Mark Moore


Posts

Shorter pieces, newest first.

73 GW queueing, and no count of the capacity sitting idle

Posted 13 November 2026 · View on LinkedIn

Ofgem’s minded-to position on DCP 412 came out on 16 October. E.ON wants high-capacity, low-use sites moved into a cheaper residual charging band, and Ofgem is minded to say no. The consultation closes on 11 December. Read the full post

Most of the money is on the transmission side, which uses the same bands. For a high voltage site in the top band, the transmission fixed charge would fall from about £523 a day to about £184 (para 7.15).

That matters for the connection queue because the bands are set on capacity holdings where the data exists (para 3.5). Whatever the residual was designed to do, in practice it is part of the cost of holding capacity a site does not draw, and cutting it weakens the reason to give unused capacity back.

Meanwhile Ofgem’s own Curate consultation counts around 73 GW of data centre demand, across roughly 315 projects, waiting to connect. The DCP 412 document does not mention data centres once.

The rules already have a route for this. Under clause 12.11C of the National Terms of Connection, if a metered site never goes above 75% of its Maximum Import Capacity in any 12 consecutive months, the network operator can propose a reduction. Nothing changes unless the customer agrees or Ofgem decides it.

Some network operators publish substation headroom, and some net off contracted demand, but I have not found a published count of how much agreed capacity at existing sites sits below that 75% line, or where it is. Ofgem’s impact assessment describes the number of eligible sites only as small (para 7.8) and had no data for extra high voltage (para 7.11).

The data could probably be gathered by asking site owners directly, or from the half-hourly readings held by the companies that supply the power.

If network operators published agreed capacity against peak draw, substation by substation, how much of that 73 GW do you think could connect sooner?

Analysis. Sources: Ofgem, DCP 412 minded-to consultation (16 October 2026); Ofgem, Curate consultation (July 2026); ENA, National Terms of Connection v19.0.

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

Graphic accompanying the post: 73 GW queueing, and no count of the capacity sitting idle

Mind the gap: the kit turns up, the grid doesn’t

Posted 12 November 2026 · View on LinkedIn

A 132 kV transformer ordered now should arrive in about 18 months, and a 400 kV unit in about 24. Those are the figures UK developers gave Baringa for its 2026 update to the DESNZ supply chain study, roughly half what the same study recorded in 2024 (the old four-year number is still doing the rounds). Read the full post

Good news, you’d think. Except a transformer that arrives early doesn’t connect early, because the date is set by the reinforcement behind it.

Grimsby to Walpole carries a licence date of the end of 2031, and National Grid’s own September update says full operation by 2033. SP Energy Networks has its new Dumfries North 400/132 kV substation contracted for 2036, while solar developers report switchgear in six to twelve months. Across 155 dated TEC Register entries at ASTI-linked substations, I found about 62% of the capacity dated beyond the licence backstop, a median of two years after the target date (an association with a regulatory date, not an engineering programme).

EGL1 is where the kit did set the date. Due in 2027, it’s now expected around April 2029, with the joint venture citing global supply and capacity problems. It asked Ofgem for 480 days of penalty relief and got 57.

On site, the order book runs the other way. GE Vernova says it’s mostly sold out through 2030, and heavy-duty turbines shipped in 2030 and 2031 will be commissioned in 2032 and 2033. Large gas engines take about two to three years from order to delivery, on my arithmetic from Wärtsilä’s 2026 orders. Both are global order books rather than UK figures.

So on the grid route the kit sits on the platform waiting for the line, and the on-site route has a queue of its own, only it’s at the factory.

Synthesis: published figures from Baringa for DESNZ, Ofgem, National Grid, SP Energy Networks, GE Vernova and Wärtsilä, with an original comparison of UK grid kit against the reinforcement it serves.

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

Friday’s board said Oslo’s platform was filling. On Monday, Statnett lowered the barrier.

Posted 10 November 2026 · View on LinkedIn

On Friday’s Departures board I gave Oslo the remark PLATFORM FILLING. I’d started that edition expecting the Nordics to be the train running on time, and the evidence didn’t support it. Read the full post

On Monday Statnett paused new grid reservations for data centres until 1 August 2027 at the latest. Its own figures show why: 800 MW of data centres connected, 4,200 MW reserved, 9,100 MW in the queue and a further 18,000 MW of requests. That’s 32,100 MW against a national peak of just over 25,000 MW, or about 128%.

Most of today’s coverage is about volume. Statnett’s stated reason is behaviour. The pause is a risk-reducing measure while it writes functional requirements for data centres, covering how large loads ride through faults and how fast they change. Plants whose requirements are approved can still connect, and some smaller plants due before August 2027 may be considered.

The same morning, ministers announced a licensing regime, with priority for defence, health, critical infrastructure and national AI, and less room for global cloud and crypto.

The GB debate has mostly been about who gets a place in the queue. Ofgem counts about 73 GW of data centre demand in it, more than GB’s forecast winter peak, and the Curate decision is still awaited. Norway has just added a second test: how the load behaves once it’s connected.

The updated board is below.

Synthesis: Statnett, ‘Midlertidige tiltak for fortsatt forsvarlig systemdrift’, 9 November 2026; Norwegian government announcement as reported by Capacity, 9 November 2026; Ofgem Curate, 29 July 2026; NESO Winter Outlook 2026/27, 23 October 2026.

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

Graphic accompanying the post: Friday’s board said Oslo’s platform was filling. On Monday, Statnett lowered the barrier.

UK data centre map, Edition 2: live today

Posted 10 November 2026 · View on LinkedIn · See the map

The second edition of my map of proposed and under-construction UK data centres is live today. 48 schemes pass the same four tests as Edition 1, each met from a primary source: a named site, a stated size in MW, a stated power supply and a dated planning status. The link is in the first comment. Read the full post

Six are rated Red because on-site or dedicated generation is planned as primary or bridging power: Quest Park at Stewartby, West Sleekburn, Thames Valley Park, Cardiff East, the SDC M40 campus and Sembcorp Wilton. Five of them name gas, with bridges running to grid dates between 2030 and about 2040.

One is rated Green: Kao Data’s Harlow campus, with up to 40 MW of private-wire solar contracted from Downing and the rest from the grid.

The other 41 are Amber, on grid supply or a clean-power claim with no named asset. One moved last week: DataVita has dropped plans for up to 1 GW of private-wire renewables at its Lanarkshire AI Growth Zone campus and told the Guardian it will rely entirely on the National Grid, through a renewable power purchase agreement. It stays Amber. Its 24 MW DV3 building next door joins the map in this edition.

On the public record, 20 of the 48 have a grid connection contracted or reserved. Another 24 state a grid supply with no agreement shown.

95 more schemes sit on the watch list, 23 of them one or two checks away. The map is a free resource. Under constant review, changes or additions will be made to the map as soon as they are processed. If a scheme is missing, send me the planning reference.

Local response follows on 20 November, and who ultimately controls the schemes on 27 November.

Six schemes would rather build their own power than wait. How many of the 41 on grid supply have a connection date they can actually plan around?

Analysis. Sources: council planning registers and applicants’ published planning documents, checked 8 November 2026. Power ratings as defined on the map page.

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

Graphic accompanying the post: UK data centre map, Edition 2: live today

Self-build fixes the last two kilometres. Who fixes the rest?

Posted 9 November 2026 · View on LinkedIn

Ofgem’s self-build consultation closes on Friday, and it reads differently next to the Great British Grid announcement. The government launched the two together on 29 September as ways to speed up connections. Read the full post

Ofgem’s proposal would let a customer build every sole-use asset between its site and the existing transmission system, well beyond today’s 2 km limit, and 36 demand projects in the queue have already registered interest. Ofgem is candid about the catch, though. For many projects, it says, “building these assets quicker will not mean an earlier connection date”, because enabling and wider works set the date and reinforcement is out of scope.

So who does the wider works? GB Grid’s only described route into network assets is bidding for new, separable transmission projects through competitive tendering, with the first tender expected in 2027 and no project yet named. Reinforcing the existing network stays with the incumbent transmission owners.

On what’s been published to date, the two don’t meet. Ofgem’s consultation never mentions GB Grid, Great British Energy or public investment, and GB Grid has no published budget. Electrical Review suggested GB Grid could fund upstream reinforcement while developers self-build, and Herbert Smith Freehills Kramer noted that a business can’t simply ask GB Grid to bid for its connection. I can’t find anything yet that routes public money to the stretch of network a self-builder is actually waiting on.

A self-builder could finish its own pipe in record time and still be waiting on the main in the street.

Synthesis: Ofgem’s self-build consultation (7 Oct 2026), the DESNZ and Great British Energy announcement (29 Sep 2026) and Ofgem’s tendering criteria, read side by side, as at 9 November 2026.

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

UK data centre map, Edition 2: live Tuesday 10 November

Posted 8 November 2026 · View on LinkedIn · See the map

On Tuesday I’m re-publishing my map of proposed and under-construction UK data centres. Edition 1 went out on 13 October with 13 schemes. Four weeks of working through planning registers later, 48 pass the same tests. Read the full post

Thanks to @Lynsey Breen of @AC Land, who asked why her scheme wasn’t on Edition 1. Fair question. It sent me back to search council planning registers directly, instead of starting from announcements.

The tests haven’t changed. Each has to be met from a primary source: a named site, a stated size in MW, a stated power supply and a dated planning status. Trade press figures are used only to flag schemes for the watch list.

Every scheme on the map carries a power supply rating. 41 rely on the grid, or on a clean-power claim with no named asset behind it. 6 plan on-site or dedicated gas as primary or bridging power. Just 1 has a named new clean supply contracted to the site, the same count as in Edition 1. Honestly, I was hoping for more green.

Behind them sits a watch list of 95 schemes that don’t yet clear every test. 23 are candidates, one or two checks away from the map; the rest are earlier or thinner. The watch list is published so readers can judge coverage for themselves.

Operating data centres are left off: the map covers the pipeline only.

The map is a free resource and will be hosted on my website. Follow the link, released on Tuesday, to get to the full map. Under constant review, changes or additions will be made to the map as soon as they are processed.

Two more layers follow. On 20 November, local response, measured by recorded objections. On 27 November, who ultimately controls the schemes.

If six schemes have already decided not to wait for the grid, how many of the other forty-one will follow once their connection dates are known?

Analysis. Sources: council planning registers and applicants’ published planning documents, checked 8 November 2026. Power ratings as defined on the map page.

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

Graphic accompanying the post: UK data centre map, Edition 2: live Tuesday 10 November

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

Posted 6 November 2026 · View on LinkedIn · Read the article

Last time the board was Euston’s. This time I’ve set London against the rest of FLAP-D, the five markets the industry groups together (Frankfurt, London, Amsterdam, Paris and Dublin), and added two more services: Oslo for the Nordics, and the Caledonian Sleeper for Scotland. Read the full post

A few from the board:

Frankfurt’s distribution operator now allocates requests of 3.5 MW or more once a year, pro rata, and expects the restrictions to last until the mid-2030s.

In Amsterdam, the expected connection date for the Vijfhuizen part of the network moved from 2031 to 2036 in January.

EirGrid considers the greater Dublin region fully constrained, so there’s no expected date to put on the board.

The Caledonian Sleeper is held at the platform. Holyrood has agreed that no decisions on hyperscale applications should be made until national planning guidance is complete, and the Scottish Government expects to publish that guidance before Christmas.

Paris is the exception. On France’s fast track the State identified large 400 kV sites first, and the first contract, signed with RTE in January, provides 240 MW by the end of 2027, ahead of a 2028 to 2029 target. Off the fast track, operators report waits of two to seven years.

The carriages show CBRE’s colocation vacancy for the first quarter of 2026. Every market with a published figure is more than 90% let.

Thanks again to Guy Massey, whose US arrivals board started the series. The full board, with every source, is in the article.

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

Provenance: Synthesis, with original comparison across markets.

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

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

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

Posted 5 November 2026 · View on LinkedIn · Read the article

DESNZ’s consultation on Wholesale Contracts for Difference closes on 9 December. It would offer existing wind, solar and hydro (up to 29.8 GW across 1,063 sites at a 5 MW threshold) a fixed price for the wholesale part of their revenue, with ROCs carrying on unchanged. Read the full post

I’m in favour, and I sent my response to DESNZ this week. On DESNZ’s estimate, gas set the power price around 90% of the time in the early 2020s, and a two-way contract on the existing fleet turns that exposure into a hedge for consumers.

The detail needs work, though. A generator with a Capacity Agreement for the delivery year starting October 2028 has to give notice by 1 June 2027 to switch. The auction is expected towards the end of 2027. It gives up its capacity payments first and learns the auction result later.

In the article I set out that and three other design points: fixed end dates if start dates are flexible, separate clearing for 5 and 10 year contracts, and publishing the bid stack once the round is over. There’s a buyer’s angle too, for data centres and anyone else hoping to sign a long-term PPA with an operating wind farm.

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

Provenance: Analysis.

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

What have data centres ever done for us?

Posted 5 November 2026 · View on LinkedIn

With apologies to Monty Python (& Dr Terri Simpkin). A village hall, an objection meeting, a show of hands. Read the full post

“What have data centres ever done for us?”

“The NHS App?”

“Fine, the NHS App. More than 39 million people are registered on it. Granted.”

“Your bank. The card you tap at the till.”

“Yes, yes, apart from the NHS App and the banking...”

“The weather forecast. The football on your phone.”

“The swimming pool. In 2023 Exmouth Leisure Centre started heating its pool from a small data centre, expected to cut the pool’s gas use by 62%.”

“Exmouth doesn’t count. But apart from the NHS App, the banking, the forecast, the football and one pool in Devon, what have they ever done for us?”

“In 2024 the government made them Critical National Infrastructure, alongside energy and water.”

“...Next item on the agenda.”

The joke only works because every answer is true and most of them are not clearly understood in the village hall. At the Data Centre Alliance’s conference in Birmingham on 20 October, Dr Terri Simpkin made the point that the public knows very little about what data centres actually do. The benefits are national and spread thin. The costs are local and concrete: a large building, a new substation, construction traffic, sometimes water, and a grid connection in the same queue as everyone else’s.

If the benefits are national and the costs are local, what does a scheme have to give the town it sits in before it wins the room?

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

Graphic accompanying the post: What have data centres ever done for us?

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

Posted 3 November 2026 · Read the article

From 2027, around 10,000 manufacturers stop paying three levies on their electricity: the Renewables Obligation, Feed-in Tariffs and the Capacity Market. Data centres keep paying them. Read the full post

This year those three come to £48.53 per MWh on a flat load. For 100 MW of booked capacity, every percentage point of load adds about £0.42m a year.

So the bill depends on how full the site is. Published analyses of UK Power Networks’ data put the average data centre at about a fifth of its booking: about £8.5m a year per 100 MW. At the 70% that planners assume, £29.8m. Flat at full booking, £42.5m.

It also depends on when the site runs. The Capacity Market is charged only on electricity drawn from 16:00 to 19:00 on winter working days, 246 hours this year. An AI training site running at 80% that steps down in those hours would cut its bill from £34.0m to £28.0m.

I’m at LCCC Connects in Birmingham today, so it seemed the right day to put numbers on it.

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

Provenance: Analysis, my own arithmetic on published 2026/27 rates.

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

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

Three lines to power, and the main one is suspended

Posted 2 November 2026 · View on LinkedIn · Read the article

Right now, a new data centre can’t apply for a transmission connection in Great Britain. NESO closed the queue to new demand in March 2025, there’s still no date for the next window, and last week’s government response puts Ofgem’s decision on its data centre reforms in the first quarter of 2027. Read the full post

I’ve drawn every route to energisation as a metro map. Three power lines (transmission, distribution, on-site gas) and two consent lines (local planning, and the national DCO route, which stops at the English border).

Planning isn’t a condition of Gate 2 for demand. Land rights alone will do, and planning becomes a milestone after the offer, with the application due 48 months before the connection date on a long-dated project.

Distribution is open, but a big enough project still needs a Transmission Entry Assessment through NESO. The government confirmed last week that this demand sits outside the gated queue, and it is weighing regional thresholds that would bring it in.

On-site gas has its own long station at the gas main: Cadent aims for 12 months on its largest connections, and National Gas’s indicative timeline for a national transmission system connection comes to about 36 months on my sum of its stages.

There’s also a short branch: a modest on-site bridge that runs part of the site until the grid arrives. Kept under 50 MW, it stays in local planning. The permit turns on a different number, 50 MWth of fuel input across every engine on site, standby sets included.

The short stations are well sourced. Offers within three months, 90 days to accept, a section 35 direction in about eight weeks. The long ones, reinforcement and build, have no published times at all, so the map shows none.

If the transmission window reopened tomorrow, would your project be ready for the first station?

The full map, with sources for every station, is in the article. If the map is hard to read (at whatever size is available to you) please feel free to message me and I will send you the original PDF.

Provenance: Synthesis, with an original schematic.

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

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

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

Posted 30 October 2026 · View on LinkedIn · Read the article

Anyone heading north from Euston knows the old concourse. No seats, thousands milling under the split-flap board, and then the rush when the platform was finally announced. Read the full post

Prompted by Guy Massey’s US arrivals board, I’ve drawn the UK data centre pipeline the same way. Nine services, the time each was meant to leave, and when the public evidence says it will.

A few from the board:

About 13 GW of large demand can connect before 2030. Another 86 GW is slated for 2030 to 2035 (NESO).

Data centre connection requests stand at around 72.8 GW through to 2039, 59 GW of it at transmission across 173 projects.

Planning has a statutory 13 weeks for a major application. The observed average for data centres is about 490 days.

Gate 2 Phase 1 offers, already extended once to 30 September, are now due by 31 October after a second extension. That’s the platform announcement, and holders then have 90 days to accept. Expect the rush.

The first Strategic Spatial Energy Plan was promised for 2026. Last week its statutory deadline moved to 31 December 2027.

Along the bottom is my own figure: every month a data centre waits for power costs its investors £65,000 to £105,000 per MW of IT load, and that’s a floor.

The newest service is the Budget’s Data Centre Covenant (28 October). Expected: TBC, because “strategic” has no definition yet.

GB Grid (28 September) isn’t on the board yet. It has no timetable, so it stays off until it gets one.

The full board, with every source, is in the article.

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

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?

Posted 29 October 2026 · View on LinkedIn · Read the article

In Part 1, Project Flowers (a UK answer to xAI’s Colossus) reached first power in about eight months with fuel cells on a council-owned site, against about twenty on an average site. Memphis took four. Read the full post

I’ve now priced those months, using the cost of delay I published on 9 October: £65,000 to £105,000 per MW of IT load per month. I’ve taken the lowest published rate, which was modelled for longer waits, so treat the results as indicative.

At 31 to 40 MW of IT load, a month is worth about £2.0m to £2.8m to a fuel cell site.

The council route saves about twelve months, worth roughly £24m to £33m. With gas engines, where engine delivery sets the pace, it saves eight months and about £6m to £8m.

Against Memphis, the UK’s best case runs four months behind, about £8m to £11m at fuel cell scale. Part of that buys protection Memphis’s neighbours didn’t get.

All of these are gross, before the plant, fuel and financing. The net figure depends on the site.

There’s a catch in the council route. Permission on a council’s own land benefits the council and any named joint developer, so the time saved becomes something the two would negotiate over.

The working, the chart and the caveats are in the article.

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.

The Data Centre Covenant: what the Budget offered data centres, and what it asked in return

Posted 28 October 2026 · View on LinkedIn

The Chancellor published a Data Centre Covenant alongside today’s Budget. I have read the full text on GOV.UK, and three points stand out. Read the full post

Temporary on-site power is now government policy. The covenant commits to a “tightly controlled” framework for on-site generation until a grid connection arrives, limited to “exceptionally strategic projects”. My own modelling puts the cost of waiting for a connection at roughly £65k to £105k per MW per month, so a year’s delay on a 50 MW scheme runs to £39m or more. Bridge power is now a recognised policy tool, for a handful of sites at least.

The queue may be reshuffled. Government intends to use Planning and Infrastructure Act powers to reserve and reallocate capacity, and to transfer pre-2030 connection contracts to strategic sites. Anyone holding a pre-2030 date on the TEC register should read that line twice.

Priority comes with a community bargain. Strategic projects will need a Local Benefit Plan with “substantial investment” linked to size, a design PUE of 1.3 or better, annual energy and water reporting, and a contribution to UK jobs and supply chains. Ofgem’s data centre commitment fee, which I responded to in September, gets explicit support.

The definitions are still missing. There is no MW threshold, no criteria and no application route; those will follow “in due course”. Without them, it is hard to see clearly how the industry will be affected. I would like to think there will be some focus on smaller projects, which are easier to complete, so that the benefit is not concentrated in a few hyperscale locations.

If temporary power and connection transfers go only to the strategic few, what happens to the projects sitting just below the line?

Graphic accompanying the post: The Data Centre Covenant: what the Budget offered data centres, and what it asked in return

Flexible connections and the AIEC, why it matters

Posted 26 October 2026

Until the Data Centre Alliance conference in Birmingham last Tuesday, I didn’t know the AI Energy Council existed. It’s a joint DSIT and DESNZ council, co-chaired by both Secretaries of State, with Ofgem, NESO, National Grid and the large cloud operators round the table. Having looked into it since, its work looks potentially vital to how data centres get connected, and very few people seem to know about it. Read the full post

The minutes of its meeting of 13 April 2026 came out in June. Read in order, they make a chain.

1. DESNZ told the council that about 125 GW of demand sits in the connections queue. The system’s past peak is roughly 45 GW.

2. Clearing out speculative applications will shrink that number, but the minutes say substantial genuine demand remains.

3. Three ways to let some of it on sooner were explored: curtailment during stress events, non-firm connections and voluntary flexibility.

4. NESO put the flexible demand it can manage at about 6 GW, broadly in line with Clean Power 2030. Going further needs more work.

5. Around the table, non-firm connections were seen as a possible barrier to investment, partly because the fallback is diesel, and diesel brings permitting problems.

6. On-site batteries and solar were raised as a way round that. Behind-the-meter solutions were parked for a future meeting.

The evidence that data centres can flex already exists in London. National Grid’s trial with Nebius and Emerald AI in December 2025 cut a GPU cluster’s demand by up to 40% across more than 200 simulated grid events. Ofgem’s June update on demand connections reform also describes non-firm and phased connections with a path to firm.

The missing link is the non-firm connection. A bridge strategy becomes significantly easier if a proportion of the requirement is supplied through a non-firm connection. Data centres seldom use the amount of power the headlines suggest, so even a small share of supply from the grid could make a significant difference to how the bridge is configured and when the site can connect.

The time saving is where the value lies.

Synthesis. Sources: AI Energy Council minutes, 13 April 2026 (GOV.UK); National Grid, UK-first trial of AI grid technology; Ofgem, Demand Connections Reform update, June 2026.

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

Graphic accompanying the post: Flexible connections and the AIEC, why it matters

Is curtailment quietly controlling the small end of the queue?

Posted 26 October 2026 · View on LinkedIn

Last week I set out the five controls now sitting on GB transmission connections. There may be a sixth, and nobody designed it. Read the full post

On-site C&I solar is usually financed on a simple structure. The customer uses most of the power, a fund owns the panels, and an export connection sits behind the deal in case the customer defaults. That export route is the funder’s backstop. Take it away and the credit committee says no.

A fund financing sites under 1 MW told me recently that export has got harder to secure, and that fewer sites now get solar than could. When an offer comes back curtailed, or without firm export, the backstop weakens and the project stalls before FID. It never shows up as a refused connection. It just doesn’t get built.

If that’s happening widely, it’s a queue control that works through the funder’s credit committee. CMP446 raised the transmission assessment threshold in England and Wales from 1 MW to 5 MW in May 2025, which should have made the small end easier. Curtailed distribution offers could be taking some of that back, and nobody’s counting it.

I don’t have the evidence yet, so I’m asking.

If you finance, develop or connect C&I schemes under 5 MW:

Has a curtailed or non-firm export offer stopped a deal reaching FID?

What level of curtailment does your credit committee treat as a deal-breaker?

Does it differ by DNO?

Comments or DMs both welcome. I’ll publish what comes back, anonymised.

Where does curtailment stop being a network problem and start being a finance one?

Six controls on the queue. One has no CMP number. Five sit in the connection codes. The sixth works through the funder, one level down: gated process (TMO4+, 2025); queue milestones (CMP376, 2023); User Commitment (CMP192, 2013); progression fee (CMP448, 2025); oversubscription fee (CMP470, proposed); curtailed export offer (no CMP number, untested), where a DNO offer with export curtailed, distribution under 5 MW, ends in no FID: never built, never counted. The sixth control has no code, no fee and no record of what it stops.

Who pays for the wires? The factories got there first

Posted 23 October 2026 · View on LinkedIn

Slough Trading Estate has had its own power station since about 1920. It began as a 2,000 kW plant for the Army’s vehicle depot, and between 1933 and 1938 the estate added new units totalling 16,800 kW, while the Central Electricity Board’s 132 kV grid was being joined up into a national system. The same estate now hosts more than 30 data centres. Read the full post

Most firms moved the opposite way. Private generation retreated through the interwar years as public suppliers offered more attractive tariffs. Southern Railway closed its "obsolete" generating plant at Eastleigh works and took a bulk supply from Southampton Corporation.

The plant that lasted tended to have a second job. ICI’s Wilton station on Teesside has supplied electricity and process steam to the chemical complex since March 1952, and still does under Sembcorp. At Slough, the old units were partly demolished from 2018, and the 55 MW energy-from-waste plant that replaced them exports power to the grid and supplies steam to a factory next door.

A century ago, firms generated because there was no grid to buy from. Data centre developers face something odder: a grid right outside the fence that is full, with connection dates years out.

In the rush to power, I feel that consideration about stranded assets has become secondary. History seems to show that a correctly configured asset can far outlive its original intended use. I have worked through how this can be done for data centres and I’m happy to talk it through one to one.

If you’re planning on-site power for a site today, what will it be doing the day the grid connection is energised?

Synthesis. Assembled from published sources; the point that the plant which lasted tended to have a second job is my own reading.

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

Graphic accompanying the post: Who pays for the wires? The factories got there first

Data centres, measured in Jackdaws

Posted 22 October 2026 · View on LinkedIn

Jackdaw received consent yesterday. Wood Mackenzie puts its 2027 output at about 140 million cubic feet of gas a day, and the government says it could heat around 1.4 million homes at peak. Read the full post

Converted to power, that’s about 4 million cubic metres a day, nearly 16 TWh of gas a year at a typical UK calorific value. In a modern combined cycle plant at around 50% efficiency, it comes to roughly 0.9 GW running every hour of the year.

Now the comparison that gets less attention. In September I estimated that the 6 GW of data centres NESO expects to connect by 2030 would draw about 1.4 GW on average, because data centres use far less than the capacity they book. So in energy terms, the whole UK data centre fleet expected by 2030 is about one and a half Jackdaws.

That’s a useful proportion to keep in mind when data centres are described as a threat to supply. One mid-sized gas field, approved today, is in the same range as the entire near-term fleet. The harder constraint is the connection, not the energy.

Analysis: own conversion of government and Wood Mackenzie figures; data centre draw from my 18 September analysis of NESO figures.

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

Data centres, measured in Jackdaws. One Jackdaw is the field’s 2027 gas output as round-the-clock power, about 0.9 GW. The 2030 data centre fleet’s average draw of 1.4 GW is about 1.5 Jackdaws; its 6 GW of connected capacity is about 6.6 Jackdaws.

Who owns the battery queue?

Posted 22 October 2026 · Read the article

I traced the 25 largest battery companies holding countersigned Gate 2 capacity on NESO’s TEC Register (5 October, 9.5 GW) through Companies House. Read the full post

They come down to 17 owner groups. Five of them, Fidra (EIG), Copenhagen Infrastructure Partners, Statera (EQT), Matrix (TPG) and Revera (Carlyle), control 52% of the capacity.

Funds control 68.8% of the ranked queue. On the batteries already built and on the register, the figure is 38%.

Three things the register doesn’t show on its face. The National Wealth Fund, owned by HM Treasury, holds 25 to 50% of the company behind Thorpe Marsh (1.4 GW), the largest entry. Twelve companies carrying Harmony or HEIT names answer to three unrelated owners. And two significant holders stopped naming any controller in August.

Ofgem’s commitment fee is aimed at capacity that may never be built. If most of the queue belongs to funds, will the fee change who builds these batteries, or only who holds them while they wait?

Original analysis. I haven’t found a published trace of who controls the countersigned Gate 2 battery queue. The built comparison credits The Renewable Guy & Cities and Vitreous Labs (September 2026).

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).

Loughton without the farmland, or the wait

Posted 21 October 2026 · View on LinkedIn

Paweł Czyżak, PhD’s model of Nscale’s Loughton site is a good piece of work, and it got me asking a narrower question. On the 1.5 ha plot itself, without the Green Belt fields next door, how soon could all 50 MW of IT be running? Read the full post

Timing is what saves money here. Every year a hall sits dark is a year of base rent the owner never collects (what a tenant pays per kW of IT capacity for the building, power and cooling plant, with electricity on top). At a hyperscale-type $130/kW/month, waiting for the 2033 grid date costs about £350m.

Taking the 45 MW of reported substation headroom as a phased connection gets 39 MW running from 2027. Adding 18 MW of gas engines and a 20 MWh battery gets the full 50 MW running from 2028, five years early. The gas connection is the long pole at 6 to 24 months; the engine permit takes about three. The bridge costs about £38m and protects about £65m of rent.

Paweł’s own core design (the same 45 MW on non-firm terms with a 360 MWh battery) could also run from 2028, but only if a network offers those terms, and the battery alone is about £74m. His solar doesn’t change the connection, as he says, and 155 MWp would need a national consent order on Green Belt land.

Running everything on gas is slower still. About 63 MW of engines crosses the 50 MW consent threshold, and those orders average three to three and a half years. Nothing runs until about 2031.

Ramps matter too. AI training loads swing by 30% or more in under a second. Below the 45 MW cap the grid absorbs that, but engines near full load take only a small instant step, so the battery does the fast work.

Everything rests on one assumption: that UK Power Networks would offer the 45 MW as a phased connection now. We know that type of contract is coming, when it does the sector could be transformed.

So would a network operator offer half a connection today, if it meant a data centre in 2028 rather than 2033?

Assumptions and sources are in the image. Indicative, not advice. Views are my own.

Please check out my archive at markdmoore.co.uk

Loughton options on the 1.5 ha plot: when each route gets 50 MW of IT running, with assumptions and sources.

A grid date is a probability

Posted 20 October 2026

I’m at the Data Centre Alliance conference at the IET in Birmingham today. If your programme rests on a connection date, this one is for you. Read the full post

The TEC Register can be read backwards. I took the edition of January 2025 and found 99 battery projects contracted to connect by now. Twenty-six have connected. Seventy-one are still in the queue, most with a later date and some still showing one that has passed. Two left.

The February and July 2024 editions give the same answer: roughly one in four.

Project status predicts the outcome. Of the batteries already under construction, 86% made it (a small group of seven). With consents approved, 35%. At scoping, 13%.

Data centres cannot be tested the same way.

NESO has said it does not maintain a register of demand connections (FOI 25/291), so nobody can publicly measure how data centre dates hold. Ofgem’s Transforming Connections Data consultation proposes a demand capacity register. It closes on 6 November.

What can be measured is the network behind the dates. The NAO reported in September that 50 of the 56 essential transmission projects are forecast to finish after their optimal date. Of 155 TEC entries tied to the accelerated transmission programme in my own register, 52% carry dates beyond the licence backstop. Data centres wait on that same network, and Ofgem puts them at about 73 GW of the demand queue.

The read-across to data centres is an inference, and the graphic marks it as one. Even so, I see no reason to expect a data centre date to be firmer than a battery’s.

If your contracted date came with a probability attached, what would you plan to?

Original analysis of NESO TEC Register editions, 2024 to 2026; NAO, NESO and Ofgem as cited. Supported by National Energy SO Open Data.

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

A grid date is a probability: of 99 battery projects contracted to connect by now on the January 2025 TEC Register, 26 connected, 71 are still in the queue and 2 left; outcome by project status; the read-across to data centres marked as an inference.

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

Posted 19 October 2026 · Read the article

Microsoft has walked away from anchoring Nscale’s data centre at Loughton in Essex. No reason given, but grid officials had reportedly told Nscale that power would not arrive until the 2030s. Read the full post

This was the site named in 2025 as the UK’s largest sovereign AI data centre. It had planning permission and a 90 MW connection agreement. It still could not get power on time.

Two numbers stood out when I went back through the policy papers. The government’s own AI Growth Zones paper calls grid connection the single biggest blocker. Ofgem’s Curate consultation found 99% of data centre capacity in the queue sits in projects above 40 MW.

So the queue is a big-site problem. A 30 MW site can be bridged with on-site power and fits on around ten acres. Owners are already trying it, from Stoke Poges to Wokingham, and every attempt runs into objections, permitting thresholds and conditions that turn an 18-month bridge into something much slower.

In the article I set out three policy changes that would let small, bridged sites go ahead now, and why GB Grid, whose operating rules have not yet been published, could be the vehicle that moves them onto the grid the moment it arrives.

Synthesis. Sources: Reuters, The Telegraph, Ofgem, GOV.UK, Future Energy Networks.

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

Two data centre sites: a 90 MW site waiting for grid power in the 2030s, and a 30 MW site running on on-site power while it waits.

Are smaller data centres actually faster to power in Great Britain?

Posted 19 October 2026

After my 5 October piece on AI labs reportedly sounding out 20 to 30MW sites, I went looking for evidence that small sites reach power sooner. I couldn’t find any. Read the full post

Plenty of people say it. Savills argues inference can be built in 5 to 20MW tranches that sit better with grid and planning constraints. NESO told the Environmental Audit Committee that smaller data centres may be built out faster. Yet NESO’s own Ten Year Outlook 2026 gives every new data centre the same seven-year ramp to full capacity, whatever its size.

Ofgem’s figures show GB’s transmission demand queue rising from 17 GW to 97 GW between November 2024 and June 2025, while distribution moved from 24 GW to 29 GW. Most of what’s actually running is far smaller. UK Power Networks counts over 80 operational data centres on its network: 10 at EHV, 56 at HV and 18 at LV.

The link between size and time is what’s missing. The Embedded Capacity Registers list generation and demand-side response contracts, so a plain data centre import connection doesn’t appear. UKPN’s Large Demand List was cleared for publication with capacity and NESO completion dates anonymised or banded. That’s a fair answer to commercial confidentiality, and it also takes out the two fields this question needs.

If size makes no difference to the wait, the case for spreading compute across smaller sites has to rest on something other than the grid.

I’ll publish what comes back, anonymised, in a follow-up.

So, if you’ve connected a data centre in GB in the last five years, how long did it take from accepted offer to energisation, and at what size and voltage?

Analysis

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

Graphic accompanying the post: Are smaller data centres actually faster to power in Great Britain?

What does the wait add up to?

Posted 16 October 2026 · View on LinkedIn · Read the article

Thoughts for the weekend. Who pays for the wires? A postscript, by popular demand (mine).
Last week I put the cost of waiting for a grid connection at £65,000 to £105,000 per megawatt of IT load, for every month of delay. This week I’ve added it up. Read the full post

Take the whole data centre queue, about 73 GW by Ofgem’s count, and the total comes to over £300 billion. That figure means nothing. Built out in full, the queue implies £693 billion of capital spending, about 23 per cent of UK GDP, and much of it will never be built.

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

The range is wide for a reason. It turns on how much of the queue gets built and how long each site waits, and neither is published. If 7 per cent is built, the central figure falls to about £22 billion. At 30 per cent it’s about £95 billion.

Britain has no public register of demand connection dates. Ofgem’s consultation on connections data proposes one, and it closes on 6 November.

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?

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 post: What does the wait add up to?

A year of data centre water, measured in hours of leakage

Posted 15 October 2026 · View on LinkedIn

There are three published figures for how much water UK data centres use in a year. Read the full post

WRc, working for MOSL, the company that runs England’s business water market, puts it at 1.88bn litres. That’s England only, and only water from the mains. Water UK told the Environmental Audit Committee 6.6m litres a day, which is about 2.4bn a year. Foxglove and The Times said at least 10bn, and called that an underestimate.

The lowest and highest are a factor of five apart, on the same question.

Now put them next to leakage. Ofwat’s figure for England and Wales in 2024-25 is 2,869 million litres a day. That’s roughly one litre in five of everything put into supply.

At that rate the pipes lose the WRc figure in about 16 hours. The Water UK figure goes in 20 hours. Even the Foxglove figure is gone in three and a half days.

Data centre water still matters locally. WRc found that six facilities account for about 65% of the total. Water UK points out that peak demand lands on hot days, when the rest of the network is already stretched. Those are real site-level questions.

The national debate is running on numbers that don’t agree with each other, though, and none of them comes from reported data. No UK data centre has to report its water use.

Which figure have you been quoting, and do you know where it came from?

Provenance: Synthesis. Figures from Ofwat, Water UK, WRc for MOSL and Foxglove. Scopes differ and are not like for like.

Graphic accompanying the post: A year of data centre water, measured in hours of leakage

Fourteen UK data centre schemes passed the test. One rated Green.

Posted 13 October 2026 · View on LinkedIn · Read the article

EnergyTag rated 21 US data centre projects on how they buy their power. I’ve built a UK version, with one rule that shapes everything: a scheme goes on the map only if its MW, its supply arrangement and a dated status can be read in a developer, council or government document. Schemes that appear only in trade press go on a watch list. Read the full post

Thirty-eight schemes were researched. Fourteen passed.

The one Green is Kao Data’s Harlow campus, which has a 40 MW private-wire solar farm being developed by Downing Renewable Developments under a PPA. It won’t generate until about 2029 and covers part of the load, but it is contracted, named and tied to the site.

Three are Red. SDC M40 Campus in Buckinghamshire is consulting on a gas energy centre of about 200 MW, with a grid-only option alongside it. Quest Park at Stewartby plans gas at every building until a 200 MW grid connection arrives in October 2034. Thames Valley Park, near Reading, filed on 28 September with gas fuel cells as its main supply until 2037.

Ten are Amber, grid supply or a clean-power claim with no named asset behind it. Google Waltham Cross has a named clean-energy arrangement but no MW figure from Google or the council, so it waits on the watch list with 23 others.

The NESO TEC Register couldn’t help. Its 106 Demand rows (64.7 GW) are all hybrids, and not one names a data centre.

The map updates on the second Tuesday of every month. The article sets out the rules.

A scheme can seek planning consent and a grid connection without ever stating, in one public place, how much load it will draw and where the power comes 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 centre schemes passed the test. One rated Green.

If the UK had its own Elon Musk, how fast could they build a data centre?

Posted 12 October 2026 · View on LinkedIn · Read the article

xAI’s Colossus in Memphis went from start to finish in 122 days. It had a ready factory shell, a gas main to hand, and a local rule that let its turbines run for up to 364 days without an air permit. Read the full post

I’ve set that against the UK rules, and size decides almost everything. In England, on-site power above 50 MW of fuel burned needs a bespoke permit, and above 50 MW of electrical output it needs a development consent order, about 16 months from submission on 2026 averages. There’s no 364-day window here either: a mobile generator that stays in one place for six months is treated as static.

So I worked out how big a fast UK build could be. With gas engines, about 8,500 to 11,000 GPUs, roughly a tenth of Colossus. With fuel cells, which don’t burn their fuel, about 24,000 to 31,000.

On a council-owned site with fuel cells, time to power comes out at about eight months, twice Memphis. On an average site it’s nearer twenty, because planning consent for a data centre averages about 490 days.

The full working, the timeline and the council route are in the article.

It’s been done here before. Tommy Flowers’ team had the first Mark 2 Colossus in service four months after the first machine started work, in time for D-Day.

Original analysis, from public sources, checked 10 October 2026.

Graphic accompanying the post: If the UK had its own Elon Musk, how fast could they build a data centre?

The latest grid queue slip came from the engineering desk

Posted 10 October 2026 · View on LinkedIn

Last month I wrote that automating the assessment of the GB connections queue would not build a substation. Sorting projects into order is the quick part. What follows is slower. Read the full post

On 2 October Ofgem agreed to move the deadline for Gate 2 Phase 1 connection offers from 30 September to 31 October. That deadline has now moved twice, and about 62 offers are caught by the latest change.

Ofgem’s letter is specific about why. NGET asked for two more weeks on its construction offers for smaller embedded projects behind grid supply points, and four more weeks for transmission-connected and large embedded projects, citing complexity and further network redesign. Another 25 or so offers were held back, mainly at the customers’ own request. Ofgem said it was disappointed with NGET. Some of the trade coverage I have read points that at NESO instead, which the letter does not.

The queue was re-sorted last December. Where the timetable now catches is the point at which someone has to work out what each project needs from the network, and that work is taking longer than planned.

Some proportion. This is one month and 62 offers, and NGET says 98% of its transmission construction offers reached NESO within two weeks of the original milestone. A construction offer is also still paper. Nothing has been built.

For scale: the licence default for these offers is 7.5 months from the application window closing in August 2025. Phase 1 now sits at about 14.

The offer clock, moved twice: Gate 2 Phase 1 connection offers. Application window closed 26 August 2025; licence default 7.5 months, to 10 April 2026; Phase 1 deadline now 31 October 2026, about 14 months. Second slip of about 62 offers, NGET construction offers late.

What is the grid worth to a data centre?

Posted 9 October 2026 · Read the article

Thoughts for the weekend. Who pays for the wires? Part 3 of 3.
In 1964 Robert Fogel asked how much poorer America would have been in 1890 without its railroads. For farm freight his answer was no more than 2.7 per cent of GNP, because rivers, canals and wagons could have done most of the work. A network is worth what it saves over the next best alternative. Read the full post

For a data centre waiting on a connection, the simplest alternative is no power at all. On sourced inputs that costs between £65,000 and £105,000 per megawatt of IT load for every month of delay, or roughly £285 million for a 50 MW site over a five-year wait. Most of it is margin never earned. Ofgem’s proposed commitment fee comes to about 4 per cent of it.

The figure is a floor. It’s also an upper bound on what the grid is worth, which is Fogel’s point. Sites that build their own power are playing the part the canals played in 1890, and every one of them caps the value of the network it’s waiting for. By how much is a harder question.

The private cost isn’t the social cost either. Some of the margin a delayed site never earns is earned by another site instead, and I haven’t found anyone who has measured how much.

If a connection date is worth £3 million to £5 million a month to a 50 MW site, what should it cost the customer who receives one?

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.

Five conferences on energy: what was promised, and what was costed

Posted 8 October 2026 · View on LinkedIn

Before this year’s party conferences began, two of the biggest bill levers had already been pulled. Since April, 75% of Renewables Obligation costs have come off domestic electricity bills and been paid by the Exchequer, for three years. VAT came off domestic electricity on 1 October, worth about £45 a year to an average household. Read the full post

I’ve put the energy content of all five conferences into one table, with a row showing where each promise overlaps what’s already in place.

Reform’s £250 a year rests partly on removing VAT and policy costs, and both have already been largely taken off electricity bills. There’s no published costing that I could find. The Conservatives would remove the Renewables Obligation altogether, but three quarters of its cost to households has already gone until 2029, and their wider plan rests on modelling of £320bn to 2050 that Carbon Brief has disputed. Labour’s VAT cut is the one pledge with a stated funding source.

What’s left on the bill is mostly wholesale gas and network costs, and network costs are the part that grows as the grid is rebuilt. Only Labour (Great British Grid) and the Conservatives (“let industrial users build power connections quickly”) said anything about connections, and neither explained how a faster route fits with the reformed queue and its Gate 2 dates. The Lib Dems and Greens put no energy figures in front of their conferences this year; in 2025 the Lib Dems had grid upgrades and £1bn for burying cables on the agenda.

If you’re developing a data centre or a battery site, which matters more to you now: a cheaper bill, or a connection date you can rely on?

Synthesis: party conference speeches and policy documents, September to October 2026. I haven’t found another 2026 comparison covering costing and connections across all five parties.

Graphic accompanying the post: Five conferences on energy: what was promised, and what was costed

The WEF’s new data centre playbook asks for something Britain is already testing

Posted 6 October 2026

The World Economic Forum and Oliver Wyman published a decision playbook on data centre sustainability and resilience this month. It’s a useful map of the metrics landscape, and one policy recommendation stood out: where projects compete for limited grid capacity, regulators should set transparent frameworks for managing and prioritising connection requests. Read the full post

Britain is already running that experiment. NESO’s connections reform has sorted the queue into Gate 1 and Gate 2, Ofgem’s Curate consultation proposes a commitment fee and milestones for data centre demand, and CMP470 proposes a commitment fee for oversubscribed generation technologies. None of it appears in the report, which mentions the UK mainly for planning and water law and uses London as its example of an urban, infrastructure-constrained market.

That matters because the UK is producing the evidence the playbook says is missing. NESO told Parliament it holds around 72.8 GW of data centre connection requests through to 2039, forecasts about 5.2 GW connected by 2030, and doesn’t expect all of it to connect. How a front-door test sorts that gap is the whole question.

One thing neither the playbook nor the UK regime measures yet is ownership. A queue position can change hands at project-company level while the name on the connection agreement stays the same, so a test applied on the day of application says little about who holds the position two years later.

If you work on either side of this, which UK measure do you expect other markets to copy first?

Provenance: Synthesis. WEF and Oliver Wyman, Data Centre Sustainability and Resilience (September 2026); NESO written evidence to Parliament (DCU0081); own reading of Ofgem’s Curate and CMP470 consultations.

Graphic accompanying the post: The WEF’s new data centre playbook asks for something Britain is already testing

Who holds the cards when the biggest AI buyers go small

Posted 5 October 2026 · View on LinkedIn · Read the article

Fifty-eight of the 315 data centre projects in GB’s connection queue sit below 50MW, holding about 2% of the queue’s capacity. On 18 September CNBC reported, citing four people familiar with the conversations, that Anthropic has sounded out deals of 20 to 30MW in the UK and the Nordics. OpenAI has reportedly looked at similar Nordic sites. Neither company has confirmed a deal. Read the full post

The same buyer was reported by Bloomberg a month earlier to have agreed around 460MW at Nscale’s West Virginia campus, also unconfirmed. Large and small at once, in other words.

Ofgem’s Curate proposals handle the small end in a particular way. The commitment fee, £237,500 to £712,500 per MW, applies only above 40MW, so most projects in that lane would post nothing. The queue milestones reach much lower, and one of them is a binding offtake deadline tied to the connection date. Developer and buyer can both see it approaching. With a frontier lab across the table, the developer’s hand at that date is weak.

Splitting is the second gap. At the proposed rates a single 250MW site would post between £59m and £178m. Ten 25MW sites carrying the same load would post nothing, and Curate has no test for what counts as one project. Ireland’s regulator does: applications below its de minimis threshold for the same project or owner are treated together.

The article covers three routes a lab could take into GB, and why the fee and the milestones are measured against different quantities.

If the largest buyers can meet their needs in blocks under 40MW, should Curate carry an aggregation test, and what should count as one project?

P.S. Researched with Claude, made by Anthropic. Yes, I used Anthropic to check up on Anthropic. It took it very well.

Analysis: draws on published reporting, Ofgem’s Curate consultation and my own consultation response.

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

Same buyer, different scale, as reported: a 460MW campus deal at Nscale, West Virginia (Bloomberg, 26 August 2026) against UK and Nordic enquiries of 20 to 30MW, midpoint 25MW (CNBC, 18 September 2026). Neither deal is confirmed.

Last time Britain let a network boom run

Posted 2 October 2026 · View on LinkedIn · Read the article

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 on the Lancaster Canal. The pub, the stables and the jetty for the passenger boats have all gone. So has the canal company: within a generation of the railway arriving, it had leased the canal to a railway company. Read the full post

In 1847 British railway companies spent £44 million building track, almost 8 per cent of GDP. As a share of today’s economy, that would be more than £200 billion in one year. Parliament was already regulating them, and one rule it wrote has outlived the companies themselves.

The Railway and Canal Traffic Act 1854 banned undue preference, so a railway or canal company couldn’t favour one customer over another. Jon Stern argues the Victorian railway model became the template for regulating electricity, and the distribution licence still carries a condition headed “Prohibition of discrimination”.

My reading is that this is where the connection queue comes from. If you can’t prefer one customer over another, the safest way to decide who goes first is the order they turned up in.

Connections reform hasn’t repealed any of that. What it has changed is what counts as undue: being ready, and being needed by the system, are now good reasons to go ahead of someone who applied first.

Meanwhile contracted demand offers went from 41 GW to 125 GW in seven months, at least 80 GW of it data centres, and developers who can’t wait are building their own power. Mark Casson showed what happened when Victorian companies were refused access: they built duplicate lines, and the surplus ended up on everyone’s fares.

After 170 years, Britain has decided that some preference is due. Who pays for the duplicates this time?

Synthesis. Peer-reviewed economic history, primary legislation and regulatory decisions; the Packet Bridge detail is first-hand; the link between the queue and the 1854 principle is my own reading.

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

GB Grid can’t buy its way up the queue

Posted 1 October 2026 · View on LinkedIn

Andy Burnham announced GB Grid at conference on Tuesday, a publicly owned investor sitting inside Great British Energy. Most of the coverage set the reported £4bn (Government hasn’t confirmed a figure, start-up costs come from GBE’s existing budget) against the up to £70bn Ofgem expects transmission owners to spend between 2025 and 2031, and concluded it’s small. I’d look at the delivery side before the size of the cheque. Read the full post

The NAO reported on transmission delivery on 11 September. As at March 2026, of the 56 ongoing projects needed for Clean Power 2030, three are expected ahead of NESO’s optimal dates, four have slipped past 2030 and 28 face significant risk to their agreed delivery date. The constraints it names are skills, the supply chain, planning (DESNZ respondents report 18 months to three years to secure compulsory land rights) and system access, where two-thirds of the outage applications NESO handled in 2025-26 (15,300 of 23,400) were cancelled or rearranged. Finance isn’t on the list.

My own register of the 26 ASTI projects points the same way. Across 155 dated TEC Register entries linked to those schemes, 52.3% sit beyond the licence backstop date. That’s an association with a published regulatory date, not a read of anyone’s engineering programme, but it’s hard to square with capital being what’s missing.

So what could GB Grid’s money buy? To get its projects built sooner it would have to make itself the priority customer for cable, transformers, outage slots and consenting capacity, and even if it managed that, supply isn’t elastic. Its projects would move up the same queue and push existing ones back.

A prediction, for the record: by 30 September 2027, fewer than three of the 56 will be ahead of their optimal date and more than four will have slipped past 2030. I’ll check it against the next published delivery figures.

If more money can’t shorten the queue, what would?

Provenance: Original analysis. ASTI register built from dated NESO TEC Register snapshots, three obtained under FOI; NAO (11 September 2026), Ofgem and GOV.UK figures as cited.

Graphic accompanying the post: GB Grid can’t buy its way up the queue

Fixing the rules before the result

Posted 29 September 2026 · View on LinkedIn · Read the article

In July I traced the owners of fifteen large solar and battery schemes, 8,850 MW in total. For at least ten of them, the developer named in the public record wasn’t the owner. Read the full post

That was a snapshot. I’ll measure the same fifteen again on 27 January and 27 July 2027, and publish whatever I find. Today’s article sets out the rules first, so they can’t be adjusted to fit the answer.

A scheme counts as changing hands when a new outside owner takes 25% or more. That’s the level at which Companies House requires owners to be declared, so anyone can check the count.

Checking the starting position turned up two things. My July record was out of date or wrong on six schemes, and the article corrects each one openly. And one result is already in: on 7 August a new owner of between 25% and 50% was notified at Coalburn 1, in a chain that ends at AXA. It will be counted in January.

Fifteen schemes is a small cohort, so the result will describe these schemes only. The register shows who owns what, and says nothing about why.

How many of the fifteen would you expect to change hands by next July, and how many would it take before you’d call it a pattern?

Provenance: Analysis.

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?

Posted 28 September 2026 · Read the article

About 90 GW of battery storage is operating or holds Gate 2 status, against roughly 29 GW needed by 2035. According to Ofgem’s own impact assessment, about a quarter of it has posted no securities or liabilities at all. Read the full post

Ofgem’s proposed fee starts at £3,000/MW and climbs to £25,000/MW only if the surplus doesn’t shrink. At the start that’s around 3% of build cost for a short-duration battery. Projects that intend to build can carry it, and those holding a position as an option get a reason to think again.

I’ve supported it in my response to Ofgem. The one addition I’ve suggested is a way to see who ends up holding the queue.

A developer facing a rising security has more than two choices. Alongside staying and leaving, it can sell the project to someone better placed to fund it, and the capacity stays in the queue under a new owner.

The register wouldn’t show that. When Alpiq bought 90% of Harmony Energy in July, for reasons that have nothing to do with the fee, Companies House stopped naming anyone as controlling the company. The TEC Register lists project companies and has no ownership field, so it recorded nothing.

Ofgem expects the competition effects to be limited. Ownership reporting alongside the securities statements would let that be tested as the fee takes effect.

The full article also covers the 9 GW of batteries looking to convert to data centres, and how Ofgem costs the liquidity the fee requires.

If the queue shrinks, how would anyone know whether it’s held by more owners or fewer?

Graphic accompanying the post: Why does CMP470 make sense?

The haulage deal nobody would sign, except everyone already has

Posted 25 September 2026 · Read the article

I’ve worked in three industries over my career: engineering, logistics and now energy. Their business models couldn’t be more different, and when I started putting them side by side, some of the terms energy customers accept every day looked very odd next to how a haulage contract works. Read the full post

Imagine hiring a haulage firm, and the contract came with three conditions.

You pay a flat daily fee for the truck, whether it leaves the depot or not. For the biggest sites connected straight to the national grid, that’s £20,829 a day, about £7.6 million a year, just to have the connection available.

If the firm needs a new truck to serve a new customer, that customer doesn’t pay for it alone. Everyone else on the local network gets a little added to their bill to help cover it.

And whatever happens, the firm earns a return on its investment set in advance by the regulator, not won in a competitive market. Performance moves it a little either way. Right now the baseline is 5.7% on the national lines and 5.23% on the local ones.

Put plainly, customers are forced to buy the truck, and the haulier still gets paid, on a truck it now owns. Much of what’s spent on it joins the company’s asset base, which is what its return is calculated on.

The only way round it is to generate your own power next door and run a private line. Even then, if you stay connected to the grid, the fixed daily charge still applies.

If you wouldn’t sign that contract with a haulage firm, why accept it from the company that delivers your electricity?

Full sources and figures in the article.

Provenance: Analysis.

Graphic accompanying the post: The haulage deal nobody would sign, except everyone already has

Five controls on the connections queue, and what each one tests

Posted 24 September 2026 · View on LinkedIn

The GB transmission queue now has five separate controls on it. Put side by side, they bite at different points in a project’s life and they measure different things. Read the full post

User Commitment (CMP192, 2013) makes a project carry part of the network cost, though lightly at first: liability starts at 25% of the Wider Cancellation Charge after a trigger date roughly three years before energisation. Queue management milestones (CMP376, 2023) test progress and remove projects that stall. At entry, the TMO4+ gated process tests readiness and fit against Clean Power 2030. Then came the Progression Commitment Fee (CMP448), approved December 2025. It sits dormant, covers projects between Gate 2 and Milestone 1, and runs from £2,500/MW to £10,000/MW if terminations pass a threshold.

CMP470, Ofgem’s minded-to consultation published on 17 September, starts at £3,000/MW and climbs to £25,000/MW. It switches on when a technology exceeds its government capacity target by more than 50%.

That trigger is the new part. The earlier four respond to a project’s own conduct. CMP470 responds to the aggregate: roughly 90GW of BESS operational or at Gate 2, against a FES-derived 29GW for 2035. A consented battery sitting below the floor has to lift its security the same as a placeholder does. Ofgem’s own impact assessment says the fee makes projects compete on liquidity, and admits it cannot predict how many will leave.

Is a mandated trigger the best approach?

If the trigger is a government target, who carries the cost when the target moves?

Graphic accompanying the post: Five controls on the connections queue, and what each one tests

Great Britain’s data centre queue, in three numbers

Posted 18 September 2026 · Read the article

My Scotland post on Tuesday asked which filter is binding on the data centre pipeline. NESO’s new Ten Year Outlook puts figures on three of them for Great Britain as a whole. Read the full post

72.8 GW of data centre connection requests sit in NESO’s 2025 pipeline data. By 2030, NESO expects 6.0 GW of data centres to be connected in total, of which about 4 GW is new. That new capacity is roughly 5% of the request. Across the whole fleet, expected average draw in 2030 is about 1.4 GW.

Three assumptions sit between those numbers. Queue success is capped at 50%, and lower for less-progressed projects. New sites take seven years to build out. And NESO has cut its utilisation assumption from around 70% to 29% of built capacity, using distribution network data. That cut is why its 2030 demand forecast fell from about 20 TWh to 12.6 TWh even as forecast capacity rose.

Metered data points the same way. TwentyForty’s analysis of 96 data centres on UK Power Networks’ network found the median site drew 18% of contracted capacity, and nearly half never passed 40% even at their busiest half-hour.

The distribution connection terms already allow a network to propose reducing capacity after twelve months below 75%. How often that happens is not published.

The workings, sources and limits are in the accompanying article.

If most connected data centre capacity is expected to sit well below its booked level for years, what should happen to it before the network builds for the rest of the queue?

Analysis. The data is NESO’s and TwentyForty’s. The arithmetic and the reading of the connection terms are mine; the clause point comes from original research not yet published.

Graphic accompanying the post: Great Britain’s data centre queue, in three numbers

What sits inside the data centre price gap

Posted 17 September 2026

In my post on Scotland’s data centre pipeline I said the published data couldn’t show which filter is binding. William Andrews answered in the comments: power at roughly £40-50/MWh elsewhere, against £70-90/MWh delivered in the UK once network charges land. Both figures can be checked against published 2026/27 tariffs, so I checked them. Read the full post

They sit on different bases. £40-50 is in the range of European energy-only PPA prices. For an illustrative 100MW transmission-connected site at a 90% load factor, the UK charges added on top of energy come to roughly £75-100/MWh before any power has been bought.

On the same basis, excluding energy: UK (Scotland) £75-100/MWh; Sweden about £27-31/MWh, most of it electricity tax; Spain, above 145kV, about £6-8/MWh; Norway about £2-12/MWh, depending on connection point.

Network charges are the smaller part in Scotland. The locational transmission charge for demand there is zero, the fixed residual comes to around £10/MWh at this size, and balancing adds about £13/MWh. Policy levies (RO, CfD, Capacity Market, Nuclear RAB, FiT) add roughly £50-70/MWh.

A renewable private wire can avoid most of those levies, but only on what flows over the wire. Anything drawn through a retained grid connection carries the full stack.

Tomorrow I’ll publish a longer piece on the network side: who the carriers are, how they’re paid, and what a data centre is actually buying from them.

How much of a flat data centre load can a Scottish private wire realistically carry, and has anyone published that figure?

Analysis: NESO, Ofgem, LCCC/EMRS and HMRC 2026/27 figures, with Nordic and Spanish tariff and tax data. Capacity Market and FiT from secondary sources; Nordic network figures are my calculation.

Graphic accompanying the post: What sits inside the data centre price gap

Scotland’s data centre applications total roughly 7,150MW. NESO has allocated between one and two gigawatts.

Posted 15 September 2026 · View on LinkedIn

Operating capacity today is 30MW. That is DSIT’s own estimate, given to Parliament in answer to PQ HL5239, covering colocation load as at autumn 2024. Against a Great Britain total of 1.6GW, Scotland holds under two per cent. Read the full post

The applied-for figure comes from a compilation of local authority planning portals. The allocated figure comes from a Scottish Government answer to a written question, and the transitional Regional Energy Strategic Plan carries 1.6GW for data centre demand in Scotland. So developers have applied for roughly four times what the people who have to build the network are planning for, and nearly twice what the whole country draws at winter peak.

Both numbers are public. I have not seen them put side by side.

Which is where the structures get interesting. Apatura has nine schemes across the Central Belt, each in a separate vehicle. The Larbert applicant is Apatura DC Project 5 Ltd. ILI has three more, branded The Stoics, at Auchtertool, Hurlford and Newhouse. A planning consent attached to a connection position is a tradeable asset, and holding one requires nobody to pour concrete.

Falkirk holds a pre-determination hearing on the 300MW Larbert scheme next Tuesday, after 6,719 objections. It will be assessed against NPF4 and the Falkirk local plan, on its own. North Lanarkshire will assess Ravenscraig on its own. Fife will assess Auchtertool on its own. The 7,150MW question reaches no single decision-maker anywhere in the system.

I’m thinking about a monthly tracker that traces ownership and progress. Would anyone be interested to see that?

Graphic accompanying the post: Scotland’s data centre applications total roughly 7,150MW. NESO has allocated between one and two gigawatts.

The seven projects the NAO singled out have 37 GW sitting behind them

Posted 14 September 2026 · View on LinkedIn

The National Audit Office published its value for money report on the transmission upgrade on 11 September (HC 589). One figure in Part Two did not make the coverage. NESO has assessed the constraint cost impact of seven projects out of eighty-eight, and puts it at up to £6.7 billion a year if they are not delivered when needed. For the other eighty-one, it has not run the calculation. Read the full post

So I went looking for what sits behind those seven.

Six are ASTI schemes I already hold licence dates for: Norwich to Tilbury (two projects), Eastern Green Links 1 and 2, Yorkshire GREEN, and the Brinsworth to Chesterfield to High Marnham upgrade. The seventh, Sea Link, I have left out, because its landing points are inferred from geography rather than named in the ASTI description.

Against the TEC Register of 25 August, those six have 69 contracted projects at their named substations, totalling 37.4 gigawatts. That is 5.5 per cent of the register. Ten of the 69 carry no energisation date at all. Drax alone accounts for 8.2 GW.

This is a narrower cut than the ASTI-wide figure I posted on 7 September, not a revision of it.

As at March 2026 the NAO records three of the seven as off track for Clean Power 2030. The other four are on track, but facing significant delivery risks.

NESO responded the same day, under the heading that strategic planning is essential to delivering value for consumers and industry. One line in it is worth keeping: consumers pay when infrastructure build and generation fall out of step. That is the mechanism, conceded by the body best placed to dispute it. The rest of the response is about coordination across government, Ofgem, network companies and NESO, and says that no single organisation holds all the levers. Reporting takes fewer levers than delivery does.

Graphic accompanying the post: The seven projects the NAO singled out have 37 GW sitting behind them

Six rulebooks set a data centre size line. Almost none measure the same thing.

Posted 11 September 2026 · View on LinkedIn

Several land at or near 1 MW. Here’s what each one is measured against. Read the full post

Ireland’s CRU decision of December 2025 starts at 1 MVA of Maximum Import Capacity, and from 10 MVA a site has to bring dispatchable generation or storage at least equal to its import. It’s final, and facing a High Court challenge.

Spain’s draft royal decree, put to public hearing on 27 August, applies from 1 MW of access power; the preamble says access power was chosen because the thing being allocated is network capacity.

Germany’s Energy Efficiency Act catches sites from 300 kW of non-redundant nominal connected load, and the government bill approved by cabinet on 24 June would move that to 500 kW of installed IT power, the EU reporting threshold. Fingrid’s draft grid code sets its large-demand requirements at 30 MW of rated demand.

GB has three numbers on two rulers. Ofgem’s Curate proposal sets the commitment fee threshold at 40 MW, and the drafted CUSC text measures that against connection capacity. The queue management milestones are consulted on at 10 MW of rated IT load. The drafted definition, which I worked through for my Curate response, follows the Cyber Security and Resilience Bill: 10 MW for an enterprise data centre and 1 MW for everything else, colocation included.

Rated IT load and connection capacity pull apart once cooling overhead and redundancy are counted. One scheme can sit above one line and below the next.

Should GB measure the fee and the milestones against one quantity, or is there a case for keeping two?

Analysis. The texts are CRU’s, MITECO’s, the German government’s, Fingrid’s and Ofgem’s. The reading of them is mine, and the GB drafting point is in my Curate response.

The original six-page carousel as a video, 37 seconds, no sound.

Ofgem’s data centre milestones may bite at 1MW, not 10MW

Posted 8 September 2026 · View on LinkedIn

Curate closes on the 17th. Most of the commentary is working from the consultation document, which is fair enough, it runs to 86 pages. The four ancillary documents carry the actual drafted law, and, if I’m interpreting it correctly, I think there’s a gap between the two. Read the full post

Chapter 7 asks whether the new queue management milestones should apply above 10MW rated IT load. That’s Ofgem’s stated preference and it’s what’s being reported.

The drafted CUSC text doesn’t set that threshold. It defines a Data Centre Service as any facility at 1MW or above, unless the facility is on an Enterprise Basis, meaning its sole purpose is serving its owner’s own undertaking, in which case the figure is 10MW. Colocation isn’t enterprise basis. Neither is anything built to lease or sell on. So for most of the commercial sector the drafted floor reads as 1MW.

The draft direction to DNOs then applies that same definition to every distribution customer sitting outside a Transmission Entry Assessment, with no MW threshold in it at all.

The fee, meanwhile, gets a proper threshold instrument. Condition E18, annual review, 28 day consultation, Ofgem approval before the number can move. The milestones get nothing equivalent.

Practically, a 2MW colocation site on a DNO connection is outside the regime as Chapter 7 describes it and inside it as drafted. That’s compute offtaker evidence, long lead procurement, an investment grade rating or committed credit support, and design certification.

Both limbs track the Cyber Security and Resilience Bill, so the drafting is faithful to the Bill. Chapter 7 describes one of them.

Have I read this correctly? Is the definition doing the work of the threshold, or does the threshold decision come later and the drafting simply runs ahead of it?

Graphic accompanying the post: Ofgem’s data centre milestones may bite at 1MW, not 10MW

Connection dates and grid upgrades don't line up the way the industry assumes

Posted 7 September 2026 · View on LinkedIn

Most people I speak to take it as read that a connection date is set by when the local reinforcement finishes. It’s sort of a cross between industry folklore and an assumption. Read the full post

Over the past few weeks I've been tracking connections in detail, which is useful but doesn't reveal the underlying pattern. So I tested the folklore against published data (base: 14-08-26).

Ofgem's Accelerated Strategic Transmission Investment (ASTI) framework covers 26 transmission projects. Each one carries a delivery date written into the transmission owner's licence, in Appendix 1 of Special Condition 3.41. That date is the optimal delivery year plus twelve months, an allowance Ofgem added in 2022 for supply chain conditions, and Special Condition 3.41.5 then permits a further twelve months before the output counts as undelivered.

A full three years sit between the year the system was judged to need a reinforcement and the date a penalty starts running.

I matched those 26 projects to the NESO TEC Register. That gives 39 substations, 186 entries and 96 GW, roughly 14 per cent of the register by volume.

Then I compared every connection date to its own scheme's licence date.

Under 10 per cent land on the licence year. Another 4.5 per cent fall inside the twelve-month grace. The median entry sits two years beyond it. And 22 per cent are dated before the optimal delivery date, so ahead of infrastructure that on any published reading will not exist yet.

I ran it twice, once taking the earliest governing scheme and once the latest, because several substations depend on more than one upgrade. Waiting for the last upgrade made the fit slightly worse.

So, connection dates don't track the published regulatory schedule. Whether they track an engineering programme that the licence date only loosely follows, I can’t tell you from the published data.

Perhaps a high drop-out rate is simply expected to do the correcting?

If you're underwriting a project on its connection date, what do you think sits behind that date? Are you still relying on folklore and luck?

Connection dates against the ASTI programme, year by year, from the TEC Register snapshot of 14 August 2026. 42 seconds, no sound.

Seven projects, 4.2 GW, gone from the register in a week. I can't tell you why from the data.

Posted 1 September 2026 · View on LinkedIn

Between 14 and 25 August, seven projects left the TEC Register. All seven belonged to the Elmya RPC joint venture. Nothing else was withdrawn from the other 2,057 projects on it. Read the full post

I traced eleven companies at Companies House, from the project vehicles up to the top of the UK structure. Every PSC entry unchanged since formation. No charges registered anywhere in the chain. The top company files a statement that it knows of no registrable person, which is correct, because the shareholder above it is Canadian and not registrable here. Management of the portfolio moved to Octopus Energy Generation in March and was confirmed publicly in June. That was a management mandate. No shares moved, so the ownership record sat still through all of it.

Then I looked for what separated the seven from the four that stayed. Gate status doesn't do it, the withdrawn projects were more likely to carry a Gate 1 label than the survivors. Connection date doesn't either, the latest date in the whole group belongs to a project that stayed.

What I think decides it isn't published. Gate 2 wants an option that reflects the operational timeline. Reform pushed dates into the 2030s, so options agreed for two or three years need extending, and by then the landowner knows what the site is worth. Annual payments replace a single fee. One of the withdrawn projects, rated 900 MW, carries a cumulative loss of about £430,000, so relatively little had been spent on it.

If you are selling into developers, how can you tell a live pipeline from a parked one?

Graphic accompanying the post: Seven projects, 4.2 GW, gone from the register in a week. I can't tell you why from the data.

The TEC Register doesn't show connections reform yet, and NESO has now put in writing why.

Posted 28 August 2026 · View on LinkedIn

Last December NESO confirmed a reformed pipeline of 283 GW of generation and storage, and its results page describes clearing a queue that had grown past 700 GW. That same page tells readers to go to the TEC Register for the latest position on connections. So I went and looked, across four dated snapshots. Read the full post

Between 2 February 2024 and 14 August 2026 the register grew from 531 GW to 685 GW, and from 1,577 projects to 2,064. Narrow it to the reform window and it's starker. Of the 1,851 projects on the 28 January 2025 file, 1,791 are still there in August 2026. Sixty had gone, carrying 18.2 GW between them. Another 273 had joined.

The explanation arrived last week in a response to an environmental information request (NESO ref FOI/26/174). The register's Gate column is populated only once an agreement has been countersigned, and stays blank until then. On the 14 August file, 400 GW of the 685 GW is blank.

So the register is a contractual record with a countersignature lag in front of it. It was never built as a scorecard for the reform, and it doesn't work as one.

That matters in both directions. I keep seeing the queue total quoted as evidence the reform hasn't bitten, and it isn't evidence of that. It's evidence that contracts take a while to sign.

Where would you go instead for a project-level view of what reform actually removed?

Note added 27 September 2026: the blank-Gate figure on the 14 August 2026 file is 395.3 GW; the post rounds it to 400 GW.

Graphic accompanying the post: The TEC Register doesn't show connections reform yet, and NESO has now put in writing why.

Grid connection dates are sliding, and I can now show it project by project.

Posted 26 August 2026 · View on LinkedIn

I've been comparing four dated snapshots of the TEC Register: 2 February 2024, 2 July 2024, 28 January 2025 and 14 August 2026. The three older files came from NESO under an environmental information request (ref FOI/26/174), so the publication dates are confirmed rather than inferred from a filename. Read the full post

My first attempt at this grouped capacity by connection year across the snapshots. That analysis was wrong and I dropped it. Built projects have their effective date blanked on the current file, so the year bands were not comparing like with like, and the totals mixed genuine date changes with projects that had only just joined the register.

So I tracked individual project IDs instead.

On the February 2024 file, 326 projects carried a connection date of 2026 or earlier, 55.5 GW between them. Two and a half years on, here is where that cohort sits. 71 projects are built, 16.3 GW. 99 still carry a near-term date, 13.9 GW. 137 have moved to 2027 or later, 22.8 GW. Another 19 have left the register altogether, 2.6 GW.

So about 41 per cent of that near-term capacity has pushed back by at least a year, against 29 per cent delivered.

Some of that is capacity genuinely arriving later. Some will be projects parking a date they have no intention of meeting. The register doesn't distinguish between the two, and I don't think anything published does.

If you're modelling revenue against a connection date on a current TEC entry, how much weight are you putting on it?

Graphic accompanying the post: Grid connection dates are sliding, and I can now show it project by project.

I went looking for proof my own thesis was wrong

Posted 24 August 2026 · View on LinkedIn

The bramble thesis argues ownership in UK solar and battery storage keeps concentrating while the developer brand on the sign stays the same. I wanted to write the counter case this week, still-independent developers, proof that new capital keeps entering the market rather than just consolidating what's already there. Five names went on the list: Aura Power, Fidra Energy, Cero Generation, TBC Partners, Starlight. Read the full post

Checking them took longer than writing the post would have.

Aura Power merged with Verdant Energy on 6 August, backed by CVC DIF, creating a platform with around 1GW operating and under construction plus a 10GW pipeline. The Aura Power brand stays. Founder Simon Coulson stays too, on a minority stake.

Cero Generation was never independent in the window I was checking. Macquarie's Green Investment Group launched it back in 2021. Enso Energy, its UK development partner, is a separate company entirely.

Starlight is NextEnergy Group's in-house development arm. Earlier this year it sold a Scottish battery project to the NEUK I Fund, owned by NextEnergy Capital, inside the same group.

Fidra Energy hasn't changed hands, but it was EIG Global Energy Partners' platform from the outset, not really independent either.

One name survived the check. TBC Partners, a founder-owned family office, still trades as itself.

The post I meant to write doesn't exist. The one I found instead makes the case better than my original research did.

Name five independent developers in this market today. How many would survive the same check?

Graphic accompanying the post: I went looking for proof my own thesis was wrong

Flexible Hybrid Grid Connection: coming to a British data centre near you?

Posted 21 August 2026 · View on LinkedIn

Finland's grid operator has made mandatory what Britain is still consulting on. Read the full post

Since 12 June 2026, any demand facility over 30 MW connecting directly to Fingrid's transmission network must be able to cut its power intake by at least 30%, to at least three separate levels, within 15 minutes of being asked. Data centres, electrolysers, electric boilers, all in scope. The obligation bites now, running ahead of the full KJV2026 grid code, which is still out for consultation and closes this Friday.

A graphic doing the rounds on LinkedIn presents this as an integrated flexible load model: battery storage, on-site generation, heat reuse into district heating and energy market participation, all pictured as though each one carried the same status. Only the curtailment capability is actually mandated. Everything else in that picture is one commercial route to meeting it.

Worth being precise about what the obligation costs, because there is an easy misreading here. The requirement is measured on import at the connection point. A facility has to draw less, and it can do that by throttling compute, by discharging a battery to cover its own load, or by running backup plant it has probably already built for resilience. None of those needs an export-capable connection. The harder problem is the controls and telemetry layer that lets the system operator dispatch the reduction reliably inside fifteen minutes, which for a colocation operator is a tenant contract question as much as an engineering one.

Britain has already signposted the same direction.

Ofgem's live Curate consultation, closing 16 September, deals with queue viability and the proposed Data Centre Commitment Fee, and is silent on curtailment. The flexibility question sits in the Operate pillar, which Ofgem has flagged for consultation in autumn 2026 with a focus on flexibility requirements and operational control for large demand users.

If the UK does go in this direction what could it mean for your project?

Note added 27 September 2026: the KJV2026 consultation closed on 21 August 2026, and Ofgem’s Curate consultation closed on 17 September 2026.

Graphic accompanying the post: Flexible Hybrid Grid Connection: coming to a British data centre near you?

Harmony Energy is still Harmony Energy

Posted 19 August 2026 · View on LinkedIn

Two weeks ago you could work out who controlled Harmony Energy Limited in about ninety seconds. Companies House listed Kipp K Ltd, registered in Knaresborough. Kipp K's own entry listed Peter Kavanagh at more than 75%. Two clicks, one name. Read the full post

On Monday I asked where you'd actually look to find out who you're contracting with on a UK battery project. This is why I asked.

I checked again this morning. Harmony Energy Limited now carries a statement that there is no registrable person or registrable relevant legal entity. Nobody is named at all.

Nothing improper has happened here and I want to be clear about that. Alpiq completed its purchase of a 90% stake in July. Unlisted overseas companies can't go on a UK PSC register, so Alpiq can't be listed. The rules then say look through to the individuals behind it, and Kavanagh's retained 10% sits under the 25% threshold. The statement is the correct filing. It's the regime working as written.

Meanwhile the visible business is unchanged. Same name, same chief executive, same team, eighteen grid-scale projects delivered and around 400MW under construction. Second ownership change on that name in about a year, after Harmony Energy Income Trust was taken private by Foresight in June 2025 through a bidco split 51/49 with one of Foresight's own portfolio companies.

So the UK's beneficial ownership regime is built to find an individual, and it does, right up until control passes to a diffusely held overseas corporate. Then the trail ends lawfully and the register names no one.

What do you actually rely on, when the register says nobody?

(BtW, no disrepect to Harmony here, it was just one of a series of recent transactions that illustrated the point that I was trying to make.)

Graphic accompanying the post: Harmony Energy is still Harmony Energy

What is social licence and how does it affect my data centre project?

Posted 18 August 2026 · View on LinkedIn

Social licence is whether a community accepts what you're building, as distinct from whether you have planning permission. The term came out of mining in the late nineties and it's been studied properly ever since. Read the full post

I've read a lot of pieces this year telling data centre developers to engage earlier and be more transparent. The advice is sound. I haven't seen anyone measure it though.

The measurement exists in other sectors. Slovic wrote about trust and risk perception in 1993 and concluded that public distrust isn't ignorance, it's structural, and once trust goes it's very hard to get back. Moffat and Zhang modelled community acceptance across an Australian mining region in 2014, and found procedural fairness was the strongest single predictor of trust, and that the quality of contact with a community mattered while the quantity of it didn't. Devine-Wright surveyed 503 residents in the South West in 2013 about a proposed high voltage power line and found trust in the developer was one of the significant predictors of opposition.

None of that work has been pointed at data centres.

Over the past few months I've spent some time tracing the fifteen largest UK data centre schemes back to whoever ultimately controls them. It's Companies House, PSC filings and planning documents, and it takes a while. Several of the largest end up at privately held platforms that most people wouldn't be able to identify without doing the same work.

What I don't know yet is whether that matters.

If you can't find out who's building the thing at the end of your road, does the opposition get harder, and by how much. I've got the ownership data. I haven't settled the design.

So, two asks. If you've run a structural equation model on community acceptance, I'd like ten minutes of your time. And if you know where scheme level objection data sits, please tell me.

Graphic accompanying the post: What is social licence and how does it affect my data centre project?

Only eight days from commissioning to a new owner

Posted 17 August 2026 · View on LinkedIn

NextEnergy UK announced on 13 August that it had acquired a 107MW/151MWh battery portfolio, three sites in southeast England. Maldon, 40MW/40MWh, already operational. Loudwater at 39MW/55MWh and Basildon at 28MW/56MWh had both just completed commissioning, eight days before the deal was reported. Read the full post

All three carry Capacity Market contracts.

The seller was Eku Energy. Two weeks before that, Eku had bought a 300MW/600MWh project at Didcot from TBC Partners, its first transmission-connected battery in the UK, due to connect in 2028.

Same business, one fortnight, roughly 107MW of finished assets out and 300MW of consented development in. That's capital recycling, it's an ordinary infrastructure model and I'm not implying anything else about it. Eku is jointly owned by a Macquarie Asset Management fund and British Columbia Investment Management Corporation.

None of it shows up at project level, though. Maldon is still Maldon, same site, same CM contract, same planning file.

What moved was who collects the revenue.

If you were running counterparty checks on a UK battery site this month, where would you actually look to find out who you're dealing with?

Or more importantly, would you even think to look?

Graphic accompanying the post: Only eight days from commissioning to a new owner

When did you last check who actually owns your three biggest customers?

Posted 10 August 2026 · View on LinkedIn · Read the article

Three things change when your biggest customer gets sold. None of them send you an email. Read the full post

Your credit assessment was done against an owner who might not be there any more. Your parent company guarantee could be pointing at a parent that's left the structure. And signing authority moves, which you usually find out about when procurement centralises and your renewal turns into a tender.

Meanwhile your contact is the same person, the letterhead is the same, the pipeline is the same, and the name in your CRM hasn't moved.

I've spent three weeks tracing ownership on large UK energy projects through Companies House. Britain's largest consented battery has had three eras of control since January 2023, behind a company rename that's filed as an amendment to details. A 400MW solar scheme sold in February resolves one step up to a Dutch entity and then stops, which isn't evasion, it's just the design limit of the register.

Everything is filed. Almost none of it is readable at the point somebody needs to make a decision.

I also corrected something I got wrong in the first article of this series.

Full piece here.

Graphic accompanying the post: When did you last check who actually owns your three biggest customers?

The BESS sector is nine years old. Some job descriptions want more than that.

Posted 5 August 2026 · View on LinkedIn · Read the article

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

DESNZ dates the first major grid-scale battery storage project in the UK to 2017. The year before, National Grid awarded 201 MW of Enhanced Frequency Response across eight projects. So the sector is nine years old. The requirements I keep seeing sit at five to seven years, which sounds modest until you count backwards. Seven years from now means you were in it before August 2019, inside the first two and a half years. Five years means before August 2021.

Annual output from UK batteries went from 0.5 GWh in 2017 to almost 2,300 GWh in 2025. Nearly everyone working in British storage today arrived after a seven year filter closes. That filter is not selecting the best people in the sector, it is selecting the earliest ones, and those are different populations.

Data centres are not the same problem, and lumping them together is where most commentary on this goes wrong. That sector is decades old. Its constraint is throughput, not age. Uptime Institute found 46 per cent of operators struggling to find qualified candidates in 2025, with operations management overtaking junior roles as the biggest gap as experienced people retire, and about a quarter of leavers simply moving to a competitor.

Then there is the awkward bit. Van Iddekinge and colleagues meta-analysed 81 samples in Personnel Psychology and found prior experience correlates with job performance at 0.06, and with staying in the job at 0.00. Under half a per cent of variance explained. Harvard and Accenture asked 2,250 executives about it and 88 per cent agreed their own processes screen out qualified people for missing exact criteria.

So the years-in-sector line is filtering hard on something that predicts almost nothing, in sectors that cannot supply the years anyway.

If you have been ruled out of one of these roles, the filter that removed you was not measuring whether you could do the work. That is not me being kind, it is what the evidence says. Keep going.

If you write these specifications, what would you put in that line instead, and would your panel actually trust it?

More detail available in the linked article below.

Graphic accompanying the post: The BESS sector is nine years old. Some job descriptions want more than that.

If you sell into the grid-scale renewables industry, how many customers can you really target?

Posted 3 August 2026 · View on LinkedIn · Read the article

Up until the weekend before last, I thought the eventual result, for vendors to this industry, would be a market that defined itself in plain sight, consolidation arriving through deals you could read about, leaving a customer list you could still count. Then I spent a weekend in Companies House, and what I found became the piece I published last Monday, fifteen of Britain's biggest consented solar and battery projects traced to whoever ultimately controls them. Read the full post

This week, the thread that matters if you carry a revenue number. Picture a sales director building next year's plan. The CRM shows fifteen live prospects across those projects. Fifteen relationships, fifteen probability weightings, fifteen doors. Now lay the ownership map over it. Those fifteen projects resolve to roughly ten ultimate decision-makers, and three of them (Macquarie, CIP, EDF) hold multiple positions each. The prospect list is the same length it was yesterday. The customer count is not.

None of this is a forecast, it has already happened, it just happened below the visibility of the market maps everyone sells from. The early signs are there if you look. Sungrow has supplied BW's storage platform in Sweden, then at Bramley, then a 1.4GWh order at Hams Hall, the relationship travels with the platform rather than the project. Macquarie's pivot of Island Green Power from selling projects to keeping them turns a fifteen gigawatt pipeline into one owner's procurement function.

Other industries have walked this road before us. Groceries ended up with a statutory code and a regulator. Car making restructured its entire supply base into tiers. The full argument, including the honest counter-forces, is in the article below.

A question for anyone running revenue in this supply chain: have you ever re-cut your pipeline by ultimate owner rather than by project name, and if you did, what happened to the number?

Graphic accompanying the post: If you sell into the grid-scale renewables industry, how many customers can you really target?

Britain's biggest data centres aren't owned by the funds I expected.

Posted 29 July 2026 · View on LinkedIn

I went looking for the same handful of international infrastructure managers behind this week's piece on renewables, assuming they'd turn up behind the AI data centre boom too. Four of the top six schemes by capacity say otherwise. Two of the largest, Elsham (close to a gigawatt) and Humber Tech Park (384MW), look unrelated in every public document, different names, different websites, nothing linking them. Trace the ownership register and both come from the same promoter, Greystoke, and both end at the same holding company, a trust incorporated in 1947. A third scheme, 600MW in east London, traces to a completely different private platform. Together that's nearly 2,000MW of Britain's largest proposed data centre capacity, and none of it belongs to the names you'd guess. Read the full post

Blackstone does originate one of the biggest schemes outright, its campus in Northumberland. EDF is the more striking case. It already owns UK solar capacity, and it isn't quietly sitting on an overlap. It has signed a memorandum of understanding to build a 1 gigawatt data centre powered by new nuclear reactors, on the site of its own decommissioned coal station, in partnership with Tritax Big Box, a FTSE 250 property company, with the Energy Secretary's name on the announcement.

So it isn't five funds quietly owning both sides. It looks more like two separate markets, mostly obscure private ownership at the very top of one, and the rare fund that does move into power doing it in the open, at government level, rather than through a structure you'd need to trace.

Which is harder to keep an eye on: the pattern nobody's watching, or the strategy nobody's bothered to add up?

Graphic accompanying the post: Britain's biggest data centres aren't owned by the funds I expected.

Who actually owns Britain's energy transition?

Posted 27 July 2026 · View on LinkedIn · Read the article

Over the weekend I continued the long process of clearing my late father's garden. Due to the intense boredom of cutting up brambles, my thoughts wandered to the state of the renewables market and the vaguely economic subject of developer concentration. Read the full post

Everyone assumes concentration arrives the way it does elsewhere, developers buying developers, deals that make the trade press. Standing in the bramble patch it occurred to me that it does not need to. Nearly every big renewable project in Britain is its own company. So the market can concentrate one project at a time, quietly, and the number everyone watches, the count of independent developers, never moves.

Brambles look like dozens of separate plants. They are usually one plant, connected under the soil. You only find out by digging. So I spent the weekend digging, through Companies House, into fifteen of the largest consented solar and battery projects in Britain, about 8,800MW.

What I found: the named developer is materially misleading for at least ten of the fifteen. The largest battery project in the country, listed everywhere as a Scottish storage firm, traces through three group companies to Washington DC private equity (with, curiously, the UK's National Wealth Fund as a minority investor in a structure it does not control). And in case after case the sale follows within a year of planning consent. Consent is the trigger.

By capacity, around 20 percent of these projects remains in genuinely UK-independent hands. I have started calling it the bramble thesis. The full fifteen, the method, and the honest counter-examples are in the article linked below.

Whose job should it be to make this visible?

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

Four questions I ask before trusting a mechanism-backed revenue model.

Posted 24 July 2026 · View on LinkedIn

On Wednesday I argued that mechanism design decides winners in this market now, not technology. A fair challenge back is what testing a mechanism actually involves. This is the list I work through, all of it built from schemes I’ve covered this summer. Read the full post

First, where does the floor sit against the full cost stack. Not the headline opex number, the whole stack, cycling, degradation, mid-life refurbishment. Worth remembering on LDES that final cap and floor levels aren’t locked until the post-construction review from 2029, so anyone modelling today is modelling a range, not a number.

Second, what triggers the cap and when does it bite. A cap that activates in benign trading conditions takes away upside you thought you’d underwritten. The fine print matters more than the headline generosity.

Third, does qualifying status survive the rules as written, not as summarised. Thresholds and aggregation rules are where projects quietly fall out. A scheme that looks open at brochure level can close completely once you combine technologies on one connection, or sit behind the wrong grid supply point.

Fourth, what happens if the problem the mechanism prices gets fixed. Constraint-driven revenue assumes the constraints persist, and the transmission reinforcements due in the early 2030s are the expiry date on that assumption. If the base case needs the grid to stay broken, that wants saying out loud in the investment committee (it rarely is).

None of this is complicated, it’s just rarely done in sequence before the money moves.

For those of you on the capital side, which of these gets skipped most often, and has one of them actually burned you? I’m curious whether it’s the same one.

Four questions I ask before trusting a mechanism-backed revenue model: where the floor sits against the full cost stack; what triggers the cap and when it bites; whether qualifying status survives the rules as written; what happens if the problem the mechanism prices gets fixed.

Graphic added for the archive; the original post was text only.

Every post I've written this summer has secretly been the same post.

Posted 22 July 2026 · View on LinkedIn

Grid connections. Constraint payments. The LDES cap and floor. Transmission residual charges. On the surface, four different stories. Underneath, one story told four ways. Read the full post

Britain has decided what energy system it wants faster than it can build it. Everything interesting in the sector right now happens in that gap, and every mechanism I've written about is really an answer to the same question: while we wait for the wires, who pays?

Constraint payments are the gap priced annually, over £1bn a year, consumers paying wind to stop and gas to start in the same hour. The LDES scheme is an attempt to pay once for an asset that absorbs the gap instead (7,645MW provisionally selected, the floor structured as insurance rather than subsidy). The connections queue, cut from 700GW to a 283GW prioritised pipeline, turned grid access from a formality into a rationed asset with holding costs. And April's residual charge increase showed what happens when a charge designed around yesterday's behaviour meets customers who adapt.

The key points here are probably that the technology arguments are mostly settled and the capital is mostly available. What decides winners now is mechanism design, the fine print of who holds which risk when the assumptions drift.

So here's the question I'd put to both sides of the market. If you're deploying capital, are you underwriting the asset or the mechanism it sits inside? And if you're designing the mechanisms, which of your calibration assumptions would survive the market adapting to them?

Graphic accompanying the post: Every post I've written this summer has secretly been the same post.

Ofgem built a network charge that is very hard to avoid, so of course people are now working out how to avoid it

Posted 20 July 2026 · View on LinkedIn

From 1 April the transmission residual rose by around 60 percent, and most of it landed in fixed standing charges rather than unit rates. The whole point was that you could not dodge it by trimming consumption, since it follows your agreed capacity, not your meter. Read the full post

Here is the part I keep coming back to. There is an old idea in economics, Goodhart's law, that any measure you turn into a target stops measuring what it used to, because people start managing the number itself. Agreed capacity was a fair enough proxy for how much network you lean on, until the charge on it grew big enough to be worth optimising. So some sites now run forensic half-hourly analysis to show their real peak sits under their agreed capacity, then apply to drop an ASC band. Others are fitting behind-the-meter batteries purely to shave that peak so the band comes down anyway, which is really a bet that the charge keeps rising (NESO's own trajectory says it will), so the market is already pricing where this goes.

To be fair, not all of this is gaming. You cannot drop a band on paper, you have to actually release the capacity or you breach and pay for it, so some of what is happening is over-declared headroom being handed back, which is efficient and probably overdue.

From the regulator's point of view, the bands were set around the capacity that the firms had already booked, on the quiet assumption those bookings would stay put. But make the charge big enough and people rebook, so the ground the whole thing stands on starts to move.

So which is it. A charge doing its job, finally flushing out capacity that was never really needed. Or a charge sold as unavoidable, slowly proving it was nothing of the sort. I would love to know which way the people who set the bands would call it.

Graphic accompanying the post: Ofgem built a network charge that is very hard to avoid, so of course people are now working out how to avoid it

The connections queue is no longer a waiting list, it is a market structure.

Posted 17 July 2026 · View on LinkedIn

Burges Salmon published a piece this week on how Gate 2 uncertainty is reshaping energy M&A. The transaction mechanics they describe (split exchange and completion conditional on the Gate 2 offer, deferred consideration tied to grid milestones) will be familiar to anyone who has worked development-stage deals. Markets always structure around uncertainty, that part is textbook. Read the full post

The structural shift underneath is the more interesting story.

Connections reform did three things at once. It rationed grid access, cutting over 700GW of applications down to a prioritised pipeline of 283GW of generation and storage. It made new entry uncertain, NESO has removed the target date for the next application window and signalled it will be limited by technology and location. And it attached holding costs to the positions that survived, with progression milestones and non-delivery charges in force since January.

Put plainly, a Gate 2 position is now a scarce, dated asset that must be actively defended with capital and delivery capability. Assets with those characteristics do not stay evenly distributed for long.

The capital flows already point this way. Global energy transition deal volumes fell around 15% last year while deal value rose over 20%, fewer and bigger bets. Infrastructure funds increasingly want multi-technology platforms rather than single projects. And last month Drax agreed to acquire Bluefield, taking a 2.9GW development pipeline into a listed generator alongside 0.9GW of operating assets. The US saw the same pattern first, where approved interconnection positions became the currency driving platform consolidation.

My working view (held loosely, the developer distress data is not in yet) is that the Gate 2 process will be remembered less for the delays and more as the moment UK development stopped being a cottage industry.

If grid position is the scarce asset, who is best placed to hold it, the developers who originated it or the platforms consolidating around it?

Graphic accompanying the post: The connections queue is no longer a waiting list, it is a market structure.

The economics of Ofgem's storage scheme improve every time a transmission project runs late, and that deserves more attention than it's getting

Posted 10 July 2026 · View on LinkedIn

While Scottish constraints persist, the scheme works as designed. Batteries earn constraint-driven spreads, floors stay dormant, and consumers save on constraint payments they'd otherwise be funding. Read the full post

Now run the other branch. The Eastern Green Link projects and wider B6 reinforcement are due in the early 2030s, around the time several of the shortlisted batteries commission. If the wires arrive on schedule, spreads compress, merchant revenues fall, and floors start binding, leaving consumers funding the reinforcement and topping up floors on assets whose revenue case has eroded. Awarding 7,645MW, the very top of Ofgem's guided 2.7 to 7.7GW range, makes that exposure larger rather than smaller.

Ofgem's implicit defence is that major transmission projects are rarely on time and the constraint need will outlast the plan. History supports them. But notice what that means: the scheme's economics quietly depend on the underperformance of a parallel programme the same regulator oversees.

Nobody outside the confidential bid data can check how floor levels compare with projected constraint savings. The arithmetic may well work. We're simply not shown it.

Should a consumer-funded programme of this size publish the numbers that justify it?

Graphic accompanying the post: The economics of Ofgem's storage scheme improve every time a transmission project runs late, and that deserves more attention than it's getting

Ofgem's new storage floor may never pay out, which is the point most of the commentary has missed

Posted 8 July 2026 · View on LinkedIn

The LDES cap and floor scheme is being read as a subsidy. The structure says otherwise. Read the full post

A battery sited behind a Scottish constraint boundary charges when wind would otherwise be curtailed, often at negative prices, and discharges when the boundary frees up or demand rises. That spread is market revenue, not support payment. If it holds, the asset earns above its floor, the floor pays out nothing, and the cap claws back anything beyond a fair return on invested capital. Consumers only pay in years the asset underperforms.

That's insurance rather than subsidy, and it isn't a novel idea either. It's the same model that financed Britain's interconnectors, applied to storage.

It also explains the durations on the 26 June minded-to list, which run from 8 to 22 hours against a merchant market that stops paying at around 4. Constraint events last days rather than minutes, and hours nine through eighteen only become financeable once somebody underwrites them. The floor is that somebody.

The whole structure rests on one assumption, though: that the constraints persist. Friday's post covers what happens if they don't.

Graphic accompanying the post: Ofgem's new storage floor may never pay out, which is the point most of the commentary has missed

Connecting is about to get easier. That is the problem.

Posted 7 July 2026 · View on LinkedIn

Last Thursday I asked whether undifferentiated BESS buildout in Britain was quietly drawing to a close. The best pushback in the thread argued the opposite, so it is worth testing properly. Read the full post

The argument runs like this. Connection studies have treated batteries as firm at system peak in both directions. A worst case, and one that inflated reinforcement assumptions and clogged the queue. NESO has conceded the point. Its connections reform material commits to remodelling storage to reflect how batteries actually run, across transmission and distribution. Model them realistically and the network has more room than the queue suggests. On this reading, the oversupply is partly an artefact of bad assumptions.

The modelling point is right. The conclusion does not hold.

Fixing the modelling lowers connection costs. Lower costs make more marginal projects viable, not fewer. The fix feeds the queue it was meant to relieve.

And the oversupply was never measured in dispatch assumptions. It is measured in capacity. NESO expects to award connection agreements to over 80GW of battery storage by 2035, roughly three times what Clean Power 2030 implies the system needs. Remodel every battery in the queue and that ratio does not move.

So the answer to Thursday's question is yes, but not at the point most people are watching. Connecting is about to get easier, not harder. What gets harder is the revenue case. When 80GW can connect and the plan has room for a third of it, a grid offer stops being the scarce asset. The scarce asset is a problem the system is paying to solve. The right duration, in the right place.

Which puts a question to anyone opening a Gate 2 offer this summer. Is that connection agreement the thing you queued two years for, or the start of competing on price with everyone else who got one?

Graphic accompanying the post: Connecting is about to get easier. That is the problem.

When wind behind the B6 boundary exceeds what the wires can carry, we pay twice for the same hour of electricity.

Posted 6 July 2026 · View on LinkedIn

NESO pays the wind farm to switch off, then pays a generator south of the constraint, usually gas, to produce the energy the wind farm didn't. Consumers fund both payments through their bills, and neither buys a single additional clean unit. Read the full post

Constraint costs passed £1bn a year some time ago, and NESO's own projections have them climbing until the major transmission reinforcements land in the early 2030s. The wind is in Scotland, the demand mostly isn't, and the wires between them are full. Building generation faster than transmission has this as its mechanical consequence, so there's no scandal here, just arithmetic that worsens with every year the reinforcements slip.

The question worth asking is whether you can stop paying annually for the problem and instead pay once for an asset that absorbs it.

Ofgem published its answer on Friday 26 June. That's Wednesday's post.

Graphic accompanying the post: When wind behind the B6 boundary exceeds what the wires can carry, we pay twice for the same hour of electricity.

Ofgem spent two years clearing the battery queue. Now it says there are too many batteries in it

Posted 2 July 2026 · View on LinkedIn

In its own connections-reform commentary, the regulator admits many more battery projects will receive offers than Clean Power 2030 needs, and it is weighing measures to push oversupplied projects out of the queue early. Read the full post

Hold that next to Friday, where the same regulator provisionally underwrote 7.65GW of long-duration storage under cap and floor, lithium-ion included, with a consumer-backed floor.

These are not contradictory. They are one signal at higher resolution. The system does not need more storage in the abstract. It needs a particular shape: the right duration, in the right place. Short-duration batteries crowded into already-deep parts of the network have tipped into surplus. Long-duration capacity sited where wind sits stranded behind constraints is scarce enough that Ofgem will pay to guarantee it.

Most of that pumped hydro sits in northern Scotland, exactly where the surplus wind is and the grid cannot yet carry it south. That placement is the whole point.

So the question shifts. Not whether a battery connects, but whether it solves a problem the system is paying to solve, where it is paying to solve it.

Is undifferentiated BESS buildout in Britain quietly drawing to a close?

Graphic accompanying the post: Ofgem spent two years clearing the battery queue. Now it says there are too many batteries in it