central Scotland with London as the comparator.

Most of it is margin the site would have earned and didn’t, between half and two thirds of the total. Around a fifth is value lost for good: the risk of a pre-let walking away, and a weaker rent when the site does open. The holding costs that get most of the attention make up the rest, including the capital tied up in land and security deposits. Ofgem’s proposed commitment fee is part of that, and at 50 MW over five years its carrying cost comes to about 4 per cent of the total.

The figure grows with the length of the wait, from £65,000 to £69,000 a month per megawatt at two years to £101,000 to £104,000 at eight. And it hardly moves with the size of the site. A 20 MW operator loses about as much per megawatt as a 100 MW one, with a much smaller balance sheet to carry it.

It’s a floor. Development overhead is left out, and the low end of every sourced range is used. The only other British figure I’ve found is in an opinion piece in Data Centre Review in July, which put the cost of each month of delay on a typical 60 MW project at up to £10 million to £11 million, or up to £167,000 to £183,000 per megawatt. That figure counts lost revenue, cost overruns and contractual penalties, and gives no source. Mine counts margin, which is smaller than revenue, and that probably explains part of the gap.

These figures are for a site that’s ready: land bought, customer signed, building finished or close to it. A queue position with nothing behind it loses far less by waiting. Ofgem’s July consultation on data centre connections includes a worked example, a hypothetical 100 MW site whose construction is deferred along with its connection. Pushing its connection back by three years costs it about £10,000 to £13,000 a month per megawatt of connection capacity. On the same basis, mine starts at about £50,000. The difference is the capital a ready site has already committed and has to carry through the wait.

What makes up the cost of waiting: margin never earned, 53 to 68 per cent; value lost for good, 19 to 25 per cent; holding costs, 12 to 22 per cent, of which the commitment fee about 4 per cent. Cost per MW of IT load per month at a two-year wait: £65k at 20 MW, £69k at 50 and 100 MW.

Fogel’s lesson

This is where Fogel earns his place in the series. The cost of waiting is what the grid is worth to a data centre only if waiting is the sole alternative, and it isn’t. Part 2 described on-site generation as the modern duplicate line, built because access to the shared network isn’t available in time. Each of those projects plays the part the canal played in 1890. It is a second-best route, and its existence caps what the first-best one is worth.

So £65,000 to £105,000 a month is an upper bound. On-site power pulls the true value of the grid down, by an amount that turns on what that power costs and how quickly it can be consented and built. That is a harder calculation, and one for another day.

There’s a second gap, and it runs the same way. A developer’s loss isn’t the country’s loss. Some of the margin a delayed site never earns is earned by another site instead, in Britain or abroad. How much depends on how much data centre demand can move, and I haven’t found a published estimate of that for the UK. The private cost of waiting sits above the social one. I can’t yet say by how much.

Who pays for the wires?

Part 1 described the network companies as carriers, paid a regulated return whoever turns up with the freight. Part 2 followed the Victorian surplus lines onto everyone’s fares. This part puts a price on the thing the carrier hands out.

If a month of delay costs a 50 MW site somewhere between £3 million and £5 million, a connection date is one of the most valuable things a network company allocates. Since connections reform it has been allocated on readiness and need as well as on order of arrival. What it hasn’t been, so far, is priced.

If a connection date is worth that much to the customer who receives it, what should it cost them?

Sources

Fogel, R. W. (1964). Railroads and American Economic Growth: Essays in Econometric History. Johns Hopkins Press. The 2.7 per cent figure as reported in Donaldson, D. and Hornbeck, R. (2013), Railroads and American Economic Growth: A “Market Access” Approach, NBER Working Paper 19213.

Hawke, G. R. (1970). Railways and Economic Growth in England and Wales, 1840–1870. Clarendon Press. Figures as reported in Bogart, D. (2013), The Transport Revolution in Industrializing Britain: A Survey, University of California, Irvine, working paper.

Leunig, T. (2006). Time is Money: A Re-Assessment of the Passenger Social Savings from Victorian British Railways. Journal of Economic History 66(3), pp. 635–673.

The Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 1993, awarded to Robert W. Fogel and Douglass C. North. nobelprize.org.

McAinsh, J. (29 July 2026). Power on paper isn’t enough: why certainty is the new capacity. Data Centre Review. Opinion piece.

Ofgem (2026). CMP470, Oversubscribed Technologies Commitment Fee: minded-to consultation.

Ofgem (29 July 2026). Proposed data centre connection reforms: Curate consultation. Chapter 4, case study: IRR impact scenario analysis.

Cost of delay: M D Moore, forthcoming paper on the cost of grid connection delay for data centres. Real 2026 pounds; 8 per cent nominal cost of capital at 2 per cent inflation; 70 per cent load factor; central Scotland, with London as comparator.

Provenance: Original analysis. I haven’t found another sourced estimate of the cost of waiting for a British data centre that breaks it into its parts; the published figures I’ve found either count gross revenue or give no source. Economic history from published work; the link to the grid is my own reading.

What is the grid worth to a data centre? The cost of waiting for a grid connection, £65,000 to £105,000 per MW of IT load per month, set against the 1890 and 2026 comparison of networks and their next best alternative.

Who holds the cards when the biggest AI buyers go small

Published 5 October 2026 · Read the post

Fifty-eight. That is how many of the 315 data centre projects in GB’s connection queue sit below 50MW, and it is the lane the most powerful buyers in the industry may now be shopping in. Read the full piece

On 18 September CNBC reported, citing four people familiar with the conversations, that Anthropic has sounded out data centre deals of 20 to 30 megawatts in the UK and the Nordics. OpenAI has reportedly looked at similar-sized Nordic opportunities. Neither company has confirmed any deal, and Anthropic declined to comment.

The small deals would sit alongside some very large ones. A month earlier, Bloomberg reported a $45 billion, six-year agreement for around 460MW at Nscale’s West Virginia campus. Structure Research’s Jabez Tan told CNBC the attraction of smaller sites is speed to usable capacity: inference can be split across separate sites, whereas training a frontier model needs chips tightly coupled in one place. Crusoe, which built the Abilene campus OpenAI uses, raised $3.9 billion the day before the CNBC report, partly to fund factory-built modular data centres trucked to sites that already have power.

Why it matters

This could have big implications for the UK market. It could be a response to Curate, Ofgem’s proposed reform of the data centre connection queue. It could equally be a response to the social licence problems data centre developers keep running into. In the US, community opposition has become a standing obstacle to large campuses, and TechCrunch notes that Crusoe’s small units could help it sidestep that backlash, at least in part. Closer to home, the SNP’s national council voted in July for a temporary halt to data centre projects without planning consent. On 16 September Holyrood stopped short of a moratorium, but backed a pause on decisions for hyperscale AI data centres until national planning guidance is published, and made environmental impact assessment mandatory above 50MW.

In any event, it is significant. The most powerful players in the industry could now be looking at the smaller projects.

What Curate actually tests

Ofgem’s Curate proposals draw two lines through the queue. A Data Centre Commitment Fee of £237,500 to £712,500 per megawatt would apply above 40MW, secured when a project accepts its connection offer and forfeited if it drops out. Below and alongside that sit new queue milestones testing offtake, procurement, and financial and technical capability. A 20 to 30MW project is inside the milestones, and close to or below the fee line.

That matters because of what the two instruments reward. The fee can be posted as a bond or letter of credit, so in practice it screens on access to credit. Elliott Meighan of Equity Energies made the point first, that the fee rewards whoever has the biggest credit facility. I put it this way in my response to the consultation: the fee will screen by balance sheet depth, and viability and balance sheet depth are not the same property. The financial capability milestone works the same way, asking for an investment-grade rating or committed credit support. Both favour the deepest balance sheets in the market, and few are deeper than a frontier lab and its backers.

The offtake milestone adds a second effect. A developer building to lease or sell must show a binding compute contract by a date fixed by its connection date, which both sides can watch approaching. That hands leverage to the offtaker at a known point in the timetable. When the offtaker is one of the most powerful buyers in the industry, shopping in a lane that holds 58 projects and 2% of the queue’s capacity, the developer’s hand at that date is weak.

Three routes in

There are three ways a lab could take 20 to 30MW in GB, and Curate treats them very differently. The first is space on a connection that is already energised, whether colocation or an existing building. No new offer is accepted, and Curate never applies. The second is new build through the queue, where everything above applies in full.

The third is to bridge: accept the grid offer, post the fee, and run the site on its own power until the connection arrives. Gas networks received 113 enquiries from data centres across 2024 and 2025, according to Future Energy Networks, with gas connections taking 6 to 24 months against up to 15 years for electricity. Aggreko has already bridged a Dublin data centre on 14MW of gas generation and battery storage while it waited for the grid. Bridging at that order of size is commercial today. Generation on site also lowers the import a project needs to request, which is another route under the 40MW line.

Two rulers

The two lines are also drawn with different rulers. The drafted code sets the fee threshold against connection capacity. The milestones follow a definition imported from the Cyber Security and Resilience Bill, set in rated IT load. The consultation narrative describes the milestone line as 10MW. My current reading of the drafted text is that 10MW applies only to a facility serving its owner’s own business, and that for everything else, colocation included, the line is 1MW.

Last month I compared six rulebooks that set a data centre size line and found almost none measuring the same thing. Fingrid draws its large-demand line at 30MW of rated demand, and the band the labs reportedly sounded out sits just beneath it.

The two quantities pull apart once cooling and redundancy are counted. Google reports a 2025 fleet average power usage effectiveness of 1.09, while the Uptime Institute’s 2025 survey puts the global average at 1.54. On those bounds, 30MW of IT means anything from about 33MW to 46MW at the meter. On that reading, a lab renting 30MW from a neocloud would be inside the milestones from the first megawatt, and could sit on either side of the fee line depending on how the site is cooled.

I made this point in my response to the consultation and proposed three fixes. Let the drafted code govern, and stop describing the fee threshold in IT load terms. Calibrate the credit support behind the financial capability milestone to requested connection capacity, so one quantity governs both instruments. And give the milestone threshold its own reviewable instrument, as the fee already has, rather than a definition copied from a Bill that reaches the Lords report stage on 26 October. Ireland’s use of Maximum Import Capacity in MVA is worth a look as the common measure.

What counts as one project?

Ten 25MW sites add up to 250MW of load, and none of them is individually above 40MW. For inference, spreading load that way has a genuine technical logic. It is also the shape that anti-splitting rules are written to catch, and Curate has none for the 40MW line.

Ofgem did think about splitting. It chose a fee that scales linearly with megawatts partly because that ‘reduces the risk of unintended incentives and gaming, such as restructuring or splitting of projects to fall within lower charging bands’ (para 4.22). Linear scaling removes the reward for splitting above the threshold. Below it, the fee drops to zero. At the proposed rates a single 250MW site would post between £59m and £178m, while the same load split ten ways would post nothing. Ireland’s regulator, by contrast, says that multiple applications below its de minimis threshold for the same project or owner must be treated on an aggregated basis, and that splitting a project to qualify will not be accepted (CRU/2025236, section 3.2.5).

If the biggest buyers can meet their needs in blocks below 40MW, should Curate include an aggregation test to de-game the threshold, and what should count as one project?

P.S. Research assistance: Claude, made by Anthropic. Yes, I used Anthropic to check up on Anthropic. It took it very well. But remember that Anthropic has Claude too, so my guess is they settled where to plug in weeks ago.

The thin lane: GB’s data centre queue by size. 58 of 315 projects (18%) sit under 50MW, holding 1,446MW, 2% of the 72,978MW in the queue. Ofgem Curate consultation, Table 1.

Last time Britain let a network boom run

Published 2 October 2026 · Read the post

Thoughts for the weekend. Who pays for the wires? Part 2 of 3.
I used to live in Bolton-le-Sands, a short walk from Packet Bridge. From the bridge you can see where the swift boats called. The jetty was on the canal, the stables for the changes of horse stood between the bridge and the Packet Boat inn, and passengers bound for Kendal or Preston waited in the pub. The stables have gone and the pub has closed. The canal company that ran the service has gone too: within a generation of the railway arriving it had leased the canal to a railway company, and in 1885 it sold it outright. Read the full piece

In part 1 I wrote about the network companies as carriers, paid a regulated return whoever turns up with the freight. This week I want to go further back, because Britain has run this experiment before, twice, and the records are unusually good.

In 1847 British railway companies spent £44 million on construction. Andrew Odlyzko puts that at almost 8 per cent of GDP, roughly twice the military budget of the same year. The same share of today’s economy would be more than £200 billion, spent in a single year. The canals had been through their own mania half a century earlier. Both booms were private capital building networks that Parliament then had to work out how to govern, and several of the answers it reached are still sitting in the electricity licence today.

The canal company owned the water, not the boats

Every canal needed its own Act of Parliament. Those Acts handed the company a monopoly over the waterway itself but not over the traffic on it; independent carriers ran the boats and paid tolls for the privilege. Dan Bogart’s survey of the transport revolution is the clearest short account of how that worked.

Anyone who read part 1 will recognise the arrangement. A distribution network operator’s licence carries a standard condition prohibiting the licensee from generating, so the company that owns the wires is kept out of the business of filling them.

Gladstone and the first price regulation

By 1844 the railways had proved themselves, technically and commercially, and were paying large dividends. Gladstone, then at the Board of Trade, took the view that competition between the companies couldn’t be relied on to protect the people using them. The Railway Regulation Act of that year gave the state powers to cap rates, to buy railways into public ownership, and to impose detailed price and quantity conditions. Iain McLean’s work on the politics of the Act is worth reading if you want to know how it got through a Parliament full of railway shareholders.

It also created the parliamentary train: one service a day each way, at least 12 miles an hour including stops, in carriages protected from the weather, for no more than a penny a mile. Plenty of companies met the letter of that and ran the train at an hour nobody wanted to travel.

Ten years on, the Railway and Canal Traffic Act 1854 went further. Railway and canal companies now had to afford all reasonable facilities for receiving, forwarding and delivering traffic, and they couldn’t give any undue or unreasonable preference to one user over another.

That language never really left. Jon Stern’s economic history of UK utility regulation argues that the model worked out for Victorian railways became the basic template for regulating electricity, gas, water and telecoms, and that electricity repeated the railway pattern of early competition followed by consolidation into regional monopolies. Section 16 of the Electricity Act 1989 puts a distributor under a duty to connect on request. The distribution licence carries a standard condition headed “Prohibition of discrimination”, covering the chapter that sets out connection terms and charges. I read that as the 1854 principle applied to a different commodity; the licence itself doesn’t cite the Victorian Act, so the link is mine.

What happened when access was refused

If you read one book on this, make it Mark Casson’s The World’s First Railway System (2009). Casson builds a counterfactual, the most efficient network that could have been built with what was known at the time, and sets it against the network Britain actually got.

His account runs like this. The companies grew into powerful regional monopolies and then contested each other’s territory. When one company was refused access to a rival’s lines, it built a duplicate line of its own. Gladstone’s plans for an integrated national system were blocked by MPs protecting local interests, each town wanted more railway than its neighbours, and too many lines were built. The cost of those surplus lines came back as higher fares and freight charges, and the whole economy paid it.

Access could also be taken rather than refused, and the canal I started with was on the receiving end. The Lancaster Canal Company had leased the Lancaster and Preston railway in 1842, two years after it opened. In 1846 the new Lancaster and Carlisle company began running its trains on over the Lancaster to Preston line without an agreement to do so. The government inspector found them operating “in defiance of” the owners’ opposition. The tolls due on those trains were held back for almost three years while the companies argued, and the matter was settled only when the Lancaster and Carlisle took the line over.

Two network booms side by side: British railway construction of £44m in 1847, almost 8% of GDP, and contracted demand connection offers in Great Britain rising from 41 GW to 125 GW between November 2024 and June 2025, at least 80 GW of it data centres.

The duplicate line today

Now look at the connections queue. Ofgem reported in July that contracted demand offers rose from 41 GW to 125 GW between November 2024 and June 2025, with data centres accounting for at least 80 GW of that. Nscale was told this summer that grid power wouldn’t reach its Essex site in time for the planned opening, and reports say it’s looking at gas fuel cells instead. Future Energy Networks, the gas network trade body, says its members are processing 46 applications from data centre developers.

My reading, and it’s only mine, is that on-site generation built to bridge a connection delay is the modern duplicate line. It’s a second set of assets, built because access to the shared one isn’t available on the timescale the customer needs.

Two things are different this time. The duplicate is generation rather than track, so it doesn’t compete with the network for traffic in the way a second railway did. And the developer pays for it directly, while the network it’s waiting for is paid for through a regulated return spread across every user. The Victorian surplus lines found their way onto everyone’s fares in the end. Whether today’s bridge power stays on the developer’s balance sheet, or eventually lands on the system through stranded assets, flexibility contracts or something nobody has designed yet, is a question I don’t think anyone has answered.

What the Victorians left us

What the Victorians left us, I think, is the queue. Once a network owner can’t give undue preference to one customer over another, the safest way to decide who goes first is the order they turned up in, and that principle ran from the canal and railway Acts into the Electricity Act 1989 and the licence conditions that followed it. Connections reform hasn’t repealed any of that; the duty to connect and the ban on discrimination are both still there. What it has done is change what counts as undue. Being ready, and being needed by the system, are now good enough reasons to go ahead of someone who applied first, and Ofgem has said plainly that first come, first served is no longer fit for purpose. The US regulator got there first: FERC’s 2023 reform argued the old first-come queue had itself become unduly discriminatory against projects that were ready to build. After 170 years, Britain has decided that some preference is due.

If the cost of the Victorians’ surplus lines ended up on the fare, where do you think the cost of today’s bridge power will end up?

Sources

Bogart, D. (2013). The Transport Revolution in Industrializing Britain: A Survey. University of California, Irvine, working paper.

Casson, M. (2009). The World’s First Railway System: Enterprise, Competition, and Regulation on the Railway Network in Victorian Britain. Oxford University Press.

Electricity Act 1989, section 16 (duty to connect on request), as substituted by the Utilities Act 2000.

Future Energy Networks, statement on data centre gas connection applications, reported by New Civil Engineer, 2026.

Garstang & District Heritage Society. Packet boats on the Lancaster Canal.

Laffan, R. M. (1848). Lancaster and Preston Railway: Report to the Commissioners of Railways, 24 August 1848.

Lancaster and Preston Junction Railway Act 1843 (6 & 7 Vict. c. iv). Lancaster and Preston Junction Railway Amendment Act 1849 (c. lxxxvii).

Lancaster Canal Trust. Our history.

McLean, I. and Foster, C. (1992). The Political Economy of Regulation: Interests, Ideology, Voters, and the UK Regulation of Railways Act 1844. Public Administration 70, pp. 313-331.

Odlyzko, A. The collapse of the Railway Mania, the development of capital markets, and Robert Lucas Nash. University of Minnesota.

Ofgem (29 July 2026). Ofgem acts to free up grid capacity by tackling speculative data centre projects. Press release.

Ofgem. Standard conditions of the Electricity Distribution Licence: condition 19, Prohibition of discrimination under Chapters 4 and 5; prohibition on generating by the licensee.

Railway Regulation Act 1844 (7 & 8 Vict. c. 85). Railway and Canal Traffic Act 1854.

Federal Energy Regulatory Commission (2023). Order No. 2023, Improvements to Generator Interconnection Procedures and Agreements. Federal Register, 6 September 2023.

Ofgem (2025). Proposed licence changes to enable TMO4+ connections reform. Consultation.

Stern, J. (2017). Competition, economic regulation and affordability in infrastructure industries: an economic history 1840-1980. CCRP Working Paper No 30, City, University of London.

Data Centre Review and Results Sense, July 2026, on the Nscale Essex site.

Synthesis. Historical figures and mechanisms are from peer-reviewed economic history, primary legislation and local history sources; the Packet Bridge detail is from my own knowledge of the village; the parallel with bridge power is my own reading, not a finding of those authors.

Last time Britain let a network boom run: the Lancaster Canal from Packet Bridge, Bolton-le-Sands. Photograph: Mark D. Moore.

Fixing the rules before the result

Published 29 September 2026 · Read the post

On 27 July I published a piece asking who actually owns Britain’s energy transition. It traced fifteen large solar and battery schemes, 8,850 MW between them, from the developer named in the public record to the owner at the end of the chain. For at least ten of the fifteen, the named developer was not the economic owner. A follow-up on 10 August publicly corrected one claim in that first piece. Read the full piece

That was a snapshot. The more useful question is whether the pattern continues, and the way to answer it is to fix the rules before looking. This piece sets out those rules. I’ll measure the same fifteen schemes on 27 January 2027 and again on 27 July 2027, and publish whatever the measurement shows.

The cohort

The fifteen schemes are those published on 27 July. None has been added, removed or replaced.

SchemeMWTechnologyIdentity
Thorpe Marsh1,450BESSFull
Springwell800Solar and BESSFull
One Earth740Solar and BESSFull
Trafford680BESSWeaker
Cottam600Solar and BESSFull
Coalburn 1500BESSFull
Coalburn 2500BESSFull
Heckington Fen500Solar and BESSFull
Gate Burton500Solar and BESSFull
Sunnica500Solar and BESSFull
West Burton480Solar and BESSFull
Gateway450BESSFull
East Yorkshire400SolarFull
Hams Hall400BESSWeaker
Pembroke350BESSFull
Total8,850

MW figures are as published on 27 July. On the national grid register snapshots of August 2026, Heckington Fen appears at 400 MW and Thorpe Marsh at 1,400 MW.

What counts as a change

A scheme counts as having changed hands when a new outside owner takes 25% or more of the project company, directly or through the companies above it. Outside means an owner that wasn’t already in the chain on 27 July. The evidence is a new person with significant control notified at Companies House.

I’ve used 25% rather than a majority for two reasons. It is the threshold at which Companies House requires an owner to be declared, so every counted event can be checked by anyone on the public register. And capital in this market often arrives as a half-share rather than an outright purchase. A majority-only rule would miss most of the transactions the first piece found.

Three further rules apply:

A reorganisation inside the same group, where the owner at the top doesn’t change, doesn’t count.

A change counts in the period in which it completes, whenever it was agreed. A stake sale agreed before 27 July that completes afterwards is a change.

Where one transaction moves several schemes, it is reported as one transaction affecting several schemes. Every result will give both figures, in the form “three schemes changed hands in two transactions”.

Why these dates

The start date, T0, is 27 July 2026, the day the first piece was published. It can be checked against the publication timestamp. I fixed it as the start date on 14 August, after that piece had gone out, and I’m stating that here.

The two measurements fall six and twelve months later. Across fifteen schemes, too few events happen in a month to measure monthly. Companies House filings also lag the events they record, and by different amounts. At Coalburn 1 the new owner was notified on 7 August and filed six days later, but three directors appointed the same day were filed on 24, 25 and 28 September, seven weeks or more after the event. Any interval much shorter than three months would miss changes that had already happened. The second measurement tests whether the first was signal or noise.

Two schemes are harder to watch

Each scheme is identified by its entry on the national grid register and its Companies House number. For thirteen of the fifteen, both can be matched. For two they can’t.

Trafford has no entry of its own on the grid register, and I haven’t been able to confirm which company holds it.

Hams Hall has no entry of its own on the grid register either. The entry under that name belongs to a different developer’s scheme at the same substation. The owner is known, but the project company below it hasn’t been identified.

Both stay in. Each result will be reported twice, side by side: for all fifteen schemes, and for the thirteen that are fully identified. If the two figures differ, the reason will be visible.

What the checks found about the starting position

Before publishing this, I checked every scheme against Companies House and the grid register. The record I held from 19 July was wrong or out of date on six schemes. Each of the following is the position as it stood on 27 July. None of them counts as a change, because none happened after that date.

SchemeRecorded on 19 JulyPosition at T0 on the register
Gate BurtonEDFStill EDF. On 10 June 2026 EDF Energy Renewables moved its holding down a level: Orion Holdings 2 Limited, which EDF Energy Renewables owns outright, became the direct owner. The owner at the top is unchanged.
CottamMacquarie, 100%The Macquarie holding companies ceased as owners on 3 July 2026 and a new holding company took their place. The company at the top of the new chain states that it has no registrable owner. A charge in favour of Shell New Energies UK Limited was created the same day and delivered to Companies House on 14 July. Whether this was a reorganisation or a sale is not established.
West BurtonMacquarieHeld through a chain administered by Foresight Group. The Macquarie attribution was wrong.
SpringwellEDF 50%, Luminous 50%EDF between 50% and 75%, Luminous between 25% and 50%.
Hams HallPenso, Luminous and BW jointlyBW ESS Asset UK Limited, 75% or more. BW ESS is part of BW Group.
GatewayOwner not resolvedResolved. The chain ends at a private individual, who is not named here.

I checked the week between the end of tracing on 19 July and the start date on 27 July separately. No scheme recorded an ownership event in that week.

One result is already known

The same checks found one change after the start date. On 7 August 2026 a new owner of between 25% and 50% was notified for Coalburn 1. The chain above it ends at AXA. It will be counted in January under the rules above.

I’m stating it now, before anyone else finds it. The rules weren’t written around it: the completion rule was settled on a different scheme, Coalburn 2, and the 25% threshold is the one Companies House already uses.

What this can’t show

Fifteen schemes is a case series. The result will describe what happened to these fifteen, and it can’t be read across to the market as a whole. It won’t show why ownership moved either. A change after consent would be consistent with consent acting as the trigger, and ordinary portfolio decisions that would have happened anyway would leave the same record. The register shows who owns what, and from when. Motive isn’t on it.

The measurement is still worth taking. Most of what is said about who owns the energy transition is assertion, and this is a small, dated test with the rules written down first.

Sources

Mark Moore, Who actually owns Britain’s energy transition?, LinkedIn, 27 July 2026 (T0)

Mark Moore, Your biggest customer might have changed ownership twice. Did anyone tell you?, LinkedIn, 10 August 2026

NESO, Transmission Entry Capacity Register, snapshots of 14 and 25 August 2026

Companies House, Alcemi Storage Developments 4 Limited (12376965), persons with significant control

Companies House, Glencoe Topco Limited (16379743), persons with significant control

Companies House, Gate Burton Energy Park Limited (12660764), persons with significant control

Companies House, Cottam Solar Project Limited (12711231), persons with significant control

Companies House, Cottam Holdco 2 Limited (17298479), charges

Companies House, West Burton Solar Project Limited (13049324), persons with significant control

Companies House, Springwell Energyfarm Limited (13484004), persons with significant control

Companies House, Hams Hall Holding UK Limited (16642921), persons with significant control

Companies House, Gateway Energy Centre Limited (07146501), persons with significant control

Fixing the rules before the result: the fifteen solar and battery schemes, 8,850 MW, with Coalburn 1 marked as the one new outside owner since 27 July 2026, and measurement dates of 27 January and 27 July 2027.

Why does CMP470 make sense?

Published 28 September 2026 · Read the post

Ofgem’s minded-to consultation on CMP470 closes on Thursday 1 October. It proposes an Oversubscribed Technologies Commitment Fee, a charge on battery storage projects for holding a queue position in a technology the system already has too much of. I’ve submitted a response supporting it. This piece sets out why, and what I’ve suggested alongside it. Read the full piece

The problem in two numbers

Around 90 GW of battery storage is either operating or holds Gate 2 status: roughly 7 GW of the first and 83 GW of the second. Estimated need by 2035 is around 29 GW. Some surplus is useful, because projects drop out and others need to be ready to replace them. Three times the requirement is well past that point, and network companies still have to plan for all of it.

Holding a place has cost very little. Ofgem’s impact assessment shows that 26% of the battery capacity at Gate 2 or already energised has posted no securities or liabilities at all. Counted by project, half have posted less than £800/MW in securities. Projects commissioning in 2027, where Ofgem presumes the full user commitment regime already applies, have posted £5,600/MW on average. At the low end, a queue position is close to free.

Why the fee works

The fee starts at £3,000/MW. If the oversubscription doesn’t fall, it rises to £5,000/MW after six months and then by £5,000/MW every six months to a ceiling of £25,000/MW. None of that bites straight away: the fee first appears in the July 2027 securities statement and is payable from October 2027.

Ofgem puts the ceiling at more than 25% of build cost for a short-duration battery and more than 7% for a four-hour one. The £3,000/MW start is 12% of the ceiling, which on the same basis works out at roughly 3% and under 1%. A developer who intends to build can carry that. One holding a position as an option has a reason to think again, and plenty of notice to do it.

That’s the job a commitment fee does. Ofgem has no direct read on whether a developer means to build, so the fee asks for money and lets the answer sort the queue.

Seeing who holds the queue

The impact assessment looks at the response to the fee mainly as a choice between staying and leaving. There’s a third option, which is to sell.

A developer that would rather not fund a rising security can sell the project company to someone better placed to. The capacity stays in the queue and the new owner posts the fee. Ofgem has seen the pressure coming. Stakeholders told it that stronger fees fall hardest on smaller, less well-capitalised developers, and the impact assessment accepts that a liquidity requirement may favour larger balance sheets. What neither document goes on to consider is a change of ownership.

Ofgem prices the interest cost of the fee at a single real rate of 4.08%, taken from BBB-rated utility bonds, and applies it to every developer. The index behind that rate only admits bonds rated BBB- or better. A smaller developer without an investment-grade rating is likely to borrow above it, so its cost may be somewhat higher than the headline figure suggests.

Changes of ownership are already hard to see, whatever the reason for them. In July 2026 Alpiq bought 90% of Harmony Energy, with co-founder Peter Kavanagh keeping 10% and staying on as chief executive; brand, team and operating model all carried over. On 9 July, as the deal completed, Companies House stopped naming anyone as controlling the company. Kavanagh’s 10% sits below the reporting threshold, and the company filed a statement that it has no registrable person or registrable legal entity. The TEC Register lists project companies and carries no ownership field, so nothing on it records the change. There’s nothing to suggest the sale had anything to do with the fee, which isn’t active yet. It’s simply an example of how little a change of control shows up.

Exits can be just as hard to read. Between the register snapshots of 14 and 25 August 2026, seven projects totalling 4.2 GW left the TEC Register. All seven sit in companies of one joint venture between Elmya Energy and Renewable Power Capital. At Companies House, the seven and a related development company were all reorganised on the same day, 31 March 2026, with identical changes of registered office, directors and company secretary. Nothing was filed around the August exit, and the companies remain active. Anyone reading the register line by line would log seven separate exits; it takes the company names, and a look at Companies House, to see a single owner’s decision behind them.

Gate 2 status is also now a deal term. In May, Gresham House agreed to buy a 480 MW battery project near Rayleigh, conditional on the project receiving a favourable Gate 2 offer.

None of this counts against the fee. If a project passes to an owner who can post the security and build it, the queue is better used for it. Ofgem concludes that competition effects are likely to be limited, and that may well prove right. Ownership data would allow it to be tested, and would show changes in the structure of the battery pipeline that the register on its own can’t.

The data centre door

The impact assessment notes that some battery projects are trying to modify their Gate 2 offers to connect as data centres instead, citing up to 9 GW. Demand is outside the fee, so conversion looks like a way out. NESO’s guidance closes most of it. A full change from battery storage to a data centre is treated with a later queue position, because a data centre imports at close to full load for far more hours. Converting costs the place in the queue.

Two things are still unclear: what happens to fee already posted when a project converts, and how a converting project sits against the separate commitment fee proposed for data centres under Curate. I’ve asked Ofgem to confirm both.

What would help

I’ve suggested two things, both modest. NESO could collect the ultimate controlling party of each fee-liable project when securities statements are issued, starting with the first one in July 2027. It needn’t be published project by project. An aggregate figure, such as the share of liable capacity held by the largest owners, would do. Then a review after the first two escalations, comparing ownership at activation with ownership at review and setting exits against sales.

With that data in hand, Ofgem could track who ends up holding the queue once the fee starts to bite.

If the fee works and the queue shrinks, how would anyone know whether the capacity that remains is held by more owners or fewer?

Sources

Ofgem, CMP470 minded-to consultation, September 2026 (paras 2.5, 3.18, 3.23, 3.24, 5.13, 5.14)

Ofgem, CMP470 Initial Impact Assessment, 17 September 2026 (paras 1.2, 2.15 and fn 15, 2.16, 2.17, 4.16, 4.17 and fn 25, 4.19; Figures 2 and 3)

FTSE Russell, FTSE UK Broad Investment-Grade Bond Index (UKBIG) factsheet, as cited by Ofgem in IIA fn 25

NESO, Guidance on Material Technology Changes (section 4.1)

Fasken, Harmony Energy completes sale of a majority stake to Alpiq, July 2026

Companies House, Harmony Energy Limited (10141078), persons with significant control

Companies House, RPC Elmya Carnation company filings

NESO TEC Register, snapshots of 14 and 25 August 2026

Energy-Storage.news, UK BESS M&A rebounds, 9 June 2026 (Gresham House, Rayleigh)

Graphic accompanying the article: Why does CMP470 make sense? CMP470: the argument in five steps, from oversubscription to the likely response, with sources for each step

What a data centre actually buys from the grid: sources, figures and what’s still open

Published 25 September 2026 · Read the post

Britain’s network companies aren’t paid the way most businesses are. They’re regional franchise carriers. Their revenue is set in advance by a five-year regulatory settlement, and any gap between what they’re allowed and what they actually collect is corrected in later years. Read the full piece

Ofgem’s own explanation is that “it is not practical to collect the exact revenue allowed owing to tariffs being set before network usage is known” (RIIO-ED1 Annual Report 2015-16, p.10). The volume risk sits with customers, not the carrier. Below is what that means for a large connecting customer such as a data centre, with the source for each figure and a note of what was left out because it couldn’t be verified.

One point first. The charges below don’t all fall on the same kind of site. A site connected directly to the national transmission network pays transmission charges but no local distribution charges. A site connected to a local network pays both. The post’s three conditions come from across the system, not from a single site’s bill.

What’s charged

For sites connected directly to the transmission network, the first charge is a fixed daily transmission residual: £20,829.24 per site in the top band (T-Demand4) for 2026/27 (NESO, Final TNUoS Tariffs for 2026/27, January 2026, Table 10). NESO charges it per site per day according to band, not according to how much power the site draws. Sites are placed in bands by capacity percentile and stay there for the whole price control (NESO, Transmission Demand Residual Guidance, pp.6-8). Locational transmission charges come on top. Over a full year the residual comes to about £7.6 million per site before a single unit of power is used.

For sites connected to local networks, the cost of reinforcement is spread across all customers. Reinforcement means upgrading substations and cables to serve a new connection. Ofgem’s Access and Forward-Looking Charges Significant Code Review (Final Decision, May 2022) removed the reinforcement contribution from demand connections entirely. It moved that cost into Distribution Use of System (DUoS) charges, which every customer on the network pays (Executive Summary, p.5; sections 3.39 to 3.41). The changes took effect from 1 April 2023 (Ofgem, charging and connections update, 27 April 2023, p.4). Generation connections keep a narrower boundary: a connecting generator still pays for reinforcement at its own voltage level, and only the higher-level work is spread across customers.

Where the spending goes

Network spending is split each year by a capitalisation rate. Part is recovered from customers in the same year (“fast money”). The rest is added to the company’s Regulatory Asset Value (“slow money”), and the company’s cost of capital is calculated as “a percentage return on the licensee’s RAV” (Ofgem, RIIO-ED2 Price Control Financial Handbook, December 2022, §3.6 and §4.1). The asset base is paid down over its working life. Until then, the company earns a return on what customers have funded and it owns.

What the carrier earns

Ofgem’s RIIO-3 price control, decided in December 2025 and in force from 1 April 2026, sets electricity transmission’s allowed cost of equity at 5.70%. That figure is on a semi-nominal basis and assumes 55% notional gearing (Finance Annex, 4 December 2025, section 1.10; confirmed against National Grid Electricity Transmission’s own Final Determinations, Table 5). It replaces the RIIO-T2 baseline of 4.30%, CPIH-real. The two figures aren’t on the same inflation basis, so the rise shows the direction of travel rather than a like-for-like increase.

A second RIIO-3 figure, 6.12% at 60% gearing, is for gas transmission and gas distribution, not electricity. At least one press report has run the two together without making that clear. They’re two sectors within the same settlement, not two estimates of one number.

On the distribution side, RIIO-ED2 set a baseline cost of equity of 5.23%, CPIH-real, at 60% notional gearing (Final Determinations Overview, November 2022, section 5, pp.29 to 33).

These are allowed returns, not guaranteed profit in the strict sense. Ofgem’s RIIO-2 performance data for 2024 reports “modest operational out-performance driven by efficiencies in operating costs, capital expenditure and incentives”, plus larger gains on finance costs where high inflation interacted with the debt allowance. The baseline itself, though, is fixed by formula on a five-year cycle whatever the network’s day-to-day performance. That’s what the post means by a set return.

A short word on producers

Across Great Britain as a whole, the average transmission charge paid by generators is held between €0 and €2.50/MWh by a retained EU regulation (Commission Regulation (EU) No 838/2010). NESO sets a negative adjustment tariff to keep the average in range (NESO, Calculation of the Generator TNUoS Adjustment Tariff, May 2026, p.1). Individual generators can pay more or less than that depending on where they are; only the national average is capped.

Since April 2023, balancing costs have been charged to final demand only. Balancing costs are the cost of keeping supply and demand matched in real time. Ofgem approved that change in April 2022 under CUSC modification CMP308.

The CfD scheme is funded by “a compulsory levy on electricity suppliers in Great Britain” (Explanatory Memorandum to SI 2016/363, §7.3), which reaches customers’ bills.

That isn’t the full picture of producer economics. It’s enough to show that the carrier doesn’t set the price of what it carries.

Private wire

The alternative to the carrier is a private wire: generating power nearby and running your own line to the site. It avoids the usage-based network charges on the power it carries. It doesn’t remove the fixed charges. Ofgem’s Targeted Charging Review (December 2019) made residual charges “fixed charges for all households and businesses”. NESO’s guidance treats a site with on-site generation that stays connected to the grid as liable, and gives a data centre as its example (Transmission Demand Residual Guidance, pp.5 and 10). Even when you run your own fleet, the truck is still on the bill.

What was left out

Three claims were cut during checking rather than softened. That stranded reinforcement costs sit on customers’ bills: no Ofgem source found. That revenue corrections carry interest: not confirmed in the price control documents. That a private wire lets a site skip the regional network: wrong, as the fixed charges stay whether or not the site draws power.

Sources

NESO, Final TNUoS Tariffs for 2026/27, v1.0, January 2026.

NESO, Transmission Demand Residual Guidance.

NESO, Calculation of the Generator TNUoS Adjustment Tariff, May 2026.

Ofgem, Access and Forward-Looking Charges Significant Code Review: Final Decision, May 2022.

Ofgem, Charging and connections prioritisation update, 27 April 2023.

Ofgem, Targeted Charging Review: Decision and Impact Assessment, December 2019.

Ofgem, CMP308: Removal of BSUoS charges from Generation, decision, April 2022.

Ofgem, RIIO-3 Final Determinations: Finance Annex, 4 December 2025.

Ofgem, RIIO-3 Final Determinations: NGET, December 2025.

Ofgem, RIIO-ED2 Final Determinations Overview, November 2022.

Ofgem, RIIO-ED2 Price Control Financial Handbook, December 2022.

Ofgem, RIIO-ED1 Annual Report 2015-16.

Ofgem, RIIO-2 regulatory performance data: 2024.

Explanatory Memorandum to The Contracts for Difference (Electricity Supplier Obligations) (Amendment) Regulations 2016 (SI 2016/363).

Graphic accompanying the article: what a data centre actually buys from the grid

Four gigawatts left a connection register this year, and the register cannot say where it went.

Published 21 September 2026 · See the trackers

Between January and August 2026, accepted-to-connect capacity on SSEN’s Embedded Capacity Register fell by 4,054.8 MW. Connected capacity over the same period rose by 197.6 MW. Those are two columns in the same file, and nothing in the register links them. You cannot say the second came out of the first. Read the full piece

I want to be precise about the gap, because the precision is the whole point. Roughly 3.9 GW is a derived residual. It is the arithmetic difference between one column falling and another rising, and it is not a measurement of anything. Some of it energised. Some of it was re-scoped, or re-sited, or given up on. Some of it may be a correction to a figure that was wrong before. The register records none of those events, because a register is not an event log. It is a statement of contract position at a moment, and the fields populate when a contract does, not when something happens on the ground.

That distinction governs everything about how these files can be read, and it is why I publish the method before I publish the series.

Here are the rules I hold to.

Diff on a stable identifier, one row per scheme, and state the counting basis wherever a total appears. The registers invite double counting. A scheme can appear twice when it is moving between grid supply points, and if you sum the rows you will count its capacity twice.

Never describe a departure as a withdrawal. A row leaving the accepted-to-connect column tells you the contract position changed. It does not tell you the project died. Treating the two as equivalent produces a story about collapse that the data does not support, and the mistake is easy to make because collapse is the more interesting story.

Never sum registers that do not share a population. The NESO Embedded Register covers Scotland only. The DNO Embedded Capacity Registers cover the fourteen licence areas. Adding them produces a number that describes nothing.

Treat dates as fields, not facts. A target energisation date is defined in the governing template as an estimate “likely to change to reflect the latest date notified by customers”. It is what the customer last said, recorded when someone got round to recording it.

Fix the snapshot provenance before anything else. NESO overwrites each publication, so a file you did not download on the day is gone. The publication date in the filename is the only provenance record, which is why I never rename a downloaded file.

None of that is difficult. It is just slow, and it has to be done the same way every week or the series means nothing.

The distribution registers are harder than the transmission one, and I spent yesterday finding out how much harder. The governing template has mandated a Unique Site Identifier since February 2026, defined as an identifier that “distinguishes each entry uniquely rather than use site name”. Seven months on, three of the six DNO groups do not publish it at all. Of the three that do, one publishes an identifier that repeats across paired rows, so it is not unique. Another publishes one that has the connection status written into the string, which means it changes the moment a scheme energises, so it cannot follow a project across precisely the transition you would want to track. One publisher, out of six, has an identifier that is both unique and durable.

There is a reason to say this out loud rather than quietly work around it. Connections reform is being run to targets. If the registers that record its progress cannot support a consistent count across editions, then the reporting on it rests on people rebuilding the same fragile joins in private, each slightly differently, and nobody being able to check anyone else’s number.

Tomorrow I publish the first edition of the Embedded Capacity Register tracker, covering all six DNO groups. It will carry the national accepted-to-connect position, the method used to arrive at it, and the places where I could not arrive at one cleanly and have said so instead.

One question I cannot answer from the files. When capacity leaves the accepted-to-connect column, somebody knows whether it energised or whether it went away, because somebody countersigned the paperwork either way. What would it actually cost to publish that, and who decided it was not worth recording?

Provenance: Primary. SSEN movement computed from its published Embedded Capacity Register, January and August 2026 editions. Template and identifier findings from the DCUSA Agreed Version 5.0 register template and the six DNO groups’ own current files, collected 18 September 2026.

Embedded Capacity Register tracker graphic, Edition 1

72.8 GW requested, 4 GW connecting, 1.4 GW drawn: the workings behind the data centre queue

Published 18 September 2026 · Read the post

This is the appendix to my post of Friday 18 September. The post gives three numbers for Great Britain’s data centre pipeline. This piece shows where each one comes from, how the arithmetic runs, what the sources assume, and what the figures cannot tell you. Anyone with the three source documents open should be able to reproduce every figure here in a few minutes. Read the full piece

All the underlying data is NESO's or TwentyForty's. The calculations are mine, as is the reading of the connection terms, and one observation comes from an original research project of my own that has not yet been published. Where that is the case, I say so.

1. The three numbers

Requested: 72.8 GW of data centre connection requests through to 2039. Source: NESO's written evidence to the Environmental Audit Committee (DCU0081), based on pipeline data collected in 2025 from transmission and distribution network operators.

Expected to be connected by 2030: 6.04 GW in total, of which 2.02 GW was already connected at 1 January 2025, so about 4.0 GW is new. Source: NESO Ten Year Outlook 2026 data workbook, sheet ED1, cells S39 (1 January 2030) and N39 (1 January 2025).

Expected average draw in 2030, whole fleet: about 1.4 GW. Source: derived from the same workbook, sheet ED1, cell S21, which gives 12,579 GWh of data centre demand in financial year 2030.

2. The arithmetic

Average draw. NESO gives demand as energy over a year, not as power. Dividing annual energy by the hours in the year converts it: 12,579 GWh divided by 8,760 hours gives 1.436 GW, which I round to 1.4 GW.

New capacity connecting between 1 January 2025 and 1 January 2030: 6.04 GW less 2.02 GW is 4.03 GW. As a share of the 72.8 GW requested, that is 5.5 per cent. The 6.04 GW total cannot be set against the request directly, because it includes the fleet that was already connected before the 2025 pipeline data was collected.

The 1.4 GW of average draw is likewise a whole-fleet figure, existing sites included. It describes how hard the connected fleet is expected to work, not how much of the queue turns into load, and I do not express it as a share of the request.

Average draw as a share of connected capacity in 2030: 1.436 GW divided by 6.04 GW is 24 per cent. That sits below NESO's own utilisation assumption of 35 per cent for 2030, and the gap is expected rather than an error. NESO's figure is measured against built capacity, and it assumes new sites take seven years to build out to their full connection, so in any given year built capacity is less than connected capacity. Measured against the larger connected figure, the same demand gives a lower percentage.

The same check runs on 2025. The workbook gives 4,603 GWh of demand for financial year 2025, an average of 0.53 GW, against 2.02 GW connected at 1 January 2025. That is 26 per cent of connected capacity, again below NESO's 29 per cent of built capacity for the same year. The figures hang together.

One further split is visible in the workbook. Of the 6.04 GW connected by 2030, 2.19 GW is at transmission level and 3.86 GW at distribution level (cells S40 and S41). Transmission-connected capacity was 0.02 GW at 1 January 2025 (cell N40), so almost all of the 2.19 GW is new. NESO's evidence to the committee puts nearly 59 GW of data centre requests in the transmission queue alone. On NESO's own numbers, about 2.17 GW of that, roughly 3.7 per cent, is connected by 2030.

3. What NESO assumes

The Ten Year Outlook 2026 assumptions workbook (sheet Demand Modelling, rows 32 to 35) sets out the four levers that turn the queue into demand.

Queue success. NESO applies a maximum success rate of 50 per cent to projects with accepted applications in 2026, falling over time and for applications that are less progressed. It records that the rate was reduced from last year because the 2026 queue data contains more speculative projects. The queue itself is listed in NESO's bibliography as not public data, so the size of the 2026 queue behind the 6.04 GW cannot be checked from outside.

Ramp. New data centres are assumed to take seven years on average to build out to their connection capacity, adding one seventh each year.

Utilisation. NESO assumes a fleet-wide average of 29 per cent utilisation of built capacity in 2025, rising to 35 per cent in 2030 and holding there. The 2025 value comes from data centre utilisation data supplied by UK Power Networks and National Grid Electricity Distribution. The increase to 2030 rests on stakeholder feedback that greater AI use should raise utilisation.

Efficiency. No net reduction in demand from efficiency improvements is assumed within the ten-year window.

4. What changed since last year

NESO's evidence to the committee described its previous ten-year forecast as around 5.2 GW of connected data centre capacity and just over 20 TWh of demand by 2030. The 2026 outlook raises capacity to 6.04 GW and lowers demand to 12.6 TWh.

The two moves pull in opposite directions for one reason. NESO records that the new utilisation data reduced its assumption from a flat value of around 70 per cent used in Future Energy Scenarios 2025 to 29 per cent. More capacity is expected to connect, and each megawatt of it is expected to do much less work.

5. What the meters show

NESO's utilisation figure is a modelling assumption. TwentyForty, the freight infrastructure body, has published the metered picture for one network. Its report Power In (July 2026) analyses UK Power Networks' half-hourly utilisation records for all 96 data centres on its distribution network, more than five million readings since January 2023.

TwentyForty found that the median site drew 18 per cent of the capacity it had contracted, and the twelve largest connections ran at a median of 7 per cent. On a peak basis, nearly half the sites never exceeded 40 per cent of their booked capacity, even in their single busiest half-hour across three years. The credit for that analysis is TwentyForty's, and anyone interested in the detail should read the report.

The two sources measure different things and should not be combined. TwentyForty's 18 per cent is average draw against contracted capacity at existing sites on one network. NESO's 29 per cent is average utilisation of built capacity across the whole fleet, used as a forecasting input. They point the same way from different methods, which is as far as the comparison goes.

6. What the connection terms already allow

This section draws on an original research project of mine on the utilisation of industrial connections, which is currently unpublished.

Large distribution-connected sites with current-transformer metering are covered by Section 3 of the National Terms of Connection, currently version 23.0, effective 22 January 2026. Clause 12.11C of that section provides that where import through a connection point does not, at any time during any twelve consecutive months, exceed 75 per cent of the Maximum Import Capacity, the network operator may, within the following month and having due regard to all the circumstances, propose a reduction. The reduction can be no lower than the highest import in that period. The provision was introduced by change proposal DCP115 and approved by the Authority in July 2015.

The test in the clause is a peak test. It asks whether import ever exceeded 75 per cent in a year, not what the average was. That makes TwentyForty's peak finding the relevant comparison rather than either average figure: on its data, nearly half of UK Power Networks' data centres never once went above 40 per cent of booked capacity in three years.

The limits need stating as plainly as the finding.

The right is discretionary, and a proposed reduction does not bind unless it is agreed, or determined under Clauses 12.12 or 14. Capacity agreed within the previous twelve months is protected by Clause 12.13.

It applies only to distribution connections on the National Terms. Transmission-connected sites, 2.19 GW of NESO's 2030 figure, are contracted on different terms. Existing bespoke agreements also displace Section 3 under paragraph F of Section 1, and large sites frequently hold them.

TwentyForty's contracted capacity is very likely the same quantity as Maximum Import Capacity, but I have not confirmed that from the report's method notes.

I have found no published figure for how often network operators use the clause. The right exists. Whether it is exercised is a separate question, and it is one my research has yet to answer.

7. A note on Scotland

In the same evidence, NESO says that with sufficiently strong locational signals a maximum of 20 per cent of future data centre demand could be located in Scotland. The roughly 7,150 MW of Scottish applications used in my post of 15 September is about 10 per cent of the 72.8 GW GB figure. The two numbers come from different compilations on different dates, so the comparison is directional only.

8. What these figures cannot show

The 72.8 GW is 2025 pipeline data. NESO's 2026 outlook is built on a later, larger queue that is not published, so the request and the forecast are not the same vintage.

The 6.04 GW and the 1.4 GW describe the whole connected fleet, including sites connected before 2025. Only the increase in capacity, about 4.0 GW, can be compared with the request.

Capacity is a snapshot at 1 January; demand is a financial-year total. That is adequate for a figure rounded to 1.4 GW and not for anything more precise.

NESO's Ten Year Outlook is, in its own words, not designed for network planning or connections. It is a view of the current trajectory, not a plan.

Utilisation in the outlook rests on two distribution network operators' data. TwentyForty's metered analysis covers one. Neither covers transmission-connected sites, where most of the new capacity sits.

Average draw says nothing about peak. A network is sized for the busiest half-hour, not the average, which is why this piece does not compare 1.4 GW with winter peak demand.

9. Reproducing the figures

Download the Ten Year Outlook 2026 data workbook and assumptions workbook from NESO's Future Energy Scenarios documents page. In the data workbook, sheet ED1, read row 21 for annual data centre demand and rows 39 to 41 for connected capacity, taking the 2025 and 2030 columns (N and S). In the assumptions workbook, read sheet Demand Modelling, rows 32 to 35. The 72.8 GW, the 59 GW transmission figure and the 20 per cent Scotland ceiling are in NESO's written evidence DCU0081 on the Parliament committees website. TwentyForty's figures are on its Power In report page, and the connection terms are published by the Energy Networks Association.

10. Sources

NESO, Written evidence to the Environmental Audit Committee, DCU0081. committees.parliament.uk/writtenevidence/166090/html

NESO, Ten Year Outlook 2026 data workbook, v001, and levers and assumptions workbook, v001. neso.energy/publications/future-energy-scenarios-fes/fes-documents

TwentyForty, 12 Pillars of Change: Power In, July 2026. twentyforty.uk/twelve-pillars/power-in

Energy Networks Association, National Terms of Connection, version 23.0, effective 22 January 2026. energynetworks.org/publications/national-terms-of-connection

If most of the data centre capacity being connected is expected to sit well below its booked level for years, and the distribution terms already provide a route to recover it, what should happen to that capacity before the network builds for the rest of the queue?

Analysis. The data is NESO's and TwentyForty's, and I have linked both. The arithmetic and the reading of the connection terms are mine; the clause observation comes from an original research project that is currently unpublished.

From queue to load: GB data centres. Requested 72.8 GW, connected by 2030 6.0 GW, average draw 2030 1.4 GW.

Your biggest customer might have changed ownership twice. Did anyone tell you?

Published 10 August 2026 · View on LinkedIn · Read the post

Almost certainly not, and there's no particular reason anyone would have. Nothing you can see changes when it happens. Your contact is the same person, the letterhead is the same, the pipeline you're quoting against is the same pipeline, and the name in your CRM hasn't moved. Underneath all of that, the company you're actually contracting with can have been sold twice in eighteen months. Read the full piece

I wrote a fortnight ago about how many customers you really have, which was a counting problem, fifteen doors resolving to about ten buyers. This is a different problem and it's the one I think is more expensive. Not how many there are, but whether the one you think you know is still the one you think it is.

Why a supplier should care

Three things move when ownership moves, and none of them announce themselves.

Credit is the obvious one. You almost certainly aren't contracting with the developer whose name is on the door, you're contracting with a ring-fenced project company, and what stands behind that company is entirely a function of who owns it. A project company under a large infrastructure fund is a different credit from the same project company under a small private platform, and the entity number on your contract doesn't change when the owner does.

Then there's the parent company guarantee. If you negotiated one, you negotiated it against a parent that may no longer be in the structure. If you didn't ask for one because the ultimate owner looked strong enough, that assessment has quietly expired.

And then authority. Who can actually approve your order, at what value, and whether that threshold survived the transaction. Procurement tends to centralise after a platform acquisition, which is usually the point at which suppliers discover it, and by then it's a live tender rather than a conversation.

This isn't anyone behaving badly

East Yorkshire Solar Farm is about 400MW, 530MWp, consented in May 2025 and sold by Boom Power to Ampyr Solar Europe. The register handles it perfectly. Control changes on 19 February 2026, the old holder ceases and ASE UK Holdings Limited is notified the same day, the outgoing director resigns that day and five new ones arrive within a fortnight. Prompt, accurate, complete.

Go up one level and ASE UK Holdings' own controller is a Dutch entity, notified in December 2021, and the UK register stops there. That's not evasion, it's the design. Name a foreign legal entity as your controller and you've discharged the obligation. So the trail doesn't get hidden, it just ends.

I've used Ampyr deliberately because they've done everything right. If I'd picked a business with a messy filing record you could fairly say I went looking for a bad actor. A well-run European IPP buying a consented project properly, and a British supplier still unable to establish who ultimately stands behind it, is the harder version of the problem. The same holding company sits above two other solar projects I found without looking hard, and carries 26 subsidiaries, so it isn't one relationship at stake either.

And it isn't rare

Gateway Energy Centre in Thurrock is 450MW, the largest consented battery in the country. Its controlling company is called Green Bess Developments (UK) Limited, which appears nowhere in the trade press, largely because until early 2024 it was called InterGen Projects (UK) Limited. The change is on the register, filed as an amendment to details, so nothing on the project company's own page tells you those two names are the same business.

Keep going and it does resolve, to a single individual resident in Prague, sole controller since July 2024. Before him, two individuals from January 2023. Before them, a UK holding company. Three eras of control in eighteen months on Britain's largest consented battery, and no fund anywhere in the chain, which is not what the market maps imply for an asset that size.

If you were selling into that project across those eighteen months, you were dealing with the same team about the same consent the whole way through.

Why you can't see it even when nothing is hidden

Here's the part I nearly overstated, so I'll be straight about the evidence.

Keeping the existing chief executive after an acquisition is normal, and it's measured. Bargeron, Schlingemann, Stulz and Zutter put target CEO retention at around 68 per cent in private equity deals, against roughly 31 per cent where both sides are public companies. That's textbook and I'm citing it to show what normal looks like, not to prove a point.

Statera is the clean example. EQT bought it from InfraRed in November 2023. Nearly three years on, the founder is still chief executive, the brand is unchanged, and there are three EQT people sitting on the board alongside an independent director. Control is entirely legible if you read the board. It's invisible if you read the brand, and it's invisible if you simply keep dealing with the person you dealt with in 2022. That combination, real continuity of people on top of real change of owner, is why this doesn't feel like anything from the outside.

What I can't tell you is that it lasts. Practice data puts chief executive retention through a full hold below 20 per cent, most replaced across five or six years. Statera at three years, and Aypa holding its founder across two owners while changing its name from NRStor C&I along the way, are unusual rather than typical. So this is a mechanism I can show you working, not a rule about how often it works.

What the research all misses

Every study I can find on retention asks the same question, which is whether keeping the chief executive helps or hurts the shareholders and what it does to the acquisition premium. That's a governance question and it's reasonably well answered.

Nobody appears to have asked what it does to the counterparties. Suppliers, lenders and contractors price risk off a name and a relationship, and both survive the transaction completely intact. There's solid work reading the ownership register at scale, Jofre and Knobel published across 5.2 million beneficial owners in May 2025, but it's aimed at corruption and laundering. As far as I can find, nobody has pointed it at a sector and asked what the pattern means for the businesses trading with it.

How far this reaches

Two things from the wider work, briefly, because they set the scale.

First, a correction. In the first article I said a small number of international funds hold positions on both the renewable and data centre side. Having closed the population properly, top fifteen by capacity, that doesn't hold at the top of the market. Four of the top six resolve to private origination platforms with no connection to the funds I named. Origination is size-stratified, and institutional capital tends to arrive after consent rather than originate anything. Wrong first time, corrected in the open.

Second, I tested whether the current wave of deals is operating platforms buying developers, which is how the sector reads it. I wrote the threshold down first, four of seven deals had to fit. One did. The other six were financial capital, a take-private, an asset joint venture, a sponsor-to-sponsor rotation, a joint venture buyout and two minority investments. So the popular reading of the market is wrong, and what's actually happening is ownership moving underneath businesses that carry on looking identical.

One more reason this is about to matter

Ofgem's Curate consultation proposes a data centre commitment fee of between £237,500 and £712,500 per MW, returnable at energisation and forfeited on early exit. Whatever you make of the calibration, a fee at that level is a credit test, and credit tests select for balance sheet. If holding a queue position starts to need capital only the largest owners comfortably have, the customer base concentrates further through a mechanism nobody designed for that purpose. It closes on 16 September.

The question

I've spent three weeks doing this by hand and the summary is that the register is transparent without being legible. Everything is filed. Almost none of it is readable at the moment somebody needs to make a decision.

So, for anyone running sales, credit or procurement in this supply chain. When did you last check who actually owns your three largest customers, and when you checked, did the answer match the name on the invoice?

Ownership traced through Companies House filings, published transaction announcements and company statements, verified 8 August 2026. Corrections welcome, as always.

Graphic accompanying the article: Your biggest customer might have changed ownership twice. Did anyone tell you?

A five to seven year BESS requirement selects the founding cohort

Published 5 August 2026 · View on LinkedIn · Read the post

I have read a lot of BESS and data centre job descriptions this year. The experience requirements stopped making sense, so I looked into why. Read the full piece

Start with the arithmetic, because it settles half the question on its own. DESNZ records the first major grid-scale battery storage project in the UK as becoming operational in 2017. The year before that, National Grid ran its first Enhanced Frequency Response tender and awarded 201 MW across eight projects, each between 10 and 49 MW, drawn from 64 pre-qualified sites. That tender is the commercial starting gun for grid-scale storage in this country. Everyone who has built, financed, connected or optimised one of these assets started, at the earliest, around then.

So the maximum grid-scale BESS experience available to anyone in Britain is roughly nine years. The requirements I keep seeing are not asking for ten, they sit at five to seven, and that is where it gets interesting, because those two numbers do very different things.

Seven years, counted from now, means you were working in UK grid-scale storage before August 2019. That is inside the first two and a half years of the sector, when the EFR fleet had only just energised. Five years means before August 2021, which is roughly the first half of the sector’s life. Seven bites hard, five is a good deal softer, and specifications that write “five to seven” as though it were one requirement are papering over the difference.

The scale point is what makes the seven-year version so narrow. DESNZ records annual output from UK batteries rising from 0.5 GWh in 2017 to almost 2,300 GWh in 2025. Output is a proxy for the size of the sector rather than a headcount, so treat it as indicative, but the direction is not in doubt. Almost everybody now working in British battery storage arrived after the point at which a seven-year filter closes. The filter is not selecting the best of the sector, it is selecting the earliest, and those are not the same population.

That founding cohort is also, by any reasonable reckoning, already employed. I have seen the retirement argument made about this group and I do not think it holds, because the people who went into grid-scale storage in 2017 were mostly early-stage developers, optimisers and fund-side analysts rather than the ageing asset-owner workforce the utilities data describes. The reason the pool is thinner than the headline suggests is simpler. Everyone in it has a job, and the sector competes for them rather than growing more.

Data centres are a different problem, and I want to be careful not to lump the two together, because doing so is where most commentary on this goes wrong.

One is a ceiling, the other is a flow

Data centres are not an emerging sector. Uptime Institute is blunt that the industry has been short of staff and skills for more than a decade, so nothing about the age of the sector constrains the candidate pool. The constraint there is the rate at which the industry builds against the rate at which it trains, and the two have separated.

Uptime’s Global Data Center Survey 2025 puts numbers on it. Around 46 per cent of operators report difficulty finding qualified candidates for vacant roles, and 37 per cent struggle to hold on to the staff they have. What changed in 2025 is where the shortage sits. Operations management roles overtook junior and mid-level positions as the largest reported gap, as experienced managers and engineers retire without enough trained successors behind them. Roughly a quarter of the people who leave are poached by competitors, and only a small proportion leave the industry altogether, which tells you the sector is recycling a fixed population rather than growing one. Uptime’s own research director makes the point that for the first time operators are finding senior recruitment harder than early-career recruitment.

The same shape shows up in the wider transition. LinkedIn’s Green Skills Report 2025, drawn from its membership base across 84 countries, records green hiring growing at about 7.7 per cent a year against 4.3 per cent growth in the share of workers acquiring green skills. In the UK the split is 7.8 per cent against 3.4 per cent. Demand is running at roughly twice supply and has been for several years.

Two different mechanisms, then. In storage the pool has a hard ceiling set by when the sector began. In data centres the pool is bounded by training throughput and drawn down at the senior end by retirement. What the two have in common is that the job description is written as though neither constraint existed.

The filter does not do what it is supposed to do

This is the part I found most useful, and it is the part nobody writing job descriptions seems to know about.

Van Iddekinge, Arnold, Frieder and Roth published a meta-analysis in Personnel Psychology in 2019 covering 81 independent samples, asking a simple question: does the experience someone brings from before they were hired predict how well they then perform? The corrected correlation with job performance came out at 0.06, across 44 samples and 11,785 people. Training performance managed 0.11. The correlation with whether someone stays with the employer was 0.00, across 32 samples and 11,676 people.

An r of 0.06 is under half a per cent of variance explained. Put plainly, the years-of-experience line is close to uninformative about the thing it exists to predict. It is not weakly useful, it is barely distinguishable from noise.

There is a fair objection to that finding and I will make it myself rather than wait for the comments. The samples are drawn from people who were already hired, and those people were selected partly on experience in the first place, which truncates the range and pushes correlations down. The authors deal with this in the paper. It does not rescue the requirement, but anyone citing the number should know the objection exists.

The employers, meanwhile, already agree. Harvard Business School’s Project on Managing the Future of Work, with Accenture, surveyed 2,250 executives across the US, UK and Germany for its 2021 study Hidden Workers: Untapped Talent. Some 88 per cent accepted that qualified, high-skilled candidates are screened out of their processes because they do not match the exact criteria in the job description. For middle-skills roles the figure reaches 94 per cent. This is not campaigners saying it, it is hiring executives saying it about their own systems.

So why does it persist

The best evidence on the mechanism comes from Modestino, Shoag and Ballance, published in the Review of Economics and Statistics in 2020. Using a large database of online postings they showed that education and experience requirements rise when the supply of available candidates rises, within the same occupations and job titles, identified partly through troop withdrawals from Iraq and Afghanistan as an external shock to local labour supply. Their estimate is that the rise in unemployment between 2007 and 2010 lifted the share of postings demanding two or more years of experience by 3.5 percentage points, and that labour supply accounts for 18 to 25 per cent of the total increase in requirements over that period.

That work is American, it covers middle-skill occupations, and it is set in a recession, so I would not transplant it directly onto senior UK energy roles. What it establishes is the mechanism. The requirement is a rationing device, calibrated against how many CVs the employer expects to sift, not against what the job actually needs.

Which leaves an obvious question. If requirements track candidate supply, thin supply should produce modest requirements, and in BESS it plainly does not. My working explanation, and I flag it as a hypothesis rather than a finding, is template transfer. The people writing these specifications came from oil and gas, thermal generation and engineering construction, where ten to fifteen years is an unremarkable line to write, and the convention travelled into a sector too young to support it.

Nor is the stated remedy arriving. The Burning Glass Institute and Harvard Business School examined 11,300 roles where employers had publicly dropped degree requirements and found the share of hires without a degree rose by 3.5 percentage points, which works out at fewer than one in 700 hires across the economy. Around 45 per cent of the firms that announced a change made no measurable difference to who they actually hired.

What I take from it

If you are being ruled out of these roles, the filter that removed you was not measuring your ability to do the work. That is not consolation, it is the finding, and it comes from a meta-analysis of 81 samples rather than from anybody’s bruised feelings.

For anyone writing the specification, the cost of the unmet requirement lands on you. It lands as vacancy duration, and per Uptime it lands hardest at exactly the seniority where the shortage is now worst. A specification that cannot be met is not a high bar, it is an unfilled seat.

The question I’m asking myself, and I do not have a clean answer to it, is what the honest replacement looks like. If years in sector predicts almost nothing, and the sector is too young to supply the years anyway, what should the line in the job description say instead, and would a hiring panel trust it?

References

Van Iddekinge, C. H., Arnold, J. D., Frieder, R. E., & Roth, P. L. (2019). A meta-analysis of the criterion-related validity of prehire work experience. Personnel Psychology, 72(4), 571–598. DOI: 10.1111/peps.12335

Modestino, A. S., Shoag, D., & Ballance, J. (2020). Upskilling: Do Employers Demand Greater Skill When Workers Are Plentiful? Review of Economics and Statistics, 102(4), 793–805. DOI: 10.1162/rest_a_00835

Fuller, J. B., & Raman, M. (2021). Hidden Workers: Untapped Talent. Harvard Business School Project on Managing the Future of Work, with Accenture.

The Burning Glass Institute & Harvard Business School Project on Managing the Future of Work (2024). Skills-Based Hiring: The Long Road from Pronouncements to Practice.

Uptime Institute (2025). Global Data Center Survey 2025.

LinkedIn Economic Graph (2025). Green Skills Report 2025.

Department for Energy Security and Net Zero (2026). Grid-scale battery storage statistics, DUKES 2026 methodology note.

National Grid (2016). Enhanced Frequency Response tender results, August 2016.

Graphic accompanying the article: A five to seven year BESS requirement selects the founding cohort

How many customers can you really target?

Published 3 August 2026 · View on LinkedIn · Read the post

Last week I published a piece tracing the real ownership of fifteen of Britain's biggest consented solar and battery projects. The finding that traveled was the map, at least ten of the fifteen are not economically what their database entry says. This week I want to pull on the thread that matters most to anyone who sells into this industry, because the map has a commercial consequence folded inside it. If the projects have fewer ultimate owners than names, then suppliers have fewer customers than prospects. And almost nobody's sales plan reflects that. Read the full piece

Fifteen doors, ten buyers

Picture a sales director building next year's plan. The CRM shows fifteen live prospects across the biggest consented projects in Britain. Fifteen relationships, fifteen probability weightings, fifteen doors to knock on. Now lay the ownership map over it. Those fifteen projects resolve to roughly ten ultimate decision-makers, and three of them, Macquarie, CIP and EDF, control multiple positions each. The prospect list is exactly the same length it was yesterday. The customer count is not.

Three things quietly break when that happens. The first is the arithmetic. Win probabilities assumed to be independent are not, two opportunities that share an ultimate owner are one bet wearing two names, and a pipeline weighted across fifteen doors is really spread across ten. The second is the negotiating table. When you price project A, a platform that also controls projects B and C knows your exposure across all three, and you may not even know they are the same buyer. That asymmetry is worth real margin and it runs entirely one way. The third is where the decision actually lives. Platforms do not buy project by project for long. Procurement moves from the project team, where relationship selling works, to the fund level, where scale and balance sheet win.

None of this is a forecast

I want to be clear about the status of the claim, because it is not a prediction. It has already happened, the reason it does not feel that way is that it happened below the visibility of the market maps everyone sells from. The platform deals of the last three years, EQT and Statera, Brookfield and Banks, KKR and Zenobē, Macquarie and Island Green Power, CVC DIF and Low Carbon, BW and Penso Power, looked like separate deals. Viewed from the supply side they were one event, the customer base consolidating. The developer-to-owner pivots looked like strategy fashion, they are what happens when consented projects become the asset the platform keeps rather than sells. And the thing mid-tier sales teams have been feeling for two years, cycles lengthening, decision rights drifting upward, good project relationships that suddenly cannot close project deals, is exactly what centralised procurement feels like from the outside when you cannot see the platform.

Other industries have made this exact journey

And it is well studied. When British grocery retail concentrated, supplier terms tightened until the state had to step in, a statutory code of practice in 2010 and a dedicated regulator, the Groceries Code Adjudicator, in 2013, both flowing from the Competition Commission's 2008 finding that the largest retailers were transferring excessive risk and unexpected cost onto their direct suppliers. The economists watching that market named the crueller mechanism underneath, the waterbed effect, better terms for the most powerful buyers worsening the terms for everyone smaller. I should say the evidence stayed contested, the Commission looked for the waterbed and declined to rely on it. It is still exactly the mechanism a mid-sized supplier should watch for, not least because by the time such things are provable they have already happened.

When car making consolidated, the supply base did not simply shrink to match, it restructured into tiers, a handful of system integrators holding the direct relationships and everyone else selling to suppliers rather than to the customer. Construction, my own former industry, shows the procurement endpoint, framework agreements as the default interface once buyers reach sufficient scale.

The early instruments are already visible in this market. Sungrow, the Chinese battery giant, has supplied BW's storage platform across its Swedish portfolio, then its Bramley project, then a single 1.4GWh order for Hams Hall, signed jointly with Penso Power a month before BW absorbed the company outright. The relationship travels with the platform, not the project. The same supplier holds the battery contract for Fidra's Thorpe Marsh and, as of financial close on 17 July, for West Burton C too, the two largest storage positions in the country, which makes a second platform relationship. Macquarie's pivot of Island Green Power from seller of projects to keeper of them converts a fifteen gigawatt pipeline into one owner's procurement function. CIP procures at fund scale across vintages. Offshore wind, one maturity step ahead of solar and storage, treats multi-project preferred-supplier arrangements as normal practice. And lenders' approved-vendor lists already do quietly what frameworks do formally, they gatekeep who is allowed to sell at all.

There are forces pushing the other way, and honesty requires them. Battery technology moves faster than framework terms, which favours spot buying. EPC wraps put a contractor between supplier and owner on many projects. And owners burned by price volatility are wary of long commitments. So the defensible claim is not that frameworks arrive tomorrow. It is that every industry whose customer base concentrated this way ended up there, the first instruments are already on the table, and a supplier's strategy should price that trajectory now rather than discover it in a lost tender.

Good news, bad news, unevenly distributed

Is concentration good or bad for suppliers? Both, and not evenly. The credit is better, a project company standing behind a Macquarie platform beats one standing behind hope. But the benefit has to be claimed, you are still contracting with a ring-fenced project company unless you negotiate the parent guarantee, what the ownership map changes is that you now know whom to ask and what their covenant is worth. The pipeline is realer, platform-held consents actually get built. Against that, the targetable market is smaller than the addressable market, the tier system rewards scale, and if the waterbed mechanism operates here, the squeeze lands hardest on the middle.

Who already knows all of this? The largest suppliers, though notice how they learned it, the expensive way, account by account, across years of deals, and notice too, on the evidence above, that the most platform-fluent of them are the largest Chinese manufacturers. Their knowledge is private relationship capital, their own buyer map, not the market’s. Who does not know it is almost everyone else. Mid-sized UK suppliers watching win rates decay without an explanation. New-technology entrants counting prospects that are really one buyer. And overseas entrants building a UK go-to-market plan from databases that still show a fragmented market which no longer exists.

So the question I would put to anyone running revenue in this supply chain is not how many customers you have. It is how many you can really target, who actually holds the decision for each of them, and whether your plan was built on the number of doors or the number of buyers. If you have re-cut a pipeline on ultimate ownership and found something different from your CRM, I would genuinely like to hear about it.

Follows “Who actually owns Britain’s energy transition?” (27 July). Ownership chains traced through Companies House filings, corporate statements and announced transactions, July 2026. Corrections welcome.

Graphic accompanying the article: How many customers can you really target?

Who actually owns Britain's energy transition?

Published 27 July 2026 · View on LinkedIn · Read the post

A few weeks ago I wrote that a grid connection is no longer a waiting list, it is a market structure. The post did well, and one question kept nagging at me afterwards. If grid positions and planning consents are now scarce, valuable assets, someone should be accumulating them. So why does every industry database still show a fragmented market of dozens of independent developers? Read the full piece

Correction, 10 August 2026: a claim made in this series about fund ownership across renewables and data centres was withdrawn. See Your biggest customer might have changed ownership twice.

I think I found the answer, and it is hiding in plain sight. The databases are not measuring the thing that matters.

The problem with counting developers

Almost every large renewable energy project in Britain is owned by its own dedicated company, set up for that single project. There are good reasons for this, it makes the project easier to finance and easier to sell. But it has a side effect. When a project changes hands, the company changes hands with it, and the project's name in every database stays exactly the same. The “developer” field never updates. You can buy half the pipeline in Britain and the market will still look fragmented on paper.

That was actually the thought that started all this, out in the garden. If concentration is coming to this market, I had been assuming it would arrive the way it does elsewhere, developers buying developers, the kind of deal that makes the trade press. But it does not need to. The projects are the companies. So the market can concentrate one project at a time, quietly, through purchases that never touch the developer's name, and the number everyone watches, the count of independent developers, need not move at all. Once that occurred to me I rather needed to know whether it was already happening.

So I did the tedious thing. I took fifteen of the largest consented or under-construction solar and battery projects in Great Britain, roughly 8,800MW between them, and traced each one through Companies House to whoever ultimately controls it. Not the name on the planning application, the actual end of the ownership chain.

Standing in the garden it occurred to me that the market looks like a bramble patch, dozens of apparently separate stems. Brambles, though, spread underground, what looks like many plants is often one root system. The only way to know is to dig. So I dug. I have started calling it the bramble thesis.

What the chains show

Take the largest battery project in the country, the 1,450MW Thorpe Marsh scheme in Doncaster. The databases say Fidra Energy, a Scottish storage company. Walk the ownership chain and it runs through three group companies to a structure sponsored by an American energy private equity house in Washington DC, which has been there since the company was incorporated, the group's own corporate statements confirm it. And there is a twist. The UK's National Wealth Fund has taken a significant minority stake, which means the British state is a minority investor in the country's biggest battery project, inside a structure it does not control. Nothing improper about any of it, but “Scottish battery firm” is not the economic reality.

Or take the quiet accumulation nobody has named. EDF, the French state utility, bought the consented 500MW Gate Burton solar project outright a year after it won consent, holds half of the 800MW Springwell project, and has three more giant solar schemes behind them, one already consented and two moving through the system. Each transaction was announced separately. Put them together and EDF is arguably becoming the largest consolidator of big English solar, and I have not seen that sentence written anywhere.

Or the pattern I found strangest of all. The Coalburn battery projects in Scotland, among the largest in Europe, are funded by a Danish infrastructure investor, which has already agreed to sell half of each one to French and Danish institutional money, with the handover happening the moment each project switches on. These assets will have changed hands twice by the time they enter commercial service. Ownership here is not a register, it is a conveyor.

The timing is the part I keep returning to. Macquarie completed its buyout of Island Green Power, the developer behind the Cottam and West Burton mega-projects, months after both consents landed, then switched the company's whole strategy from selling projects to keeping them. The East Yorkshire solar farm was sold, as a company, nine months after its consent. Half of the One Earth project changed hands while the application was still awaiting its decision. In case after case the corporate event follows the consent event. Consent is the trigger.

The actual numbers

Across the fifteen projects, my figures, now checked link by link (the limits below still apply):

The named developer is materially misleading for at least ten of the fifteen. By capacity, around 20 percent of these 8,800MW remains in genuinely UK-independent economic hands, and I can name every megawatt of it. The rest terminates in Washington, Stockholm, Copenhagen, Paris, Sydney and Singapore.

And the counter-examples matter, so here they are. Ecotricity's 500MW Heckington Fen project is exactly what it says it is, a UK company that consented its own project and is building it. So is the Sunnica joint venture, and RWE's Pembroke battery is transparently on a listed utility's balance sheet. Interestingly, the projects that are what they appear to be cluster at the smaller end of my fifteen. The bigger the asset, the more likely the label is wrong.

One more finding I did not expect. In early 2024, when the InterGen gas power station business changed hands, the deal specifically excluded InterGen's consented 450MW battery project at Gateway in Essex, along with its pipeline of Scottish battery schemes. Those were kept back by the selling group and now sit in a separate company, and Spanish capital has since bought into the Scottish ones. A consented battery position was carved out of the sale of the entire business around it. The market is telling you, in transaction structure, that these positions are now a separate asset class.

I should be straight about the limits. The public ownership register only captures stakes above 25 percent, contractual control (framework agreements, forward funding) is invisible to it, and a couple of my chains end at structures I could not resolve from public records at all. So my concentration figure is a floor, not a ceiling. The real number is higher.

I would add one thing. Along the way I found four errors in the industry's own visible data, including three “developers” who had already sold up. That was one weekend of methodical checking. Which makes me wonder what a proper job would find.

Why would anyone hide this? Mostly, they aren't

I want to be careful here, because the tempting conclusion is the wrong one. Nobody designed these structures to hide ownership. Project companies exist because they make projects financeable and sellable, the opacity is a side effect, not a strategy.

Everything I found is on the public register, nothing was hidden from me. It is like hiding a key in a bunch of keys, to find it you have to know to look for it. The register is transparent, it just is not legible, and those turn out to be very different things.

But a side effect can still pay out, and this one pays out very unevenly. The professionals on the capital side are not confused by any of this, they trace these chains as a matter of routine, it is priced into every deal. The people who cannot realistically trace them are everyone else. The council weighing a planning application from a company with a local-sounding name. The policymaker designing support schemes on the assumption of a diverse, competitive developer market. The landowner negotiating an option without knowing the same fund holds the option next door. Nobody built that information wall deliberately. It is just that everyone on one side of it does rather well out of it standing.

And this is not a renewables story, it is a pattern with form. After 2008, American institutions bought hundreds of thousands of family homes through one-company-per-house structures, and for years the market looked like thousands of small landlords until researchers joined the companies up. Closer to home, investigations into English fishing quota found a handful of families controlling a dominant share through what looked like many independent vessel companies. In both cases the official picture stayed fragmented long after the reality had concentrated, because the measurement layer was pointed at the wrong thing. I would gently suggest we are watching the same film again, in a sector we are betting the country's energy security on.

And in energy itself, half of this count has already been done. Common Wealth put overseas ownership of British offshore wind capacity above eighty percent a few years ago, and that finding travelled widely. What has not been mapped is the consented onshore pipeline, where the ownership moves after consent, one project company at a time, below the visibility of every database. That is the gap this weekend was about.

So the question I would rather ask is this. If the ownership map matters for consent decisions, for policy design, and for anyone selling into this market, and if right now that map effectively exists only inside the institutions doing the buying, who should be responsible for making it visible? Because at the moment the answer is nobody, and that suited everyone fine until someone spent a weekend in the garden checking.

Method note: ownership chains traced through Companies House filings (persons with significant control, filing histories and charges registers), cross-checked against planning records, corporate statements and announced transactions, July 2026. Corrections welcome, genuinely, this is exactly the kind of work that improves with more eyes.

Graphic accompanying the article: Who actually owns Britain's energy transition?

Five predictions about UK energy

Published 14 June 2026 · Six posts, 14 to 19 June 2026

Five predictions about UK energy this week, one a day, and I'll mark my own homework on every one in the autumn. Read the full piece

Published as six posts, 14 to 19 June 2026. Text as posted, one typo corrected. To be scored in public in the autumn.

Sunday 14 June: the five questions

Most energy commentary describes what already happened. That's useful, but it's safe. So this week I'm doing something a little more exposed: five specific, dated calls on where the UK system goes next, posted one a day from Monday. No hedging, no 'it depends'. Each is the kind of claim that ends up clearly right or clearly wrong, and before long we'll know which. It's also important to point out that these are my own opinions, there's no connection to ENWCML my current employer.

Note added 27 September 2026: accurate when posted in June 2026. ENWCML is now my most recent role; see the Profile.

Here's the running order, as questions for now:

Monday: do offshore wind strike prices finally turn back up in AR8?

Tuesday: how large are the first long-duration storage awards, and who wins them?

Wednesday: does the door for new grid connections reopen this year?

Thursday: will Ofgem put a real price on data-centre speculation?

Friday: how big does the 2026 constraint bill actually get?

I have a firm answer to each, and a few of them cut against the consensus. I'll also come back in the autumn and score myself honestly, hits and misses included, because a prediction nobody ever checks is just noise.

Before I start posting, I'm curious where you'd bet against me. Which of the five do you reckon I'm most likely to get wrong? Call it now, and we'll see who reads this right.

Monday 15 June: My first call for the second half of 2026: offshore wind gets more expensive, not less.

This week I'm putting down five dated predictions for where UK energy goes next. Here is the first, and it cuts against the prevailing mood.

AR8, the next CfD auction, opens in July. Most commentary assumes strike prices keep drifting down, because that has been the direction of travel and because more competition is meant to push costs lower. I think that run is over. My call is that offshore wind in AR8 clears above where it landed in AR7, which was around £91 per MWh for fixed-bottom projects.

The reason is that the forces pushing prices down have largely run their course, while the ones pushing them up have not. Turbine and cable costs, financing rates and supply-chain tightness are all still elevated, and developers have spent two years saying the economics no longer work at the old numbers. AR7 already cleared below its administrative strike price, which suggests the headroom for further falls is thin. A government keen on volume, and AR8 was deliberately brought forward to keep volume moving, tends to get that volume by paying for it rather than by squeezing harder.

What would prove me wrong: a large budget paired with a low price cap that forces bidders down, or a clearing price held flat by sheer competition among offshore developers. Both are possible. We will know when the results land early next year.

So here's the question for the offshore people in particular. Am I reading the cost pressure correctly, or is there more room to fall than I'm giving credit for? Tell me where this call is weak.

Tuesday 16 June: Call two of five: the first long-duration storage awards land small, and pumped hydro wins.

My second prediction for the second half of the year, and this one resolves fast, possibly within days of this post.

Ofgem is about to make the first awards under the long-duration storage cap and floor regime. The indicative range it set was wide, 2.7 to 7.7GW of capacity. My call is that the first window comes in toward the bottom of that range rather than the top, and that large pumped-storage hydro takes the bulk of it, ahead of lithium-ion.

Two things point that way. Ofgem has said openly it will weigh technology diversity and may set aside some battery projects in favour of pumped storage or flow, which is a clear signal about where it wants the money to go. And the eligible pool is dominated at the top end by a handful of very large hydro schemes, the likes of Coire Glas at 1.45GW and Earba at 1.8GW, where a single award moves the whole number. A regulator running a brand new scheme tends to start cautious, prove the model, and scale later, rather than commit the full range on day one.

Where I could be wrong: Ofgem might decide the Clean Power timeline demands volume now and award nearer the top, or lean into batteries for speed of delivery. Both are live possibilities.

A note for anyone reading this after the fact, since the list may already be out: if it is, you can mark this one straight away, which is rather the point of making it.

For the storage people, what's your read? Does Ofgem go cautious and hydro-heavy, or surprise everyone on volume?

Wednesday 17 June: Call three: the door for new grid connections stays shut for the rest of 2026.

Halfway through the week, and a prediction I'd put real money on.

The connections reform reopened the queue for projects already in it, and offers are flowing again, area by area. But the route in for anyone new runs through the next application window, and NESO has quietly removed the date it once attached to that window. There is no longer even a 'no earlier than' marker. My call is that this does not change before the end of 2026. No firm opening date gets announced this year.

The basis is partly that the machinery is fully occupied. NESO is working through Gate 2 offers in tranches that run into early 2027, and you cannot easily reopen the front door while still processing everyone already inside. Partly it's that there's no obvious pressure forcing a date, since the projects that matter politically are mostly the ones already in the queue. New entrants are the constituency with the least leverage, and they're the ones left waiting.

What would falsify this: a government push to show the queue is open for fresh investment, perhaps tied to data centres or to a Clean Power progress story, could put a date on the calendar before December. It's possible. I just don't see the forcing function yet.

This one matters most if you're trying to get into the queue rather than through it. If that's you, I'd genuinely like to hear it. How are you planning around a window with no opening date?

Thursday 18 June: Call four: Ofgem puts a hard price on data-centre speculation this year.

Fourth of five. This one is about the demand side, which gets far less attention than generation but is where the next fight is.

The connection queue has been swamped by data centre requests, something like 50GW of them, more than the country's entire peak demand, and much of it speculative. Ofgem has been consulting on what to do, and has floated tougher financial tests, self-build, and a possible levy aimed squarely at filtering out projects with no real intention of being built. My call is that it goes beyond guidance and lands something with teeth this year: a financial commitment test or a levy that makes parking a speculative request genuinely expensive.

The logic is that soft measures won't shift behaviour when the prize, a connection in a constrained system, is this valuable. Guidance asks nicely. A levy or a forfeitable deposit changes the maths. Ofgem has signalled a further consultation for the summer, and the language in its recent output has hardened noticeably.

Where this could miss: regulators often consult, soften under industry pressure, and land something more modest than the early signals suggest. That pattern is real, and it's the main risk to the call.

Here's a question worth throwing open, because reasonable people land on different sides of it. Is pricing out speculation the right tool, or does it just hand the grid to whoever has the deepest pockets, the hyperscalers, and squeeze out everyone smaller? I lean towards it being necessary, but I can see the objection clearly. Where do you sit?

Friday 19 June: Call five: the 2026 constraint bill blows past anything we've seen, and I'll mark it at £1.5bn-plus.

Last of the five, and the one where I'm most confident about the direction and least confident about the exact number. So I'll say that plainly rather than dress it up.

Constraint payments are what we pay to switch generators off when the grid can't move their power, and then pay other plant to run in their place. In 2025 they had already passed £690m partway through the year, up sharply on the year before. NESO itself has warned the figure could reach £7bn by the end of the decade if the network doesn't catch up in time. My call is that the full-year 2026 number comes in well above 2025, and I'll put a marker down at more than £1.5bn for the year.

The direction isn't really in doubt. More renewables keep connecting in places the network can't yet fully evacuate, and the wires lag the generation by years. Every quarter the build programme runs late, the constraint meter runs faster. That part I'm confident about.

The precise figure is another matter. Constraint costs are volatile, weather-dependent, and sensitive to a few big network outages, so £1.5bn is a marker rather than a forecast I'd defend to the decimal. I'd rather name a number and risk being visibly wrong than hide behind a vague 'it will rise'.

So tell me if the marker is too bold or too timid. If you model this for a living, where would you put the 2026 total? And that's the five for the week, I'll come back and score myself on all of them in the autumn.

Five calls on UK energy, made in June 2026: offshore wind in AR8, long-duration storage awards, the next connections window, a price on data centre speculation, and the 2026 constraint bill