I am standing almost next to the entrance to the old steelworks, just a few hundred yards away from a proposed data centre location.

I am home for a few days, Stoke on Trent, The Potteries, previous home of Shelton Bar steel works and at some point, 90 per cent of the world’s production of table wear, porcelain and pottery. At its height the city had 2,000 bottle kilns, a Michelin Tyre factory, several other large energy users and an enormous bar steel works. The steel works were there because Stoke on Trent had 5 coal mines to feed it and around 10,000 people worked there at its height and as my father would say when the furnace was open a red glow would cast light over the whole night skyline. The plant closed the main works in 1978 and the rolling mill in 2000, 160 years of steel making on one site.

Two miles from Blyth, on the Northumberland coast, an aluminium smelter drew about 2,500 GWh of electricity a year for forty years. Blyth’s 19,224 households drew just 49.7 GWh in 2024 on DESNZ’s meter-level figures, less than 2 percent of the aluminum plant. The smelter used the domestic electricity of 50 Blyths, in plain sight of Blyth, and the only day it made the national news was 6 March 2012, when Rio Tinto said it was closing.

Blackstone is now spending £10bn on a data centre campus at Cambois, on the same stretch of coast, with a 720 MW substation and a full build-out due by 2035. When it is running it will draw something between 75 and 90 Blyths and that number is being called unprecedented in planning objections and in the press, and it is roughly what the smelter’s neighbours lived beside without comment from 1972 to 2012.

Figure 1: “50 Blyths”. The real outline of Blyth’s built-up area, tiled once per 49.7 GWh, until the smelter’s annual electricity is filled. Then the same for Cambois in the lead accent. One shape is the town for a year.

The industrial history of Britain is a series of loads arriving, running, and leaving. The loads have not really changed much; they have shifted technology but the size is similar. A generation has grown up without one in the next town, and has lost the sense of scale that comes from living beside a Works. When a data centre developer says a campus will use as much power as a city, the reaction is disbelief. The correct reaction is recognition.

Shotton village has hosted a steelworks, a paper mill and a gap long on one site but the grid connection still lives.

British Steel closed the hot end at Shotton on 31 March 1980. 2 blast furnaces and 12 open hearth furnaces stopped, 6,500 people went, and it is still the largest single-day industrial redundancy in Western Europe. The works had made 1,542,000 ingot tons in its best year, 1965, and was running below that by the late 1970s. An open hearth integrated works of that vintage used something near 30 GJ of energy for every tonne of crude steel, almost all of it coal, coke and oil, with electricity at perhaps 400 to 600 kWh a tonne. Reconstructed from output, the hot end consumed in the region of 10,000 GWh of total energy a year and drew about 550 GWh of electricity. The households of Shotton and Connah’s Quay together used 39.7 GWh of electricity in 2024. In electricity alone, the furnaces were 14 Deesides. In total energy, they were 250.

Five years after the furnaces went cold, a Finnish consortium built a newsprint mill on the reclaimed land next door. Shotton Paper opened in 1985, ran 2 machines, and by the end was making 260,000 tonnes of newsprint a year from recycled fibre on UPM’s figures. A deinked newsprint mill draws around 1 MWh of electricity per tonne, so the mill pulled between 250 and 350 GWh a year from the grid and its own generation, roughly the same electricity as the furnaces it replaced. It employed 190 people when UPM sold it in 2021, against the 6,500 the steelworks had lost. Nobody in Flintshire marched against the paper mill. It was a mill. Mills use power. That is what they are for.

UPM stopped the newsprint machines on 30 September 2021 and handed the site to Eren, a Turkish packaging group, which is converting it to containerboard. The site’s biomass combined heat and power plant kept running throughout, because the Welsh Government’s own energy report lists it as one of the two largest in Wales. The land has now carried an open hearth steelworks, a newsprint mill, a demolition, and a packaging mill under construction, and the grid connection that fed the melting shop in 1979 has outlasted every one of them.

Figure 2: “One site, four loads”. Shotton 1960 to 2030, each offtaker as a band scaled to annual electricity, the Deeside households as a hairline underneath, the 2021 to 2024 gap in white.

The whole country did the same thing, on DUKES’s own numbers

Industrial energy consumption in the United Kingdom was 62.3 Mtoe in 1970 and 17.3 Mtoe in 2025 on DUKES 2026, a fall of 72 per cent.

That headline flatters the electricity story. Most of what went was coal, coke and fuel oil burned in furnaces and boilers. Industrial electricity was 73 TWh in 1970, rose to 116 TWh in 2005, and is now 76.7 TWh, the lowest since 1983. The sectors that carried it are named in the tables. Between 2005 and 2025 iron and steel went from 5.0 TWh of electricity to 1.7 TWh, non-ferrous metals from 7.7 TWh to 3.8 TWh, paper and printing from 13.2 TWh to 8.0 TWh, chemicals from 21.1 TWh to 13.3 TWh.

Set that against the data centres. DESNZ matched every data centre it could find to its meter for the first time this year and found 4.5 TWh of consumption in Great Britain in 2024, 2 per cent of grid demand, three quarters of it in London and the South East. The paper industry alone shed more electricity than that in the last twenty years. Non-ferrous metals used more than that in 2005 on their own. Every data centre in the country, added together, draws less than the smelters and the paper machines Britain stopped running while nobody was counting.

Figure 3: “What left and what arrived”. Small multiples, one per DUKES sub-sector, 2005 to 2025 electricity, with data centres 2020 to 2024 on the same scale.

The queue tells a different story about the future, and it should be read with the same scepticism as any connection queue. NESO held around 173 grid connection requests for new data centres in May 2026, adding to more than 100 GW, on the Senedd research service’s count, and Ofgem has said in terms that there are too many of them to connect. Most of that will never be energised. The projects that do get built will be the Cambois size, and there will be a few dozen of them, and each one will arrive in a town that has forgotten what a load is.

Stoke forgot faster than anywhere

My home town of Stoke-on-Trent is the case where the forgetting is most complete because the industry that left was everywhere and made of gas. The potbanks ran on coal until the Clean Air Act of 1956 emptied the bottle ovens, and then on gas-fired kilns for the forty years that followed. Wedgwood at Barlaston, Royal Doulton on Nile Street and a hundred smaller works burned gas at temperatures above 1,200°C, day and night, in the middle of residential streets. Shelton Bar, the city’s steelworks, blew out its furnaces in 1978 and closed the rolling mill in 2000. DUKES puts the mineral products sector, which is where ceramics sit, at 2.9 Mtoe in 2005 and 2.2 Mtoe in 2025, and the bulk of the pottery closures came before that series starts. The city’s 112,162 households now draw 321.6 GWh of electricity a year. A single potbank of the 1960s would not have been remarkable in that company, and there were dozens.

The objections that survive are jobs and where the money goes

Two objections to a data centre survive contact with the record. Neither concerns the amount of power.

The first is jobs. A 1970s comparison is unfair, because the steelworks that employed 6,500 would employ a fraction of that if it were built today, and the honest comparison is with a modern works. Even so, the gap is real. Northumberland County Council’s own framing of Cambois in April 2024 was 1,600 direct jobs, of which 1,200 are construction, leaving something near 400 permanent posts on an average draw of about 470 MW. Lynemouth had 515 on about 290 MW. Shotton Paper had 190 on about 35 MW. Per megawatt the data centre is the thinnest employer Britain has ever hosted, by a factor of 2 against the smelter and 6 against the mill. That is a fair thing to say out loud, and it is a separate argument from the one about electricity.

The second is where the value lands. The smelter sold aluminium to the world and paid Rio Tinto’s shareholders, and it also paid 515 wages into Ashington and Blyth, paid business rates, and bought its power from a station that employed another 130. The paper mill did the same for Deeside. A hyperscale campus buys electricity in Northumberland and sells compute in Virginia, and the margin between the two settles in Delaware. The load is local. The product leaves. Norway’s oil had a foreign customer too, and nobody in Stavanger minded, because the state took 78 per cent of the margin at the well and the payroll stayed in Rogaland. Britain takes rates on the roof. So the question a planning committee should be asking is the one it never asks. If the electricity is normal, and it is, what does the town get for hosting it, and should something leave with the data that stays behind, a levy on the compute sold from British ground or a rate on the connection that cannot be moved to Delaware? That is a question about capture, and it is worth an argument. The megawatts are not.

Those are the arguments worth having, and they can be had in planning committees and in Parliament with the numbers on the table.

The argument that the campus uses too much power cannot, because the power is the same power the country always used, in the same places, on the same wires, and the only thing that has changed is who remembers, we skipped 2 generations and they need reminding.

What the next paper mill looks like

The developer stands on 100 hectares of flat ground at Cambois, on top of a colliery, next to the substation that was built for a battery factory that never opened, and says the campus will need as much power as a city. The people listening think they have never heard anything like it. Their grandparents heard it in 1972, when Alcan said the smelter would need its own power station, and their great-grandparents heard it in 1958 when the first turbine at Blyth turned. It was always true. It was never news.

Whether the data centre is a good use of that power is a question about jobs and about where the money goes, and it deserves a straight answer. Whether the power itself is unusual is not a question at all. Britain has forgotten what industry looks like, and a data centre is what it looks like when it comes back.

Subscribe now

A note on the numbers. Household electricity is DESNZ’s domestic consumption by Middle Layer Super Output Area for 2024, published 18 December 2025, aggregated to the ONS built-up areas for Blyth, Shotton and Connah’s Quay, and Stoke-on-Trent through the ONS best-fit lookup, so each town total is within about 10 per cent. National series are DUKES 2026, tables 1.1 and 1.1.5, energy supplied basis, with a methodology break in 1994. The data centre figure is DESNZ’s Energy Trends special feature of June 2026. Lynemouth’s electricity is reconstructed from Alfed’s 180,000 tonne nameplate at 13 to 15 MWh a tonne. Shotton’s steelworks load is reconstructed from late 1970s output at typical open hearth intensities and is the softest figure in the piece. Shotton Paper’s load is reconstructed from UPM’s 260,000 tonne capacity at deinked newsprint intensities. Cambois is stated on its 720 MW substation at 60 to 70 per cent steady draw once complete, about 4,100 GWh a year. Average draws in the jobs comparison are annual energy divided by 8,760 hours. Every reconstructed figure is labelled as such on the charts. Rio Tinto’s own metered figure for Lynemouth and the Iron and Steel Statistics Bureau’s returns for Shotton will replace the reconstructions if they can be obtained before publication.

Subscribe now

The two Balancing Mechanism units are really just 0.9 MW apart on the register but they spent the twelve months to August 2026 doing completely different jobs. NESO accepted 2,211 MWh of upward movement and 68,816 MWh of downward movement at Blackhillock in Moray, a ratio of 31.1 to 1. At Clay Tye Farm in Essex it accepted 24,710 MWh up and 25,058 MWh down, a ratio of 1.01 to 1. The registered capacities are 50.4 MW and 49.5 MW. Where they are plugged in impacts the rest and it shows why the Scottish wind industry has this as a high priority.

Battery storage gets discussed as though megawatts are the asset. A developer has 50 MW. Another developer has 50 MW. Both have roughly two hours of storage. Apply a revenue curve, a degradation assumption and a cycling profile, and the two lines in the model converge. However, the British electricity system has other ideas.

The reason is pragmatic and physical and it’s because wires have a limit. Britain has obviously built its wind where the wind is, which is largely Scotland, and its demand is where the people are, which is mostly definitely not Scotland. The result, when more power tries to move south than the transmission system can carry, the system operator has to fix it in the last hour before delivery, paying somebody to generate less at one end and somebody else to generate more at the other. Batteries are among the things it pays.

Convert the accepted volumes into hours at full registered power, which is accepted MWh divided by registered MW, and the gap gets easier to hold in the head. Blackhillock took 44 hours of upward instruction against 1,365 hours of downward. Clay Tye took 499 and 506. Two units of the same size, and the grid only ever wants to push one of them in one direction.

What the we are counting on each unit

The Balancing Mechanism is the market NESO uses in the last hour before real time to keep supply and demand matched and to keep power flows inside what the wires can physically carry. Generators and batteries tell the system what they intend to do. NESO then pays them to move away from that intention. Simply, a movement up puts more power into the grid. A movement down puts in less.

For a battery, up means discharging harder or charging less than planned. Down means charging harder or discharging less. The industry calls the two directions offers and bids.

That word “up” is where readers go wrong, and the tooling does not help. The VitreousLabs panels these figures come from label the two sides “paid to put power in” and “cost of taking it out”. Power in there means power into the grid, so a battery discharging. Read it as charge going into the battery and the whole comparison inverts.

The second thing to hold on to is the baseline. Every accepted volume is measured against the Final Physical Notification, the position each unit told the system it would run at before real time. A downward acceptance of 40 MWh puts a unit 40 MWh below wherever it said it was going. Whether that came from charging out of idle or from discharging less than intended depends on the plan it started with. The physical charge and discharge cycle sits behind these figures and never appears in them.

So the numbers describe how NESO used each unit against its own declared position. Total throughput is a separate question, and these panels do not answer it.

Blackhillock sits behind Britain’s biggest bottleneck

If we look at where the wires run and Blackhillock stops being surprising.

Scotland has built a great deal of wind a long way from the places that use electricity. The power wants to travel south. The transmission system cannot always carry it. NESO divides the Scottish network into boundaries and publishes what each one can move. The B2 boundary has a base capability of around 2.8 GW. B4 has around 4 GW. NESO expects the generation arriving north of both to keep pushing required transfers above those figures.

When the wires fill up, something has to give. NESO reported that 13% of hypothetical British wind production was curtailed in 2024/25, with Scottish network congestion a major cause and planned transmission outages making it worse. The work has not stopped since. NESO’s Summer Outlook describes significant outages across B2 and B4 while the East Coast Upgrade, the Tealing to Kincardine reinforcement and other programmes proceed.

Curtailment means paying a wind farm to stop generating. It costs money and it looks bad. The cheaper answer, where one exists, is to find something north of the bottleneck that will consume the surplus instead.

A large battery does that on instruction.

Zenobē sited Blackhillock between Inverness and Aberdeen to relieve congestion associated with Viking, Moray East and Beatrice, and phase one entered operation in 2025 as a transmission connected asset. Set that against 68,816 MWh of accepted downward movement and 1,365 hours at full registered power, and the shape of the year reads straight.

One caution. The data cannot prove the reason behind any single acceptance. What it shows is a unit whose accepted profile matches its stated purpose and its position on the network, in a year when the system operator documented severe constraints in that exact part of Scotland. That is an inference. It is a strong one, and it is still an inference.

There is also a catch built into the revenue. Blackhillock’s downward volume exists because the wires north of it are full. They are being reinforced. The East Coast Upgrade and the Tealing to Kincardine works are designed to move more power south, and every new circuit cuts the number of hours the boundary binds. An asset whose profile depends on a constraint is underwriting a problem with a funded solution and a delivery date attached. That is a different risk from the one a merchant battery in Essex carries, and it should carry a different discount rate.

Clay Tye works in a very different system

Essex gives a different answer.

Clay Tye Farm is a 99 MW, 198 MWh scheme built from 52 Tesla Megapacks, connected to UK Power Networks’ distribution system and optimised through Tesla’s Autobidder. E_CLAYB-2 is one of its two Balancing Mechanism units, and it sits a few miles from one of the largest concentrations of electricity demand in Europe.

Its accepted movements come out close to symmetrical. 24,710 MWh up, 25,058 MWh down, a difference of 1.4%. Sometimes the system wants more energy from it and sometimes less. Over a year in the south east those calls roughly cancel.

The instruction pattern differs too. Clay Tye took 45,946 acceptances against Blackhillock’s 23,187 while moving less total energy. The average acceptance at Clay Tye was 1.08 MWh against 3.06 MWh at Blackhillock. One unit gets tapped constantly in small amounts to trim national imbalance. The other gets leaned on hard in one direction to hold a bottleneck open.

The cash line is a trap

Blackhillock’s Balancing Mechanism cash looks alarming. £261k received on upward movement, £3.7m paid on downward movement, and a net of £3.4m negative. Clay Tye shows £2.7m in, £1.8m out and £926k positive.

Anyone reading those two columns as a scoreboard will reach the wrong conclusion.

The £3.7m is a purchase. Blackhillock bought 68,816 MWh through the mechanism at an average of £54/MWh and sold 2,211 MWh back at £118/MWh. The other 66,605 MWh left the site by routes this view does not cover, priced in markets it does not show. A negative Balancing Mechanism line on a unit that buys nearly all its energy there and sells almost none of it back is what the arithmetic demands.

The price gap is the more interesting number. Clay Tye’s average charging price through the mechanism was £72/MWh against Blackhillock’s £54/MWh, a third higher. That is the constraint showing up in cash. Surplus energy north of a full boundary is cheap because it has nowhere else to go.

One more tile on those panels deserves a warning. The figure reading £55.37 per kW per year at Clay Tye and £5.20 at Blackhillock is gross upward revenue divided by registered capacity. It counts one side of one market. Set it beside a published BESS revenue index, which stacks wholesale trading, ancillary services, the Capacity Market and balancing together, and it will read as a collapse in earnings at Blackhillock that has not happened. That is the number most likely to travel without its definition attached.

Blackhillock also earns well outside energy balancing. Its grid forming inverters supply inertia, short circuit level and reactive power, the stabilising properties that used to arrive free with large spinning power stations, and it holds a Stability Pathfinder contract for that purpose. Zenobē has named EDF and Kraken in its optimisation and service stacking.

None of which makes this a controlled experiment. Different connection voltage, different optimiser, different contracts, different technology. Geography is one variable among several. The honest claim is that it is the one doing the most visible work.

Duration makes location matter more

A one hour battery could once be analysed almost entirely through service prices. It sold frequency response, the market saturated, the price fell, and where the asset stood barely entered the conversation.

Two, four and six hour assets pose a different question. A long duration battery earns by moving energy across the day, and the value of moving energy across the day depends on where the cheap hours and the expensive hours sit. That is a question about the network as much as about the clock. Behind a constrained boundary the cheap hours are cheaper and there are more of them. In front of a demand centre the expensive hours are dearer.

The two effects do not occur in the same place. Britain’s constraint is in the north and its demand is in the south, and a battery cannot be in both.

That is the trade a developer makes when they pick a site. Very little in a standard revenue model shows it.

Anyone who listens knows my mantra, location is the critical piece.

Britain has one wholesale electricity price and many electricity networks.

The wires have weak points, export regions, import regions, voltage problems and places where new generation is arriving faster than reinforcement. The Balancing Mechanism is where those differences surface as instructions and as cash. A national revenue forecast averages them away. A national utilisation rate averages them away. A national cycling benchmark averages them away.

NESO has put a figure on the value at stake. Siting an additional 5 GW of flexible demand behind Scottish transmission bottlenecks could save consumers as much as £5bn on its own analysis. That is not a rounding item.

It also changes what a connection offer is. The industry treats one as permission to inject and withdraw power at an agreed capacity. What it confers is access to a particular piece of electrical geography, and the system operator has already worked out which pieces it needs.

The questions that follow belong in investment committees next to capex and augmentation. Where will surplus energy appear. Which boundary contains it. How often does that boundary bind. How much more renewable capacity is coming behind it. When does reinforcement arrive, and what happens to the constraint when it does. What else is competing for the same job.

That last pair matters most for anyone underwriting a Blackhillock-shaped asset. A constraint is a revenue source and a problem the system is trying to remove.

The queue is really measured in the wrong unit 50MWs do different things

Britain describes its future energy system almost entirely in megawatts. So many gigawatts of batteries, of offshore wind, of solar. The connection queue is ordered the same way.

50 MW at one node and 50 MW at another are different products. A battery on the wrong side of a bottleneck cannot relieve a constraint on the other side of it, whatever the register says. The connection point carries information about how an asset will behave years before it energises, and that information sits underneath a queue date and a headline capacity where nobody has to look at it.

The industry has got very good at asking how much storage Britain needs. Where is the harder question, and the more valuable one.

Blackhillock and Clay Tye have already answered it. The register says 50 MW twice. The Balancing Mechanism spent a year disagreeing.

Subscribe now

Sola Strand

The hotel sits on a beach ten minutes from Stavanger airport, and it is full of Norwegian pensioners on coach trips. It is not meant to be a fancy place, its standard Norwegian fayre.

The building is immaculate. The food is as good or better than you will find in a European capital. The spa runs the length of the ground floor and out onto a terrace facing the white sand beaches and the North Sea. Outside, coaches unload people in their late sixties and seventies who have come for four days of sea air and tasting menus, and who treat the whole thing as unremarkable.

A British pensioner on a coach trip gets a carvery in Torquay at best, its terrible food and tiny rooms in a poor hotel in Llandudno. That comparison does the work no chart can. Nobody had to explain to me what Norwegian oil money looks like at the end of a working life, because it was checking in around me with wheeled suitcases.

These are the oil generation, precisely. Anyone who is 72 today was 21 in 1975, four years after Ekofisk started producing and four years before Statfjord. They entered the labour market as the state began collecting, and they have spent an entire career inside a fiscal settlement funded by a resource most of them never saw as it was miles and miles off the coast in the North Sea. A spa hotel on a beach is what the harvest looks like once the harvesting is done.

The woman checking them in is about twenty. She will retire around 2071. Whether she gets her own Sola Strand experience is the only question this piece is about.

Then there is the walk. From the hotel to the airport is a little under 45 minutes on foot and the path is perfect. Newly laid tarmac. Cut grass to the edges. Signage that has been designed rather than procured. Cycle lanes with their own kerbs. The airport car park is a wall of Teslas, ID models and Polestars, because 95.9% of new cars sold in Norway last year were fully electric on OFV’s count, which the country could afford because it never had to ask what the subsidy cost.

All of it is paid for. That is the point of the walk, and it is the least examined fact in Norwegian public life.

Oil is sadly the perfect industry and here passes five tests, nothing else passes more than two, let me explain

Most countries that get rich get rich on something good. Norway got rich on something perfect, when you go looking for a successor that perfect scenario starts to matter. Five properties have to be held at once.

Margin. Once a field is developed the cash margins are indecent by the standards of any normal business. Johan Castberg cost NOK 86bn in 2024 prices and paid that back, on Equinor’s own account at first oil on 31 March 2025, in under two years against a thirty year field life. Johan Sverdrup phase one came in at NOK 83bn, NOK 40bn below the plan of development estimate and two months early, with a full field breakeven Equinor put below $20/bbl at first oil and now puts under $15, and an expected lifetime return to the Norwegian state of more than NOK 900bn. Spend NOK 200m on an exploration well, and if you hit, the arithmetic stops being a business case and becomes a windfall.

Duration. Fifty five years and counting from Ekofisk. Production peaked in 2024 at 240.7 million standard cubic metres of oil equivalent and the Norwegian Offshore Directorate expects it to hold roughly flat through 2026 before declining. Most booms give you a decade. Spice routes, gold rushes, textile towns and tech runs were either short, brutally extractive on the domestic population, or never large enough relative to the state to fund an entire welfare model. Norwegian oil has now paid for four working lifetimes.

Invisibility. It is all offshore. There is no tannery at the end of the street, no slag heap, no cracking tower on the skyline of the fourth largest city. The externalities either sit under 300 metres of water or leave on a tanker and are burned in somebody else’s air. Norway collected petrostate revenues and kept postcard scenery.

Fiscal capture. The state takes 78% at the margin: 22% ordinary company tax, plus a special tax with a formal rate of 71.8% levied on a base net of ordinary tax. On top sits the State’s Direct Financial Interest, which returned NOK 243bn to the treasury in 2025 on Petoro’s figures, and a 67% holding in Equinor that paid the state NOK 34.7bn in dividends the same year. The industry does not resent any of it, because the terms were set in the 1970s before anybody had got rich. That sequencing cannot be copied retrospectively, and it is why Norway has a fund and Britain has a memory.

The customer is abroad. This is the property nobody mentions and the one that kills every proposed successor. Oil imported the world’s money into Norway rather than moving Norwegian money around inside it. Offshore wind, absent export cables, would sell power to Norwegian citizens (does sell to UK citizens in the UK) and Norwegian industry, which redistributes domestic wealth rather than harvesting foreign wealth. A wind farm returning 10% and an oil field returning 10% are not the same object in the national accounts, because in one case the payer is a refinery customer in Rotterdam and in the other it is a paper mill in Trøndelag and the woman on reception at Sola Strand.

Norway has also closed the export route on itself. NorthConnect, the 1.4 GW link to Scotland, was not shelved. The Ministry of Petroleum and Energy refused the concession on 16 March 2023 and the King in Council upheld the refusal on 10 November 2023, having conceded the project was socio-economically profitable and rejected it anyway. On 3 February 2025, after the Centre Party walked out of the coalition, Jonas Gahr Støre and Terje Aasland committed to approving no new interconnectors before at least 2029 and told Statnett not to plan any. Norwegian offshore wind is not going to be an export industry in this parliamentary term or the next.

Nothing else passes more than two tests. Offshore wind passes duration and invisibility. Carbon capture and storage passes invisibility and fiscal capture, and fails margin so badly the state has to pay for it. Hydrogen passes nothing yet. Seabed minerals might pass margin and fiscal capture and would fail invisibility loudly enough to stop the programme, which is roughly what happened when the December 2024 budget deal blocked the first licensing round.

The state already knows, and has written it down

Norway’s budget documents say the turn is coming, in the flat language departments use when they would rather not make a headline.

The National Budget 2026, presented on 7 October 2025, puts the turning point for public finances, where expenditure growth exceeds revenue growth, “probably already in the next parliamentary term”. That is 2029 to 2033. Perspektivmeldingen 2024, published on 9 August 2024, puts the financing gap at 0.0% of mainland GDP in 2033 and 6.2% by 2060, revised to 5.9% in the 2026 budget. The same document halves annual fiscal room from NOK 24bn across 2025 to 2029 down to NOK 10bn across 2030 to 2033, and notes that once demography and the defence build-up are paid for the discretionary remainder averages about NOK 3bn a year.

The physical curve underneath is the Offshore Directorate’s. Its 2024 resource report has base case production falling from 243 million standard cubic metres of oil equivalent in 2025 to about 83 million by 2050, which is 34% of today. The low scenario goes to near zero. The high scenario, which assumes everything undiscovered gets found and developed, still lands at half of today.

Set the state’s petroleum cash flow against the state’s oil money spending and the scissors are already closing. In 2022 the shelf paid in NOK 1,457bn in 2026 kroner against structural spending of NOK 377bn, a ratio of 3.9x. For 2026 the National Budget in October had cash flow at NOK 521.3bn against structural spending of NOK 579.4bn, which is spending above revenue for the first time since 2021. A May price revision to NOK 686bn put the shelf back on top. Norway’s fiscal position this year was rescued by the oil price moving from $67 to $91 between October and May.

The fund now finances just over one krone in every four of budget spending, on the Revised National Budget’s own phrasing. The structural non-oil deficit has gone from 6.0% of trend mainland GDP in 2014 to 13.1% in 2026, and from NOK 220bn to NOK 579bn in fixed 2026 kroner. The fiscal rule has held in percentage terms only because the fund grew fast enough to flatten the percentage.

Nothing is flashing red yet. Productivity growth in market oriented mainland Norway ran at 0.9% a year over 2017 to 2024 against 2.8% a year in 1992 to 2003, and total factor productivity at 0.4%, which is the sort of number that produces a working group rather than a crisis. The mood on the ground has moved the other way. ONS is running in Stavanger this week, 24 to 27 August, under the theme “Courage”, and the courage on display is mostly the courage to keep drilling. The strategy, whether or not anyone will say it into a microphone, is one last big harvest.

The pretenders carry oil’s risk profile without oil’s reward profile

Offshore wind, carbon capture and hydrogen all require capital committed years before the first unit of output, and all three deliver utility returns at best.

The Norwegian numbers are unusually explicit because the subsidies are capped and published. Sørlige Nordsjø II, won by Ventyr in March 2024, cleared at 115 øre/kWh, about €98 per megawatt hour, on a fifteen year two sided contract for difference settled against the monthly NO2 spot price, with the state’s exposure capped at NOK 23bn for up to 1.5 GW. SINTEF calculated the cap is exhausted after about six and a half years of that fifteen. Utsira Nord carries a NOK 35bn ceiling for one 500 MW floating project expected to produce about 2 TWh a year, which is NOK 70m per megawatt, or NOK 17.5 of state support for every kilowatt hour of first year output.

Set that against what energy costs to make. NVE’s 2024 estimates put onshore wind at €36 a megawatt hour, large hydropower at €37, bottom fixed offshore wind at €99 and floating at €132, all at a 6% real discount rate. Norwegian power sold for €50 in 2025, and €34 across 2015 to 2019 in today’s money.

The onshore figure will look low to anyone who has financed a wind farm, and it is. NVE costs the plant before tax, without the grid connection it includes in the offshore numbers, and against a baseload price that wind does not capture. The whole Norwegian onshore fleet paid €0.9m (NOK 10.4m) of resource rent tax for 2024, its first year of assessment, which is not what a fleet earning €14 a megawatt hour of rent would pay. Read €36 as a floor and I personally dont buy that one bit!

Oil sits on the same axis. Brent averaged $69 in 2025, which is €36 for every megawatt hour of energy in a barrel. Johan Sverdrup produces at under $15, or €8. That gap of €28 is the rent, and the state takes 78% of it. Floating offshore wind runs the arithmetic backwards, costing €132 to make something that sells for €50.

Norway’s fiscal model runs on taxing resource rent, at 78% on petroleum and 45% on hydropower. Neither rate does anything where there is no rent.

The programme is not going well either. The government dropped the planned bottom fixed tender at Sørvest F in February 2025 because the grid solution could not be made to work at any sensible cost. On 9 June 2026 a Storting majority of Høyre, Frp, KrF and Rødt ordered an external quality assurance review, a socio-economic analysis and a technology assessment of the NOK 35bn Utsira Nord scheme, reporting by mid 2027, over the objections of a minority Labour government. On 24 August, four days ago, the Ministry of Energy granted Ventyr a twelve month extension on its licence application for Sørlige Nordsjø II, moving the deadline to 15 October 2027, after the consortium cited “a number of unexpected challenges in the development and planning” and “the difficult macroeconomic environment in offshore wind”. Norway has 94.6 MW of offshore wind operating at Hywind Tampen and nothing commercial under construction.

Equinor, they are still building. Turbine installation at Dogger Bank A finished in February 2026 and commissioning should finish by the end of this year. Dogger Bank B took first power in early March. Empire Wind 1 off New York survived two federal stop work orders, the April 2025 one lifted in May and the December 2025 one struck down by preliminary injunction on 15 January 2026, and Equinor guided in March to first power late this year. Offshore construction on Bałtyk 2 and 3 started on 18 February with first power expected in 2027.

The ambition has been dismantled in four moves.

• On 5 February 2025 Equinor retired its target of putting more than 50% of gross capex into renewables and low carbon by 2030 and cut the 2030 capacity ambition from 12 to 16 GW down to 10 to 12 GW.

• On 4 February 2026, with the 2025 results, it took USD 4bn out of the 2026 and 2027 capex plan, “mainly within power and low carbon”, against a full year renewables operating loss of USD 1.6bn including a USD 1.4bn impairment on Empire Wind and the South Brooklyn Marine Terminal.

• On 16 June 2026 it scrapped the capacity target altogether, replaced it with a power output figure of more than 20 TWh by 2030, put USD 1bn back into oil and gas, and set a 2028 to 2030 capex split of roughly 60% Norwegian shelf, 30% international oil and gas, 10% power. It closed the Japanese offshore wind business on 26 June.

• On 17 August it paid USD 940m for a controlling stake in a 1,483 MW gas fired plant in Pennsylvania which on its own generates about 1.6 times Equinor’s entire 2025 power output.

Delivery without ambition. That is what a bridge looks like from underneath. Equinor is finishing what it started because the contracts exist and the reputational cost of walking is real, not because it can see a gold rush on the other side. BP and Shell stepped into offshore wind and stepped out again because they could. Equinor cannot, because it is 67% owned by a state that needs something for Stavanger to stand on, and a national assignment is not a commercial conviction.

The three obvious objections, answered

The fund is the new engine. It is a bank account with a very good year behind it. Transfers now cover one budget krone in four, and the NOK 1,753bn return in the first half of 2026 exceeded the whole year’s expected petroleum cash flow by 2.6x. Saving is not generating. A fund employs nobody in Rogaland, where Offshore Norge counts more than 65,000 people in the petroleum value chain out of a county population of 508,922. What the fund does is convert Norway from a producer into a shareholder in everyone else’s production, which makes national income a function of American equity multiples rather than of anything a Norwegian does at work. A fund that finances consumption rather than productive capacity buys a long and comfortable drawdown.

What about salmon, hydro, Kongsberg, seabed minerals. Real, and not big enough. Seafood exports hit a record NOK 181.5bn in 2025 against NOK 1,010.5bn of oil, gas and condensate, so petroleum is 5.6x the entire fishing and farming industry in its best year ever. Metals were NOK 100.2bn, chemicals NOK 84.1bn, electricity NOK 26.6bn.

The instructive part is what the state did when salmon started earning. Norway introduced a resource rent tax on aquaculture with effect from 1 January 2023, proposed at 40% effective in consultation, cut to 35% in the bill, adopted at 25% by the Storting on 31 May 2023. Spring estimates put the yield at NOK 16bn to 17bn a year. It assessed NOK 868m for 2023 and NOK 469m net for 2024, with yield per tonne falling from NOK 1,454 to NOK 343 in a single year as the industry hived the taxable sea phase into standalone entities and moved value creation outside it. The state went looking for the next rent to tax and could not even find that one. An institution shaped by fifty years of collecting rent reaches for a rent, because it has forgotten how to do anything else.

Somebody will find the next one. Denmark thought it had. Novo Nordisk was worth more than Danish GDP in May 2024, Danmarks Statistik has stated flatly that without pharmaceuticals Danish growth would have been negative in 2022 and 2023 and less than half what it was in 2024, and the B shares peaked at DKK 1,033.20. They were DKK 304.95 this week, down 70.5% from the top, after a disappointing CagriSema readout in December 2024, a chief executive change, 9,000 job cuts announced in September 2025 and a guidance cut in February that took 18% off the price in a day. Denmark did find a magic industry. It has taken twenty months to discover it may not be a durable one, and you cannot tell in advance, which is the whole problem with waiting for the next perfect industry to turn up.

Fortnum’s to Marks and Spencer’s

The honest version of Norway’s future is relative decline, which is harder to talk about than poverty because it sounds like ingratitude.

If you have always shopped at Fortnum and Mason and you are one day expected to shop at Marks and Spencer, the rest of us still think you are a lucky bastard. From the inside it is a real downgrade, and pretending otherwise does not make the adjustment easier.

Denmark is the shape of it. Denmark had North Sea oil too, peaking at about 390,000 barrels a day in 2004, and its best year of state revenue from it was 2008 at DKK 35.9bn, which was 3.7% of Danish general government revenue. Norway’s 2022 was 49.2% of state revenues. Denmark never built a fund, went net importer in 2018, set an end date of 2050 and carried on being a solid, high value, well run economy. Impressive. Not exceptional.

The convergence has already happened underneath the headline. Strip petroleum out and mainland Norwegian GDP per capita was $62,807 in 2024 against Denmark’s $60,773, in constant 2015 dollars, a gap of 3.3%. In 1990 mainland Norway sat 22% below Denmark. The mainland economy has done well and it has done Danish-well. Everything above that line is the shelf.

Ending up as Denmark would be an excellent outcome by any normal standard. Many countries have become rich. Very few have stayed rich on new terms, and none of them managed it while the pension bill was rising underneath. Folketrygden spending goes from NOK 670bn in 2023 to NOK 778bn by 2033 in fixed 2025 kroner, from 15.8% to 17.4% of mainland GDP, and it already takes 34.5% of the state budget. The pensioners at Sola Strand are not only the beneficiaries of the harvest. From the 2030s they are also the largest single line item in the budget that has to replace it.

What I am not going to pretend

I run numbers on offshore wind for a living. I have spent years building models that turn wind speeds and capex curves into an internal rate of return, and I believe in the industry enough to have made a career of it.

It is not the answer to this question. Offshore wind is necessary, it is real, it will get cheaper, and it is a utility business with utility returns selling to domestic customers under regulated prices. It cannot do for Rogaland what Ekofisk did, and no amount of enthusiasm in a conference hall changes the arithmetic. Almost nobody in the post-oil commentary will say that out loud, and the reason is that the people who understand the numbers are the same people being paid to be optimistic about them.

Which leaves the question where it started. The woman on reception at Sola Strand is twenty. She will work for fifty years in an economy whose central revenue engine will have shrunk to roughly a third of its current output by the time she stops, and nobody in the Norwegian state, the Norwegian energy industry or the Norwegian commentariat can currently tell her what replaces it.

If you are twenty in Stavanger this week, what do you tell yourself? If you are fifty three, what do you tell your grandson?

That the question hangs there, at a conference called Courage, in the richest small country in Europe, in the middle of its best half year on record, is the clearest signal available.

Sources. Statistics Norway StatBank tables 11012, 08792, 08809, 09174 and 11342. Norwegian Ministry of Finance, National Budget 2026 (Meld. St. 1 (2025–2026), 7 October 2025), Revised National Budget 2026 (Meld. St. 2 (2025–2026), 12 May 2026), Perspektivmeldingen 2024 (Meld. St. 31 (2023–2024), 9 August 2024). Norwegian Offshore Directorate, Resource Report 2024 and The Shelf in 2025. norskpetroleum.no government revenue and production datasets, updated 12 May 2026. Norges Bank Investment Management half year report 2026, 12 August 2026. Petoro full year 2025 results, 19 March 2026. Skatteetaten petroleum and resource rent tax assessments. NVE, Kostnader for kraftproduksjon, 2024 update, and the strategic impact assessment of offshore wind, 28 November 2024. Statistics Norway table 09365 for wholesale power prices. US Energy Information Administration for Brent, Norges Bank for the dollar and euro rates. SINTEF on the Sørlige Nordsjø II subsidy budget. Equinor fourth quarter and full year 2025 results (4 February 2026), Capital Markets Day 2026 (16 June 2026) and annual report 2025 (19 March 2026). offshoreWIND.biz on Ventyr, Utsira Nord and Empire Wind. Norwegian Seafood Council, 6 January 2026. Danish Energy Agency and Eurostat gov_10a_main. Danmarks Statistik NYT releases on the pharmaceutical contribution to growth. World Bank NY.GDP.PCAP.KD. NAV, Utviklingstrekk i folketrygden. OFV new vehicle registrations 2025.

Everything Europe is anxious about converges on one Estonian county of 130,156 people. In the eighteen months to mid-2026, Ida-Viru, a small county in the very northeast of the country, watched its grid cut loose from Russia, its Finnish cable severed by a shadow-fleet tanker’s anchor, its Russian citizen residents stripped of the local vote by 93 votes to 7 in parliament, a “Narva People’s Republic” hawked at it on Telegram in the packaging Russia used before Donbas, a €400m state oil plant open on its soil, €194.5m written off that industry’s future by the state’s own auditors, and Europe’s first rare earth magnet factory open in Narva with money meant to bury the old economy.

Meanwhile the county’s actual product left quietly by sea. In 2025 Estonia exported 775,300 tonnes of shale oil to Singapore, 87 per cent of the trade, up from 359,200 tonnes two years earlier, while the share going into Rotterdam’s blending tanks collapsed. Shipping entered the European Union’s carbon market in 2024 and the FuelEU rules followed in 2025. When Europe priced emissions at sea, the oil sailed beyond the pricing, and the county’s wages sailed with it.

Figure 1: “The seam runs under the border”

Let’s start with the rock itself. Kukersite is a marine shale laid down 450 million years ago. It runs in a belt across Ida-Viru and continues under the Narva river into Russia, where the mining town on the far end of the seam is called Slantsy, the Russian word for shale. The border was drawn through the middle of an orebody and its company towns to follow the river which as always is convenient for cartographers and difficult for anthropological reasons. The workforce was settled in the region from across the Soviet Union to dig it, which is why the county is 68.5 per cent Russian today and Narva itself around 95 per cent Russian speaking, with roughly a third of the city holding Russian passports (difficult enough for the Estonians to manage in simple times). The interwar industry piped shale gas to Leningrad / St Petersburg. The Soviet industry fed a secret uranium works at Sillamäe, a closed city that appeared on no map, from the same geology. The two halves of the seam then ran a natural experiment. The Russian half decayed after separation in 1991. The Estonian half became the most shale dependent power system on earth, and is now paying, dearly to try and unpick its dependance.

The rock outlived its power stations

Oil shale supplied about 76 per cent of Estonian electricity as late as 2018, and about 70 per cent of the country’s carbon dioxide. Then the carbon price rose. The plants fell off the market in the summer of 2019 and never came back. In the third quarter of 2025 the fossil units at Narva generated 57 GWh, down 83 per cent in a year, and Eesti Energia burned less oil shale in that quarter than in any quarter of its history, a good thing.

But here the secret starts to unveil itself. The plants did not close though, because Estonia cannot afford to lose them. In February 2025 the Baltic states cut themselves off the Russian frequency area. The old turbines at Narva are most of the spinning mass Estonia has left, so from January 2026 the grid operator Elering pays about €59.5m a year for the fleet to stand ready. The replacements are all late. The regulator’s own arithmetic says Estonia needs about 1,000 MW of new plant by 2035. New gas plants cost more to build than the old shale blocks cost to keep. The nuclear law only entered force in 2027 and a reactor is a decade away at best. A tender for new gas will be launched in the late 2026 period. So the dirtiest fleet in Europe is now three things at once: a climate liability, a strategic asset, and the cheapest insurance on offer.

Figure 2: “The rock changed customer”

And the rock keep coming out of the ground. It changed customer. While the power business shrank into a paid reserve, the state finished a new third oil plant at Auvere. Enefit 280-2 cost close to €400m, produced its first oil in April 2026, and can make up to 250,000 tonnes of shale oil a year. VKG, the private producer at Kohtla-Järve, makes about 680,000 tonnes of oil products a year, mostly a low-sulphur blending component for ship fuel. About 90 per cent of Estonian shale oil has been exported since output first passed a million tonnes in 2017. Estonia promised the European Commission it would take oil shale out of its energy system by 2040 and it will but the oil will simply go elsewhere to be burnt, blended and destroyed for more atmospheric CO2. It opened a new extraction capacity in 2026. The product now leaves on tankers.

The grubby half of a clean transition

The carbon accounting is the whole point. Burned at Narva, the rock counted in Estonia’s national inventory and made the country one of the worst per-capita emitters in Europe. Turned into oil and sold as ship fuel, it burns in international waters, where until recently no inventory reached. Estonia’s territorial emissions fall and the wages stay in Ida-Viru. The carbon dioxide lands in nobody’s column.

Figure 3: “One county’s rock, the world’s ships”

Then Europe extended its rules out to sea, and the customs returns show what happened next. Exports to the Netherlands, the Rotterdam blending pool, were 280,700 tonnes in 2023. In 2024, 187,500. In 2025, 43,900. Singapore over the same three years took 359,200 tonnes, then 443,000, then 775,300, which was 82 per cent of everything Estonia shipped. The buyers filling the gap are Lomé, Lagos, Cotonou, Dakar, Walvis Bay, Maputo and Argentine ports, the bunker stations of the unpriced world. Germany and Britain bought 63,800 tonnes in 2023 and nothing since. Shipping entered the European Union’s carbon market on 1 January 2024. FuelEU Maritime, which penalises fuels with heavy lifecycle emissions, followed a year later. Shale oil carries one of the heaviest lifecycle footprints afloat, because retorting the rock consumes serious energy before the oil ever meets a ship. Proving motive would need the traders’ books. The customs data only shows the timing, and the timing is exact.

Figure 4: “When Europe priced the sea, the cargoes left”

The ships close a loop nobody in Tallinn enjoys discussing. In December 2024 the tanker Eagle S, part of Russia’s shadow fleet, dragged its anchor across Estlink 2 and cut the county’s main cable to Finland. The bunker pool Estonian shale oil feeds is the pool the shadow fleet sails on. No customs return will ever put an Estonian molecule in a sanctions runner’s tank. The geometry stands anyway. The county’s grid is attacked from the sea by the same anonymous shipping economy its one export serves.

There is one more story buried in the trade table. In 2021 Estonia exported 4.8m tonnes under the shale oil heading, because its terminals were then a transit corridor for Russian fuel oil. Sanctions removed about 4m tonnes of that business in two years. The tank farms at Muuga and Sillamäe were sized for Russia’s exports. They now handle only Estonia’s own. Estonia is exporting its CO2 problem to the rest of the world.

The money is going before the politics

Everything under the trade gave way in the same year the new plant opened. The world fuel oil price sagged through 2025 and shale oil sagged with it. Eesti Energia’s achieved price fell 7 per cent with hedging, and would have fallen 17 per cent without. Output dropped 9.5 per cent while the older retorts sat in repair. The new plant reached the market with no forward book, because shale oil sells twelve months ahead and a plant with a slipping start date cannot promise cargoes. Then the auditors weighed the price outlook against the carbon rules closing on the product and wrote €194.5m off the oil assets. Without the write-down the group made €111.9m in 2025. With it, a loss of €82.6m. The business still runs. The future has been repriced.

The technology export died first. Eesti Energia spent two decades selling its retorting know-how abroad. The Utah venture was liquidated in January 2024. The Jordanian one, a $2.1bn oil shale power station at Attarat in which the Estonian state company holds 10 per cent, spent four years in arbitration after Jordan’s state power buyer attacked its $0.17 per kilowatt hour tariff as grossly unfair, at $13m in legal fees, before the tribunal ruled in August 2024. Whatever the award’s fine print says, a flagship whose only customer spent four years trying to escape the contract is a poor advertisement. The world looked at the Estonian model and returned one liquidation and one lawsuit.

The state, meanwhile, pays both sides. It granted €125m toward the new oil plant, put €100m of share capital into Eesti Energia in May 2025, and pays the €59.5m standby fee into the same industrial complex. The European Union’s Just Transition Fund has put €354m into ending the industry those payments extend. Both columns of money arrive in the same county in the same years. The column keeping the industry alive is not the smaller one.

A community fighting for oxygen

Underneath all of it is a workforce that has heard every phase-out speech twice. Fossil industry generates close to half of Ida-Viru’s output. About 5,800 people worked in the sector at the last firm count, some 16,000 lived in households that depend on it, and the sector wage of €1,663 a month runs well above anything else the county pays. The average worker is 47. In 2016 low oil prices forced VKG to sack 500 people, a quarter of its staff. In 2020 the carbon price took another 1,033 jobs in a single year. This community knows what a market verdict feels like before any minister stands up to announce one. Of about 7,500 unemployed in the county in early 2025, more than 3,500 lacked working Estonian. The transition arithmetic is honest about its own limits: €354m, most of it to retrain around 11,000 workers, against a target of about 1,100 new jobs. Ten retrained for each job promised.

The same community lost its vote in the same window. In March 2025 parliament amended the constitution by 93 votes to 7 and removed the local franchise from Russian and Belarusian citizens. Stateless residents voted once more in October 2025 and will not vote again. In 2021, 139,281 third-country residents held that vote and 61,224 used it. Narva’s mayor called the change a declaration that Estonians do not trust their Russians. Moscow works the wound daily, pushing a “Narva People’s Republic” on Telegram and TikTok in the branding it used before occupying eastern Ukraine. The fieldwork keeps finding the opposite of the Kremlin’s story. Loyalties in Narva run local and Estonian, and everyone on the promenade can see that Ivangorod, across the river, lives worse. The pressure is real all the same. A wave of redundancies among Russian-speaking industrial workers on the NATO border is the one outcome three successive governments have paid any price to avoid. The oil exports are that price. They keep the wages moving while the power stations wind down, which is why a state that promised to end the industry built one more plant for it.

One proof exists that the story can end differently, and it stands in Narva. In September 2025 Neo Performance Materials opened Europe’s first rare earth magnet plant there, fed from its separation works at Sillamäe, the only such facility running outside Asia. The Sillamäe site started in 1926 as a Swedish-financed shale oil plant. It spent the Cold War as a closed Soviet uranium works. It turned to rare earths in 1970, and it now anchors the European Union’s critical materials plans. One plot of Ida-Viru ground has carried shale oil, uranium, rare earths and magnets inside a century. The transition promised to the rest of the county has happened exactly once, ten kilometres from the retorts. It took a hundred years and three changes of empire.

It looks like the ending is already on the balance sheet

The politics will grind on. The 2040 phase-out sits quietly in the recovery plan paperwork, placed there without consulting the company it binds. The newest oil plant holds an environmental permit to 2035. The trade will keep finding ports beyond the carbon rules for as long as such ports exist, and the wage bill will keep making it untouchable in Tallinn.

The market is not waiting. The price fell. The customers moved beyond Europe’s reach. The flagship export project ended in arbitration. And the state’s own accountants marked the oil assets down by €194.5m in the year the newest plant produced its first barrel. Estonia built its last retort into a falling market, and its own auditors priced the ending before the first cargo sailed. Nobody has yet priced the county.

A note on the numbers. Trade flows are UN Comtrade, Estonia reporting, customs heading 2707, net weight, 2021 to 2025, cross-checked against Statistics Estonia at the eight-digit level before publication. Destinations are as declared to customs at departure and are consistent with independent marine fuel testing. Estonia has no meaningful coal tar trade, so the heading is in effect the shale oil line. Company figures are Eesti Energia’s unaudited 2025 results and quarterly releases. Generation figures are Ember data via Our World in Data, with Estonian oil shale classed under oil and other fossil fuel. Employment is stated at its last firm count, January 2020, from the European Commission’s Territorial Just Transition Plan documents and the Coal Transitions regional fact sheet. The Attarat award of August 2024 has no published text and press accounts of the outcome conflict, so no claim is made about which side prevailed. Sillamäe and Narva plant history is from Neo Performance Materials, Invest in Estonia and Benchmark Mineral Intelligence.

The Green Revolution animation, Europe 1970–2027

Watch Europe’s grid build itself and one thing jumps out. Solar goes from almost nothing in 2000 to the largest single source of capacity on the continent by 2023, and by 2025 it stands at roughly 400 GW against 90 GW of coal. Run the same film for Germany, Spain or Poland and the shape is the same. Coal fades, gas holds, and a wave of amber dots covers the map in barely fifteen years. France is still hanging in with nuclear but that too is changing fast.

Then look at what happened last year. European solar additions fell for the first time since 2016. In Poland, tenders were barely a fifth subscribed. In Spain, solar power purchase agreements are the cheapest in Europe and merchant projects have stopped clearing investors know why. In the Nordics and Baltics, developers cannot find anyone to sign for fixed price solar at all. Utility scale solar in mature markets has not slowed because panels got expensive. Panels have never been cheaper, the price is still dropping and I suspect that if they had not we would have seen this stall two years earlier. It has slowed because the electricity it produces has stopped being worth anything at the hour it produces it.

The numbers are stark. Sweden’s SE2 zone saw 679 negative-price hours last year, Spain 569, Germany and France well over 100 in April 2026 alone. Solar capture factors, the share of the average price a solar plant actually receives, sat at about 0.26 in Germany and 0.28 in Spain this spring, and 0.10 in France. A generator earning a quarter of the market price does not service debt.

The rather obvious answer is storage of course, catch the midday surplus and sell it at seven in the evening and, up to a point that is exactly right. Europe is nowhere near that point yet but the yet is not far off. Batteries sized for the daily cycle pay for themselves in most of these markets, and the fact that we have built so few of them is a scandal of grid queues and permitting rather than economics. But storage has a ceiling. The evening bucket is only so big. A bigger battery costs more and cycles less, and once the daily cycle is served the next battery is chasing surplus that only exists on a few dozen days a year. It will not pay, and the market already knows it, which is why standalone battery projects have stalled in the very markets with the most surplus and again, falling BESS prices have and are masking this problem.

So if cheap panels and cheap batteries are not the constraint, what is? Two things, and both were built for a world that no longer exists.

The first is how we pay for generation adn I am talking right down to a household level. Nearly every scheme that built the projects in the animation pays per megawatt hour produced. Feed in tariffs, as produced or as forecast PPAs, contracts for difference on metered output.

A generator paid per MWh is indifferent to price. It keeps producing into an oversold market, the glut gets deeper, and then the costs land on consumers and the understandable caution next project’s business case. That was harmless when clean electricity was scarce.

A useful measure here is the capture factor. What a solar plant actually earns per megawatt hour, divided by the average market price. It falls as solar grows, because every plant sells at the same hour and drives the price down together, and it is lowest in spring, when the sun is strong, demand is soft and the reservoirs are full. This April it was about 0.26 in Germany, 0.28 in Spain and 0.10 in France. Below roughly 0.4, two things happen at once. Even the cheapest solar built do not typically service debt on market revenue alone, and any support scheme that pays per megawatt hour ends up paying mostly for the hours when the price is zero or negative, which is to say it pays plants to deepen the glut. That is the point at which as produced support stops being a hedge and becomes part of the problem.

Support has to shift to paying for what the system is short of: capacity that is available in the evening and in winter, storage attached to generation, firmness and shape. Poland’s capacity market and Italy’s storage auctions are early versions and, not coincidentally, those are the two markets where batteries earn best.

The second is how we charge for consumption, right down to the household. Most Europeans still pay a flat price for a product whose cost swings from negative at noon to very high at dusk. Smart meters are barely deployed in Germany. Where time of use exists it is a blunt hammer shaped tool in the form of a three band shape rather than a price. This matters less for the kettle after Corrie or at half time in the football than for what is coming, which is millions of electric cars and heat pumps that will either arrive flexible by default, charging and heating when power is free, or arrive as fixed load that makes the evening peak worse. Which of those we get is decided by the tariff structure in place before they connect, including network charges and an end to storage being taxed twice as both consumer and generator.

There is a third condition, and it is not optional. Germany has over 200 GW of connection requests against no capacity until nearly 2029. Spain approved 4.5 GW of 40 GW requested. Sweden’s surplus is trapped in the north while the south pays four times the price. Perfect support design and perfect tariffs produce a queue of well-financed projects that cannot connect. Grid has to be built ahead of generation and the queue has to reward projects that bring their own flexibility.

Do those three things and solar and wind keep growing in mature markets. But it will be a different product from the one that filled the map between 2010 and 2023: solar with a battery, solar with a west-facing share, solar sold under a contract that pays for shape, and wind where it complements it. The residual surplus on the sunniest afternoons will be curtailed, and that is not waste, it is the cheapest thing on the site being thrown away in favour of not building a battery that would sit idle for 300 days a year.

The green revolution is not over. It has simply run into the price system that was designed to start it, and that system now has to be rebuilt to finish it and people like me need to education governments on how to move faster to allow this to happen.

Data: Global Energy Monitor plant trackers; Ember; Eurostat; pv magazine and Pexapark capture-price analyses, spring 2026.

This week, the game stepped up, Robertas Kaunas, Lithuania’s defence minister, told the news portal 15min on 6 August that Lithuania has data on Russia weighing unconventional kinetic attacks on critical infrastructure in the Baltic region, and that:

“Russia is likely to use Ukrainian production drones to do so.”

Ukraine launches long range strike drones at Russian targets. Some are shot down and some crash. Russia collects the parts, and Kaunas said Russia is capable of assembling them into a working device. He called this the most realistic scenario currently facing the region.

Of course, he was careful about the limits of what he knew. No decision has been taken. No target or timetable has been identified. Lithuania sees no build up of conventional forces on its borders and faces no conventional military threat. Anyone reporting the statement as an imminent warning has inflated it. But he publicly stated it; WHY.

For a Defence Minister to make such a public statement tells us something about both the level of confidence and how he is looking to head off such an attack by highlighting it. It was also his second warning in eight days. On 30 July he said Putin was weighing escalation as the advance in Ukraine stalled, and that the Baltic region and Poland were the closest options to hand. The day before he spoke, an unidentified first person view drone attacked a Ukrainian cargo plane on the ground at Leipzig airport.

March, at quarter to four in the morning, is it a warning?

On the night of 25 March 2026 a drone came in over the Estonian border at Narva and struck the chimney of the Auvere power station at 03:43. Auvere sits just two kilometres from Russia. Estonia’s Internal Security Service went out, recovered the wreckage, examined it, and… confirmed it was Ukrainian. Ukraine had been running its heaviest campaign of the war against the Russian oil export terminals at Ust Luga and Primorsk that month, and Russian electronic warfare had been pushing the attacking drones off their tracks. One of them ended up in Estonia but nobody publicly called it an attack.

Two nights earlier a drone had come down at Lake Lavysas in the Varėna district of Lithuania. The same night as Auvere, another landed in a field at Dobročina in Krāslava municipality, Latvia. Weeks later on 7 May a second stray exploded at an oil storage facility in Rēzekne. On 19 May a Romanian F-16 flying Baltic Air Policing shot one down over Põltsamaa, the first such engagement in Estonian skies. The following day Lithuania issued its first red level air threat alert and told Vilnius to shelter - and of course we did!

Estonia and Latvia judged all six of the located events correctly. The judgement was easy, because the drones hit a chimney, a field and an oil tank, and importantly nothing of consequence was lost although the oil take could have been tricky. Wreckage of exactly that kind at the foot of a converter transformer at Alytus would support the same innocent reading, and for the first forty eight hours there would be nothing in the evidence to prefer the guilty one.

^{GRAPHIC 1: Nine, out of thirty eight thousand}

The trick or worry, is in the numbers here

Russia launched more than 38,000 Shahed type drones at Ukraine during 2025, most of them aimed at the power system. Nine crossed into Baltic airspace in the same year, two of those flown in deliberately from Belarus. Twenty four incidents have been logged across the three states between January 2025 and 21 May 2026, fifteen of them in the first five months of this year alone, according to counts kept by Novaya Gazeta Europe.

Against a campaign of that size, six or seven strays a year is statistical noise. That is precisely what makes them useful. A staged strike would produce Ukrainian drone fragments, a Ukrainian flight controller, and an eastern approach during a live Ukrainian strike window on Leningrad Oblast. For the first forty eight hours it would be indistinguishable from March. Kaunas said the object of the exercise would be to make NATO waver and cut its support for Ukraine, and the headline “Ukrainian drone hits NATO power station” does that work before any serial number has been read.

Lithuanian intelligence also believes Moscow has convinced itself that the Baltic states have been letting Ukraine use their airspace for strikes on Russia. I can find nothing to say they have. They have not. And government acting on some false premise it genuinely holds is more dangerous than one acting cynically, because it thinks it is retaliating.

What is actually within reach

^{GRAPHIC 2: Nothing is out of range}

I pulled every substation at 110 kV and above in the three Baltic states out of OpenStreetMap and measured each one against the nearest point of Russian or Belarusian ground. There are 610 of them. The median sits 88 km from hostile territory. The furthest, at Ventspils, is 291 km away, which a propeller driven Shahed can cover in under two hours. A hundred and seventy two of them are inside 50 km.

Distance in this context is warning time rather than reach. Everything in the Baltic states has easily been within range of a Geran-2 since the day the type entered service. What the numbers describe is how long anybody would have to react, and for more than half the network that answer is under forty minutes.

Lithuania’s own view of the problem became public on 18 June, when the armed forces began assisting the Public Security Service in guarding four sites: the Klaipėda LNG terminal, the LitPol Link interconnection at Alytus, a transformer substation in the Alytus district, and the Kruonis pumped storage plant.

Read that list as an intelligence product. Kruonis is the only pumped storage plant in the Baltic states, 900 MW for twelve hours 10GW of storage! and roughly 94% of Lithuania’s emergency reserve. Klaipėda is the only LNG import point in the region, a single floating unit at a single jetty, and the route by which Lithuania walked away from Russian gas. The two Alytus sites are the converter station and the transformer substation that together carry one 400 kV overhead line to Ełk in Poland. Since 9 February 2025 that line has been the only synchronous connection holding Estonia, Latvia and Lithuania inside the Continental European grid. Harmony Link, the second one, arrives in 2030. NordBalt and the two EstLink cables move 1,700 MW of trade between the Baltics, Sweden and Finland, and being direct current they can do nothing at all about frequency.

Five hundred megawatts of alternating current holds three countries in the European system, and two of the four sites Lithuania is guarding with soldiers stand in the same district.

What would give it away

It is highly unlikely that a drone of this class will destroy a converter station. Realistic damage runs to winding damage, a radiator bank, perhaps the cooling system or a control building, a fire. The recovery is the harder problem, since replacement lead times for large power transformers now run to years, and a synchronous island on a single tied cable has nowhere comfortable to put a forced outage at the connection point.

If it happens, the forensic argument gets won or lost inside a week, and they would need four pieces of evidence to prove it.

1. Approach geometry comes first. A genuine stray runs on a Ukrainian axis towards Ust-Luga, Primorsk or Leningrad Oblast, arriving from the south east during a known strike window. Anything launched from Kaliningrad or Belarus arrives from a different quadrant, and Baltic air surveillance will know which.

2. Fuel state comes second. A drone that has flown a thousand kilometres and been spoofed off course arrives near the end of its endurance. One launched from a hundred kilometres away does not.

3. Serial provenance comes third. Flight controllers, airframe batches and component lots can be traced back to where they were built and, in principle, to where they were last used. Unsurprisingly Estonia has not published that work for Auvere. What it gave within hours was a visual identification of the airframe as Ukrainian, described at the time as preliminary, and four months on the prosecutor's proceedings have produced nothing public. On the easy case, with no casualties and no damage to the plant, that was enough. On a contested one it would not be.

4. The last is the one Moscow cannot easily fake, and it is the precision of the damage. A drone that has genuinely lost its way hits a chimney, a field, a storage tank. It does not find the single asset in a district whose loss would matter. If the wreckage turns up at a converter transformer or a specific 400/330 kV bay, the target selection confesses.

None of this says an attack is coming. It says Russia holds an option that is cheap, deniable, and pre-laundered by eighteen months of genuine accidents. Ukraine’s entirely lawful campaign against Russian oil exports created the conditions. Russian jamming shaped them. Baltic restraint in refusing to call March and May attacks, which was the correct and responsible call at the time, finished the job.

So the trap is built out of everybody else’s good behaviour, and the only way out of it runs through the forensics. Attribution has to move faster than the news cycle. Right now it does not.

But still for a Lithuanian Defence Minister to publicly state this out loud, we should all take a moment to consider it and ask again, why.

Sources: Lithuanian Ministry of National Defence and 15min, 6 August 2026; LRT; Euronews; Bloomberg; Kyiv Independent; Aerotime; ERR; Baltic News Network; Centre for Eastern Studies; Estonian Internal Security Service; Novaya Gazeta Europe; Missile Defense Advocacy Alliance; Litgrid; KN Energies; Ignitis Group; OpenStreetMap contributors.

At dawn on Thursday 30 July, the sea off El Tarajal filled with people. For ten days the numbers coming round the breakwater had been climbing, 150 to 200 a day by the count of Ceuta’s own government, young Moroccan men for the most part, swimming the final stretch from Fnideq. That morning the trickle became a flood. By Friday evening, between 50,000 and 60,000 people had entered Ceuta, a Spanish city of 84,000 on the northern tip of Africa, in roughly 48 hours. At least 72 died, drowned or crushed against the breakwater. The confirmed toll is still rising and local officials believe it already understates the dead.

Subscribe now

Within three days, almost all of the living had gone back.

Both political readings of last week collapse on that fact. An invasion does not withdraw voluntarily inside 72 hours. An open door does not lead 48,300 people to turn round and walk home through it. What happened at Ceuta was a failure with three authors: a Moroccan economy that cannot employ its young, a Spanish legal system that opened a seam without meaning to, and the smuggling networks that sold the seam as a door.

A fence that ends in the sea

Ceuta is a peninsula of 18.5 square kilometres on the Moroccan coast, under Spanish rule for more than four centuries. With Melilla, 250 miles east, it forms the European Union’s only land border with Africa. Mainland Spain lies 25 kilometres away across the Strait of Gibraltar, visible on a clear day. Morocco claims both cities and has never recognised Spanish sovereignty over either.

The land border is roughly eight kilometres of double fence, raised to ten metres in its most exposed sections, running from Benzú on the north coast to Tarajal on the south. At each end the fence stops at the shoreline and a breakwater carries the line into the sea. A competent swimmer can go round it. From the beaches at Fnideq, the Moroccan border town on the southern side, reporters have documented swims of up to five kilometres. From Belyounech in the north the water gap is shorter. In late July the Alborán Sea is warm and usually calm. Everything that followed turns on that arrangement: a fortified land border, and beside it, water.

Why they came

The demand existed long before the trigger. More than half of Morocco’s population is under 35. Unemployment among 15 to 24 year olds stands at around 36 per cent, and roughly one graduate in five is out of work. An Afrobarometer survey published in June found that more than half of Moroccans under 35 had considered emigrating. In the north, a prolonged drought has gutted the farm economy around Tetouan and Fnideq, and inflation has eroded what work remains.

The politics of that frustration surfaced last autumn. Between late September and mid October 2025, the youth movement known as Gen Z 212, organised on Discord and named after Morocco’s dialling code, brought protests to Rabat, Casablanca, Tangier and Marrakesh. Its grievance was specific: heavy state spending on stadiums for the 2030 World Cup, which Morocco co-hosts with Spain and Portugal, while public services decayed. The deaths of eight women in childbirth at a hospital in Agadir became the movement’s touchstone, and “stadiums are here, where are the hospitals?” its slogan. The protests were contained with a mixture of concessions and arrests. The conditions that produced them were not addressed.

Spain, meanwhile, was moving against the European grain. In April the Sánchez government completed a regularisation of undocumented workers already in the country. Government figures reported in July put applications at 1.17 million, roughly twice the initial estimate, with some 610,000 approved. Restrictionist groups have argued the amnesty acted as a pull factor for Ceuta. No evidence has been produced for that link, but the scheme fed a broader perception, carried through diaspora and family networks, of a Spain processing people in rather than out.

What the court actually said

On 29 June the Fifth Section of the Supreme Court’s Administrative Litigation Chamber issued judgment 814/2026. It was published on 8 July. It settled a question so narrow that few outside immigration law registered it.

Since 2015, Spanish law has permitted what officials call border rejection and critics call devoluciones en caliente, hot returns: a person caught attempting to overcome the “containment elements” of Ceuta or Melilla, meaning the fences, can be handed directly back to Moroccan authorities without identification, legal counsel or an asylum hearing. The Guardia Civil had routinely extended the same treatment to people intercepted in the water.

The court ruled that it could not. A swimmer has overcome no physical barrier, so the summary regime does not apply and sea arrivals must enter the ordinary procedure: identification, due process, an asylum claim if one is made. The judgment legalised nothing. It guaranteed nobody the right to remain. Moroccan asylum claims in Spain almost never succeed, and the ordinary procedure ends, for most, in a documented return rather than an immediate one.

That is the ruling. The version that travelled through TikTok and Moroccan social media in the fortnight after 8 July was shorter: swim to Ceuta and they cannot send you back. Spanish officials say the translation was deliberate. “Trafficking mafias took a self serving interpretation of a Supreme Court ruling,” Pedro Sánchez said, one that “spread like wildfire”. Rights activists in Morocco doubt that many of those who crossed had heard of the judgment at all, which, if anything, strengthens the government’s account: what moved people was not the law but a rumour about the law.

Two hundred a day

The build-up can be dated precisely. Between 20 and 29 July, around 1,500 people reached Ceuta by sea, against a normal baseline of a handful a week. Most were young Moroccan men, alongside unaccompanied minors, some families and a scattering of Algerians. The city’s shelters for unaccompanied children, which Ceuta rather than Madrid must fund, were reported at roughly sixteen times their design capacity. On 29 July the authorities began moving people to the mainland, a first group of 39, most of them Sudanese, shipped to Algeciras.

^{the build-up chart}

Ceuta’s president, Juan Jesús Vivas, asked Madrid for troops and a formal emergency declaration. The interior ministry sent police reinforcements and declined the declaration, arguing migration falls outside the legal grounds for one. The Spanish news site The Objective has since claimed that the intelligence services warned the government of possible mass crossings days in advance. The claim rests on unnamed sources and a single outlet, and remains uncorroborated.

Forty-eight hours

On 30 July the border gave way. Crowds poured round both breakwaters, swimming and wading; television footage showed people kneeling and kissing the ground on the Ceuta side. On the Moroccan side, security forces fought the crowds with water cannon, tear gas and warning shots, and burnt-out vehicles lined the road through Fnideq. Melilla saw parallel violence at the Beni Ensar crossing. Vivas declared an “absolute humanitarian and social emergency”. Spain deployed the army to Ceuta.

The dead were concentrated where the crowd was: at El Tarajal, the breakwater that tens of thousands attempted to round in two days. Divers recovered bodies through the weekend. The confirmed toll stood at 67 on Saturday and 72 on Sunday. Local officials have put it at 86, and bodies had been washing ashore near Ceuta for weeks beforehand that may never be attributed. Morocco has published no count of its own.

The walk back

What followed was the part nobody scripted. Ceuta could not feed, shelter or process a crowd two thirds its own size, and the people who had crossed discovered it within hours: no accommodation, little food, extreme heat, and a city whose shops had shuttered. By Friday evening the interior ministry said more than 48,300 of the roughly 50,000 it had counted in had already returned to Morocco, most voluntarily, walking back through the Tarajal crossing. Officials estimate that around 90 per cent had gone within days. On Saturday morning the Guardia Civil began installing a 500 metre floating barrier along the breakwater. The engineering is legal as much as physical: if a barrier now exists in the water, the summary regime the court switched off may apply once more. Madrid has said the barriers will stay.

Europe reacted to the images rather than the outcome. Donald Trump, at a cabinet meeting at Camp David, called the crossing an invasion. Italy imposed 30 days of checks on air and sea arrivals from Spain, France reinforced its frontier controls, Giorgia Meloni called for Spain’s suspension from Schengen, and 22 member states signed a demand for coordinated border action. Sánchez replied with a letter condemning what he called a “selfish, polarising and unlawful reaction”, and Spanish officials pointed out what the maps already show: Ceuta sits outside Schengen’s free travel area, and passage to the mainland runs through identity checks at the port and airport.

The Morocco question

In May 2021, when around 8,000 people crossed into Ceuta in two days, Moroccan border guards conspicuously stood aside. Madrid had just admitted the Polisario Front’s leader to a Spanish hospital, and the message was understood on both shores. The comparison has hung over last week ever since, and the evidence this time is genuinely mixed. Rabat’s forces visibly fought the crowds at Fnideq, and its ambassador in Madrid, Karima Benyaich, insisted the crisis was “not wanted by the Kingdom of Morocco”. Yet Sánchez had visited Algeria, Morocco’s rival, earlier in July, a trip that analysts in Madrid say was poorly received in Rabat, and Spain’s Congress had just approved nationality rights for descendants of Sahrawis. “One way of handling that is to send a not very subtle message,” Haizam Amirah-Fernandez of the Elcano Royal Institute told the New York Times. A Moroccan human rights worker has suggested the Tarajal border was opened under unusual circumstances. There is precedent for the theory and, so far, no proof.

What the record does support is narrower and harder. A state with 36 per cent youth unemployment sits beside a fence with a legal seam in it. A rumour told sixty thousand young people the seam had opened. They went, 72 or more of them died at a breakwater, and the rest walked home hungry within three days because nothing waited on the other side. The floating barrier now moored off Tarajal answers the Supreme Court. It answers nothing in Fnideq. More than 850 unaccompanied children remain in Ceuta’s care, the durable residue of every episode since 2021, and the next rumour will find the same reservoir waiting.

Sources: Spanish Interior Ministry, Government of Ceuta, Supreme Court judgment 814/2026, AP, Reuters, RTVE, CNN, Al Jazeera, Euronews, Afrobarometer, Morocco’s HCP. Death toll as confirmed 2 August; local officials cite higher figures.

Subscribe now

This spring, Iranian drones were striking commercial shipping in the Strait of Hormuz and Qatari LNG cargoes were disrupted within days (it seems to be starting again as no agreement is reached). Russia has spent a fourth winter methodically bombing Ukraine’s power grid. “Somebody” keeps dragging anchors across the cables on the Baltic seabed, and NATO now runs patrols to watch for it and the Baltic States remain on a fairly high alert for border issues as does Poland. The United States, meanwhile, has become Europe’s biggest gas supplier as well as its least predictable ally, sometimes in the same week. Energy security has moved from the seminar room to the front page of most reasonable news outlets, and underneath the headlines sits an old question: how much of the energy your country burns comes out of somebody else’s ground, what would happen in that simply stopped, if the electrons stopped flowing or the gas didn’t arrive?

Graphic 1: Europe buys its energy; the lucky few sell it

For most of Europe the answer is quite uncomfortable. Ireland buys in 80 percent of the energy it uses, Italy 74, Germany 67; the EU as a whole imports 57 per cent. The sellers are a small club: Norway, Canada, Australia and lately the United States, which produces about 110 per cent of what it consumes. Little of this was earned by policy. It is geology. Look at Denmark in the lower panel, though: a net seller in 2000, a substantial buyer now. Britain’s story, played out on a smaller stage.

Graphic 2: Britain spent its buffer

Britain ran the full experiment within one lifetime. In 1970 half its energy came from overseas, and the oil shocks made the country a difficult place to live in. The North Sea then delivered two decades of surplus a real bonanza; in 2000 Britain still sold more energy than it bought. Production peaked in 1999 at 298 million tonnes of oil equivalent. The flip back to buying came quietly in 2004, dependence peaked at 48 per cent in 2013, and it stands at 43 per cent today, but the oil and gas fields today only produce a third of what they once did and an honest measure of dependence nets those flows off: in 2025 the UK consumed about 168 mtoe, produced about 95, and the gap of 73 mtoe is what had to come, net, from abroad.

[Graphic 3: How Europe rewired its gas supply]

Europe’s response to 2022 shows what money and urgency can do, and what they cannot. Russian gas fell from 150 billion cubic metres in 2021 to 36 last year. Nord Stream was sabotaged, Yamal was closed, and Ukrainian transit ended on New Year’s Day 2025; of the four huge Russian pipes, only TurkStream still flows. The replacement came mostly by sea. American LNG rose fourfold, and nearly half of Europe’s imported gas now arrives on tankers rather than through pipes. A ship can be rerouted, which pleases the planners. It can also be outbid by Asia or held up in a strait, which should worry them. Sadly though that simply means that dependence was not ended so much as diversified, and the continuing shambolic American presence in the Gulf was a reminder of the difference.

[Graphic 4: Nobody can embargo the wind]

The fourth chart is the hopeful one (I had to squeeze Lithuania as its such a big part of what I am doing). In 2013 only 33 of every 100 units of electricity Britain used came from British sources; in 2025 it was 53, the most in more than 20 years. Renewables did nearly all of that work. Take them away and the figure falls to 27. Wind and solar carry their own dependencies, in turbines, panels and minerals, but those are purchases of equipment, made once, not fuel bought daily from whoever holds the tap.

No serious country pursues self sufficiency for its own sake, and diverse suppliers, storage and interconnection all count large. But the lesson of the past four years, and of these four charts, is a plain one. A country either owns its resources, or it should accelerate the building the one kind of energy that no blockade, chokepoint or hostile counterparty can touch.

If you liked the article please share and restack. the graphics and the research take a great deal of effort and its nice if more people can read and understand a least some of the energy transition and how difficult it is.

Sources: Eurostat; DUKES (DESNZ); Council of the EU; Bruegel; EIA; Statistics Canada; Geoscience Australia; Energy Institute via Our World in Data; ECIU.

Figure 1. The reachable half

European industrial heat demand by temperature, against the reach of technology you can order today.

Somewhere between 47 and 55 per cent of the heat European and British industry makes by burning fuel sits below 500 degrees, where electric kit is already running in reference plants and already beats a gas boiler in the countries that price power properly. That fuel is gas, coal, oil, biomass and black liquor, and the last two count as clean only by an accounting convention. Almost none of it has been converted. The Commission put the same finding more carefully as recently as July, just one month ago, estimating that around 60 per cent of industrial fuel based energy demand could be electrified with heat pumps alone. Technology is not the constraint here. A European factory pays 3.04 times as much for a unit of electricity as for a unit of gas, against a Commission target of 2.0 for 2030, and a British one pays 4.80.

Share

That ratio is the whole of part two.

Household electricity in the European Union averaged €0.2896 per kilowatt hour in the second half of 2025 against €0.1228 for household gas, a ratio of 2.36, which already clears the 2.5 the Commission has set for households in 2030. Industry is the harder case by some distance. Non-household electricity averaged €0.1837 against €0.0605 for non-household gas, and 3.04 against a target of 2.0 leaves industry 52 per cent above the line with four years to run. The home is closer to its target than the factory is to its own, which is the reverse of how the two are usually discussed.

Britain is worse than the European average for a reason that gets almost no attention. Take the European ratio apart and swap in one British price at a time. British industrial electricity is about 7 per cent above the European Union average, which moves the ratio from 3.04 to 3.25. British industrial gas is about a third below the European average, and that alone moves it to 4.49. So roughly four fifths of Britain’s worse position comes from cheap gas rather than dear electricity. The whole British argument about energy costs is conducted as though the electricity price were the variable, and for this particular decision it is the smaller half of the problem.

Start with the dullest machine in the building

An electric boiler is a resistance element in a pressure vessel. It has no moving parts worth the name and no development risk, and any half decent engineering shop could make a basic one with minimal research. An ordinary steam plant runs at about 250 degrees, a packaged superheated boiler up to 300, and a really specialised industrial unit will reach 540. That range covers the great majority of what a brewery, a paper mill, a food plant or a chemical works actually needs. Industrial heat pumps reach 150 degrees at large scale on the International Energy Agency’s assessment, at three to four units of heat per unit of power rather than one.

The Agency is blunt about why neither is being installed at any useful rate. It names unfavourable electricity to gas price ratios, long grid connection lead times and the absence of clear policy frameworks. Not one of those is a technology problem, and not one of them is being fixed at the speed the 2030 targets assume.

Think about what a ratio of 3.04 or 4.80 does to a boiler swap. An electric boiler runs at close to 100 per cent efficiency against perhaps 85 per cent for the gas unit it replaces, so the fuel bill rises by roughly the ratio divided by 1.18. In Europe that is about 2.6 times. In Britain it is about 4.1. No plant manager signs that off, whatever the carbon price does, because the carbon price on industrial heat is nowhere near €150 a tonne and the electricity penalty is being paid every hour of every shift. A heat pump at a coefficient of performance of 3.5 turns the European ratio into a running cost saving of about a quarter, which is why the 150 degree band is the one seeing real orders while the steam band is not. At the British ratio the same machine costs 17 per cent more to run than the boiler it replaced, and nothing below a coefficient of 4.08 breaks even.

Figure 4. Two ways out

Where each machine sits on what it returns and what it pays.

A heat pump moves you along one axis and a heat battery moves you along the other. They are not competing technologies, and Britain closes the first of the two routes.

The heat battery buys in the market that is drowning

The most interesting thing built in this space over the last two or three years is a box of hot bricks or sand, and the reason it works is a market asymmetry that the storage debate keeps missing. Let us call it solid thermal storage.

A lithium battery buys power and sells power. It arbitrages a spread inside the electricity market, and every product it sells is a power market product, and all of them are saturating as the fleet grows. A thermal store buys power and sells heat. It never sells a megawatt hour back into the pool. The surplus that is crushing battery revenue is the raw material it wants, and the market it sells into has no competing storage in it at all. It competes with the marginal boiler.

Germany recorded 576 hours of negative electricity prices in 2025, up on 2024 by a quarter. Every one of those hours is power the system is paying somebody to take. A heat battery charges in exactly those hours and delivers steam at three in the morning in February. Rondo Energy started commercial operation of a 100 megawatt hour brick battery in California in October 2025, charged entirely off an on-site solar array and delivering 24 hour high pressure steam, at a claimed round trip efficiency above 97 per cent. Kyoto Group has a 56 megawatt hour molten salt unit running in Hungary on about 30 gigawatt hours of clean heat a year.

Figure 5. The year, hour by hour

Every hour of 2025 across six markets, with the cheap ones lit.

The cheap hours are not scattered. Across the continent 85 to 89 per cent of them fall between nine in the morning and three in the afternoon, which is a timetable a plant can schedule against. Britain is the exception at 60 per cent, because its cheap hours come from wind rather than solar and arrive at any hour of the night.

In Europe the pipeline is now becoming real rather than promised, and it is probably a bit larger than most people think.

Subscribe now

• Kraftblock has a 70 megawatt hour system frying crisps for PepsiCo north of Amsterdam on wind contracted from Eneco.

• TheStorage put a 250 megawatt hour sand battery into a Finnish brewery in January 2026, which on completion will be the largest sand based store anywhere.

• Brenmiller has a 10 megawatt hour unit with REBI in Spain.

• Rondo broke ground with Covestro at Brunsbüttel in January 2026 on a 100 megawatt hour battery due online by the end of this year, which will carry a tenth of the site’s steam and take out 13,000 tonnes of carbon dioxide a year.

• A further 100 megawatt hour unit is going into Heineken’s brewery and malting plant near Lisbon, charged by 7 megawatts of on-site solar and a 25 gigawatt hour a year power purchase agreement, with EDP selling the steam as a service rather than the hardware as an asset.

That last structure is the one that accountants and engineers would look hard at, more than the sand or the bricks. If we can move heat as a service capital away from the factory’s balance sheet we can then put the price risk on a utility that can hedge it. The single most common reason a plant does not electrify is that the capital committee ranks a boiler replacement below a production line upgrade every single year. Take the capital decision away and the objection goes with it. Factory managers, take note.

BloombergNEF put the global average installed thermal storage capital cost at $232 per kilowatt hour against $304 for four hour lithium (author’s note, I consider both numbers outdated already), which is the number that should be ending the argument about whether this is a serious asset class. The same physics is already running on district heating networks in Finland through a genuinely cold winter, at power prices that swung from €3 to €373 per megawatt hour inside a single week.

Figure 2. The dot

What Europe has built against the band it addresses, on one honest scale.

Set every one of those units end to end and the European fleet comes to 638 megawatt hours of nameplate storage across six countries, of which 438 is running today. Cycle all of it once a day for a full year and it delivers 0.23 terawatt hours of heat, against 706 terawatt hours a year of European industrial demand between 100 and 500 degrees. That is one part in three thousand. Multiply the fleet by ten and it is one part in three hundred. There is no unit anywhere in the United Kingdom.

The pipes are already in the ground

District heating is the part of this that needs no invention at all. The networks exist across the Nordics, the Baltics and much of central and eastern Europe, they are already aggregating thousands of buildings into a single thermal load, and they are still overwhelmingly fired by combustion. A thermal store on a district heating network is a boiler replacement with a storage asset attached, and the customer is a utility that already understands heat as a product.

In what should be another easy win, the Commission wants district heating and cooling to supply 15 per cent of total European heating and cooling by 2030, with networks growing 6 to 7 per cent a year, and it wants waste heat recovery to reach 11 per cent of European heat demand by 2050 against a technically available resource of about 300 terawatt hours a year. A third of that waste heat sits below 200 degrees, which is heat pump territory, and another quarter sits between 200 and 500 degrees, which is where a thermal store or an upgrade cycle earns its keep. Britain has almost none of this infrastructure and is the weaker case for it, with district heating serving a low single digit percentage of demand and a heat network zoning regime that only began to bite in 2025.

Nobody is looking after the middle

The forgotten group in all of this is commercial property, and the reason it is forgotten is a measurement artefact.

The Eurostat band the Commission’s industrial ratio is calculated on covers non household consumers using 500 to 2,000 megawatt hours a year. That is a supermarket, a hospital, a hotel, a district office, a mid-sized works. It is not a steel plant, which sits four bands higher and buys on a negotiated contract with network charge relief attached. So the 3.04 is a fair description of the commercial middle and a poor description of heavy industry, in the same way that part one’s household band described the house that had not switched.

That middle gets neither of the two protections on offer. It does not get the household measures, because it is not a household. It does not get the energy intensive relief, because it does not qualify. Britain shows the shape of it precisely. The Network Charging Compensation scheme raises its discount from 60 to 90 per cent from April 2026 and reaches 586 firms. The British Industrial Competitiveness Scheme cuts electricity bills by up to 25 per cent, worth up to £40 per megawatt hour, for 7,000 to 10,000 manufacturers from 2027. Eligibility for the older support runs on electricity intensity above 7 per cent, trade intensity above 4 per cent and electricity costs above 20 per cent of gross value added. Nesta’s assessment of it is that this excludes many manufacturers outright and every hospitality and retail business in the country, and the scheme is funded by a levy on the bills of the businesses it excludes.

A supermarket chain is one of the easiest electrification wins available anywhere. The heat demand is low grade, the refrigeration plant is already rejecting waste heat at a useful temperature, the estate is standardised so one engineering solution replicates across hundreds of sites, and the capital is centralised. It is being taxed as though it were a chemical works and supported as though it were a household, which is to say not at all.

Above 500 degrees is a different argument

None of the above touches cement, lime, primary steel or the high temperature end of glass and ceramics, and pretending otherwise is how this subject loses credibility with the people who run those plants.

Figure 3. How full is the easy half

Industrial heat made by burning fuel, and the share of it already in reach of electric kit.

Above roughly 500 degrees the process chemistry starts to matter more than the heat source. Cement makes carbon dioxide from limestone whatever fires the kiln. Primary steel needs a reduction agent rather than a temperature. The parity dates set out in part one land where they land for good reasons, with glass, ceramics and paper reaching parity in 2031, chemical process heat in 2036 and cement and lime in 2041, all beyond the targets written to hit them.

The temperature itself is no longer the obstacle it was, and the piece that says otherwise is out of date. Rondo’s bricks charge to 1,500 degrees and its California unit has been discharging above 1,000 in daily commercial operation since October 2025. Antora stores above 1,800 degrees in solid carbon and commissioned a 5 gigawatt hour system with Poet in South Dakota this year. What remains thin is the record of any of those systems delivering process heat to a third party at the top of that range, and the economics against hydrogen in the 600 to 1,200 degree window turn on a hydrogen price nobody can currently underwrite.

The risk in the high temperature debate is that it swallows the policy attention that belongs to the easy half. Cement is hard, so cement gets the research budget, the demonstration funding and the ministerial visit. The brewery that could have swapped its boiler for an electric one in a fortnight gets a price ratio of 3.04 and a two year grid connection.

What is actually slowing this down

Four things are slowing us down, and sadly only one of them is technical , although capital deployment is a problem.

The price ratio is first and largest. At 3.04 the European factory is being charged a premium to do the thing every published strategy asks it to do. The Commission has proposed the right direction with network charge reform, a rule that electricity should not be taxed more heavily than gas, and the phase out of fossil fuel subsidies. None of that is law yet.

Grid connection is second. A 20 megawatt electric boiler is a 20 megawatt new connection, and the queue for that in most of northern Europe and Britain is measured in years rather than months. The Agency names it explicitly. No amount of capital support moves a plant that cannot get the amps.

Capital allocation is third, and heat as a service is the answer to it that is already being demonstrated at Lisbon rather than argued about in a paper.

Fourth is that the support architecture is pointed at the wrong tenth of the problem. The Industrial Decarbonisation Bank carries €100bn and the Innovation Fund is running a second industrial heat auction, both of which are welcome and both of which are shaped around the hard cases. The easy half needs no innovation funding whatsoever. It needs the electricity to cost less than 3.04 times the gas in Europe, and less than 4.80 times in Britain.

Europe and Britain have spent a decade arguing about how to decarbonise the half of industrial heat that is genuinely difficult, while the other half burns gas because the tax code says it should. The kit is in the catalogue. The reference plants are running. The only thing missing is a price that makes the obvious decision the profitable one, and every year that stays missing is a year of the easy half burning gas it did not need to burn.

Sources: Eurostat nrg_pc_204, nrg_pc_205, nrg_pc_202 and nrg_pc_203, household and non-household electricity and gas prices, second half of 2025, extracted April 2026. Household prices include all taxes, levies and value added tax. Non-household prices exclude recoverable taxes and value added tax, so the household and industrial ratios are calculated on different tax bases and are not strictly comparable. Department for Energy Security and Net Zero, Quarterly Energy Prices, March 2026, manufacturing industry averages for the fourth quarter of 2025 excluding the Climate Change Levy, at 16.8 pence a kilowatt hour for electricity against 3.5 for gas. European Commission, Electrification Action Plan, COM(2026) 595, 17 July 2026. International Energy Agency, Renewables for Industry, and Energy Technology Perspectives 2026. Rondo Energy on the Holmes Western and Covestro Brunsbüttel installations. European Commission Tripartite Agreement for Energy Storage, 26 June 2026, on the Heineken, EDP and Rondo project in Portugal. Energy-Storage.News on Kyoto Group in Hungary and on Kraftblock with PepsiCo and Eneco in the Netherlands. ESS News on TheStorage in Finland. Brenmiller Energy on the REBI project in Spain. Antora Energy on carbon block storage temperature and on the Poet installation. BloombergNEF, May 2024, on installed thermal storage capital cost. Nesta on British industrial electricity prices and scheme eligibility. Department for Business and Trade and HM Treasury on the British Industrial Competitiveness Scheme, 16 April 2026, and on the Network Charging Compensation uplift from April 2026. Hourly day-ahead prices for 2025 from the Fraunhofer Energy-Charts mirror of Bundesnetzagentur SMARD data, CC BY 4.0, and from Elexon market index data for Great Britain.

A note on the British ratio. The 4.80 is not on the same basis as the 3.04. Eurostat covers non-household consumers using 500 to 2,000 megawatt hours a year and excludes recoverable taxes and value added tax. The British figure is an average across all manufacturing size bands and excludes only the Climate Change Levy. The gap is far too wide for that difference to explain, but the two are not strictly comparable. The decomposition uses a conversion of 1.17 euros to the pound and the split shifts a few points between 1.10 and 1.20, though the 4.80 itself is currency free. Both fuels fell in price over the year to the fourth quarter of 2025, gas by 11.6 per cent against electricity at 7.9, so the ratio widened while energy got cheaper.

A note on boiler temperatures. The 250, 300 and 540 degree figures describe ordinary steam plant, a packaged superheated boiler and a specialised industrial unit respectively. They are the practitioner range rather than a single published envelope, and they sit above the International Energy Agency’s summary figure of steam to 350 degrees at around 70 bar. Where a single citable source is needed, the Agency figure is the one to quote and the range is the one to believe.

A note on the negative hours. The 576 for Germany is the count of hours whose day-ahead price averaged below zero across 2025, reconstructed from the hourly series and matching the published figure. The day-ahead market moved to 15 minute settlement on 30 September 2025, so quarter hourly prices after that date are averaged into hours to keep the whole year on one basis. Counts on a quarter hourly basis are higher.

A note on the European fleet. The 638 megawatt hour total covers named units with a traceable source in Germany, Hungary, the Netherlands, Finland, Spain and Portugal. It excludes Avery Dennison in Belgium, whose capacity is not published, and one Finnish unit serving a district heating network rather than industrial process heat. The list is not confirmed exhaustive and an earlier count of it came to 308 megawatt hours, so treat the total as a floor. Annual delivered heat assumes one full cycle a day for 365 days, which is generous for a site with a continuous steam load and conservative for one charging only on cheap hours. The conclusion survives either assumption by three orders of magnitude.

A note on the temperature split. The share of European industrial heat demand sitting below 400 to 500 degrees is given as a range because the two standard studies disagree and both use 2012 as their base year. Naegler and others, published in the International Journal of Energy Research in 2015, put demand below 400 degrees at 53 per cent of the European Union total on one approach and below 500 degrees at 47 per cent on a second. Rehfeldt and others at Fraunhofer, published in Energy Efficiency in 2017, put process heat below 500 degrees at 47 per cent of process heat for the European Union plus Norway, Switzerland and Iceland, rising to 55 per cent once industrial space heating is counted. No more recent bottom up estimate at this resolution has been published, and the industrial mix has changed since 2012, so these shares should be treated as the best available rather than as current.

Ratios quoted are computed from the Eurostat European Union averages and are arithmetic, not weighted by the mix of firms within each band. The 3.04 describes the medium sized non-household consumer, 500 to 2,000 megawatt hours a year, and does not describe an energy-intensive site buying on a negotiated contract with network charge relief. Sweden at 0.91 and Finland at 0.87 both sit below parity on this basis, and both have very little gas consumption, which flatters the ratio.

Germany has done well. It spent 2025 becoming the first big market where the heat pump outsold the gas boiler. In July it repealed the rules that got it there. Sitting under the record and under the repeal is the same tax code, which charges households more for a unit of electricity than for a unit of gas, and that arithmetic is about to do far worse damage in the factory than in the home.

Subscribe now

Europe and the United Kingdom have been legislating to wean heat off gas since the 2009 renewables directive while charging households more than twice as much for the electricity that replaces it in almost every market. That multiple, the spark gap, is the number the whole argument runs on. Germany demonstrated both halves of it in a single year.

German households bought 299,000 heating heat pumps in 2025, up a massive 55 per cent on the German Heat Pump Association’s count. The German energy agency DENA puts them at 48 per cent of all new heating systems sold against 44 per cent for gas boilers, the first year the heat pump has taken the larger share, up from 27 per cent twelve months earlier. In July the Bundestag passed the Building Modernisation Act. It removes the rule that any new heating system had to draw at least 65 per cent of its energy from renewable sources, which in practice meant a heat pump, a district heat connection or a biomass boiler. Households may now fit a gas or oil boiler again. The Act arrived in the same month the European Commission published a plan asking every capital to nearly double the number of heat pumps it installs each year by 2030.

Heating and cooling is close to 50% of European final energy demand and the hardest half to decarbonise. Sales rise where governments cut the price of electricity against gas. Sales stall where they do not. Almost no government in Europe or the United Kingdom has cut that price.

Figure 1. Fossil boiler restrictions across the European Union and the United Kingdom, 2024 to 2026. Two tiles move in three years. Greece adds a ban and Germany repeals one.

What Germany actually did

Stay with me, the detail matters. The Building Modernisation Act replaces the Building Energy Act the coalition had promised to abolish. The 65 per cent renewable rule for new heating systems is gone. In its place the government has put an obligation on gas and heating oil suppliers to raise the green share of what they sell from 2029, and a requirement that all heating systems run on emissions neutral fuels by 2045. The ban on installing new fossil boilers from 2045 was dropped.

That is a transfer of the obligation from roughly 41 million German households to a few hundred fuel suppliers, and it is easier to police at the supplier end. It also lands the obligation on an industry with no supply to meet it. German biomethane output runs near 10 TWh a year against gas demand for heating of roughly 300 TWh, and green hydrogen for the domestic network exists in pilot volumes. A quota on a fuel nobody can source is a quota that moves the failure date to 2029 and the argument to a different room.

Support was cut in the same package. The Federal Funding for Efficient Buildings scheme still pays a base grant of 30 per cent of eligible costs to any owner, with bonuses on top to a ceiling of 70 per cent and a maximum of €21,000. What changes is the eligible cost cap, which falls from €30,000 to €28,000 on relaunch and then by €750 every six months until it reaches €22,000 in the second half of 2030. The 30 per cent income bonus for households with taxable income below €40,000 survives. A low income household keeps most of what it had. A middle income household loses a slice every six months for five years. The government expects the change to save over €2bn by 2030. The German Heat Pump Association warned that the two changes together would push buyers towards the cheaper fossil option at the point of sale, which is the only thing that matters when a boiler stops working on a cold February evening.

Grants alone were never going to be enough, and Germany has just shown why. The 299,000 units came against a target of 500,000 for 2025 set by the previous government. A grant cuts the price of the box on the wall once. It does nothing to what the household pays to run it for the next 20 years, and that running cost is where German electricity loses to German gas.

The measure built to fix the running cost has been delayed. From 2027 the European Union was due to start charging a carbon price on heating and transport fuels through a second emissions trading scheme, separate from the one that has covered power stations and heavy industry since 2005. It puts the cost on the fuel supplier, who passes it to the household at the meter, which is what would finally make gas heating dearer than electric heating. Poland, Czechia and others pushed for delay during the negotiations on the 2040 climate target, and the start date moved to 2028. Germany will freeze its own national carbon price through 2027 to match. The one measure designed to raise the price of burning gas at home now arrives a year late, into a market that has just been told it may keep buying boilers.

Figure 2. The year each technology stops needing a subsidy to beat its fossil competitor, buildings and heat picked out.

A few weeks ago I built an electrification timeline for this Substack, plotting the year each technology stops needing a subsidy to beat its fossil competitor. On it, new build heat pumps and induction cooking crossed that line in 2024. The retrofit heat pump does not cross until 2033, and the row carries one annotation, gated by carbon price. The gate is the second emissions trading scheme, and it has just moved a year to the right. The row that has not crossed is the whole existing housing stock on both sides of the Channel.

The market it repealed was recovering

The European Heat Pump Association’s full market report, published in June, puts 2025 sales at 2.9 million units across 21 countries, up 13 per cent, taking the installed stock in those countries to 29.3 million. Eighteen countries reported 2025 data and 14 of them grew. The four that fell were Austria, France, Norway and Poland, France because its support scheme was caught in the budget crisis and Poland because its subsidy programme kept changing shape.

The pattern is consistent enough to be predictive. Sales grew where a government left its grant scheme unchanged for two years and cut a tax or levy off the electricity bill. Sales fell where a government reopened its grant scheme and changed the rates. Installers will not hire or hold stock against a subsidy that is rewritten every twelve months, and a household making a €15,000 decision reads the newspapers and makes evidence based choices.

Set that against the requirement. The European Commission’s Electrification Action Plan, published on 17 July, asks for 4 million installations a year across the European Union by 2030, from 2.4 million in 2025 on its own narrower basket. The market is growing at roughly half the rate the target needs.

Brussels has set a target it cannot measure

The Electrification Action Plan asks member states to bring the household electricity to gas price ratio below 2.5 by 2030, and the footnote that establishes the problem does not survive a check. Footnote 20 says that only Finland and Sweden sit below a ratio of 2, on the Commission’s own calculations from Eurostat data for the second half of 2024. Footnote 25 of the same document says that Finland is absent from the relevant Eurostat dataset for households. Both cannot hold. Finland has no household gas price in the Eurostat series at any consumption band, in any recent period, so no Finnish ratio exists to calculate. Poland has none either, its gas series stopping after the second half of 2023. Cyprus and Malta are missing too.

Run the Commission’s stated basis and six countries come out below 2 rather than two. The Netherlands at 1.26, Sweden at 1.24, Bulgaria at 1.88, Portugal at 1.92, Italy at 1.96 and Austria at 1.97, on band DC electricity against band D2 gas with all taxes and levies in. On the most recent half year, the second half of 2025, the group is Sweden at 1.30, the Netherlands at 1.49, Portugal at 1.73 and France at 1.78, with Italy at exactly 2.00. Twenty four band pairings were tested and none returns a set of two.

The Dutch omission gives it away. The Netherlands sits at or near the bottom of the European table in both periods, and the plan praises the Dutch by name in its taxation section for limiting the gap between the two carriers, then leaves them off the list of countries that have closed it.

France is the other awkward case, and for the opposite reason. It charges excise duty of €33.70 per megawatt hour on electricity against €17.16 on gas, taxing the electron at twice the rate of the molecule, and it still lands at 1.78 because the nuclear fleet makes the underlying electricity cheap enough to absorb the penalty. A tax gap and a price ratio are separate quantities, and the plan sets a target on the second while writing most of its policy against the first.

The measurement problem runs deeper than the missing countries. Band DC covers a household using 2,500 to 4,999 kWh of electricity a year, which is a household that does not heat with electricity. Move up to the 15,000 kWh band, where a heat pump household actually sits, and Denmark falls from 2.63 to 1.71, the largest proportional fall of the eight countries where both bands can be calculated, because the reduced Danish duty on power used for heating only bites once the heating load is on the meter. Ireland falls furthest in absolute terms, from 3.11 to 2.10. Germany falls from 3.16 to 2.68 and Sweden from 1.30 to 0.91. Belgium moves from 3.90 to 3.86, which is what a country that gives heating no reduced rate looks like. The ratio the Commission is targeting describes the household that has not switched.

Figure 3. Electricity to gas price ratio on band DC against the 15,000 kWh band where a heat pump household sits, eight countries, with the 2030 ceiling of 2.5 marked. Ireland and Denmark clear the ceiling on the change of band alone.

Then there is the question of whether the ratio predicts anything. Set the price ratio against sales per 1,000 households for the six countries where both numbers exist and the rank correlation is minus 0.37, pointing the right way and nowhere near strong enough to carry a policy target. Lithuania has the highest sales rate in Europe, 46 per 1,000 households, at a ratio of 2.86 that already breaches the 2030 ceiling. Belgium sells 21.2 per 1,000 at 3.90, almost five times the British rate at an almost identical price gap. Germany’s record year computes to 7.2. The explanation is most likely product mix, because the Lithuanian and Nordic markets run heavily on air to air split units and the association’s own headline pairs Lithuania with Norway, where sales fell in 2025. A split unit is a €2,000 appliance bought against a comfort problem. A hydronic retrofit is a €15,000 decision bought against a running cost, and it is the second one the ratio governs.

Figure 4. Electricity to gas price ratio against heat pump sales per 1,000 households, with the Commission’s 2030 ceiling of 2.5 marked.

Rates for Lithuania and Norway are as published by the European Heat Pump Association. Belgium, France, Italy, Germany and Great Britain are compiled from published national unit sales over Eurostat household counts for 2025. Ratios are Eurostat for the second half of 2025 with all taxes and levies included. The British figure is the Ofgem cap for the third quarter of 2026 with standing charges spread over comparable consumption, so it is close to but not the same measurement.

The Commission has set four numbers and rules against the problem. Household electricity must cost no more than 2.5 times household gas by 2030, and industrial electricity no more than 2.0 times industrial gas. No member state should tax a unit of electricity more heavily than a unit of gas. Network charges are to be redesigned so that a household adding an electric heating load does not carry a disproportionate share of grid costs. And subsidies to fossil fuels are to be phased out. The direction is right and the instrument is a target on a statistic that four member states do not report, that is measured on the wrong household, and that has yet to be shown to predict a sale. That second number, the industrial 2.0, is where part two begins.

The price ratio and the grant do different jobs, and confusing them is why the German year looks strange. The ratio decides how large the market can eventually get, because it sets what the household pays every winter for the next 20 years. The grant decides how quickly the market gets there, because it cuts the price of the installation once. The scheme Germany inherited pushed hard on the grant through 2025 and produced a record year. The government that took office then cut the grant and left the ratio untouched.

Denmark and Britain, same problem, opposite answers

Denmark charges a reduced rate of electricity duty on power used for household heating, and has done for years, and the Commission’s own plan records that it has gone further, cutting electricity taxes to the permitted minimum for two years. The effect shows up in the place the headline ratio hides it. A Danish household on the standard band pays 2.63 times as much for electricity as for gas. A Danish household heating with electricity pays 1.71. Danish sales rose 36 per cent in 2025, the second strongest rise in the survey behind Germany, from a ratio that looks unremarkable until you measure it where the heat pump lives.

British sales reached 125,037 units in 2025, up 27 per cent, across a stock of 28.4 million homes. At that rate it would take 227 years to replace the current fleet. The Climate Change Committee’s seventh carbon budget requires 450,000 installations a year by 2030. That is a 3.6x gap against a run rate that took a decade of policy to build, and even 450,000 a year would need 63.1 years to get through the existing stock.

The British instruments are more stable than the German ones and several are good. The Boiler Upgrade Scheme pays £7,500, rising to £9,000 for homes on oil or liquefied petroleum gas from 21 July 2026, and stays universal with no means test. The insulation condition that blocked applications on an unimproved certificate went in April 2026. The Clean Heat Market Mechanism obliges boiler manufacturers to match a rising share of their sales with certified heat pump installations, 6 per cent in the first year and 8 per cent from April 2026, at £500 for every missing credit. The Warm Homes Plan carries £15bn across the parliament.

What the plan declined to do was move policy costs off electricity bills. Nesta’s modelling of it has a household fitting a heat pump, solar panels and a battery cutting its total energy bill from around £1,670 a year to around £670. On a standard tariff with no solar and no battery, a heat pump costs about what a gas boiler costs to run.

Then on 21 July a new government moved the ratio for the first time, and did it in the one shape that cannot work. Value added tax on domestic electricity falls from 5 per cent to zero on 1 October, worth about £41 a year on Ofgem’s typical consumption and 4.8 per cent off the electricity bill. It leaves gas alone, so it genuinely moves the number this market runs on. It also expires on 31 March 2027, excludes Northern Ireland under the Windsor Framework, and costs the Treasury about £850m.

The size of the move is smaller than the timing makes it look. On the cap running from 1 July, 26.11p per kilowatt hour for electricity against 7.33p for gas, the British household ratio is 3.91 once standing charges are spread across comparable consumption. Zero rating takes it to 3.72. The Commission’s ceiling is 2.5. Most of the fall from the 4.67 of the first quarter of 2026 happened on 1 July, when the cap put gas up 24 per cent against 5 per cent on electricity, which is the Strait of Hormuz moving the ratio by more than the Treasury did. Northern Ireland stays at 4.22 on the regulated Power NI and firmus tariffs, because the zero rate does not reach it. The next cap lands on 26 August and will move all of this again.

No household will replace a boiler on the strength of six months of 4.8 per cent, and no lender will price it. As a winter cost of living measure £41 is worth having. As heat policy it is worse than doing nothing, because it spends £850m of the money that would have paid for a permanent levy shift and buys the government a line about cutting the cost of electricity. A heat pump bought this autumn is a 20 year decision meeting a six month tax cut, and the Treasury has now made the structural fix harder to fund than it was in June.

What 60 million heat pumps do to a December evening

Europe has 28 million heat pumps today and 60 million on the REPowerEU target, and every one of them works hardest on a still December evening. That is the hour the continent is building 68 GW of new gas capacity to cover, and the hour German wind fell to 3.1 GW against a seasonal norm of 19.2 GW on 12 December 2024 while the intraday price touched €1,000 per megawatt hour.

A heat pump fleet is a large controllable load carrying two to four hours of usable thermal inertia. Sweden has built around 1.6 GW of demand response out of it through aggregation. Almost nowhere else has, because the flexibility value is not monetised for the person who owns the appliance. The distribution operator watches the peak arrive and reinforces against it. The household sees a flat tariff and heats when it feels cold.

Every heat pump installed without a controllable tariff and a settlement arrangement that pays for shifting adds to the winter peak, and somebody then buys firm capacity to cover that peak at gas plant prices. The housing ministry that paid the grant does not see that bill. The transmission operator picks it up a decade later, in connection charges and capacity payments, long after the appliance is on the wall and impossible to unpick.

Now the easy half is over

The painful point here is that the home is the part of this that should already be finished. The machines are in mass production and 300 European factories could build three times as many as they currently sell. The installers exist. The grants exist. European Union subsidies for standalone fossil boilers ended on 1 January 2025 and the buildings directive sets a phase out for 2040. On all of that, the European Union manages 2.4 million units a year against a requirement of 4 million, and the largest market in the group has just repealed the rule that made it the largest.

Industry has none of the above. No mass produced product, no installer base, and no grant scheme worth the name. Electricity is 33 per cent of European industrial energy use and has not moved in a decade, and process heat is around half of all industrial energy use.

Look at that parity timeline again. Food, brewing and textiles cross in 2028. Glass, ceramics and paper in 2031. Chemical process heat in 2036. Cement and lime in 2041. Every one of those dates sits beyond the 2030 targets written to hit them, and the entire known stock of industrial heat pumps in Germany, the continent’s biggest manufacturing economy, comes to 130 MW.

Part two is the factory.

Sources: European Heat Pump Association, market report 2026, published June 2026, covering 21 countries with 2025 data from 18. Deutsche Energie-Agentur on 2025 German heating market shares, reported by Clean Energy Wire. German Heat Pump Association on 2025 unit sales and its response to the Building Modernisation Act. German Heating Industry Association on the 500,000 unit target for 2025. KfW programme 458 for the Federal Funding for Efficient Buildings grant rates and the revised eligible cost caps. European Commission, Electrification Action Plan, COM(2026) 595, 17 July 2026, including footnotes 20 and 25. Eurostat nrg_pc_204 and nrg_pc_202, household electricity and gas prices, second half 2025. Ofgem, energy price cap 1 July to 30 September 2026. Utility Regulator Northern Ireland, regulated tariffs from 1 July 2026. Clean Energy Wire on the Building Modernisation Act and the German carbon price freeze. European Environment Agency on the delay of the buildings and transport emissions trading scheme to 2028. Climate Change Committee, seventh carbon budget. Department for Energy Security and Net Zero, Warm Homes Plan, and Ofgem Boiler Upgrade Scheme guidance. Nesta analysis of household energy bills under the Warm Homes Plan. HM Treasury announcement of 21 July 2026 zero rating domestic electricity from 1 October 2026 to 31 March 2027. Magnus Energy, Heat Pump Attractiveness Report 2026, on demand response potential and on the German industrial heat pump stock.

Figure 1 is compiled from the European Heat Pump Association fossil fuel boiler tracker of October 2025, updated for the German Building Modernisation Act of July 2026, Scotland’s New Build Heat Standard in force 1 April 2024, and England’s Future Homes Standard, laid 24 March 2026 and in force from 24 March 2027, which is therefore not counted. Countries absent from the tracker are shown as no restriction recorded and are not inferred to have none. The Netherlands cancelled a planned 2026 obligation for existing homes in 2024, and the United Kingdom dropped a proposed 2035 phase out in January 2026. Neither had entered into force, so neither is shown as a removal.

A note on baskets and on the numbers. The association’s preliminary release of March 2026 covered 16 countries and put 2025 sales at 2.62 million. Its full market report of June 2026 supersedes that with 2.9 million units across 21 countries, 13 per cent growth, and 18 countries reporting 2025 data. Figures here use the June report. The Commission’s 2.4 million covers the European Union alone, and any figure measured against the Commission’s 2030 target uses the Commission basket. Sales per 1,000 households are as published by the association for Lithuania and Norway. Belgium, France, Italy, Germany and Great Britain are compiled from published national unit sales over Eurostat household counts for 2025 and are flagged as compiled rather than published. Denmark, Finland, Sweden and Poland have no published rate and are not estimated.

Price ratios are Eurostat for the second half of 2025, band DC electricity against band D2 gas, all taxes and levies included, on a common per kilowatt hour basis. Finland, Poland, Cyprus and Malta have no household gas price in that dataset and no ratio can be calculated for them. The British ratio is the Ofgem cap for 1 July to 30 September 2026 with standing charges spread across 3,500 kWh of electricity and 12,000 kWh of gas, which is close to but not identical to the Eurostat treatment. The Northern Irish ratio is the Utility Regulator’s Power NI and firmus Ten Towns tariffs of 1 July 2026 over its own comparison consumption. The next cap is published on 26 August 2026 and will move every British figure here.

German biomethane output and heating gas demand are rounded order of magnitude figures and need firming from the German gas industry association before publication. The 68 GW gas figure is an illustrative build out anchored to announced tenders and pipelines, gross of retirements, and not a forecast.

Subscribe now

Last week AST and SVK transmission network operators (TSO) in both Latvia and Sweden signed a joint assessment document to understand the economic viability of a 700MW DC connection between the respective countries.

In February 2026 and according to ENTSO-E numbers a megawatt hour of day-ahead power averaged €155.40 in Latvia and €106.56 in southern Sweden. Those markets sit around 200 kilometres apart across the Baltic Sea. The €49 gap was the widest either side had seen since December 2022, and theoretically, for that month a 700 MW cable between them would have carried roughly €23m of value but to date that cable doesn’t exist.

The proposal to build it is called LaSGo: a two-section 500 kV VSC-HVDC link, 700 MW per section, running from the Swedish mainland to Gotland and from Gotland to the Kurzeme coast of Latvia. A feasibility study by the Italian energy consultant CESI S.p.A reported positively for both countries in early 2025. THEMA Consulting Group is now working through the economic benefits assessment, due by the end of this year. Both transmission system operators, AST and Svenska kraftnät, remain formally engaged.

My view, stated up front: the project is worth backing, the commercial case rests on value rather than necessity, and the decision will be taken in Stockholm, by a government that does not yet exist, for reasons that are mostly about Sweden.

The spread is structural, and it is already eroding

Pull the day ahead data from ENTSO-E for the last seven and a half years and the pattern is stable. Lithuania has out priced SE4 every single year: €46 against €40 in 2019, €34 against €26 in 2020, €90 against €81 in 2021, €230 against €152 in 2022, €94 against €65 in 2023, €87 against €50 in 2024, €85 against €60 in 2025. Latvia tracks Lithuania to within a euro or two throughout due to the strong interconnection of the markets. Of the 90 months since January 2019, 96% priced the Baltics above southern Sweden.

The cause is not mysterious. SE4 sits downstream of a Nordic system stuffed full of hydro and nuclear, Lithuania has imported as much as three quarters of its power since Ignalina closed, and even after the renewables build out imports covered roughly a third of all Baltic consumption in 2024, with gas or imports setting the marginal price in most hours. February 2026 was the pattern at full stretch: really cold (it was very cold I can tell you), still weather, €155 in all three Baltic zones, €137 in Finland, €116 in Poland, €107 in SE4.

The premium is decaying, and this spring the decay became a change of regime. The monthly average spread was €38 in 2024, €25 in 2025, and €12 across the first half of 2026, as Baltic solar arrived in volume. Since March, 53% of all hours have cleared with the two zones within €2 of each other, against 31% of hours in 2024. For planning purposes we now assume a winter-concentrated premium: €20–40/MWh in the cold months, near-parity across the solar season. The future eroders are not equal, and it pays to be precise about which is which. Solar has done the observed damage. Offshore wind will do less than commonly assumed because the spread lives in low wind hours, so new turbines mostly add output to hours the spread has already left. The wires are different. Harmony Link and LaSGo move power in every weather, including the still, cold hours where the remaining premium is concentrated. The projects that erode the spread hardest are the cables themselves. Every one of those projects eats the spread that justifies it. An interconnector erodes its own revenue, although we need to find different mechanisms to enable more with different models. The commercial question is whether it does so before or after it has paid for itself and so looking at the NordBalt can give us some useful insights.

NordBalt already ran the experiment

There are a few sceptics of LaSGo but they need to explain NordBalt, because NordBalt is the same trade a decade earlier: 700 MW of HVDC between SE4 and the Baltics, Klaipėda to Nybro, 450 kilometres, built for €552m and energised in December 2015.

The hourly flow data describes an asset nobody quite intended. In 2024 NordBalt moved an average of 605 MW against a 700 MW rating, 87% utilisation, above 90% of capacity in two thirds of all hours, toward Lithuania in 95% of the hours it flowed at all. It behaves, or behaved like a mid-sized power station lying on the seabed, and Lithuania treats it as such. Availability was 97.4% in the first half of 2025, the gaps mostly Swedish side emergency disconnections.

The economics follow on and seem to work. Take the hourly price difference between the two zones, multiply by the hourly flow, and the implied congestion value on NordBalt comes to roughly €1.1bn between 2019 and 2025, €218m of it in 2024 alone. That is a market value measure rather than an audited revenue line, and the accounting reality is smaller and messier, but the reported figures point the same way, for example, Litgrid booked €17.2m of congestion management revenue in the first half of 2025 and recycled it into tariff reductions, and it credited NordBalt and LitPol with cutting Lithuanian market prices by 13% when they entered service. On any of these measures a €552m cable repaid itself within a decade.

Since November 2025 the SE4–LT border also carries the first cross-border Electricity Price Area Differential (EPAD) product in Europe, with at least 10% of NordBalt’s capacity, around 70 MW, released annually for hedging. Small, but it is the start of a hedgeable Baltic spread market and it should start to change what a second cable would mean for anyone financing Latvian generation assets.

The round trip pays Sweden’s hydro

The 2024 exhibit is however already dated, and we should look closely at what replaced it. Through 2025 NordBalt stayed a conveyor at lower intensity: 72% utilisation, 4.4 TWh moved, still almost all one way. In March 2026 the pattern changed. Since then the cable has started to run a daily cycle; it hasn’t seen a decade of operation. 200–300 MW toward Sweden through the middle of the day, imports at night and through the evening peak. Exports were now a third of H1 2026 volume, from 1% in 2024.

The obvious reading is that Lithuania now sells cheap solar west at noon and buys cheap Nordic power back at night but only half of that is really true. The first part, the midday leg, is real. You can see that Baltic solar clears €8–13 under SE4 in the solar hours and the wire carries that lower price solar west, sadly though the return leg pays no such margin, because by nightfall the two zones price look to be within a euro of each other, and in the evening peak Lithuania is slightly the dearer zone and imports anyway, meaning that the energy is coming back at full price.

The intraday margin has not disappeared. It has moved inside Sweden. Reservoir hydro holds back water or even charges while cheap Baltic imports flood the midday market, then generates into the €110–140 evening peak. The wire has become the Baltic grid’s access to the Nordic battery, and the battery’s owner keeps the round trip. The other winners are quieter: Latvian and Lithuanian solar owners, whose midday price floor is precisely that 700 MW exit (Lithuanian prices went negative in only 1.8% of hours this spring, despite the build-out), and SE4 consumers, who take midday power roughly €10 under their alternative. The TSOs’ congestion rent, meanwhile, is collapsing into a winter product: about €48m of implied rent in January and February 2026 together, and €16m across the four and a half months that followed.

That seemingly rather than a permanent spread, is what a second cable now buys: a curtailment backstop that should allow Baltic renewables financeable, access to hydro flexibility for a solar heavy system, and a scarcity conveyor that earns most of its money in exactly the weeks Stockholm cares about SE4 prices.

So adequacy is the wrong test?

Does Latvia need LaSGo to keep the lights on? No, not at all. Post-synchronisation, the Baltic system balances. One structural point needs stating precisely. Latvia is not short of interconnectors. Its borders with Estonia and Lithuania are real interconnections between separate bidding zones, and they bind: the Latvia–Lithuania border split prices in under 1% of hours in 2024, 2% in 2025, and 6% so far in 2026, with splits as wide as €145/MWh. What Latvia lacks is an interconnector that leaves the Baltic pool. Four exist today: two Finnish cables into Estonia, NordBalt and LitPol into Lithuania. Harmony Link makes a €923m fifth from 2030, again into Lithuania. Those external links are the doors, and Lithuania will hold 1.9 GW of them. Latvia holds none. Everything it buys from the west crosses at least one more border that can congest on the way. Any adequacy analysis will say the region can manage without a Latvian cable, and opponents will quote that analysis correctly.

The case for LaSGo rests on 4 quieter arguments.

Geography first. Every one of those interconnectors lands in Lithuania or Estonia. Power bound for Riga from the west transits networks that congest, and the internal north/south corridors were never designed to feed Latvia from Klaipėda. LitPol makes the point, since synchronisation its commercial capacity has averaged roughly 170 MW export and 150 MW import, recovering toward 500/353 only from 2027 under Litgrid’s published concept. Latvia’s access to western power is a function of its neighbours’ grids and their own congestion management, Riga gets no say in this..

Bankability second. AST’s position is that above roughly 1,000 MW of installed variable renewables, Latvia requires new interconnection. The Artelys modelling shows curtailment emerging at 2,000–3,000 MW of installed capacity under 2030 conditions. It’s difficult to finance a Latvian wind farm in a market where surplus hours have no exit. A 1,400 MW corridor west is the difference between a renewables pipeline and a renewables press release. CESI’s study puts Latvian power-sector CO2 down 35% with the link in place.

Resilience third. A Latvia to Sweden subsea fibre cable was cut in January 2025, and NATO expanded Baltic Sentry in response. Harmony Link runs overland past the Suwałki corridor. Route diversity across the Baltic seabed is now a security argument that finance ministries accept, and LaSGo is a new route, not a reinforcement of an existing one.

And fourth, the winter spread, which is the subsidy free part of the case and, as the previous section showed, the part that survives the solar build out. The precondition on the Latvian side is that the 330 kV Ventspils–Brocēni–Varduva reinforcement has to be built for the new interconnector to be worth landing.

Stockholm holds the cards

None of the above matters if Sweden says no, and it seems that sadly Sweden has recent practice at saying no. On 13 June 2024 the Swedish government rejected the concession for Hansa PowerBridge, a 700 MW link from Hörby in SE4 to Gustrow in Germany, and Svenska kraftnat formally ended the project. The stated logic was specific: southern Sweden runs a production deficit, Germany’s single bidding zone makes flows inefficient, and the cable risked importing German prices into SE4. The government left the door open should conditions change. Conditions in Germany have a precondition of their own, bidding zone reform, which Berlin is yet again resisting, so a Hansa style revival if it ever comes will at best be a mid-2030s project more than likely north of €800m–1bn.

Hansa is the precedent that the team running LaSGo has to answer, because as far as I can understand, at the surface the argument is identical. Why would a government that refused to export from SE4 to Germany approve exporting from SE3 and SE4 toward the Baltics?

Three reasons, in the order a Swedish minister would hear them.

• Gotland comes first. Section 1 of LaSGo is a mainland-to-Gotland cable, and Gotland’s existing mainland connection is old, thin, and inadequate for the island’s own electrification plans. Svenska kraftnät gets a domestic reinforcement it needs anyway, part-funded by an interconnector budget.

• Security comes second. A high-capacity link to a NATO ally on the far shore of the Baltic reads very differently in 2026 than a commercial cable to Germany did in 2024. Price effect comes third, and it is smaller than the Hansa case: the premium is decaying, the counterparty is a market of 1.9 million people rather than the German industrial base, and NordBalt now delivers Baltic solar into Sweden through the middle of every spring day. A cable into a renewables-heavy Baltics is partly a Swedish import asset, which no honest Hansa brief could claim about Germany.

• The timing is the delicate part. Sweden holds a general election in September 2026. The THEMA study lands weeks after the new government takes office. A permitting decision on a power export cable will be among the first energy files on a new minister’s desk, framed by whatever SE4 prices did that winter. We put the probability that LaSGo clears the Swedish gate this decade at 40–50%: better odds than a Hansa revival, precisely because the Gotland framing lets Stockholm approve it as domestic policy with a Baltic dividend.

What I would do

For a Latvian or Baltic energy business, four actions follow.

Back the study loudly. The THEMA assessment is the last cheap phase of this project. Supportive intervention now, with real load and generation data, costs little and buys standing later.

Push the Brussels track. LaSGo is a textbook Project of Common Interest candidate: cross-border, positive for both parties per CESI, security-relevant. PCI status and Connecting Europe Facility funding in the 2027–28 window are what turn a study into a project, and the presence or absence of a PCI application will be the earliest reliable signal of whether the TSOs mean it.

Fight the cost allocation. The welfare gains land asymmetrically: Latvian consumers and Swedish producers win, Swedish consumers lose a little. A default 50/50 split would be a poor reflection of that, and the cross-border cost allocation negotiation in 2027–28 is where the project’s politics and its economics meet. We should rightly expect Latvia to pay more than half, and treat that as acceptable up to the point where the consumer surplus says otherwise.

Hedge meanwhile. Energisation, if every gate opens, is 2034–36 against a cost of €1.0–1.3bn and a congested global HVDC supply chain. That is eight winters of exposure with no new steel in the water, and winter is where the price risk now lives. The SE4–LT EPAD market is thin, but it exists, and scaling it, particularly for the Q1 products, is the only tool that manages Baltic price risk on a timescale shorter than a permitting cycle.

The all hours spread is going; what remains is a winter scarcity product, a midday price floor for Baltic solar, and access to Nordic hydro flexibility, and that bundle is worth a cable. NordBalt shows a Baltic HVDC link pays back inside a decade even on conservative measures, and its 2026 data shows the two-way future arriving early. The binding constraint is a Swedish cabinet decision in 2027 or later, and the single most useful thing anyone on this side of the water can do is help Stockholm see the cable as Gotland’s, with Latvia attached.

Prices and flows computed from ENTSO-E Transparency Platform hourly data, January 2019 to July 2026. Project facts from AST, Litgrid, Svenska kraftnät, CESI, and the Swedish government’s decision of 13 June 2024. Probability judgements are the author’s personal thoughts on the topic.

No take the page and turn it upside down. That inverted hump becomes a trough where the peak used to be and that is now the price of power across most of Europe. Where the sun is highest the price is lowest, and in the middle of a German day it is close to nothing. One day in four in the first half of 2026, the market at one o’clock cleared at €1 a megawatt hour or less. A megawatt hour of German solar now earns half the average power price. A Spanish one earns 44 per cent of an average that has itself nearly halved. In almost every corner of the continent, generators are working out that midday is worthless.

^{[Figure 1: “The same shape, upside down”]}

The day has split into two markets that no longer have much to do with each other. A midday that has come loose from gas, and the hours on either side still loosley bolted on. The single most-quoted number in European power, the baseload average, hides the split completely. Germany’s baseload mean fell just 4 per cent across the first half of 2026, from €103.4 to €99.3 per megawatt hour. Over the same period the gap between the cheapest stretch of the day and the most expensive more than doubled, from €42.7 to €93.5. The average sat still while the day came apart underneath it.

Share

Midday came loose from gas

The cleanest way to see the split is to measure how tightly each hour tracks the overnight block, the 00:00 to 06:00 window where gas plants set the price and little else runs. In the first half of 2023 the midday hours still moved with the night, at a correlation of around 0.5 to 0.6. Cheap gas pulled the whole day down and expensive gas pushed it back up, and midday came along. By the first half of 2026 that link had fallen to near zero or zero. The hours from 10:00 to 16:00 no longer know nor worry about what gas costs.

They no longer need to. A gas plant in the first half of 2026 needed something like €100 to €110 a megawatt hour to cover its fuel and carbon, gas around €45 and carbon near €80, and three quarters of the midday window now clears below €100. So gas stay off, and something else sets the price. That something is solar, bidding at near zero to keep running, and the price it produces is a number no thermal plant would ever quote.

So what does that look like from a trading desk is a midday that clears near nothing on any bright day. At 13:00, one day in four in 2026 settled at €1 or less. Across the half year the German market spent 221 hours strictly below zero, against 66 two years earlier. (expect this to get worse, if not through commercial scale solar, domestic and balcony will continue to destroy these hours) The deepest single hour averaged minus €250, and inside it sat a single 15 minute period that documented a minus €413.8, both real at different resolutions. Since the day ahead moved to quarter hour settlement in October 2025, those spikes bite much keener and more often than the hourly figures alone let on. Negative prices were in the past a simple matter of a winter storm and holiday event. They are now very regular fair weather lunchtime one, exaggerated by global warming and hotter European summers.

The cause is not in dispute. German solar generated 43.8 TWh in the first half of 2026, up 48 per cent on the 29.5 TWh of 2023. All of it loads on to the network inside the same six hours, and it arrives whether the market wants it or not. The price it earns while running, its capture price, fell from €77.3 to €49.6, down 36 per cent. Measured against baseload, the capture rate dropped from 75 per cent to 50 per cent. Solar built almost half as much output again and sold it into a market that now pays it half the average. That is the whole of the cannibalisation story, and two years ago it was the argument of an earlier piece.

^{[Figure 2: “More power, less money”]}

What has changed since then is the shape of the loss. It used to smear across the day, a few euros off every hour. It now concentrates in the hours the sun is up and leaves the rest alone. The average is intact. The distribution has simply come apart.

The evening still runs on gas

The other end of the day went the opposite way. The evening block, 18:00 to 22:00, still runs on gas, and its mean average rose from €129.5 to €146.6. At 20:00, one evening in ten in 2026 cleared above €225. When the sun drops and demand holds through supper, Germany falls back on the same plants it always did, and those plants price off a European gas market that prices off the world.

That is the Hormuz line in a German electricity bill. Gas in the first half of 2026 was carrying a Gulf risk premium that had nothing to do with the weather in Bavaria. Strikes on Qatar’s Ras Laffan complex and the effective closure of the Strait of Hormuz took close to a fifth of the world’s liquefied gas off the market inside a few weeks, they were paying an American war tax. The Dutch benchmark ran above €70 a megawatt hour in March before easing back toward €40 to €50 as the immediate crisis passed. A tanker scare in the Gulf still reaches a Munich supper. It no longer reaches a Munich lunch, because at lunch the gas plants are switched off and the price is set by panels that never heard of Hormuz.

The widened spread is easy to blame entirely on solar, and that reading is wrong. The midday floor falling out is solar. The evening ceiling lifting is gas. Both ends of the day moved, in opposite directions, for unrelated reasons, and the spread is the sum of the two. Anyone selling this as pure renewable cannibalisation is giving you half the picture and the less interesting half.

Spain shows where this ends

Spain is the same process a few years further on, and it shows two different failures wearing one label. Spanish solar generated 29.1 TWh in the first half of 2026, up 46 per cent, and its capture price fell from €69.6 to €22.1, down 68 per cent. Capture against baseload went from 79 per cent to 44 per cent. The count of hours below zero went from none at all in 2023 to 620.

^{[Figure 3: “Two markets, two failures”]}

On a chart the Spanish and German troughs look alike, and the resemblance ends there. In Spain the median negative hour is minus €0.60 and the floor is minus €12.6, with 121 hours sitting at exactly zero. Spanish solar earns almost nothing at midday because there is simply too much of it arriving at once with nowhere to go. The price falls to the floor and rests there. More wires out of the peninsula and more storage inside it would clear most of the problem, and both are being built, slowly.

Germany is the harder case, and the more expensive one to fix. The median negative hour is minus €12.2 and the floor is minus €250, far below anything Spain reaches. At German midday solar is worse than worthless, because a large subsidised rooftop fleet keeps pushing power into deeply negative prices when a rational plant would switch off. It does so because it cannot see the price and has no reason to answer it. The support scheme just keeps paying for generation rather than for value, so the panel keeps generating while the market pays to make it stop. Storage helps at the margin, a bit. What it needs is a change in how the fleet is paid, and that is a political question, its certainly not; as constantly claimed… an engineering one.

A growing share of that fleet sits behind the meter, where the distribution operator just cannot see it. It shows up only as demand quietly disappearing at midday. Every month more of it connects, the hole in the visible load curve deepens, and the operator’s forecast drifts a little further from the day the network actually gets. Nobody can dispatch what nobody can see, and nobody can pay it to stop either. The market is being reshaped by capacity that appears on no screen, and the reshaping compounds monthly.

Storage is the missing bridge

What would connect the cheap middle of the day to the expensive end of it is storage, and there is almost none of it. The arbitrage sits in plain sight. A battery that charges near €1 around 13:00 and discharges above €225 around 20:00 captures more than €200 a megawatt hour on a good day, every day the weather cooperates, which in summer is most of them. No other trade in the European power market is this large or this repeatable.

^{[Figure 4: “The unbuilt bridge”]}

It is not being taken at anything like the scale required. Europe installed 77 GWh of battery storage by the end of 2025, a tenfold rise in four years on SolarPower Europe’s count, and still nowhere near enough. The same body puts the requirement for 2030 near 750 GWh, another tenfold on top. The gap between the two markets is an unbuilt asset worth billions, and the meter runs on every day it stays unbuilt. Part of the delay is connection queues clogged with projects that will never be energised. Part of it is that the revenue stack for a battery is still mostly frequency response and balancing, thin markets that saturate fast, rather than the clean midday-to-evening arbitrage the price curve is now begging for. The signal is there. The wiring between it and the concrete is missing.

Batteries also carry less of the answer than the current enthusiasm assumes. They move energy across a few hours, and the whole 750 GWh target amounts to a few hours of European demand on a summer afternoon. The trap that has closed on Germany and Spain is already sprung in the Netherlands and is opening under every market that builds solar faster than it builds flexibility, and it opens quicker each year: Germany went from 66 negative hours to 221 in three years, Spain from none to 620. Against that pace, a solution set on a 2030 timetable is a solution for a different problem. The full fix is wider than storage: a support scheme that pays for value rather than volume, demand that moves to where the power is, wires that move the power to where the demand is, and batteries on top of all three. Every part of it is currently slower than the thing it is meant to fix.

None of this makes the midday power worthless to everyone. Worthless is the generator’s word for it. To anyone on the other side, a battery, a flexible load, an interconnector, that near-zero price is an invitation, and the value that left the panel lands on whoever can absorb it. The catch is how little of it storage can or is catching. A battery earns by cycling, and against this shape it gets one good cycle a day, charge in the cheap middle, discharge into the dear evening, and by one o’clock it is full with a single evening to sell into. It banks that one spread and leaves the rest of the floor untouched. The surplus is an energy volume measured in terawatt hours, and a fleet rated in gigawatt hours taking one cycle a day is the wrong tool to move all of it. Storage narrows the two ends of the day. The cheap middle stays cheap.

The money has not caught up

For anyone lending against a solar project, the number that clears the loan is the capture price, and the capture price has moved a long way from where the models put it. A merchant Spanish project underwritten in 2023 assumed it would earn something close to 80 per cent of baseload for the life of the debt. It now earns 40-45 per cent, and baseload itself has fallen 45% per cent. The revenue line that comfortably passed the debt service test three years ago is out by close to a factor of two, and it is out in the downward and pain filled direction.

Some of that is already surfacing in refinancing conversations that are going quietly badly. Most of it is still sitting undisturbed in models nobody has reopened, because the headline baseload number looks reassuringly stable and few people have pulled the capture price out from under it. A power purchase agreement signed against the old capture rate now looks generous to the offtaker and thin for the developer. The next wave of merchant projects will be underwritten against these numbers rather than the old ones, which means fewer of them clear, which means the build rate that caused the glut starts to slow on its own. The market is beginning to price its own cure, painfully and late.

The baseload average, the most-quoted number in European power, now blends two markets that have nothing to do with each other and reports the blend as if it still described a single day. In Germany that average fell 4 per cent. The middle of the day fell to nothing.

A note on the numbers. Prices are ENTSO-E day-ahead for the DE-LU and ES zones, first half of 2023 against first half of 2026, local clock time. Since 1 October 2025 the day ahead has settled in 15 minute periods, and every hourly figure here is the arithmetic mean of its four 1/4s. Counts of hours below zero are strictly below, on those hourly means. Germany’s deepest hour, minus €250, is the hourly mean that contains a single 15-minute period of minus €413.8. The day-shape figures are drawn from the hourly series itself, every day of the half year, and the block means quoted in the text are as published. Solar capture is the generation weighted price measured against a time weighted baseload. Storage figures are from SolarPower Europe’s EU Battery Storage Market Review 2025.

Subscribe now

The country that fills up

I started to write this as a study into what happens when the DNO doesn’t pick up the shift fast enough, when it does not see that almost everything from April to October (longer in sunny and warm climates) will happen out of their sight and behind the meter. Some are seeing it, but it’s the disruptors not the institutions. That is a subject we can come back to for sure and it’s something I have spoken about a few times to those willing to listen.

What I found in the data I was looking at really shocked me and its a people movement and migration story that will impact the way our electrical networks should and will change in the future.

I live in a country that is filling up, and I have lived in it long enough to have stopped noticing. I detest the term because it’s been overtaken by the right wing and i dont want nasty politics in my writing. However, the schools are fuller than they were, the trains are fuller, the estates on the edge of every town have crept a field further out and none of it feels like a trend. It feels like weather. It is only when you draw it that you see it is not weather at all, but something closer to a tide, and that the tide is not coming in everywhere, in fact like the tide, it’s out in an awful lot of places.

Shade every council in the United Kingdom by how its population changed between 2011 and 2024 (the useful dataset I had), and the whole country turns green. Not the cities alone, not the prosperous south alone, but nearly every corner of every nation. England glows and grows almost throughout. Wales and Scotland are mixed but mostly rising. Northern Ireland is green. Of the 361 principal councils, 340 grew. Only twenty-one shrank, a thin scatter of the remotest countryside and the most expensive squares of central London.

This is the picture my friends and family live inside, and because they live inside it, they assume it is the picture everywhere.

It is not the picture everywhere. To see that, I had to stop looking at my own country and climb until the whole continent sat in one frame.

^{Every UK council, shaded by population change 2011 to 2024, on the colour scale used for the rest of Europe in this essay.}

Where the green stops

From high enough up, the single wash of green breaks clean in two. There is a line, and it falls diagonally across the map from the north west to the south east, and on one side of it almost every one of the places are filling and on the other they are draining away, population falling away at a rather alarming rate. I need to add that Spain is also suffering badly.

The filling side is the one I come from. Britain, Ireland, the Nordic capitals, the Low Countries, Switzerland, the Atlantic coast of France and then, away to the right of the frame there is Turkey. Then the colour turns. Inland Spain reddens. The Italian south reddens fastest of anywhere in “old Europe”. The whole of the former eastern bloc, the interiors of the Baltic states(with a mention to Tallinn, Riga and Vilnius, Romania, Bulgaria, the Balkans and Greece are losing people region by region, a slow haemorrhage that no headline now reports because nothing happens on any single day and honestly we have stopped watching. Now there are some that will point to recent success, Vilnius is growing and Lithuania is attracting back more of its diaspora but its slow and fleeting. growth is also fuelled here largely by recent immigration to its capital.

This was the thing I found hardest to believe. Half of Europe is not slowing down. It is emptying.

I wanted to understand what was driving this data set, I wanted to understand why my own side of the line fills so completely before I crossed to the other. Britain is not a typical case. It, with it seems, Ireland is the extreme case, and the extreme case is where the mechanism shows itself most plainly.

Migration became the whole story

Population change is not a mystery. It follows one plain identity. A place gains people through births and loses them through deaths, which together are natural change, and it gains or loses everyone else through migration, from abroad and from elsewhere in the same country. That is all. There is nowhere else for a person to come from.

If you do the sum for the United Kingdom across these thirteen years and the answer is severe. Of a total official growth of 5,996,292 people, net international migration supplied 4,509,935, just over three quarters of it. Natural change, births minus deaths, supplied only a quarter, and it is fading in front of us. In 2023 it turned negative for the first time in the modern record. More people died than were born. In the year to the middle of 2024 the population rose by 755,000, of that 750k a measly 16,000 came from births outnumbering deaths. Everything else was migration!

^{What drove UK population change each year. Natural change, in navy, collapses to nothing. Net migration, in green, climbs above 900,000.}

The births are failing quietly and almost everywhere. The fertility rate for England and Wales fell to 1.41 children per woman in 2024, the lowest since the series began in 1938. Draw natural change across the country and you can watch the growth engine stall and, it seems that successive governments have understood this and rather than laying it out have chosen to open the UK to immigration in a hope to keep production moving forward. That thesis, that quiet idea has ultimately failed. whilst our population has increased, growth per capita continues to slide unabated.

^{Where deaths now outnumber births, 2011 to 2024. Red across Scotland, Wales, the South West and the coasts. Green survives only in young city cores and around London.}

In 160 of the 361 councils, more people now die each year than are born. If two thirds of a country can no longer replace itself, and yet almost all of it is growing, the growth is arriving from outside. It is arriving from the world, and it lands hardest in the cities, London and its surroundings, Bristol, Birmingham and Manchester. Almost every region is up, with just the National Parks and The Highlands seeing no real change

^{Net international migration as a share of each area's population. The gains concentrate in the cities, London and its surroundings, Bristol, Birmingham and Manchester.}

A machine with two valves

Here the simple story needs one correction that politicians will have, and it is one the most revealing thing in all the data. Nationally, growth beyond births is international migration, because movement between British places cancels out across the whole country. But locally it is not a filling at all. It is a churn, people are not “running to the cities” like we think, the cities are filling from international immigration.

Growth is not the same as youth

I had assumed all this movement would make the growing places young. It doesn’t, quite, and the way it fails to is a subtle part of the story.

Migrants are overwhelmingly young, so the areas that receive them do skew younger, and their births hold up better against their deaths. Plot the two against each other and the link is clear.

^{Each council: international migration against natural change. A clear positive link, correlation +0.45.}

But the mechanism is not the one the headline implies. Migration correlates more strongly with a low death rate, at minus 0.50, than with a high birth rate, at plus 0.35. In regions where migration is highest a natural surplus runs because they are young and hardly anyone dies in them, not because everyone is raising large families. Youth is the hinge, and it works chiefly by holding down deaths.

And growth on its own does not make a place young at all. Only 35 per cent of the 340 growing councils are younger than the national median age of 41. Growth arrives in two entirely different forms. One is migration driven and young. The other is retirement driven and old. North Norfolk, Dorset and the Isle of Wight are all growing, and all greying.

^{Population change against median age. The deep-green young growers sit low; the pale amber retirement growers sit high. Colour is international migration.}

The whole of this fits on one nice single ring. Every council, sorted by growth, coloured by age.

Every metropolitan French department, ringed the same way as the UK. A long shrinking tail through the top left where the British ring had almost none.

France has held a higher birth rate than most of the continent, so its decline is gentler than Poland’s or Italy’s. But the split is real and it is widening, a milder rehearsal of the whole European picture. Beyond it, further down the red half of the map, the emptying grows absolute. Villages in inland Spain and the Italian Mezzogiorno and the Bulgarian interior where the youngest resident is past fifty, where the school closed a decade ago, where the last shop keeps the lights on out of habit. The map tells a real story here.

Two Europes

Stand back from all of it and one shape remains. Europe is separating into two populations. A north western band, Britain most of all, grows almost everywhere, no longer through its own births, which have collapsed, but through migration from abroad, which now does all of the work. A southern and eastern band, from inland Spain across Italy and the Alps to Poland, the Baltics, the Balkans and Greece, ages and empties, its young leaving and too few arriving to take their place.

Britain is not exceptional because it grows. It is exceptional because it grows so evenly, in every corner, in a way that only heavy and sustained migration can produce against a fertility rate of 1.41.

Take the migration away and Britain would drift toward the red half of the map within a generation, because it can no longer make its own people maybe this is teh idea, to keep migration high for a growth economy. if this is the case something needs to change. The population growth is real, it is broad, and it is borrowed. Whether that is a triumph or a warning is not a question a map can answer. What the map can do is show you, plainly and in one colour scheme, that the two Europes are pulling apart, and which side of the line each place is on.

Which brings me full circle, back to the wires

Every journey I take ends at the same place, at the wires over head and beneath the road, because in the end demography is a load flow problem waiting to be recognised. I believe, and I believe it firmly, that we are about to watch almost all domestic generation and storage disappear behind the meter. Rooftop solar, a battery in the garage, a car that charges overnight and discharges at the peak. To the household it is liberation. To the distribution and transmission operator it is a blindfold. The demand they have modelled for a century starts to vanish from view exactly as the shape of demand becomes harder than it has ever been.

This is where the population maps stop being a curiosity and become a planning input. A network operator who reads the tile map as a picture of the past is already lost. It is a forecast. Where people pile in, in the migrant cities and the growing south, domestic connections, heat pumps, electric cars, warehouses and data halls will race one another to the front of the queue, and local networks built for a smaller, older, thinner load will strain and then fail to keep up. Over-population, in the plain sense of more people and more machines than the copper was sized for, becomes the defining stress on regional networks, and it will not arrive evenly, because the people are not arriving evenly.

Fewer births and shifting migration do not only mean fewer homes. They mean fewer businesses and less industry, because industry cannot run on a workforce that is not being born or is moving away.

That is the half of the story the network rarely prices in. The emptying regions do not simply lose households. They lose the commercial demand that sits on top of households, and then they lose the industry that cannot replace its retiring workers, and the load curve sags at all three levels at once. What is left behind is a paradox that ought to interest every planner: a grid with spare capacity, cheap land, and no residents left to object.

Which is the beginning of an answer, not just a lament. The demographic voids are the natural estate of the things that need power and space but not people. Data centres do not need a maternity ward or a labour market. They need a substation, a fibre route, water, and quiet. The very regions that the young are leaving, across the north, across the emptying east and south of Europe, are the regions where the headroom is opening up. If the population is going to concentrate the strain in one set of places, the sensible response is to route the loads that do not care where they live into the other set, and to plan the network deliberately for that redistribution rather than being surprised by it.

So the two maps, the one of people and the one of power, are really the same map seen twice. A distribution or transmission operator who wants to know where to reinforce, where to hold back, and where the next gigawatt of demand can be planted without a fight, could do worse than start with net migration and the birth rate, because those two numbers now tell you more about the future of the wires than the load history ever will.

In a Baltic energy crisis the power to act at the speed of the shock is national, not European, and the fastest, most capable national actor is Poland, which is also the geographic hinge through which the Baltics reach the Continent. Poland will spend 4.8 per cent of its output on defence in 2026, the highest in NATO, fields the largest army in the Union, and is moving toward roughly 1,100 tanks. That speed makes Poland either the region’s stabiliser or the single market’s sharpest stress, depending on whether it holds the connection open or protects its own supply first. Europe’s answer to the moment, solidarity, exists in law for both halves of the system, gas under the 2017 regulation and electricity under the 2019 risk-preparedness regulation, and both share the same flaw. Each fires only through pre-agreed bilateral arrangements and negotiated compensation, each falls back to ad hoc agreement in the crisis, and neither has ever been triggered. The electricity regulation forbids members from curtailing cross-border flows, the exact move a cornered government makes, and gives itself no fast way to stop them. The technical clock can be fast, as Ukraine’s weeks-long synchronisation under fire showed. The political clock is the permanent weakness. A compiled open-source record of infrastructure incidents since 2022 shows the tempo rising sharply and Poland, the hinge, as the most-targeted country in it.

This is a stress test, not a forecast.

By the time a Baltic energy crisis is a week old, the market has done its work. The price has moved, the bill has landed, and the fork between an expensive month and a systemic one is open. The live question is who can end it. The actors who can move at the speed the crisis demands are national governments, and the fastest and most capable of them sits exactly where the crisis begins.

Poland is the hinge in two senses at once. Physically, the Baltic states reach the Continent through Poland and almost nowhere else. The single synchronous tie runs Lithuania to Poland, the GIPL gas pipe runs Lithuania to Poland, and the land corridor NATO would reinforce through crosses Polish soil. Cut Poland out and the Baltics are an island. Politically, Poland is the fastest-moving national-security state in Europe. It will spend 4.8 per cent of its output on defence in 2026, the highest share in NATO, ahead of the Baltic trio at 4, 3.7 and 3.4 per cent and far ahead of France and Germany at 2. It fields the largest army in the European Union and is buying its way to roughly 1,100 tanks, more than Germany, France, the United Kingdom and Italy combined. When a crisis turns on how fast a government can act, Poland has built the machinery to act fast.

That speed cuts both ways, and which way it cuts decides the crisis. A Poland that moves fast to hold the corridor, keep the Baltic connection live and coordinate the eastern flank is the stabiliser the region has, the one actor able to do in hours what the Union does in weeks. A Poland that moves fast to protect its own supply and its own consumers first is the fragmenter, because in a coupled market every national defence exports the shock next door. The same national-security reflex that makes Poland the region’s fastest shield makes it the single market’s sharpest stress. The crisis resolves in Warsaw rather than Brussels, one way or the other.

What that looks like in the first days is a series of Polish decisions that no one else can make in time. Whether the power keeps flowing east across the LitPol tie or is held back to steady the Polish system. Whether gas keeps moving north through GIPL or is kept for Polish storage. Whether Polish reserves are shared into the Baltic balancing market or hoarded against a hard Polish night. Each of those is a national call, taken in hours, and each one either holds the region together or begins to pull it apart. None of them is Brussels’s to make, and by the time Brussels could form a view the calls have already been taken.

This is where the crisis collides with the thing Europe promises for exactly this moment, which is solidarity. On paper the Union has two solidarity mechanisms, one for each half of the energy system. Gas solidarity, under the security-of-gas-supply regulation of 2017, obliges a member state to send gas to a neighbour in an extreme emergency. Electricity assistance, under the risk-preparedness regulation of 2019, obliges member states to offer each other help in an electricity crisis, in the same declared spirit of solidarity. Between them they cover both waves this series has followed. And they share a flaw that runs to the root.

Both rest on the same slow machine. Neither fires automatically the moment a crisis is declared. Both require member states to have agreed, in advance, the bilateral or regional measures and the technical, legal and financial terms, including who pays whom and how much, before any help moves. Where those agreements do not exist, both fall back to the same thing: agree ad hoc, in the crisis, and settle the compensation afterwards. Gas solidarity was plumbed so thinly that the Union had to write in default rules and a compensation formula across 2024 and 2025, because so few countries had signed the bilateral deals that make the promise real, and it has never once been triggered. Electricity assistance rests on a non-binding Commission recommendation from 2020 and the same web of member-state agreements, with a fresh cycle of national plans only due at the start of 2026. Solidarity, in law, is a framework for negotiated mutual assistance over weeks. It cannot reach into the first afternoon.

The electricity regulation goes further, and the further it goes the clearer the problem becomes. It was written to stop member states doing the very thing a cornered government will do, curtailing cross-border flows to keep power at home, and it says so plainly, in the language of ruling out inappropriate state interventions and avoiding undue curtailment of cross-zonal capacity. The rule against fragmentation is real and on the page. What it lacks is any fast way to enforce itself against a national government acting in hours to protect its own voters. Europe has written down that members must not break the single market in a crisis and has given itself no way to stop them at the speed a crisis moves. That is the paper shield again, in its purest form.

The crisis this series tests would not arrive as a clean invasion with a marked start line. It would arrive as hybrid pressure, the kind already running against the Baltic states and Poland, and the pattern is documented. In late October 2025 Latvia’s security service referred four people for prosecution over sabotage of critical infrastructure directed by Russian services, including an arson attack on a site working for the Latvian defence ministry. The same year brought the arson plots traced to Russian military intelligence that reached Lithuania, repeated damage to undersea cables and pipes in the Baltic Sea, years of GPS jamming, and drones and balloons crossing the border. A compiled open-source record of incidents against European critical infrastructure since 2022 shows the tempo climbing steeply, from a handful of cases that year to dozens by 2025, and it counts only about half as carrying a high-confidence link to Russia, leaving the rest honestly as unproven. That record holds one detail this series should not miss. Poland, the hinge, is the most-targeted country in it. The state that everything passes through is also the state most often struck. The Robert Lansing Institute sets these on an escalation ladder that runs from steady-state pressure, through an acute hybrid crisis, to open conflict and an Article 5 trigger. The energy system sits on the lower rungs, below the threshold of war, and that placement is the point. Infrastructure is a target chosen precisely because striking it does not obviously cross the line, and the zone below the line is the zone where Article 5 is most ambiguous and the collective response is slowest. A hybrid attack does more than exploit the decision gap. It is designed to live inside it.

If the weakness is speed, the obvious response is to spend on resilience, and here the analysis has to be honest about what actually helps. The temptation is to reach for batteries, and batteries are the wrong headline answer. The Lithuanian adequacy work shows why. The stress events that matter run to as long as 157 hours, and a battery fleet sized and priced for that duration fails on its own economics, while the new thermal plant those same findings imply is negative in net present value by construction and would run only on out-of-market support. The investments that shorten a crisis like this one are duller and slower to build. A second synchronous tie to the Continent, so the Baltics no longer hang off one line. Firm, dispatchable capacity that can run for days rather than hours. Gas storage and terminal headroom the region does not have. Demand that can be cut on instruction without economic collapse. And running under all of it, the sober fact that published grid-connection queues overstate the deliverable pipeline by four to six times, so the capacity that looks contracted is mostly a set of tradeable options rather than steel in the ground. The ranked list sits in the apparatus. Its top line is that the cheapest resilience is the one nobody sells, which is the ability to decide fast.

The other half of resilience is legal rather than physical, and it is the half nobody costs. A government that wants to act inside the first day needs powers ready before the day arrives: to declare an energy emergency, to instruct curtailment, to release national stocks, to direct the regulator, to set emergency tariffs, to prioritise supply. Most of these exist somewhere in national law. The question the crisis asks is whether they can be reached in hours rather than weeks, and whether reaching for them breaks the single market. Export controls and price caps are the fastest to grab and the most corrosive to the Union, because they protect the home market by damaging the neighbour’s. The powers that steady the region without fragmenting it, coordinated curtailment, mutual reserve sharing, joint public communication, are the ones that need agreement in advance. And agreement in advance is exactly what solidarity in law keeps failing to produce.

There is one reason for hope, and it is instructive. The technical clock can be made fast. Ukraine synchronised its grid with Continental Europe in a matter of weeks in 2022, under bombardment, a switch that had been planned to take years. When the will was there, the engineering compressed to fit. The weakness the whole series keeps finding is in the decision. Everything that runs on engineering can be made to move in the time a crisis allows. Everything that runs on twenty-seven governments agreeing moves slower, and so far will not keep pace.

So the series ends where it began, on a single gap. Europe has built energy coordination. What it has not built is energy command. It has written the rules, funded the wires and promised the solidarity, and left out the one thing a war would demand, a hand fast enough to act while the crisis is still an afternoon old. Until that hand exists, the crisis will be run by whoever can move at its speed. Most likely that is a national government, a Polish one, deciding alone what Europe has never agreed to decide together.

A Baltic power crisis reaches Germany through the gas price rather than the electricity grid, and that is what makes a regional military scare a Continental energy event. The shock moves as two waves. A fast electricity wave reprices the Baltic, and connected Swedish SE4 and the Polish market within hours and then dies at the congested German border, so actually Germany barely feels it as electricity.

A slow gas wave lifts the TTF benchmark, and through the merit order the benchmark lifts power prices across the already gas exposed Continent at the speed of a cargo voyage. The German day ahead price tracks gas at about 93 per cent, and when TTF spikes, German and Italian electrical power runs to €120 to €150 per megawatt hour while France and Iberia hold nearer €60 to €80.

The Baltic electricity market is already violent, having touched €500 per megawatt hour after synchronisation in February 2025 and swung from about €0.12 to €160 per megawatt hour in a single day that August. Such a security scare amplifies that and adds a defensive reflex that costs money before there is any shortage. The gas wave, amplified by a Hormuz tightened LNG market into which Klaipėda and Poland must bid for scarce cargoes, is the one that reaches Frankfurt and eats the storage buffer held in Latvia.

The first week reprices in a fixed order, risk premium, electricity, gas. The first month forks into contained, stressed and systemic, decided less by the gas market than by whether governments move together. The crisis is a transfer from industry and households to whoever holds flexibility and molecules, and no EU instrument moves at the speed of the price.

All of this assumes the hardware holds, and the assumption deserves stating rather than smuggling in. The Baltic system has about four single points of failure. Inčukalns, the one underground store, carries all three national markets through winter. Kruonis, Lithuania’s pumped storage of around 900 megawatts, supplies the fast balancing the region leans on hardest when it runs alone. The floating terminal at Klaipėda is the only seaborne gas route. LitPol Link is the only synchronous tie. Lose one and the character of the crisis changes.

A missile into Inčukalns or Kruonis is not this scenario, because an open strike on NATO-territory infrastructure ends the ambiguity the coercion depends on. The live threat is deniable, an anchor across a pipe, a fire at a compressor, a drone over a substation, the pattern already running against these states. Below that line the hardware is a target of opportunity.

The market does not wait to learn which it is. A store that might be sabotaged is priced as a store that might be lost, and the premium that reprices in the first minutes is partly the price of that uncertainty. If the hardware holds, the crisis stays a price event that reaches Germany through the gas benchmark. If Inčukalns goes down in a January cold snap, or Klaipėda is disabled with the store already drawn low, the Baltics meet the one thing the paper adequacy tests rule out, a real supply emergency in three countries at once. That tail is thin in a stress test. It is not zero, and it is the difference between an expensive winter and a rationed one.

A Baltic power crisis reaches Germany through the gas price rather than the grid, and the gas price travels everywhere at once. When the Dutch TTF benchmark moves, the German day-ahead power price moves with it, at a correlation measured near 93 per cent over recent years, because gas-fired plant still sets the German price in most of the hours that decide the monthly bill. That is the mechanism this part follows, out of a Baltic control room and into a German one. It is why a shock on a single 500-megawatt line in Lithuania can be read on a trading screen in Frankfurt inside the hour, while the electrons never leave the region.

The crisis moves as two waves on two clocks. One is fast and stays local. A disturbance in the Baltics reprices the Baltic, Swedish SE4 and Polish power market within hours, then runs into a physical wall at the German border and stops. The other is slow and travels everywhere. The same event lifts the gas benchmark, and through the merit order that benchmark lifts the power price across the gas-exposed Continent at the unhurried speed of a cargo voyage. Over both sits a risk premium that reprices every European energy and financial market the moment the news breaks, and respects no border at all.

Take the fast wave first, the one that looks most alarming and carries least outside the region. The Baltic power market was already violent, and it grew more so the day it left Russia. In the fortnight after the three states synchronised with Continental Europe in February 2025, the day-ahead price on the Nord Pool exchange touched €500 per megawatt hour on peak-demand evenings, driven by still air, a cold snap and a fault on the EstLink 2 cable to Finland that shut off cheap Nordic imports. On 7 August 2025 the day-ahead price ran from roughly €0.12 per megawatt hour before dawn to about €160 in the evening peak, a swing of more than a thousandfold. The joint Baltic reserve market that opened alongside synchronisation ran from a few cents to several thousand euros per megawatt hour. Estonia’s grid operator, Elering, spent about €100m on reserves in 2025 against a budget of €60m, because balancing the system is now a cost the Baltics carry themselves, where the Russian grid used to absorb it. This is the ordinary weather. A security scare amplifies that volatility and adds a defensive reflex, because operators holding capacity back for the night are not selling it into the day.

The fast wave then stops, and where it stops matters. The Baltics reach the Continental grid through Poland, and Poland reaches Germany across a border that runs congested in normal operation. A Baltic and Polish price spike propagates into Sweden’s SE4 zone and up against the German frontier, and there it dies, because no clear wire carries it further and the interconnection is already full. The fast wave is real, expensive for everyone inside the Baltic, Swedish and Polish corner, and regional. Germany barely feels it as electricity.

Germany feels it as gas. This is the slow wave, and it is the one that turns a regional military scare into a Continental energy event. The mechanism is the merit order, the rule by which the wholesale power price in every hour is set by the most expensive plant needed to meet demand. When wind, solar and nuclear cover the load, the price is low. When they fall short, a gas turbine is the last plant called, and its fuel cost sets the price for everyone on the system. Gas supplies under a fifth of Europe’s electricity, around 17 to 18 per cent in 2025, yet it still sets the price in roughly half of all hours, and in the expensive ones almost always. The European Commission’s spring 2026 forecast put the share of hours in which fossil fuel sets the EU power price at about 50 per cent, down from 70 per cent in 2020, and that remaining half governs the bill. There is a recent picture of what a gas-set spike looks like. On 12 December 2024, a windless day, the German day-ahead price neared €1,000 per megawatt hour, the highest hourly price in a decade, with a gas turbine setting the market for fourteen hours straight.

What moves the gas price, in this scenario, is already loaded. The series is set in an autumn where the Strait of Hormuz is still disrupted and the global market for liquefied gas is tight and nervous. The Baltic states buy their gas through one floating terminal at Klaipėda, and Poland is both their backstop and a large buyer in its own right, so in a crisis both are bidding for scarce cargoes into that tight global market rather than drawing on slack. That pull lifts TTF, and TTF lifts the power price wherever gas is on the margin. IEEFA’s work through 2026 traced the divergence this produces. When the benchmark spikes, day-ahead power in gas-reliant systems such as Germany and Italy runs to €120 to €150 per megawatt hour, while France on its nuclear fleet and Iberia on renewables hold nearer €60 to €80. The same gas price, and a different bill, decided by the generation mix each country chose a decade ago.

The gas wave is slower than the electricity wave, because a cargo takes days to redirect and weeks to arrive, but it is the one that reaches Frankfurt, Milan and beyond, and it is the one that eats the buffer. Late October is a good moment to be hit, because storage sits near full ahead of the winter deadline, but a sustained crisis draws that buffer down, and the refill next spring lands into a market that is still tight and still expensive.

The trend runs the other way, and that belongs in the account. The share of hours in which gas sets the European price has fallen from about 70 per cent to about 50 per cent in five years, as wind and solar have grown, and the 2024 market reforms push more volume onto long-term contracts and contracts for difference that weaken the link between the spot price and the household bill. The system is slowly becoming one a gas shock can touch less. The crisis in this series exploits the gas that is still there, in the hours that still matter, and those are the cold, still, high-demand hours a late-autumn emergency would fall on.

In order, the first week has a clear shape. The risk premium goes first, in the minutes after the news, as gas and power forwards, oil and utility equity all reprice on the screens. The electricity wave goes second, within hours, as the Baltic, SE4 and Polish day-ahead clears sharply higher and the Baltic balancing market runs hot. The gas wave goes third, over days, as TTF settles at a higher level and drags German and Italian power up behind it. By the end of the week the electricity spike in the east has come and gone in the evening peaks, and what persists is a higher gas price sitting under everyone’s power bill.

The first month is where price turns into politics, and where the account moves from markets to governments. It forks three ways. In the contained case, the buffer holds, the coordination groups meet, prices settle high but supply is never in doubt, and the episode ends as an expensive month. In the stressed case, high prices persist for weeks, governments reach for the national toolkit, capping retail tariffs, leaning on regulators, preparing industry to curtail, and the friction between those national moves and the single market begins to show. In the systemic case, the crisis runs long enough to draw storage down hard, force real curtailment in the energy-intensive plants at the sharp end, fertilisers, chemicals, steel and glass, and push the gas price toward the territory it last saw in 2022, when TTF peaked above €300 per megawatt hour and German power followed it up. The three cases are branches rather than predictions, and which one the crisis takes depends less on the gas market than on how fast governments move and whether they move together.

A price shock is a transfer, and it is worth naming who sits on each side. The losers are the energy-intensive industries that cannot pass the cost on and must either pay or stop, the households that meet it in the next bill, the unhedged retailers caught between a fixed sale price and a soaring wholesale one, and the systems most exposed to gas, which in this crisis means Germany and Italy far more than France or Spain. The winners are the flexible assets that earn most exactly when the system is short, batteries, pumped storage and gas peakers, the owners of the interconnectors who collect the congestion rent a price gap creates, the traders who move cargoes, and the exporters of American gas, whose Henry Hub benchmark trades structurally below TTF and is far less exposed to a chokepoint on the other side of the world. A Baltic crisis, filtered through the gas price, is a machine for moving money from European industry and households to whoever holds flexibility and molecules.

The market war is therefore also a solidarity test, and this is where it meets the framework Part 1 described. Every national move to soften the blow at home tends to export it next door. A country that caps its price distorts the flows across its borders. A country that restricts exports to protect its own supply is holding gas its neighbour was counting on. The single market that spreads cheap power efficiently in good times spreads the shock just as efficiently in bad ones, and the legal promise meant to hold it together, gas solidarity, has never been triggered and rests on bilateral agreements still being completed in 2025. Whether it holds the first time one member state must give up gas for another, under price stress and with its own voters watching, is the question the third part turns on. So is the behaviour of the one member state most able to act alone and at speed.

The uncomfortable finding of the market war is that Europe needs no wire from the Baltics to Germany for a Baltic crisis to reach Germany. It has a single gas price, and that price is the fastest thing in the system. Every instrument built to manage a crisis moves in days or weeks. The price moves in minutes, and it decides who pays before anyone in Brussels has decided anything at all.

First a disclaimer, this was a mammoth undertaking, its taking half my holiday to research and format and there is more LLM work that I would like and whilst I have tried to edit it all out but it may not be as clean as usual just because of the amount of text.
Second thing though, Fable 5 is fantastic, but its not necessary fro everything, i’ve been testing Kimi K2.6 and GLM-5 they are both decent models. none can come close to ChatGPT or Midjourney for graphics.

Executive Summary

A war on Europe’s eastern edge would hit European energy in an afternoon, and the collective answer would take weeks. That gap is the whole risk. The Baltic states are the trigger, because since February 2025 they have run on the Continental grid through a single tie to Poland, but the weight of the event falls on the ripple. The shock reprices power across half a dozen countries within the hour, lifts the price of gas for next winter, and reaches Germany as a price long before any collective measure is drafted. Europe has built energy coordination. The European Grids Package agreed as a Council negotiating position on 26 June 2026 is a planning and permitting measure that is not yet law and moves no supply for years. The storage rule was loosened in 2025. The solidarity mechanism has never been used. Europe does counter, and 2022 shows it counters in weeks, with an emergency gas demand-reduction regulation agreed in about a fortnight that August. In the first afternoon only national governments can act. This is a stress test and not a forecast. Estonia’s foreign intelligence service judges no attack likely in 2026, and probably not in 2027.

Share

To borrow an overused phrase from our out of favour American cousins, its HOGWASH.

Subscribe now

On the 26th June 2026 EU energy ministers agreed a watered down Council negotiating position on the European Grids Package. It still contains process obligations, but the hard delivery levers have largely been softened. Tacit approval is voluntary, digital permitting portals are optional, planning remains heavily national, and there is no binding obligation on any member state to build a specific grid line, deliver a defined interconnector capacity, remove a congestion bottleneck, or spend a fixed amount by a fixed date. All whilst we witness continued Russian aggression in Ukraine, its rebuilding of force capacity around NATO’s eastern and northern flank and Belarus is being turned into a more usable Russian military platform.

There will be lots of talk about NATO over the coming week as the Ankara summit gets underway, I wanted to shed a bit of light on what more hand wringing and time wasting from Brussels could mean. I have been war gaming this all week, using various Monte Carlo analysis of various situations around invasion, this is what WELT used as its scenario in their latest theory. It’s a subject obviously close to me both living and working here in Lithuania.

It is a Tuesday morning in late October 2026, and following a week of movement where Russia both puts troops on the goods train supplying Kalliningrad and lands them inside Lithuanian Territory and then cancels the same train therefore claiming a humanitarian Crisis inside the Oblast, Russian units are finally moving into the strip of land that joins Poland to the Baltic states, the only road NATO has into Lithuania, Latvia and Estonia. Nobody in Warsaw, Berlin or Brussels yet knows whether this is pressure or the first hour of a war.

The price moves before anything else does. Inside the first hour, before any government has decided anything, day-ahead power jumps in half a dozen countries at once. Traders in Leipzig and Paris do not wait to learn what Moscow means. Gas for next winter reprices across the Continent. In Mannheim a plant manager watches the cost of running his lines that afternoon double on the screen in front of him.

A war on Europe’s eastern edge reads as a local event, and that reading is wrong. The fighting, if it comes, would be confined to the corridor. The energy shock would run the length of the Continent. It starts in the Baltics and leaves them within the hour, moving first at the speed of a headline and then at the speed of a gas tanker, until it reaches Germany and the United Kingdom as a line on an invoice. The three Baltic states set it off. Where it travels is the story.

Anyone who runs an energy system for a living should find the arithmetic uncomfortable. Europe has spent fifteen years and about €1.6bn wiring the Baltics into the Continental grid and building the plant to hold it steady. On the one thing this shock hits first, speed, it has spent almost nothing. The Union can coordinate a market over weeks. Commanding one over an afternoon is a different capability, and that is the one Europe lacks. The distance between the two runs from an afternoon to a fortnight.

Geography explains why the spark lands here. Since February 2025 the three Baltic states have run on the Continental grid through a single high-voltage tie to Poland. Every other line out of the region carries direct current to Sweden or Finland, which moves power and money but cannot hold the frequency in step. Three national systems hang off one connection to the Continent, mapped in the first exhibit below. A security event in this exact place, and almost nowhere else, reaches straight into the European power system.

The lights would probably stay on. ENTSO-E rates the Baltic systems adequate for a normal winter, they hold generation and reserves built for exactly this contingency, and they have already run alone once, for a day in February 2025, in the hours between leaving the Russian system and joining the European one. Supply holds. What gives way is something else.

Confidence, security posture and the speed of a decision give way first, and all three move faster than any collective response can form. Confidence goes in the market. The Baltic, Nordic and Polish bidding zones clear through the same day-ahead coupling as the rest of the Continent, so the market prices fear in minutes, before a single flow reverses. Posture goes next, and it costs money. Every megawatt a system operator holds back against an uncertain night is a megawatt it cannot sell, and the round figure a link shows on a map is never what clears once the operator is holding headroom for security. Public confidence is the most fragile. A household or a factory that hoards fuel and cash on a rumour has started a run, and the only guard against a run is fast, clear official word, which is harder to produce under pressure than to promise in the calm beforehand.

None of that is where the European rulebook operates. Europe has built a great deal, and all of it sits at the slow end of the clock. On 26 June 2026 EU energy ministers agreed the Council’s negotiating position on the European Grids Package, a serious revision of the network and permitting rules that plans cross-border grids as one system and, for the first time, marks out priority projects for security and resilience. It carries no legal force yet. It is one weak step in a process that will not conclude before the end of 2026 and will not move a watt for years after. Dan Jørgensen, the Commissioner who owns it, told ministers to their faces that it was about long-term planning.

The instruments that look like crisis tools read the same way once opened. Call it the paper shield. The one hard rule, gas storage 90 per cent full before winter, was loosened in 2025, the firm November deadline replaced by a window into December and a permitted shortfall of up to 15 per cent in a difficult market. Gas solidarity, the promise that a country in an extreme emergency will be sent gas by its neighbours, is binding on paper and has never been used. Its plumbing was so thin that the Union had to add default rules and a compensation process in 2025, because too few countries had signed the bilateral deals that make the promise real.

Set against that, Europe does counter, and faster than the legislative clock suggests. 2022 is the evidence. Faced with the gas shock that summer, the Council agreed an emergency demand-reduction regulation in about a fortnight that August, and moved on storage and price measures over the months that followed. Weeks to months is the honest speed of a collective answer. Years is the clock for the Grids Package and the new interconnectors, the physical fix, and that is where years belong. The crisis answer is quicker than the physical fix and still far too slow for the first afternoon. By the time any collective measure is drafted the market has repriced and a run has begun, and the only hand that can reach into that first window is a national one.

The gas map matters even though the first test is about electrons, because gas is the medium the crisis travels down to Germany, and the Baltic gas system carries no spare copy of anything. One import terminal handles the seaborne gas: the floating unit at Klaipėda, which regasifies about 3.75 bcm a year, the only working LNG terminal in the Baltics, and written into Lithuanian law to run until at least the end of 2044. One pipeline reaches the wider Union, GIPL into Poland at roughly 2 bcm a year, live only since 2022. One large store holds the winter, Inčukalns in Latvia, about 2.32 bcm of working gas, and it carries three national markets because neither Lithuania nor Estonia has underground storage of its own. That last point is geology rather than politics. Latvia sits on porous sandstone at a workable depth, Estonia’s layer is too shallow and Lithuania’s too deep, so one Latvian store run by Conexus Baltic Grid holds the region through the cold months. The exposure has already been tested in the water. An anchor dragged across the Estonia to Finland pipe, Baltic connector, in October 2023 and left it out of service until the spring.

One arrow on this map points the other way, and it has quietly stopped working long before Russian sanctions. Lithuania carries Russian gas to Kaliningrad through the old Minsk to Kaunas line, and since the date of its inception “Independance” the Lithuanian degassing vessel has had that transit card. Vilnius could in principle supply about a third of the gas entering Lithuania’s largest pipe fed the exclave.

However, Russia removed most of the Lithuanian transit leverage over Kaliningrad gas in stages. It commissioned the Kaliningrad underground gas storage site in 2013, launched the Marshal Vasilevskiy FSRU and offshore LNG receiving terminal on 8 January 2019, and planned to expand the local storage system toward about 800 million cubic metres by 2025. It proved the point in February 2025, holding Kaliningrad for five days on its own gas in the days the Baltics left the Russian grid. The Gazprom transit contract with Amber Grid then expired at the end of 2025. The vessel that guarantees the exclave now works the Baltic with heavy machine guns bolted to its bridge wings, photographed by the Estonian border guard, which is how the other side reads the same map.

On paper the region passes the test. The EU’s security-of-supply rule, the N-1 standard, asks a country to meet peak demand with its single largest piece of gas infrastructure out of action. Lithuania clears it at 153 per cent before GIPL and above 200 per cent since the Polish link opened, on the IEA’s numbers. The standard measures the wrong failure. It assumes one pipe or terminal breaks while the market around it goes on working normally. It is silent on the case this scenario describes, where every terminal in Europe bids for the same scarce cargoes at once, in a panic, as fighting reaches EU soil. Then Klaipėda has no slack to draw on. It competes for LNG in a global market already assumed to be tight, while three countries draw down the single store at Inčukalns together. The hardware copes even so. What gives way is the price, and price is the channel that carries the crisis to Germany.

Two recent events show the gap under test, one imagined and one real. The imagined one is the war game the German paper WELT ran with the Bundeswehr’s war-gaming centre in December 2025, set in October 2026, in which a Russian move cuts the Baltics’ only land link and a fictional German government spends the crisis unable to act, held up by its own federal machinery and by an American ally slow to confirm it is bound to help. The game is fiction and should be read as such. Estonia’s foreign intelligence service judged in February 2026 that Russia has no intention of attacking a NATO state in 2026, and probably not in 2027, which is why this is a fire drill and not a forecast, and why its lesson is about the speed of a decision rather than the odds of a war.

The real event makes the same point with nothing invented. When the Strait of Hormuz closed in late February 2026 and the TTF gas benchmark ran from around €32 to above €60 per megawatt hour, with analysts warning it could pass €90 in a full closure, the Union answered with a letter to ministers, an observatory to watch the data, a run of coordination meetings, and, by late June, ministers still weighing whether to release reserves. Four months of coordination for a chokepoint it could see coming for weeks. If Europe decides this slowly on an oil shock with a long fuse, the burden of proof sits with anyone who assumes it would decide quickly on a Baltic shock that arrives in an afternoon.

The reason is that the parts of the system run on clocks that do not agree, and the second exhibit below sets them side by side. Automatic reserves answer in seconds, with no human in the loop. The coupled market reprices in minutes to hours. A national government can declare an emergency in hours to a day. The coordination groups meet in days. The collective emergency answer, on the 2022 evidence, forms in weeks. The physical and legal fix takes years. A shock that spreads in seconds and reprices in hours meets an apparatus that answers in weeks and years, and the crisis lives in the gap between the clocks.

What a national government does inside that gap, and what Brussels can make it do, are two different questions with two different answers. A Baltic or Polish prime minister on the first morning can declare an energy emergency, lean on the regulator, release what national stocks exist, warn industry to prepare to curtail and tell the public to stay calm, and every part of that is national power used within hours. What Brussels can require in the same window barely bites. It can convene meetings, share information and appeal for restraint. The solidarity the treaties promise is real over weeks and untested in a fight, and a stressed member state can hold its gas, keep its spare capacity and limit its exports faster than the Union can agree a reply. Fast national power alongside slow collective coordination is the fault line the rest of this series runs along.

The first day is the narrow part of the story, because it is about control, and control is mostly national and mostly technical. What comes next is wider. The Baltic shock does not stay in the Baltics. It travels as two waves on two clocks. A fast electricity wave reprices the Baltic, Nordic and Polish market and then runs into a physical bottleneck at the Polish-German border. A slow gas wave lifts the TTF benchmark and, through it, the price of power across the Continent, at the unhurried speed of a cargo voyage. Part 2 follows both waves out of the Baltics and into the German market, and prices them.

Europe built the machinery to coordinate a market over weeks and years, and built it well. A war would ask it to command an energy system in an afternoon. No one has built that, and the afternoon is exactly when it would be needed.

Five conclusions

1. The real gap runs from an afternoon to a fortnight. The decade long clock is the one for interconnectors and grid law, a separate problem. On the 2022 evidence Europe counters in weeks, and weeks are still too slow for the first afternoon.

2. The Baltic systems are adequate and have already run alone, so the lights hold. Confidence, posture and the speed of a decision are what a crisis breaks first.

3. The Baltics are the trigger because they hang off a single tie to Poland. The ripple is the story, and it runs to Germany as a price.

Why a model at all

The writing itself hides some of the assumptions. Write “a Baltic crisis would push gas prices up and strain solidarity” and the argument sounds firm, but every load bearing judgment made is buried inside the sentence, how far, why, in what region, when? A model forces each one into the open, gives it a number and a source, and then lets you see which ones actually matter. It is a discipline against our own bias.

What it does, plainly

I did not model a war or guess its odds. I assumed the crisis and asked what follows. Ten things that would shape it, from the depth of the gas shock to how fast Poland acts to how quickly Brussels agrees, were each given a range and a most likely value. The computer then played the crisis out 10,000 times, drawing a different combination each run, and recorded how bad each one got. This is called a Monte Carlo exercise. It is not a forecast. It is a way to find out which assumptions drive the result.

What it found

That the trigger barely matters. How hard the Baltic corridor itself is hit ranks near the bottom sadly for energy its not overly important. What decides whether a local scare becomes a European crisis is the electricity and gas price, the posture Poland takes, the depth of the market and the speed of the collective answer. Which is the argument of the whole series, reached here by a different road.

How to read the four charts

1. The drivers. Each bar is one assumption, and its length is how much that assumption moves the risk of a full European crisis. The gas channel, in green at the top, dwarfs the rest. The three physical triggers, including how hard the Baltic corridor is hit, sit at the bottom on the shaded panel. Read it as a ranking of what to worry about.

2. The gas cliff. The risk of a systemic outcome against the size of the gas shock. It stays flat at zero for four fifths of cases, then goes vertical. Drawn twice, once assuming Poland’s own system is calm and once assuming it is strained, the two lines land on top of each other, which tells you the gas shock, not the Polish condition, is doing the work.

3. The political levers. The two things governments actually control. On the left, the better Europe coordinates, the lower the risk, a clean downward line. On the right, three dots for Poland’s choices, with the self-protective option in red carrying about four times the risk of the supportive one. This is the fork the series ends on.

4. The front-load. How severe the crisis is over time, in the first day, the first week, the first month. The band is widest and highest on day one and settles after. The shock lands before anyone can answer it, which is the point the whole series keeps returning to.

The honest caveat

The assumptions are mine, so the model can only be as good as my judgment and the articles I read, and I publish the full assumptions table so you can argue with it. One number in particular, the split between broad outcomes, moves depending on where we draw the lines, so we lead with the ranking of what matters, which holds steady, and not with any single probability.

The triggers are the tell. Britain’s government fell over its defence investment plan. Latvia’s collapsed after Ukrainian drones crashed into a fuel depot and the prime minister sacked her defence minister. Lithuania’s broke apart and reformed on a pledge of defence resilience. The churn is not random. It clusters on the exposures, energy and defence, that a serious continent would be gripping, and it leaves the capitals that should be acting too busy forming governments to act. Europe is fighting rear-guard actions in a dozen parliaments and calling the sum of them a policy. The meters, which do not care who holds office, have been recording the result for five years.

Europe used less electricity in 2025 than in the year its economies were shut by decree. Metered demand across 27 European markets came to 2,744 TWh last year against 2,755 TWh in 2020, on the ENTSO-E Transparency Platform’s own numbers. Germany fell 3.7%. France fell 2.0%. Poland fell 3.9%. Electricity demand is the most honest growth statistic a modern economy produces. It cannot be revised, seasonally adjusted or talked up in a press release, and Europe’s has now flatlined for five years.

Set that against the official story. Eurostat has real EU output more than a tenth higher than in 2020. An economy that grows a tenth while its power consumption falls is doing one of two things. Either it has pulled off an efficiency miracle, or the part of the economy that uses electricity, the part that makes physical things, is shrinking while services and government spending carry the GDP print. The country detail settles which.

Divide each market’s metered load by its population and the geography is legible. A Norwegian used 24.4 MWh of electricity in 2025. A Finn used 15.0 and a Swede 12.2. The industrial core of Europe sits between 5.5 and 6.9. Great Britain, on NESO’s transmission demand of 229 TWh, comes in at 3.4 MWh per person, above only Romania at 2.8. One Norwegian consumes as much electricity as seven Britons.

This is not a wealth ranking. Britain and Italy are rich and sit at the bottom because their homes burn gas and their cars burn petrol. Norway is at the top because its heat, its water and increasingly its cars run on the wire. Demand per person measures how much of a country’s energy life has been electrified, which makes its change over five years the single best test of whether the transition’s demand story is happening. Mostly, it is not. Demand per head fell in thirteen of the 27 markets between 2020 and 2025, and the fallers include almost every large economy: Germany down 3.7%, France 2.0%, Poland 3.9%, Sweden 2.3%, Austria 2.7%. The eastern edge is shedding load outright, Romania down 7.5%, Slovakia 7.6%, Serbia 11.8%. I wish we could say that is energy efficiency taking hold, sadly its simply industry leaving. Probably just walking away from Europe for a long time.

Seven markets grew, and they share a signature. Denmark rose 13%, Ireland 9.6%, Portugal 9.0%, the Netherlands 8.4%, Finland 8.2%, Croatia 8.0% and Norway 3.2%. Denmark, Ireland and the Netherlands are where the hyperscalers landed, and Irish all-island demand is being dragged upward by data centres that already take more than a fifth of the country’s power on the Irish statistics office’s figures. Finland is the cleanest case of the other force. Olkiluoto 3 entered service in 2023, Nordic wholesale fell to about $40/MWh in the first half of 2025, the cheapest power in Europe on the IEA’s figures, and Finnish demand grew faster than any market on the continent bar Denmark. Portugal grew into the cheap hours of the Iberian solar build. Croatia grew 8% and resists a single label. Growth went where electricity became worth buying.

Mario Draghi’s competitiveness report put it to Brussels in September 2024: the continent’s firms pay two to three times American electricity prices and that this is dismantling its industrial base. The meters agree with him, market by market. EU wholesale averaged about $95/MWh across 2025, still roughly twice American levels on the IEA’s Electricity 2026 figures, German wholesale ran a little under $100/MWh in the first half and German demand fell with its industrial output. Britain is the limit case. British industry pays the highest electricity price in the developed world, 26.6p per unit on the government’s own figures, more than double the European average. Households get the same signal from the other side. A unit of electricity costs more than four times a unit of gas. A heat pump saves carbon and loses money. A household that turns one down is not confused about the transition. It is doing sums.

Now put the supply side next to it. The generation map cleaned itself over the same five years. Coal thinned to a historic low of 9.2% of EU power in 2025, solar spread to a record 13% and overtook it, and wind and solar together out-generated fossil fuels for the first time, on Ember’s full year figures. That is the half of the transition the decade promised, and it happened. The other half did not. The plan was substitution, primary energy falling because a heat pump at 350% efficiency and an EV at 90% replace a boiler at 85% and an engine at 25%, so the wire grows while the barrel and the pipe shrink. The wire did not grow. Cleaner supply met flat demand, which means the energy Europe stopped using was not substituted. It was surrendered. Europe calls that a transition. The meters call it a contraction, and nobody voted for it.

The electrification timeline explains why the demand is late and why late is not the same as absent. Sales lead, the stock lags, demand tracks the stock. An EV sold today displaces petrol for fifteen years but adds its 2.5 MWh to the meter from day one, a heat pump likewise, so demand climbs one installation at a time and the installations compound. Norway is the proof the staircase tops out, 90% EV sales, an electric heat stock, 24.4 MWh per head. And the first step has just been retaken. European heat pump sales rose 10% in 2025 to 2.62 million units across 16 markets on the European Heat Pump Association’s preliminary figures, after two falling years, and the association is explicit about why. Six countries moved taxes off electricity bills in 2025. The market turned where they did: Britain up 27% to 125,000 units on the Boiler Upgrade Scheme, Germany at a record with heat pumps taking almost half of all heat generator sales, while Poland and France, where support was cut or left in flux, fell. The experiment has a control group. Poland and France are it.

Which is also why the fix is knowable, because seven markets have already run it. Demand grows where the electron is priced to be used. In practice that means three moves. First, take the policy levies off electricity and put them on gas or general taxation, since taxing the clean carrier to subsidise everything else is the single most self-defeating price signal in European policy, and the six countries that ran the shift in 2025 got a sales turn for it inside twelve months. Britain’s government conceded the wider point in April 2026 when it moved to break the link between gas and electricity prices through long term contracts at fixed prices with low carbon generators. Second, build the cheap firm supply that lets wholesale settle low for years, which is what Olkiluoto 3 did to Finland and the hydro stock did to Norway. Third, let retail customers see the cheap hours, because a heat pump and an EV are storage devices attached to a price signal, and a flat tariff blindfolds them. None of this is exotic. All of it is visible in the seven markets that grew.

Growth did not leave Europe. It moved to where electricity was worth buying, to Espoo, Odense and Dublin’s data sheds, and it will keep moving there while the continent’s biggest economies tax the wire to protect the pipe. Electrification is not a policy. It is a price.

Sources

ENTSO-E Transparency Platform, actual total load by bidding zone, 2020 and 2025. Raw zone data integrated interval by interval, deduplicated, normalised for the 2020 leap year. Ireland (all-island SEM) annualised from 66% year coverage.

Golfech showed it on 22 June 2026. EDF stopped the plant’s last running unit because the Garonne was about to reach 28 degrees, and nothing was broken. The river was doing the warming by itself. The plant’s own contribution, the heat and the part the law actually controls, had been squeezed to almost nothing, simply no space for any additional heat outflow.

The same thing happens most hot summers now, and the temperature on the news is the wrong one to watch, it’s not interesting to a nuclear plant. A reactor does not stop because the air is at 40 degrees, a nuclear plant stops because a river crosses a legal line, and that line is written about water temperature, not air temperature

.

The number is a river temperature, not the air

The licence isn’t interested in how much the French are sweating in their non air-conditioned homes or offices. The licence sets two numbers, both about the river. The first is an absolute ceiling: the Garonne river, measured downstream of the plant after the discharge has mixed in, must not pass 28 degrees. The second is a limit on the lift: in summer Golfech may warm the river by at most 1.25 degrees. Both exist to protect the water. Warm water holds less oxygen, and a river driven past its threshold starts to kill the fish and plant life it carries.

Those two numbers are why the upstream river is what operators actually track. Golfech runs cooling towers, so in normal conditions it warms the Garonne by about 0.2 degrees, far inside the 1.25 it is allowed. That slack disappears the instant the river turns up hot. Once the Garonne reaches about 27.8 degrees by itself, the plant’s 0.2 degrees is enough to break the 28-degree ceiling, and the station has to back down. On 22 June the river was already at 28 before the plant added a thing, so no output level was low enough to comply. The unit had no room left and came off.

It reaches that point because three things land at once. A heatwave travels with drought, and drought drops the river’s flow, so there is less water to absorb whatever the plant returns. The river is already warm, because water holds heat for weeks. And the ceiling is absolute, not a margin that bends. Put a shrunken, sun-warmed Garonne against a fixed legal number and the plant’s own heat is the last straw. The few degrees in the air do the damage through the river, never directly.

Cooling towers do not move the line

The usual defence for this kind of problem is a sloppy one, that only the old once through plants get caught, the ones that draw river water, warm it several degrees and send it straight back, and that towers solve the problem. But that’s not the case because both kinds are measured against the same in river 28 degrees benchmark. A cooling tower shrinks the heat the plant adds to a few tenths of a degree, which hardly moves the dial and buys a little headroom. However, in any event it does nothing to the ceiling and nothing to a river that arrives already at or over this threshold. Both Golfech and Bugey have cooling towers and both were shut down in 2025, alongside the once-through units at Tricastin and Saint Alban. The cooling design sets how fast a plant reaches the wall. It does not move the wall.

The reason it is nuclear and not the gas plant down the road is heat. A reactor turns about a third of its thermal output into electricity and dumps the other two thirds. A modern combined-cycle gas plant runs at 55 to 60 per cent and sheds far less for the same power sold. For every unit of electricity it delivers, the reactor pushes roughly twice the waste heat into the same river. Build the hottest-running technology you have on an inland river, decades ago, and the river is where the limit bites.

Why nobody has fixed it – Environmentalists close your eyes!

Nothing has been done because the loss looks trivial on paper. EDF puts the average output lost to warm rivers and low flows since 2000 at 0.3 per cent a year. In 2025, a summer of front pages, it came to about 0.1 per cent. No board approves a cooling tower retrofit at several hundred million euros a unit, or a move to the coast, to recover a tenth of a per cent. The sums have always said leave it up until now

Moving an existing plant is not on the table, it would be a new build. You cannot carry an inland station to the sea, and fitting closed loop towers to an old once through reactor means new structures, new pumps and a slice of efficiency surrendered for the life of the plant, on a site never laid out for them.

So the real choice has been to wait, or to ask for forgiveness when the river runs hot.

Forgiveness is the other half of the system, and it gets used. When a plant matters to the grid, EDF simply asks the regulator to lift the discharge limit for a few weeks under “closer monitoring”, and the regulator agrees. ASN did exactly that for four plants in 2025, as it had in 2003 and 2006. The river takes the hit. ASN’s own checks afterwards found more algae and plankton around Bugey and a knock to fish numbers below Saint-Alban. The trade is made in the open, and it is made for the megawatts.

So far that is defensible. A tenth of a per cent is a seasonal nuisance at an awkward time, not a hole in the baseload case. France was able to run a nuclear surplus through the 2025 heatwave and kept exporting, because luckily the cuts fell at weekends and around midday, when demand was soft and record solar was filling the gap. The price spike, when it came, was an evening event: French day ahead prices roughly doubled across the heatwave against the pre-heat baseline, and the squeeze landed after dark, once the sun dropped and the cooling load stayed on. The curtailment barely showed in the yearly figures.

The bill is back-loaded

What changes the verdict is the trend. The ceiling is fixed and the river is warming. At Golfech the Garonne is projected to run about 1.4 degrees warmer by 2050, and the share of output lost there climbs from roughly 0.2 per cent over recent decades to about 2 per cent by mid-century, on the modelling. The Cour des comptes expects forced shutdowns to multiply three or four times by 2050. A line crossed once a decade in 2003 and 2006 now appears most summers, and it will not stop there. The ceiling will not rise to meet the river.

That is the case against the existing fleet, and it is weak, because those plants were sited and cooled for the climate of the 1980s and most retire before the worst arrives. The case against the new fleet is the one that holds.

Built into the new concrete

France will build six EPR2 reactors at three sites. Two pairs, at Penly and Gravelines, sit on the coast, where the sea is a heat sink that does not run dry and carries no 28-degree line in a dry July. The third pair goes to Bugey, inland, on the Rhône, one of the rivers that forced curtailment in 2025. EDF added an extra cooling stage to lower the Rhône discharge only after the public debate pressed for it, not as part of the original design. First concrete at Bugey is due in the mid-2030s, and the reactors are meant to run into the 2090s. They are being specified now, against the river of the 2030s, to work in the climate of the 2080s.

Is that a failure of planning? On the old fleet, no. On a reactor drawn up today for a river that every forecast says will be hotter and lower when the plant is middle-aged, the answer is harder to wave away. The fix was never exotic. Coastal siting and closed-loop cooling both cost money at the drawing board, and both look like overcaution in a temperate decade, which is exactly why they get cut.

So this was never an argument against nuclear power. It is an argument about where you put it and how you cool it, and about being honest that the bill for getting it wrong arrives slowly and late. The plants going up now will not be cooled by the climate we have. They will be cooled by the one we are building. The cheapest time to solve the cooling was the drawing board. The next cheapest is before the concrete sets.

Sources: the 18 November 2006 order setting Golfech’s downstream limit at 28 degrees and the summer rise at 1.25 degrees; EDF and RTE output and loss figures; ASN on the 2025 derogations; ENSI on Beznau (Switzerland), where the Aare limit is 25 degrees; Ember, Heat and Power 2025, on prices; the Garonne and Meuse projection study, 2025; Cour des comptes on the 2050 trend; the EPR2 programme at Penly, Gravelines and Bugey. Reactor thermal efficiency 33 to 37 per cent, combined-cycle gas 55 to 60 per cent.

The Energy Guru

References and further reading

The sources behind the piece, grouped by theme, for anyone who wants to go further.

The limit and the law

Autorité de sûreté nucléaire (ASN). Arrêté of 18 November 2006 setting Golfech’s downstream Garonne ceiling at 28 degrees and the summer rise at 1.25 degrees. asn.fr.

European Union. Freshwater Fish Directive (2006/44/EC), on the temperature rise permitted at the edge of a thermal-discharge mixing zone, 1.5 degrees in salmonid waters and 3 degrees in cyprinid waters.

ENSI. The 25-degree Aare limit at Beznau, Switzerland. Swiss Federal Nuclear Safety Inspectorate, ensi.ch.

The 2025 and 2026 shutdowns

MIT Technology Review. “Europe’s extreme heat is shutting down power plants,” 24 June 2026, carrying Ember data on the roughly seven gigawatts forced off in July 2025. technologyreview.com

World Nuclear News. “Heatwave forces temporary change to water-discharge rules in France,” July 2022. world-nuclear-news.org

ASN. Temporary derogations relaxing thermal-discharge limits in 2003, 2006, 2022 and 2025, with the ecological monitoring that followed. asn.fr.

The record and the losses

Cooling under fire. “Can nuclear power remain thermodynamically resilient in a warming, water-constrained world?” (2026), the first event inventory of weather-linked nuclear curtailments, 2003–2022, with EDF losses and the Saint-Alban 3-degree summer rise. ScienceDirect. sciencedirect.com

Qvist, S. “Curtailment of nuclear power output during extreme heat waves: the European case,” Energy for Humanity and Clean Air Task Force (2019). energyforhumanity.org

Clean Air Task Force. “The 2022 French nuclear outages: lessons for nuclear energy in Europe” (2023). catf.us

The 2050 projection

Guénand, Y. et al. “Future nuclear power outages in a changing climate: a case study on two contrasted French power plants,” Energy (2025), finding average losses of 0.3 per cent across 2000–2020 at Golfech and Chooz. ScienceDirect. sciencedirect.com

Energy Connects. “Heat waves are a growing threat to Europe’s nuclear power supply,” August 2025, reporting the Garonne and Meuse projection: Golfech’s losses rising about tenfold and Chooz losing up to 14 per cent in the worst years by 2050. energyconnects.com

Cour des comptes. The 2023 audit projecting forced shutdowns three to four times more frequent by 2050. ccomptes.fr.

van Vliet, M. T. H. et al. “Vulnerability of US and European electricity supply to climate change,” Nature Climate Change (2012), a fall of 6 to 19 per cent in European thermoelectric summer capacity for 2031–2060. nature.com

The fleet and the new build

IAEA. Power Reactor Information System (PRIS), operational reactor counts. International Atomic Energy Agency, pris.iaea.org.

Eurostat and Ember. Nuclear’s share of EU electricity, and France’s share of EU nuclear generation.

EDF. The EPR2 programme at Penly, Gravelines and Bugey, and the climate-adaptation budget

Subscribe now

The Energy Guru · June 2026

The week the trade did not happen

Its not the first time I have written this but finally clear evidence is starting to break out. In the first week of March, the price of gas in Europe did something it had not done in three years. The Strait of Hormuz closed at the end of February, Qatar pulled back the production that feeds roughly a fifth of the world’s traded gas and about fifteen per cent of Europe’s, and the Dutch TTF benchmark rose seventy six per cent in a week. One Tuesday it gained thirty five per cent in a single day. Every desk that trades flexibility had the same instinct. This was the week the batteries should have been printing greenbacks, but they did not, not even close.

The German storage fleet booked one of its weakest months in a year, down about ⅕ on January. The largest move in the fuel that “sets the marginal price” for much of the day had arrived, and the assets that were built to monetise this exact volatility scenario earned less than they had before it. Across the Channel the story was the same. Modo Energy put British battery revenue at about forty seven thousand pounds per megawatt year in December, the weakest month of 2025 and down forty five per cent on the same month a year earlier. The largest gas move in three years did nothing to arrest the slide.

That is the puzzle, and the answer is not a quirk of one month or one country. It is structural, and it is happening across the continent. A battery does not care what power costs. It cares what the difference between the cheapest hour and the most expensive hour is worth, and that difference has been shrinking even as the price level has swung.

What a battery is actually paid for

A battery does not earn the price of power. It earns the gap between the cheapest (or in most cases what it paid, hopefully somewhere near the cheapest) and dearest hours of the day. The idea being that the spread it can buy low into and sell high out of. A gas shock lifts the whole price curve. It does not, on its own, widen the gap. The level can rise seventy six per cent and the spread a battery trades can sit exactly where it was, which is what happened in March.

There is more to the income than the day-ahead spread, and the rest of it matters more every year. A modern battery earns across four or five markets at once. It trades the day-ahead and intraday auctions. It bids energy into the balancing mechanism when the system operator needs to move power in real time. It sells frequency response, the fast reserve that holds grid frequency at fifty hertz. And where a capacity market exists, it collects a fixed payment for promising to be available at the winter peak. Frequency response paid the early movers handsomely and no longer does. In Great Britain the fast frequency products were the making of the first fleets, and Modo Energy now puts that stream at around twenty five thousand pounds per megawatt year, well below its peak, because every megawatt of new storage competes for a service the grid needs only so much of. What was a headline revenue line in 2021 is a thin one in 2025.

The skill is no longer in any one of these markets. It is in moving between them. A battery that sits in frequency response misses a wholesale spike. One that empties itself in the morning auction has nothing left when the balancing mechanism calls at six in the evening. The operators who earn the top of the range run software that re-optimises the asset through the day, switching a megawatt from one market to the next as prices move, and accepting that the right answer at nine in the morning is wrong by noon. That is why two batteries on the same site, trading the same prices, can earn numbers thirty or forty per cent apart. The hardware is a commodity. The work and the real skill are in the optimisation where it really matters.

EXHIBIT 1 · WHERE A BATTERY STILL EARNS

Mean day-ahead price spread by market, 2025. Source: ENTSO-E Transparency Platform via European BESS Index.

The gap is not the same everywhere. Across Europe it varies more than threefold, from around sixty euros per megawatt hour in Finland to nearly two hundred in Romania. And the pattern is difficult to swallow. The deepest spreads sit in the thin, inflexible systems of the emerging east, Romania, Greece and Bulgaria, where almost no storage has been built, developers here are cashing in for certain, a prize for being the first to market and leading the risk management. The shallow ones sit in the crowded centre of the continent, Germany, the Netherlands and France, where the most has. Its clear that the money is widest where the batteries are not.

The relentless build

The reason the crowded markets are crowded is the speed of the build, and that speed is the heart of the problem.

EXHIBIT 2 · MORE STEEL, LESS SPREAD

European battery fleet against the value of a megawatt. Sources: SolarPower Europe and Wood Mackenzie via European BESS Index; Modo Energy. 2026 and 2029 are forecasts.

Europe held under eight gigawatt hours of grid storage in 2021. It holds more than seventy seven now, and on current pipelines it will triple again by the end of the decade, from 8 to 250 GWh in 9 years! That is the fastest deployment of any flexible asset class on the system. As it has happened, the value of the same megawatt has fallen rather than risen. The German day-ahead spread has compressed from about one hundred and twenty euros per megawatt hour in 2022 to about eighty five now. Solar capture, the share of the wholesale price a solar plant actually banks, has fallen from ninety eight per cent in 2022 to fifty four per cent in 2025.

This is cannibalisation, and it is not linear. Every gigawatt that charges in the cheap hours lifts those prices a little, and every gigawatt that discharges in the high price hours softens those. As I have said before, the fleet eats its own lunch. The more of it there is, the flatter the day it leaves for the next entrant, which from a regulator perspective is the idea and something to welcome. The clearest reading of that flattening is the count of hours when power sells for less than nothing. Germany ran three hundred and one negative price hours in 2023 and four hundred and fifty seven in 2024, a record. Britain logged sixty six negative hours in June 2025 against thirty five in the same month a year before. Each negative hour is a charging opportunity, which is the case for building, but the run of them tells you the middle of the day is collapsing toward a floor.

The defence is duration. The first British fleets were one hour assets built to chase frequency response. As that service saturated and spreads compressed, the economics moved to longer machines that can hold charge across a flatter day and arbitrage the evening peak against a midday trough that now lasts hours rather than minutes. The standard battery clearing the British capacity market is now a two-hour asset, but here it comes, as promised to anyone that would listen! Developers are bidding four, six and eight-hour systems. Fidra Energy alone took fifteen-year awards for the 1.4 gigawatt Thorpe Marsh and 500 megawatt West Burton C projects. The build is not slowing. It is lengthening, which simply moves the day the squeeze finishes its work.

The build is also moving behind the meter and onto shared sites (morte on this another day). Co-location, a battery wired to the same grid connection as a solar or wind farm, is now close to the default for new projects in the mature markets, because it doubles the use of a scarce connection and lets the battery soak up generation that would otherwise be curtailed or sold into a negative price. It is a sensible answer to the squeeze and it deepens it, because every co-located battery is one more buyer in the cheap hours and one more seller in the dear ones, added to a connection that cost little extra to secure.

Behind all of this sits the grid itself. As wind and solar outrun the wires that carry them, the volume of generation curtailed because the network cannot move it has climbed, and the cost of managing those constraints has become one of the largest items the British system operator carries. A battery sited on the right side of a constraint earns from easing it, charging where power is trapped and cheap and discharging where it is needed and expensive. That revenue is real and it is growing as developers realise, and, in thai case, it depends on where the asset sits rather than how cleverly it trades, which is one more reason the location of the next gigawatt matters as much as its size.

The gas alibi

There is an obvious objection to all of this, so let’s address that here as well. Gas has fallen a long way. The Dutch benchmark that peaked above three hundred and forty euros per megawatt hour in 2022 traded between thirty and fifty for most of 2024 and 2025. Gas sets the price of the high priced hours, so a marginal fuel that has dropped roughly eightfold drags the top of the spread down with it. A good part of the compression measured against a 2022 baseline is the war premium unwinding, or being managed well, rather than batteries eating their lunch, and anyone modelling off the 2022 spread is half measuring the gas crisis.

Three things say the rest of it is structural.

• The first is the week this piece opened with. Gas rose seventy six per cent in March and the spread did not widen, so the price level cannot be what holds battery revenue down. If it were, the shock would have paid.

• The second is frequency response. That market buys a fixed volume of a single service, and its price collapsed because too many batteries turned up to sell it, which has nothing to do with the price of gas.

• The third is the floor of the day. Germany ran four hundred and fifty seven negative price hours in 2024 against three hundred and one the year before, and solar capture fell to fifty four per cent, both while gas was already low and falling. The cheap end of the spread is collapsing for reasons gas cannot explain.

The honest qualification is weather. 2025 was a quiet year for wind and for price volatility, and a stormier one would lift spreads and revenue and soften the picture for a while. The structural case does not rest on a single soft year. It rests on the build, the saturated ancillary markets and the lengthening midday trough, none of which a return of volatility undoes.

The wave behind the fleet

What is built is not the whole story. What is consented and waiting is larger, and the gap between the two is where most of the confusion in this market lives.

EXHIBIT 3 · CAPITAL FOLLOWS THE BUILD, NOT THE SPREAD

Spread against installed capacity, bubble size the connected pipeline. Sources: ENTSO-E via European BESS Index; SolarPower Europe and Wood Mackenzie; national TSO disclosures.

Britain and Germany alone hold more than half of what Europe has built, and both already show compressed spreads and saturated frequency markets. Behind the installed fleet sits a pipeline two to three times its size, most of it consented in the same handful of mature markets. The thin eastern systems with the widest spreads carry comparatively little. So the next several years of building will, in the main, pour more capacity into precisely the markets where the gap is already closing. Capital is following the build, not the spread. We are about to witness a new solar grid buy back but this time it comes dressed as a battery project.

The pipeline number deserves a warning, because a connection offer is not a battery. For years the queues read as commitments when they were closer to call options, cheap to hold and easy to abandon, and they overstated the deliverable build by a wide margin. Britain has just admitted as much. In late 2025 the National Energy System Operator tore up the connections queue and removed about three hundred and eighty gigawatts of projects that could not prove they were ready to build, around a hundred and fifty gigawatts of it battery storage. The survivors passed a readiness test and hold land, planning and a credible date. The cull did not change the physics of the squeeze. It changed the honesty of the pipeline. Britain has moved from a development market to an execution market, and the projects still standing are the ones that will actually compress the spread.

The size of the old overstatement is worth stating plainly. Before the cull the connected and contracted queue implied a build several times larger than anything the supply chain, the planning system or the balance sheets behind it could deliver, and a working rule on most desks was to discount the headline pipeline by a factor of four to six to reach a number worth planning against. The reform did that discounting in public and in a single pass. What remains is not small. It is simply real, and real is what compresses a spread.

Arbitrage alone is finished

Put the spread compression and the build wave together and the conclusion for an operator is blunt. A battery that trades a single market is no longer a business.

EXHIBIT 4 · ARBITRAGE ALONE IS FINISHED

Estimated revenue per MW per year, day-ahead arbitrage only against full multi-market stacking, thirteen markets. The shaded band is the indicative breakeven for a two-hour merchant system, around €70 to €90k/MW/year and higher where grid fees bite. Sources: European BESS Index benchmarks; breakeven from Enspired and Capstone DC.

Across thirteen European markets, an asset trading day-ahead arbitrage alone earns somewhere between forty and seventy thousand euros per megawatt year. An asset that stacks several streams, intraday, the balancing markets, frequency response and capacity payments where they exist, earns between roughly ninety five and one hundred and sixty thousand. The premium for trading everything is between two and two and a half times, in every market on the chart. The single-market merchant battery, the model that underwrote much of the early investment case, no longer pays for itself. Set against a two-hour merchant breakeven of roughly seventy to ninety thousand euros per megawatt year, and higher where grid fees bite, day-ahead arbitrage reaches the foot of that band in only the two widest markets and full stacking clears it in all thirteen.

And it is not only the spread. Frequency response, the service that made the first fleets and once paid more per megawatt than arbitrage ever did, has saturated harder than anything else, because the grid needs only a fixed few gigawatts of it and the fleet sailed past that point years ago. The cash cow went first. A British battery earned about one hundred and fifty thousand pounds per megawatt year from frequency alone in 2022. It now earns around twenty five thousand. What looks like a story about thin arbitrage is really a story about every single market going thin at once, the fallbacks included. Stacking is not a clever optimisation. It is the only way to build a living out of markets that have each gone too shallow to stand on alone.

Great Britain shows what the stacked number really looks like once the year is averaged out. Modo Energy put the British fleet at about seventy two thousand pounds per megawatt year across 2025, up from roughly fifty one thousand in 2024, and the monthly profile was violent, from fifty five thousand in the depth of summer to the high seventies in winter. The capacity market does less than its headline suggests. The four year ahead auction for 2028 and 2029 cleared at sixty pounds per kilowatt year, but a two-hour battery is heavily derated, so that payment lands on a slice of its nameplate and covers a slice of the build. The problem is that the money that keeps an asset alive is the merchant stack, and the merchant stack is the thing the build out is compressing.

Nowhere is the move toward real time skill clearer than in the British balancing mechanism, the market the system operator uses to shift power minute by minute. Battery dispatch into it hit record volumes through 2025, and the assets that win there are the ones bidding the right price at the right second, not the ones with the most megawatt hours. Being dispatched, what the trade calls the in merit rate, has become a measure of how well an asset is run rather than how large it is.

The contracted floor

So the market is splitting in two, sorted by who can guarantee a return rather than chase one. The operators winning are not the ones with the cleverest single trade. They are the ones who can switch between markets in real time and the ones who have put a floor under the merchant upside.

The floor takes three forms. The oldest is the tolling agreement, where an offtaker pays a fixed fee for the right to dispatch the battery and keeps the trading risk for itself. The second is a revenue floor, where a counterparty guarantees a minimum and shares the upside above it. The third is new and tells you where the market is heading. Modo Energy’s British revenue index is now authorised by the Financial Conduct Authority, which means a developer can sell a fixed for floating swap against it and swap a known number against an unknown one, the way a generator hedges power or an airline hedges fuel, it’s a whole new topic to explore. A merchant asset is becoming a hedgeable one, oh and Lenders have noticed. The pure merchant project, financed on a 2022 spread and a hope, is no longer bankable, if it ever truly was! And the debt is migrating to assets with a contracted revenue underneath them.

The repricing runs through the debt. A pure merchant battery could maybe once be financed on a forecast spread, with lenders sizing a loan against a revenue line nobody could guarantee. That market has closed. Banks have now started to size debt against the contracted floor and treat the merchant upside as the reward to equity, which means an asset without a floor raises less debt, at a higher price, against a more sceptical model. A tolling deal or a revenue floor is no longer a way to trade upside for safety. It is the difference between a project that funds and one that does not. The contract has become part of the asset.

The read across

For Great Britain, the most mature market in Europe, this is not a forecast but a description. Frequency response revenue collapsed as capacity doubled, the capacity market funds very little on its own once de-rating is applied, and the assets that survive are those with sophisticated optimisation and contracted revenue beneath them. The connections reform has turned a speculative queue into a buildable one, which means the compression has further to run, not less.

The same film is running elsewhere at different speeds. Italy crossed two gigawatts of storage in 2025 and its MACSE auctions are pulling development south toward a target of as much as fifteen gigawatts by 2030, a capacity-led build that will do to Italian spreads what merchant building did to German ones.

The congestion underneath

There is a second reading of the squeeze in the heart of Europe, and it complicates the first. Germany trades as a single bidding zone, one price for the whole country, which the market treats as a copper plate with unlimited wires inside it. It is nothing of the sort. The wind is in the north, the demand and the industry are in the south, and the lines between them are full and have been for years, no matter the political protest and handwringing. So the national day ahead spread that a battery trades is a fiction in one important sense. It ignores where the battery is actually located.

The cost of that fiction is visible elsewhere. Germany spent about two point eight billion euros in 2024 ordering wind farms in the north to switch off and gas plants in the south to switch on, the redispatch bill that keeps a congested grid from falling over, and that bill has risen roughly fifteen fold in a decade. A machine on the right side of a southern bottleneck is worth far more to the system than its national arbitrage spread suggests, because it can charge on trapped northern wind and discharge into southern scarcity, and do for nothing what redispatch pays gas plants billions to do. The squeeze in the centre is partly real saturation and partly a pricing system that hides the value rather than destroys it.

Whether that value reaches the battery is now a live question. The European regulator ran a review of Germany’s bidding zone through 2025 and judged the country would run more efficiently split into five zones, a change it values at something like four hundred and fifty to five hundred and forty million euros a year. Germany, for reasons that are political rather than technical, does not want to be cut in two. If it ever is, southern spreads widen and a slice of the compressed merchant value reappears in the price where a battery can trade it. Until then it stays on the redispatch bill, and the operators who reach it are the ones wired into the congestion mechanisms directly rather than the day-ahead market.

The lesson is uncomfortable for anyone modelling merchant arbitrage off a 2022 spread. The headline number that sold the boom is compressing, the build that compresses it is accelerating, and the returns are migrating from the price of volatility to the skill of trading it and the strength of the contract beneath it. The gas shock did not break the case for batteries. It revealed what the case had quietly become.