28a The value of renewable energy as it scales

Editor’s note: this chapter is new in the 2026 revision. It is not David MacKay’s writing. It extends his balance-sheet method into a question the 2008 book leaves largely open: what happens to the economic value of renewable electricity as it grows to the scale the balance sheet calls for. Written by Örjan Lundberg; every figure is sourced.

MacKay’s book answers one question with great care. Can sustainable sources physically supply as much energy as we consume? The balance sheet adds up kWh per person per day on each side and asks whether the green stack can match the red. By 2026 the answer, for electricity at least, is yes in many places. The physical potential is real, and the cost of building it has fallen a long way: solar modules cost about a twentieth of their 2008 price, utility-scale solar electricity about a tenth of its 2010 cost, and onshore wind and battery costs have fallen steeply too.¹

The British grid MacKay described has already changed in one respect that matters for the rest of the book. In 2008 coal was the incumbent baseload at about 124 TWh; the last coal power station closed in September 2024, and by 2025 coal supplied essentially none of Britain’s electricity. Wind grew twelvefold over the same period, solar from nothing, and bioenergy fourfold, while gas and nuclear both fell.² MacKay’s arithmetic is untouched, since it concerns physical potential and holds whatever the starting mix. What moves is the baseline of several later chapters: the five energy plans of Chapter 27, the nuclear question of Chapter 24 and the fossil-fuel discussion of Chapter 23 were written against a coal-heavy grid that no longer exists. Where a chapter’s argument leans on that baseline, this edition flags it.

A second question then takes over, and the 2008 book barely touches it, because in 2008 there was not yet enough weather-dependent generation for it to bite. The question is what a unit of renewable electricity is worth once a great deal of it has been built.

A wind turbine or a solar panel has no fuel cost, so it bids into the electricity market at close to zero and sets the price in every hour it runs. Add more panels and the price falls in exactly the sunny hours when they all produce at once, because their output is correlated. The revenue each panel earns per unit of energy therefore falls as more panels are built. This is not a market failure. It is the market working. The effect has a name, cannibalization, and a measure, the capture price: the average price a generator actually realises, weighted by its own output, rather than the flat average price.³

The gap is already large. In the German solar market the capture price has fallen as installed capacity has grown, and in a detailed study of the Swedish system a wind-dominated 2035 shows one thousand to fifteen hundred hours a year at a price of zero.⁴ Weather-dependent generators earn the lowest per-unit prices of any source, because they are all selling at once.

Britain shows the same pattern in measured data rather than in a model. Across 2025, GB wind captured about 90% of the average market price and solar about 82%, the only two sources below the average, while gas earned a premium of almost 20% by running in the scarce hours and nuclear sat at the average.⁵

This changes the useful question. MacKay asks how much we can build. Once it is built, the question becomes what absorbs the surplus, because whatever drains the cheap hours decides where the value lands. A study of the German energy system to 2045 makes the point sharply: a system short of flexibility does not deliver cheap power. It delivers a wider spread of prices, with both more zero-price hours and a higher average price, because the peak still has to be covered by plant that now earns its keep in fewer hours. Constraining flexibility raises annualised system costs by 6.9% and the electricity price by €14/MWh, and lifts hydrogen and e-fuel imports by 47%. Supplying it generously moves the cost and the price the other way, by 3.3% and €3 to €5/MWh, and cuts the backup capacity the system must keep available by 22%, some 22 GW. Where flexibility is plentiful only 7% of hours price at zero.⁶

The surplus does not reach demand on its own. Between the cheap hours and the demand that would use them stand two gates. The first is a price gate. A flat per-unit electricity tax puts a floor under what a buyer pays, so however deep the surplus goes the delivered cost stops falling, and the strongest signal a market can send, a negative price, never reaches the meter. Sweden and Finland run the cleanest natural experiment on this. With comparable district-heating hardware but opposite tax treatment, Finnish electric boilers supplied about 8% of district heat in 2025 while the larger Swedish fleet sat almost idle.⁷ The second gate is physical. A project can be economic on every hour of the year and still wait years for a grid connection, or be refused one.⁸

Storage, and the demand that arrives with the supply

There is a way to undo much of the cannibalization, and MacKay half-describes it already. If something consumes the surplus in the hours it appears, the trough is filled, the price recovers, and the capture price stops falling. Storage does this from the supply side. A pumped-hydro scheme like Dinorwig, which MacKay treats in Chapter 26, “Fluctuations and storage”, buys power when it is cheap and sells it when it is dear, which is the same as buying in the surplus hours and selling in the scarce ones. Heat storage does it from the demand side. A district-heating system with a hot-water store can run its electric boilers when power is cheap and coast on the store when it is not, which is why Finland’s boilers earn their keep while Sweden’s, taxed at a flat rate and with little storage behind them, sit idle.⁹

Storage does not escape the trap it is meant to cure. A battery earns from the spread between cheap and dear hours, and as more batteries are built they all charge in the same cheap hours and discharge in the same dear ones, narrowing the spread they live on. The revenue figures already show it: in California the arbitrage revenue a battery earns has fallen year after year, and the Australian market has reached record lows.¹⁰ A generator competes away its capture price and a battery competes away its spread; the effect is general, and it reaches every resource whose value comes from a price signal. No single store answers every duration, either: lithium-ion is the cheapest way to shift energy across a few hours and among the dearest to shift it across a hundred, so the merit order of storage inverts as the duration lengthens. Batteries cover the daily swing, and the seasonal balance needs another technology, or demand that can wait.¹¹

This is why a purely merchant market under-builds the system as a whole, and why some planning guidance is needed. A wind farm, a battery and a flexible boiler each hold their value only if the others are there, yet none is paid to consider the others, and each meets the same falling return as it scales. What is missing is coordination: sizing storage and flexible demand against the generation being built, rather than hoping the market assembles the parts in the right proportion after each has separately learned that it cannot. In Donella Meadows’s ranking of where to intervene in a system, a tax rate or a queue rule is a weak lever near the bottom, and the goal the system is steered toward is a strong one near the top. A market told only to minimise the cost of each megawatt-hour will not build the storage and demand that make the megawatt-hours worth producing. It has to be given that goal.

Not every flexible consumer is equally worth building. Chapter O sets out Michael Liebreich’s hydrogen ladder in full. It ranks the uses of clean hydrogen from the unavoidable, fertiliser and refining and methanol, down to the uncompetitive, domestic heating and cars.¹² The lesson carries over to flexible demand in general: the load worth pairing with intermittent supply is the one whose end product is genuinely valuable, not any load that can be switched on when the wind blows. Electrolysis is a good sponge for surplus power where it feeds a factory that needed the hydrogen anyway, and a poor one where it is a roundabout way to heat a house. The German study cited above puts a number on this, by one measure. Judged by how many hours price at zero, the timing of electrolysis matters more than the timing of cars: running electrolysers and synthetic-fuel plants on a flat schedule leaves 22 to 23% of hours priced at zero, against 15% for taking away flexible vehicle charging altogether. That is a single metric from two single-lever scenarios, and the same study’s shares of daily flexibility put stationary batteries at 38% and flexible vehicle charging at 30% by 2045, with electrolysis not ranked at all, so it is one line of evidence rather than an ordering of flexible loads.¹³

The corollary is economic rather than physical, and it is the sharper half of the point. Because the capture price falls as intermittent supply grows, each new tranche of wind or solar pays only if something is built alongside it to absorb the surplus and hold the price up. To keep adding intermittent producers you have to keep adding flexible consumers: storage, electrolysers, heat that can wait, cars that charge when the wind blows. The demand has to arrive with the supply. Where it does not, the realised price falls, and the build-out continues only because a support payment makes up the difference.

Policy is starting to force this into the open. On 29 July 2026 the German government agreed to phase out the fixed feed-in tariff for new small rooftop solar, so that new installations must sell directly on the market rather than at a guaranteed price. The economy minister, Katherina Reiche, framed it as ending the “produce and forget” principle, under which curtailment costs are borne by the public “regardless of whether the electricity is needed”.¹⁴ Stripped of the guaranteed price, a new panel is exposed to exactly the cannibalized market price described above, and its owner now has a reason to care whether the electricity is worth anything in the hour it is made.

Sweden had already done it, and said the same thing in plainer terms. On 17 September 2024 the finance ministry announced that the tax deduction for installing solar cells would fall from 20% to 15% from July 2025, and that the separate tax reduction for mikroproduktion av förnybar el — the credit a household earned for selling its surplus back to the grid — would be abolished outright from January 2026.¹⁵ The reasoning given was not that solar had failed but that it had worked: installations had “ökat kraftigt”, past 250 000 systems by the end of 2023, so the support was no longer needed and solar should stand på marknadsmässiga grunder, on market terms, with the electricity price left to guide where the next panel goes. Withdrawing the microproduction credit alone was reckoned to be worth 680 million kronor a year to the treasury.

Two governments of quite different politics, two years apart, reached the same conclusion by the same route: a payment that is indifferent to when the electricity arrives stops being a subsidy for clean energy and becomes a subsidy for ignoring the market. That is the argument of this chapter, arrived at from the fiscal side rather than the physical one.

It has a second effect, and this one supplies the missing half rather than merely exposing it. A household with panels faces two quite different prices. What it earns for exporting a surplus kilowatt-hour is roughly the spot price; what it saves by consuming that kilowatt-hour itself is the full retail price — spot plus energy tax plus network fee plus VAT. Self-consumption has always been worth several times export, and the microproduction credit existed to narrow that gap. Removing it widens the gap again, and the gap is precisely what a battery is paid to arbitrage: it turns exported kilowatt-hours into self-consumed ones. The same flat tax that, on the industrial side of this argument, floors the price and blunts the signal reaching an electric boiler, works the other way at a domestic meter, because avoided consumption avoids the tax too.

Sweden’s tax code now says this outright. Under the grön teknik deduction the rate for installing solar panels is 15%, while the rate for installing storage of self-produced electricity is 50%, as it is for an EV charge point, up to 50 000 kronor per person a year.¹⁶ Support has been shifted, by more than a factor of three, from adding intermittent supply to adding the things that absorb it. In practice that means batteries of the order of 10 to 20 kWh for a single house and perhaps 100 kWh for an apartment building, sized to move an evening peak rather than to ride out a still week — which is the right scale for the daily cycle that cannibalizes solar, and the wrong one for the seasonal problem that chapter 26 is about.

So the reform does more than expose the owner of a panel to the cannibalized price. It changes what the household is rewarded for building next, from another panel to the storage that makes the panel worth more. That is the mechanism this chapter has been asking for, working at the smallest scale in the system. Whether it also works at the scale of a smelter is the harder question, and there the answer still runs through the grid connection and the industrial tariff rather than through a household’s tax return.

What a twenty-year guarantee costs when the value falls

The German reform did not come from a change of heart. It came from a deadline. The European Commission’s state-aid approval for the EEG expires on 31 December 2026, so a compliant replacement has to be in force on 1 January 2027 or new plant cannot lawfully be supported at all.¹⁷ That is worth saying plainly, because it changes what the decision was. A country that had paid the subsidy for a quarter of a century did not stop because it concluded the subsidy was wrong. It stopped because the permission ran out and something had to be written in its place.

The status matters for how firmly any of this should be read: what follows is a ministry draft, sent to consultation on 20 July 2026 and agreed in cabinet on 29 July, now going through the parliamentary process, and the numbers may move. As drafted, what replaces the guarantee is modest. New systems get a transitional payment set below the old tariff and capped at 36 months. Separately, systems under 25 kW get a direct-marketing bonus of 1.5 cents per kilowatt-hour for their first four years in direct marketing — the two run for different periods, which is in the draft rather than a slip here. The obligation to sell directly reaches systems of 50 kW and above in 2027 and steps down year by year until, by 2030, it covers every new system except plug-in and mini-PV units of 2 kW or less.¹⁸

The reason the reform is contentious is the size of what is being wound down. The EEG has lately moved about €18 billion a year. In 2024 the EEG paid solar producers roughly €10 billion above what their output earned on the market — about twice the top-up wind received, while solar generated 75.9 TWh against wind’s 139.8, a little over half as much.¹⁹ The explanation is not that solar is inefficient but that the payments still being honoured are the guarantees written in the 2000s, when a rooftop system was awarded 40 to 57 cents a kilowatt-hour for twenty years. Averaged across the whole installed fleet, German solar still drew about 17.4 cents per kilowatt-hour in 2024, while a system commissioned that same year received 6.5 to 10.7.

Now set that against what the electricity was worth in the hours it was actually made. In April 2026, 123 of Germany’s 720 market hours cleared below zero, and 105 of them — 85% — fell between 10:00 and 16:00, which is the solar block. The mean price across those negative hours was −€36/MWh, and the deepest single hour reached −€413.8/MWh.²⁰

That juxtaposition is this chapter’s argument standing in one place: a fixed payment of 17.4 cents against a market value that is not merely low but negative, in precisely the hours the asset produces. But it needs one qualification, and the qualification is the interesting part. The generator on a legacy guarantee does not pay anyone to take its output. The EEG account does, which is to say the public does — exactly the “produce and forget” arrangement Reiche described above, in which curtailment and negative-price costs are borne by the public “regardless of whether the electricity is needed”.

Germany has been closing that door for two decades, but from the top down by size and only forward by vintage. Large plants have lost their payment during runs of negative prices since the 2016 vintage, and the trigger has tightened from six consecutive hours to four, then three, and reaches one hour in 2027. The Solarspitzengesetz of 25 February 2025 went much further: it removed the size floor that had confined the rule to plants of several hundred kilowatts, so that a system as small as 2 kW now loses its payment in every single quarter-hour the day-ahead price is negative.²¹ What all of this shares is that it looks forward. The 40-to-57-cent contracts from the 2000s — the ones the €10 billion is mostly paying — sit below every size threshold ever set and predate every vintage band, and keep their money whatever the price does. Cannibalization at its limit does not show up as a low price. It shows up as a bill, and the question is only whose name is on it.

One argument to be careful with

The German debate about this is fierce, and one move in it is worth naming because a reader will meet it often. It is frequently observed that after twenty-six years of the EEG, Germany still takes about 80% of its primary energy from fossil fuels, “almost unchanged”. The actual figure for 2025 is 76.4%, with renewables at 20.6%. But the gap between 80 and 76.4 is not the problem with the claim. The problem is that it does not test what it appears to test, because the EEG is an electricity law and primary energy counts the petrol, diesel, heating oil and gas boilers it never addressed. On the quantity the law was actually aimed at, the renewable share of gross electricity consumption rose from 6.3% in 2000 to 55.1% in 2025.²²

Both numbers are real, and only one of them is a measure of the policy. This is worth flagging in a book built on unit arithmetic, because it is the characteristic failure that MacKay wrote against: not a wrong number, but a right number answering a different question. The case against the cost does not need it. The case against the cost is the €10 billion, and it stands on its own.

Whose money, and whose bill

The €10 billion is where an old observation becomes exact. Adam Smith observed in The Wealth of Nations (1776) that the directors of a joint-stock company, “being the managers rather of other people’s money than of their own”, cannot be expected to watch over it as carefully as they would their own. The hazard he described appears wherever the person deciding what to build is not the person who bears the loss if the thing turns out to be worth less than it cost, and a guaranteed tariff puts exactly that distance between the two. The panel is built once, the payment arrives whether or not the electricity is wanted in the hour it is made, and the cost of the mismatch falls on a levy account that nobody making the decision feels. The way out is the storage and the flexible demand that let the energy earn its own value, so the investment rests on the price it commands rather than on the payment that currently stands in for it.

There is a reading of all this that connects back to the reason MacKay wrote the book. Cheap energy, if it reaches industry, is what keeps energy-intensive industry alive, and that industry is a large part of what an economy runs on. The same surplus that starves a generator’s revenue is, read at the buyer’s meter, cheap input, but only if the gates let it through. Germany shows the loop run backward: after its gas price rose, energy-intensive output fell to about 82% of its 2021 level and stayed there.²³

What it costs to build against what it earns

The capture price becomes a decision rather than an observation when it is set beside the cost of building the thing. That comparison is rarely drawn, because the two numbers come from different places: the cost from engineering studies, the earnings from settlement data.

Figure 28a.4. Cost to build against price realised, Great Britain. Cost is the Department for Energy Security and Net Zero’s levelised cost for projects commissioning in 2035, in 2024 prices, with the low-to-high capex range shown. Earnings are the price each technology actually realised on the GB market in 2025, weighted by its own output.²⁴

£/MWh	Cost to build	Earned in 2025
Large-scale solar	44	65.9
Onshore wind	41	72.0
Offshore wind, fixed	59	72.0
Offshore wind, floating	91	72.0
Gas CCGT	45	94.8
Gas with carbon capture	101	94.8

Four of the six clear their costs on today’s prices. Floating offshore wind does not — 91 against 72 — which is the quantified version of the proviso chapter 10 records MacKay repeating from wind experts in 2008, that deep offshore is prohibitively expensive. Eighteen years later it is still about a quarter short. Gas with carbon capture does not either, at 101 against 94.8, which is chapter 23’s finding in one line: capture works and has not been built, for reasons of cost rather than physics.

Before reading anything into those margins, one asymmetry in the chart has to be admitted. The two columns are not measured on the same boundary. The right-hand column is a system quantity: a capture price is what the market actually paid, in the hours the plant actually produced, with everything the system was doing at the time already in it. The left-hand column is an asset quantity: DESNZ’s levelised cost counts capital, fuel and operations inside the fence, and excludes the balancing capacity a variable source needs, the curtailment it suffers, and the capacity utilisation it takes away from the dispatchable plant that must stay available for it.

That asymmetry has a direction. It leaves the cost of a variable source understated relative to a dispatchable one, so the margins in the table are too generous to solar and wind and too harsh on gas — by an amount nobody in this chart has measured. Chapter M sets out the methodological case and one attempt to quantify it: a Norwegian onshore wind farm costed on a system boundary came out 19% above its asset figure, 34% once uncertainties were carried through, and up to 3.5 times if the capacity model was left at nameplate. Those numbers come from a single case with a strong assumption — it firms one wind farm to a dispatchable standard using dedicated gas, which no system operator actually does — so they are an upper bound rather than a correction to apply. The direction, however, is not in doubt.

Which means the honest reading of the chart is not “four of six clear their costs by these margins” but “four of six clear their costs, and the variable ones by less than shown.” The £13/MWh that fixed offshore wind clears by is the number most exposed to this, and it is the number the rest of this section leans on.

And the chart is a snapshot of a moving thing, which is this chapter’s whole point. The right-hand column is what the existing fleet earned. Build more of a technology and its capture price falls, because the new plant produces in the same hours as the old. So a project appraised against today’s £72 will not receive £72 once it and its neighbours are running: wind’s value factor of 0.90 has further to fall.

Read that way the figure states the investment problem exactly. Fixed offshore wind clears its cost by about £13/MWh, and its own expansion is what closes that gap. The CfD exists to bridge it, which is why chapter 4’s note on AR6 matters: a strike price of about £82 in today’s money against a captured £72 is the public purse paying the difference between what the market gives and what the machine costs.

And it explains why the strike price and the levelised cost do not match. AR6 cleared at roughly £82/MWh indexed while the levelised cost here is £59. A bid below the published cost of building is not a contradiction; it is what happens when a contract removes the price risk, because the discount rate a project is financed at falls when its revenue is guaranteed. The CfD’s real product is cheap capital rather than a high price. That is the same insight as chapter 28a’s opening, arriving from the financing side: what a generator is paid for is not only energy but certainty.

The same comparison, for a country that imports its fuel

The table above is a British question, and it takes a particular form because Britain has a liquid wholesale market: the thing a generator is measured against is the price other generators were paid. For most of Asia the comparison is a different one, and it is much less forgiving of fossil fuel.

An importer is not choosing between two ways of earning £/MWh in an auction. It is choosing between building a machine once, in its own currency, and buying fuel every year, for ever, in someone else’s — through a strait that chapter N describes being closed in March 2026. Asia has 54% of the world’s population and 2% of its oil reserves, imports about $1.1 trillion of fossil fuel a year at over 3% of regional GDP, and has solar and wind resources sufficient for at least fourteen times its total energy demand. Set beside those numbers, the levelised cost of solar is not competing with a capture price. It is competing with an import bill and a currency risk.

That is why the bet is being placed, and the exceptions are what confirm it. The countries paying most — Thailand at around 7.5% of GDP, then Taiwan, South Korea, Singapore, Viet Nam, Pakistan, the Philippines — have the most to gain from displacing the bill. The countries sitting near zero on the same measure are Malaysia and Myanmar, and they are near zero because they export. An exporter’s arithmetic runs the other way: every kilowatt-hour of domestic solar displaces a barrel it would rather have sold. Russia is the limiting case of that position: the larger the export revenue, the weaker the domestic reason to displace it.

The caution is that this cuts both ways as an explanation. Import dependence explains the urgency, not the whole of the behaviour — Malaysia and Brunei are both exporters and both ahead of the United States on electrification, because the share of final energy delivered as electricity is driven by what a country builds and sells as much as by what it must buy. And the fourteen-times resource figure is a technical potential, the same class of number as MacKay’s own green stack in chapter 18, which he was careful to say was not a forecast. What the trade balance predicts is which countries have a reason to hurry, and on that it predicts well.²⁵

Three things that are not the same thing

There is a claim in wide circulation that follows naturally from everything above, and it is worth testing because it is testable. The claim is that a system which closes dispatchable plant and builds weather-dependent generation in its place must pay steadily more to keep itself in balance, since the machines that used to hold the frequency have gone.

Before it can be tested, three quantities have to be separated, because the argument runs them together and they are not the same. Rotational energy, the spinning mass of synchronous machines, governs how fast the frequency begins to fall in the first seconds after a large sudden fault. Ancillary services are the products a system operator buys so that something arrests that fall and then restores the frequency. The balancing market is where energy is actually activated, quarter-hour by quarter-hour and area by area, to close the imbalance that remains. Only the first of the three is a question about spinning mass. The second is a procurement question and the third is a market-design question, and as the rest of this section shows, it is the second and third that the money went to.

Start with what sizes the fast reserves, because it is not what the argument assumes. Svenska kraftnät sizes FCR-D up against the Nordic dimensioning fault and names it: Oskarshamn 3 at 1450 megawatts, assumed unchanged to 2030. It sizes FCR-D down against a fault of 1400 megawatts, being full export over either NordLink or North Sea Link. The two largest instantaneous events the Nordic system plans for are therefore its largest nuclear unit and a high-voltage cable. A wind farm does not disconnect 1450 megawatts in one step; a single reactor does, and commissioning a larger one would raise the requirement. On this product the relationship between nuclear plant and fast reserves runs the opposite way to the one the story assumes.

Then look at the volumes, which Svenska kraftnät also publishes, product by product, out to 2030. Sweden’s share of the Nordic requirement is 224 megawatts of FCR-N, 542 of FCR-D up and 524 of FCR-D down, and all three are flat every year to 2030. What grows is the restoration side: aFRR doubles from 150 to 300 megawatts in each direction, and mFRR up rises from 800 to 1400, close to the 1450 megawatts of the reference incident. The containment reserves do not grow at all: 1290 megawatts in 2025 and 1290 in 2030, of which the part sized directly by the largest single fault, FCR-D up and down together, is 1066. The restoration reserves grow by about half, from 2090 to 3150. But the tidy reading of that contrast — fault-driven reserves flat, imbalance-driven reserves growing — does not survive the dimensioning rules. Under the Nordic methodology the total FRR requirement is the sum of aFRR and mFRR for normal imbalances plus mFRR for disturbances, and the disturbance component must cover at least the control area’s reference incident: the largest instantaneous loss of a single generating unit, a single demand facility or a single HVDC interconnector. mFRR alone carries that component. So the same 1450 megawatt reactor that sets FCR-D up also sets the floor under Sweden’s mFRR, reducible only by reserve sharing between the Nordic operators, which is how a 1450 megawatt incident can sit above a 1400 megawatt requirement. The dimensioning fault is present twice in these volumes rather than once. What is genuinely imbalance-driven is aFRR plus the normal-imbalance share of mFRR, and that share is not published separately.²⁶

The cost has moved the same way. Of the 4 281 million kronor by which Svenska kraftnät’s gross ancillary-service bill rose in 2025, mFRR accounted for 6 123 million while the frequency-containment reserves fell by 1 723 million. The figure above is on the net basis, after the revenue Svenska kraftnät earns from reselling these services, which is why its totals run 500 to 1 100 million lower and its year-on-year steps differ; the direction is the same on either basis. What that does not license is the conclusion that the money went to a purely weather-driven product: mFRR is where the cover for losing the largest plant or the largest cable sits, alongside the cover for forecast error. What can be said is narrower and still worth saying. The product whose volume follows the reference incident directly, FCR-D, is flat in volume and falling in cost, and the year the bill jumped was the year the automated energy-activation market opened, in March 2025, not a year the fleet changed.

If that decomposition is right, then the containment bill should not have risen anywhere as dispatchable plant closed, and over a long enough run it is possible to check. Germany is the only one of the four systems in this section whose regulator has published the same measurement, on the same definition, for long enough to read as a series. The Bundesnetzagentur reports each year what the four transmission operators pay to hold the three balancing reserves — primary, secondary and minute reserve, FCR, aFRR and mFRR — available. Over the years in which Germany shut ten reactors and 11 gigawatts, that cost fell from 825 million euros in 2009 to 123 million in 2018, a decline of 85%. It then rose with the gas price to 644 million in 2023 and fell again to 518 million in 2024. Across the whole phase-out, seventeen reactors and 20.5 gigawatts between 2011 and 2023, the bill ended 37% below where it started. The phase-out does not appear in the series. The fuel price does.

Put that number next to the demand it serves and the reason it stayed invisible becomes clear. Germany’s 518 million euros is spread across about 457 terawatt-hours, which is 1.1 euros per megawatt-hour, or a tenth of a penny per kilowatt-hour. Whatever is making electricity expensive, it is not the cost of holding reserves, and the next section puts a figure on what is.²⁷

Sweden is the country where the claim looks strongest, and its own accounts are the reason to be careful with it. Its ancillary-service bill rose 78% in a single year, from 5 455 million kronor in 2024 to 9 736 million in 2025. Svenska kraftnät attributes the mFRR share of that to the automated energy-activation market introduced on 4 March 2025, which moved balancing to fifteen-minute settlement and to activation within each price area, scaled procured volumes up sharply, and did so into a market whose supply, in its own words, had not yet had time to adapt; prices fell again later in the year. Area-based activation is also why a megawatt of reserve in the wrong bidding area is not a substitute for one in the right area, which is a transmission constraint expressing itself as a balancing cost.

The containment side ran the other way for the reason chapter 26 gives. Pre-qualified volume for FCR-D down went from about 100 megawatts in the third quarter of 2021 to nearly 2500 by the third quarter of 2024, against a requirement of 524: hydro supplies about 1500 megawatts of it, with batteries and wind the fastest-growing entrants. Roughly five times the required volume now competes for it, and FCR-D up cleared at 6.32 euros per megawatt per hour in September 2025 against the 60 to 80 euros of 2022. That is chapter 26’s argument arriving in the accounts. The reserve price was never a fact about batteries; it was a fact about how few of them there were.

Finland is the case that decides the question, because it has no closures to blame. It shut nothing, and in April 2023 it commissioned 1 600 megawatts at Olkiluoto 3. Its reserve bill still nearly tripled, from 68 million euros in 2021 to 187 million in 2022. What happened in May 2022 was that imports from Russia stopped, taking with them about 1 300 megawatts of firm capacity that the Finnish system had leaned on for decades. So dispatchable capacity does move the balancing bill. The mechanism in the one clean case was a transmission link, not a reactor, which is a reminder that a system’s firm capacity and its own generating fleet are different quantities.

The price of the same product across Europe is the last check, and the widest one available. Over the four weeks from 1 to 28 September 2026, a megawatt of upward aFRR capacity cost 1.01 euros an hour in Finland and 124 in Poland, with fifteen more systems in between: a factor of a hundred and twenty-three for a service defined by the same European regulation and held to the same standard. Germany, which carries more weather-dependent generation than any of the others, sits in the middle of that range at 12.24 euros, and it buys more of the product than anyone — close to 2 000 megawatts of upward capacity and 1 900 of downward in every hour of the four weeks, 1 344 gigawatt-hours against Spain’s 848 and France’s 588. Its whole aFRR bill over those weeks comes to 1.07 euros per megawatt-hour of demand, which is what the entire German reserve bill came to in 2024. Whatever sets the price of standing ready, it is not the share of wind and solar on the system. If it were, the ranking would run the other way.²⁸

Sweden is the exception on that chart twice over. It publishes no procured aFRR capacity to the platform at all, so its row comes from Svenska kraftnät’s own data service; and what that shows is a system which barely buys the product. It held upward aFRR capacity in forty of the window’s 672 hours, 0.32 gigawatt-hours of it against Germany’s 1 344, and its whole aFRR bill is four eurocents per megawatt-hour of demand against Germany’s 1.07 euros. Downward it bought fifty times as much, 17.3 gigawatt-hours across 348 hours, and paid more for it: 20.59 euros against 16.83 upward. Paying more to absorb than to deliver is not peculiar to Sweden — Germany, Austria, Estonia and Finland all do, five systems of the seventeen — but Sweden pays the most of them, and buys the least upward of any. That shape is this section’s argument arriving from the accounts rather than from the dimensioning rules: Sweden’s restoration is bought as mFRR, which is where the 2025 bill went, and the reserve it is shortest of is the downward kind, which is the one the export cables size.

What has risen, everywhere, is the other bill. German congestion management ran at 2.9 to 3.4 billion euros a year through 2023 to 2025, six times the reserve cost; British constraint payments went from 397 million pounds in 2017/18 to 2 211 million in 2025/26, while the frequency and reserve categories over the same period were essentially unchanged, 146 million against 149 million, with a gas-crisis bulge in between. Those are payments to move power past a grid that is full, or to pay a wind farm not to generate. They belong to the wedge this chapter has been describing, and they are a fact about transmission rather than about balancing.

There is a cost-allocation question underneath all of this, and this chapter of all chapters should not leave it out. The reserve that covers the reference incident is bought by Svenska kraftnät and recovered from the system as a whole, through the network tariff and the balance-responsibility fee. Revenue from the latter rose from 3 977 million kronor in 2023 to 6 408 million in 2025 on rate rises of about 30% in 2024 and 25% in 2025. The requirement itself is created by whichever single unit happens to be the largest. A 1450 megawatt generator obliges the system to hold cover for 1450 megawatts, and is charged through the network tariff in proportion to its size and its energy like any other user, and through a balance-responsibility fee whose base component does not vary with either. Neither charge varies with the contingency the unit creates. Nothing in the Nordic dimensioning rules prices the reference incident to whoever sets it; Article 7(3) lets the system operators share the burden between themselves, which lowers the total, but it does not send any part of the bill to the unit whose size is the reason for it.²⁹

That is the same shape as the complaint this chapter makes about weather-dependent generation. A resource whose behaviour the system must accommodate does not meet the cost of accommodating it, and the cost lands on everyone through a tariff. The difference is which case gets discussed. The wind version is argued about constantly; the unit-size version is argued about almost never, although the Nordic rules state the principle explicitly and Svenska kraftnät puts a megawatt figure on it.

All of which changes the question worth asking. The division of labour in a Swedish-style market is clear enough: a balance-responsible party answers for its own economy and its own imbalances, and the system operator answers for the system. If the operator needs a property — inertia, containment, automatic or manual restoration, voltage support — its job is to specify the requirement and buy it, and then hydro, wind, batteries, flexible demand and anything else that can meet the specification compete to supply it. That is a market working as intended, and the Swedish pre-qualification figures are what it looks like when it does. So the useful question is not how much wind a grid can absorb before it becomes unstable. It is which properties the operator actually needs, how large each volume is, and which resource can deliver it most cheaply. Svenska kraftnät publishes all three, which makes the question answerable rather than rhetorical — and the published answer is that the containment volumes are flat, the restoration volumes are growing by about half, and the cheapest supplier of the first kind is changing fast. It is not an answer about weather, because the reference incident that sizes both FCR-D and the disturbance half of mFRR is a reactor and a cable, neither of which is weather. The one genuinely inertia-linked product, FFR, is the exception: it has no flat published figure, and Svenska kraftnät expects to need more of it.

And Britain’s industrial price is the standing version of it

Britain’s version of the value problem is not a shock but a standing condition, and it is worse than Germany’s negative-price problem described earlier in this chapter. Of the 28 IEA countries that reported industrial electricity prices for 2024, the United Kingdom’s were the highest, counting taxes and levies. Large British industrial users paid about 25.3p per kWh, some 125% above the EU-14 median of 11.3p.³⁰ For a book whose argument is that Britain should electrify — heat, transport, industry — that is the most awkward number in the entire subject, and MacKay’s method cannot see it, because it is not a fact about joules.

Set it beside the wholesale figure from earlier in this chapter and the shape of the problem becomes clear. The GB market price averaged £79.9/MWh in 2025, which is 8.0p per kWh. Large industrial users pay roughly three times that. The difference — some 17p per kWh — is network charges, policy levies, supplier margin and the cost of firming, and almost none of it is the cost of generating the electricity.

That is the reconciliation this chapter has been circling. The complaint of the generator, that wind captures only 90% of a falling average, and the complaint of the smelter, that power costs three times what the market says it does, sound like opposite grievances. They are the same wedge measured from its two ends. Cheap generation and expensive delivered power are not in contradiction; the surplus is real, and it is being consumed somewhere between the turbine and the meter. A country can cannibalise its generators and impoverish its industrial consumers at the same time, and Britain currently does both.

So the argument of this chapter has a corollary it would be dishonest to leave out. Storage and flexible demand let renewable energy earn its own value, which is the answer to the first complaint. They do nothing whatever about the second. If 17p of a 25p industrial price is network, levy and margin, then no amount of cheap wind reaches the factory until that wedge is addressed, and it is addressed by policy rather than by physics or by batteries.

None of this argues against building renewables. MacKay’s arithmetic still holds, and the build-out has largely happened. It argues that the balance sheet needs a second page. The first page asks whether we can produce enough clean energy. The second asks who bears the cost of turning an intermittent surplus into firm, useful supply, and whether policy lets the cheap hours reach the demand that would value them. A fuller treatment, with the feedback loops drawn out and the Swedish and European data behind each figure, is at oluies.github.io/elmix.³¹

---

1. Regenerated by this edition’s data-refresh script (Refresh.scala, DuckDB) from Our World in Data. Solar module cost fell from USD 4.61/W (2008) to 0.26/W (2024). Global weighted-average LCOE (IRENA), USD/MWh, 2010 to 2024: solar PV 417 to 43, onshore wind 113 to 34, offshore wind 208 to 79. Battery pack price, cited separately (BloombergNEF): about USD 1200/kWh (2010) to 115/kWh (2024).

2. Figure regenerated from Our World in Data (Ember and the Energy Institute) by this edition’s data-refresh script (Refresh.scala, DuckDB), in TWh: UK coal 124 (2008) to 0 (2025), wind 7 to 86, solar 0 to 19, bioenergy 10 to 41, gas 176 to 91, nuclear 52 to 36.

3. Capture price = the sum of (production times spot price) divided by the sum of production. Source list at https://oluies.github.io/elmix/modell/referenser.html .

4. Goransson, Johnsson and others, “Tre elsystem som kan mota omstallningen av industri- och transportsektorerna”, Chalmers / Mistra Electrification, 2025.

5. Computed by this edition’s data-refresh script (Refresh.scala, DuckDB) from Elexon BMRS: half-hourly GB generation by fuel type and the market-index price (APXMIDP), 2025. Capture price = sum(generation × price) / sum(generation); value factor against the 79.9 GBP/MWh time-weighted average. Gas 1.19, Biomass 1.08, Nuclear 1.01, Wind 0.90, Solar 0.82.

6. Julian Geis, Michael Lindner and Tom Brown, “Managing the mismatch: The role of flexibility on the path to a carbon-neutral energy system”, Advances in Applied Energy 23 (2026) 100284, from the Department of Digital Transformation in Energy Systems at TU Berlin. A high-resolution, sector-coupled model of the German energy system run to 2045. Daily flexibility needs rise by a factor of 3.7 between 2025 and 2045, driven mainly by solar; by 2045 stationary batteries supply 38% of daily flexibility and flexible vehicle charging 30%. Against the base scenario, constrained flexibility raises annualised system costs by 6.9%, the electricity price by €14/MWh and hydrogen and e-fuel imports by 47%, and gives both the highest average price and the widest interquartile range of any scenario; the high-flexibility scenario lowers system costs by 3.3%, average prices by €3 to €5/MWh and backup capacity by 22%, or 22 GW. Zero-price hours run at 7% in that high-flexibility case, 15% without flexible battery-vehicle charging, and 22 to 23% with inflexible electrolysis and power-to-X. Two cautions the authors state themselves: the demand-side management is modelled with zero activation cost, instantaneous ramping and unconstrained activation frequency, which overstates what a real system would get from it, and the cost assumptions for iron-air batteries, vehicle-to-grid and industrial demand-side management remain highly uncertain.

7. Energiateollisuus district-heating statistics, 2025; Fernqvist, Broberg, Toren and Svensson, “District heating as a flexibility service”, Energy Policy 172 (2023) 113332.

8. Energimarknadsinspektionen finding of 11 June 2026 on connection times; the Power2Earth refusal at Lulea, autumn 2025.

9. MacKay, Chapter 26, “Fluctuations and storage”. Finnish and Swedish figures and the role of thermal storage: Energiateollisuus 2025; Fernqvist and others, Energy Policy 172 (2023) 113332; the Arlanda aquifer store, Swedavia.

10. California battery arbitrage revenue: CAISO Department of Market Monitoring, and Modo Energy’s series of about 103, 80, 51 and 38 USD per kW-year. Australian NEM battery revenues at a record low: Modo Energy, 2026.

11. The storage merit order by duration: installed cost per kWh at a four-hour duty against a hundred-hour duty, where the spread widens to roughly a factor of 22 and lithium-ion falls from second-cheapest to dearest. https://oluies.github.io/elmix/lagring.html .

12. Michael Liebreich, “Clean Hydrogen Ladder Version 5.0”, 19 October 2023, https://liebreich.com/hydrogen-ladder-version-5-0/ .

13. Julian Geis, Michael Lindner and Tom Brown, “Managing the mismatch: The role of flexibility on the path to a carbon-neutral energy system”, Advances in Applied Energy 23 (2026) 100284, from the Department of Digital Transformation in Energy Systems at TU Berlin. A high-resolution, sector-coupled model of the German energy system run to 2045. Daily flexibility needs rise by a factor of 3.7 between 2025 and 2045, driven mainly by solar; by 2045 stationary batteries supply 38% of daily flexibility and flexible vehicle charging 30%. Against the base scenario, constrained flexibility raises annualised system costs by 6.9%, the electricity price by €14/MWh and hydrogen and e-fuel imports by 47%, and gives both the highest average price and the widest interquartile range of any scenario; the high-flexibility scenario lowers system costs by 3.3%, average prices by €3 to €5/MWh and backup capacity by 22%, or 22 GW. Zero-price hours run at 7% in that high-flexibility case, 15% without flexible battery-vehicle charging, and 22 to 23% with inflexible electrolysis and power-to-X. Two cautions the authors state themselves: the demand-side management is modelled with zero activation cost, instantaneous ramping and unconstrained activation frequency, which overstates what a real system would get from it, and the cost assumptions for iron-air batteries, vehicle-to-grid and industrial demand-side management remain highly uncertain.

14. “Government agrees to phase out Germany’s landmark fixed feed-in tariffs for new small solar PV”, Clean Energy Wire, 29 July 2026.

15. “Förändrade skattesubventioner för solceller”, Finansdepartementet, 17 September 2024: https://www.regeringen.se/pressmeddelanden/2024/09/forandrade-skattesubventioner-for-solceller/. The changes were part of an agreement between the government and Sverigedemokraterna; the rate cut was estimated at 100 million kronor in 2025 and 200 million a year from 2026, the abolition of the microproduction credit at 680 million a year from 2026.

16. Rates for the skattereduktion för grön teknik, Energimyndigheten’s Solelportalen: https://www.energimyndigheten.se/effektiv-energianvandning/guider/solelportalen/vilka-stod-och-intakter-kan-jag-fa/skattereduktion-for-gron-teknik/. As of 2026: 15% for solar panels, 50% for storage of self-produced electricity, 50% for an EV charging point, capped at 50 000 kronor per person per year. The battery sizes quoted are typical installed capacities rather than a recommendation.

17. The EEG’s approval under EU state-aid law runs only to 31 December 2026, so an EU-compliant successor regime must take effect on 1 January 2027; without one, new plant could be built and connected but not supported. Industry bodies including the BDEW and the BEE have pressed for a rapid parliamentary conclusion precisely so that the Commission has time to review and approve the replacement. Note that it is the state-aid approval that expires, not the statute itself.

18. Draft EEG reform sent to consultation by the federal economics ministry under Katherina Reiche in July 2026: guaranteed feed-in ends for new rooftop systems, replaced by a transitional payment set below the previous tariff and capped at 36 months, plus a direct-marketing bonus of 1.5 ct/kWh for the first four years for systems under 25 kW. Direct marketing becomes mandatory for systems of 50 kW and above from 2027, extending to smaller classes each year until it covers all new systems except plug-in and mini-PV units of 2 kW or less by 2030. Summary at https://www.pv-magazine.de/2026/07/20/reiche-schickt-entwuerfe-fuer-eeg-und-netzpaket-in-die-verbaendeanhoerung/. The industry association rejected the transitional terms as a “wertloses Trostpflaster” — a worthless sticking plaster — arguing they cannot replace the roughly 7 ct/kWh paid to partial feed-in systems and 10 to 12 ct to full feed-in systems. Those are the association’s own characterisation of the rates the reform removes, spanning the vintages currently in payment, and are not directly comparable with the 6.47 and 10.68 ct/kWh applying to systems commissioned in 2024 quoted in the previous note; EEG tariffs degress downward, so a figure above the 2024 rate refers to earlier commissioning years. Figures here are from a draft in consultation and may change in the parliamentary process.

19. The primary source is the Bundesnetzagentur’s Monitoringbericht, which has not been consulted directly here. The EEG flow of about €18 billion a year, the €10 billion paid to solar producers in 2024 against roughly half that for wind, and the fleet-average 17.4 ct/kWh are as reported from that monitoring report by Daniel Wetzel, “Die Ökostrom-Lobby hängt an den Subventionen wie ein Junkie an der Nadel”, Welt, August 2026. That piece is a signed opinion column arguing a position, in vigorous language, and is used here for its checkable figures rather than its framing; the generation comparison and the per-kilowatt-hour inference are made here from the Energy Institute’s Statistical Review of World Energy 2026, which gives German solar 75.9 TWh and wind 139.8 TWh in 2024. Rates for systems commissioned in 2024 (6.47 ct/kWh and 10.68 ct/kWh by segment) are the Bundesnetzagentur’s published tariffs. The €10 billion and the 17.4 ct figures are on different bases — net EEG payment above market revenue, and gross remuneration rate — so they should not be divided into one another; gross receipts on 75.9 TWh at 17.4 ct would be about €13 billion, not €10 billion.

20. April 2026: 123 of 720 hours cleared below zero on the German day-ahead market, 105 of them (85%) between 10:00 and 16:00 CEST; mean clearing price across the negative hours −€36.0/MWh; deepest single hour −€413.8/MWh. German negative-price hours in the first half of 2026 have been reported at 299 against 389 in the same period of 2025 — down year on year, not a trend within 2026. Note that these German figures run opposite to the Europe-wide count, which roughly doubled in the first quarter of 2026: the two are different geographies and should not be read as one series. See https://www.euronews.com/2026/07/29/wind-and-solar-overtake-fossil-fuels-in-germany-but-negative-electricity-prices-still-rife for the German half-year figures and https://www.pv-magazine.com/2026/05/08/europes-negative-electricity-price-hours-double-in-q1-amid-renewables-surpluses-market-imbalances/ for the European Q1 count. The April 2026 hourly detail predates both and should be checked against the exchange’s own published data before being relied on.

21. Under § 51 EEG the payment lapses when the day-ahead price is negative for a run of consecutive hours, with both the run length and the size floor depending on when the plant was commissioned. Plants of 500 kW and above commissioned from 2016 under the EEG 2014/2017 faced a six-hour trigger with no compensation. Under the EEG 2023, for plants of 400 kW and above, the threshold is four hours for the 2023 vintage, three for 2024 and 2025, two for 2026 and one from 2027. The Gesetz zur Änderung des EEG 2023 known as the Solarspitzengesetz, in force 25 February 2025, replaces the run with the first negative quarter-hour and removes the size floor: systems of 100 kWp and above immediately, and all systems from 2 kWp once a smart meter is fitted. Compensation for lost hours runs under § 51a, which extends the remuneration period — but it does not reach the oldest cohorts, and the 2025 reform replaced the flat hour-for-hour extension with a time-contingent model weighting photovoltaic losses at 0.5 times the negative quarter-hours actually incurred. Installations on the original 2000s guarantees are below every size floor these rules have used and are not reached by any of them.

22. The renewable share of German gross electricity consumption rose from 6.3% in 2000 to 55.1% in 2025; renewables reached 23.8% of gross final energy consumption in 2025. For primary energy, AG Energiebilanzen’s estimate for 2025 puts the fossil share of primary energy at 76.4% and renewables at 20.6%; the largest components are mineral oil at 35.7% and natural gas at 26.9%, with hard coal and lignite making up the remainder. The 76.4% is AGEB’s published total and is not re-derived here from the rounded components. Sources: Umweltbundesamt, “Renewable energies in figures”, for the electricity shares; AG Energiebilanzen’s 2025 annual estimate for the primary-energy shares. The distinction matters because primary energy includes transport fuels and heating fuels that the EEG does not address.

23. Destatis, production index for energy-intensive branches, 2025.

24. Levelised costs are from Electricity Generation Costs 2025, Department for Energy Security and Net Zero: central capex estimates for projects commissioning in 2035, in 2024 real prices — large-scale solar 44 (range 36–50), onshore wind 41 (33–55), fixed offshore wind 59 (50–72), floating offshore wind 91 (65–121), gas CCGT 45 (38–53) and gas with CCUS 101 (85–117), the two gas figures at a 30% load factor. Capture prices are for 2025 from this edition’s data-refresh script over Elexon settlement data. Three cautions. The two sides are not on the same footing: a 2035 commissioning cost in 2024 money against prices realised in 2025 by plant already built, so this is an order-of-magnitude comparison rather than an appraisal. The capture figure for wind combines onshore and offshore, so fixed offshore is measured against a blended price. And DESNZ notes that levelised cost omits system effects entirely — transmission, balancing and the value of being dispatchable — which is precisely what separates the gas column from the wind column and is the subject of this chapter.

25. Ember, Electric Asia: How Asia is leading the electric age, 11 June 2026. The figures, and the cautions that go with them — that Ember advocates for electrification, that its “Asia” excludes Russia, that the “share of Asia” percentages have an unstated denominator, and that the resource multiple is a technical potential rather than a forecast — are set out in full in the corresponding note to chapter N.

26. Volumes are Svenska kraftnät’s own published requirement for Sweden’s share of the Nordic need, from its Framtida volymbehov and Behov av reserver idag och i framtiden pages: FCR-N 224 MW, FCR-D up 542 MW and FCR-D down 524 MW, each unchanged in every year from 2025 to 2030; aFRR 150 MW rising to 300 MW in each direction; mFRR up 800 MW rising to 1400 MW and mFRR down 990 MW rising to 1150 MW, the last four carrying wide uncertainty ranges in the source, which the figure does not draw — mFRR up, for instance, is given as 580 to 1300 MW for 2025 and 1100 to 1850 MW thereafter. The plotted numbers are Svenska kraftnät’s own point estimates, and they are not the midpoints of the ranges it publishes beside them, which for mFRR up would be 940 and 1475. On the bounds the growth could be much larger or could reverse, since 1300 MW in 2025 against 1100 MW in 2030 is within the stated ranges. So the aggregate growth of about a half is a forecast point estimate rather than a measurement, and the word “about” is doing real work. The dimensioning statements are quoted from the same pages: FCR-D up against “det nordiska dimensionerande felet, Oskarshamn 3 på 1450 MW”, FCR-D down against “det nordiska dimensionerande felet på 1400 MW, med full export via antingen Nordlink eller North Sea Link”. Note the direction: it is loss of full export that raises the frequency and calls for down-regulation, not import. For the restoration products the governing document is the Nordic synchronous area methodology for the FRR dimensioning rules under Article 157(1) of Regulation (EU) 2017/1485, as amended and approved by Energinet, Fingrid, Kraftnät Åland, Svenska kraftnät and Statnett. It defines normal imbalances as “the continuous mismatch between generation, demand, import and export”, including stochastic and deterministic imbalances and forecast errors, and disturbances as occasional instantaneous deviations whose largest case is the reference incident: “the largest imbalance that may result from an instantaneous change of active power of a single power generating module, single demand facility, single HVDC interconnector or from a tripping of an AC line”. Article 7(2) requires that each control area’s positive FRR for disturbances “cover at least the positive reference incident for the control area”, reducible by sharing between TSOs under Article 7(3), and the total FRR requirement is the combination of aFRR and mFRR for normal imbalances plus mFRR for disturbances. mFRR therefore carries the whole disturbance component, which is why the dimensioning fault sizes mFRR as well as FCR-D. Sweden’s mFRR up requirement of 1400 MW sits close to the 1450 MW reference incident, which is suggestive, but no published source consulted here states that the 800 to 1400 MW step is the disturbance component being taken up in full, so the correspondence is left as a correspondence. The normal-imbalance and disturbance shares of mFRR are not published separately, so no split between weather-driven and fault-driven mFRR can be quoted. FCR-N, although a containment product, is sized on historical imbalance rather than on the dimensioning fault, which is why the fault-dimensioned subtotal quoted in the text is 1066 MW and not the full 1290. FFR is left out of the figure because it is the one product with no flat forward number — its volume is derived from expected rotational-energy levels together with the dimensioning fault, Sweden carries 38% of the Nordic need, which is 113 MW for 2026, and the source gives only scenario ranges for the 2030s. That omission cuts against the argument made here rather than for it, since FFR is the product most directly tied to inertia and the one Svenska kraftnät expects to need more of as low-inertia hours become more common. Pre-qualified volumes, the roughly 100 MW in the third quarter of 2021 against nearly 2500 MW in the third quarter of 2024 and the hydro share of about 1500 MW, are from Svenska kraftnät’s own news item of 2024 on the growth in suppliers; the article gives hydro’s figure but not separate figures for batteries and wind, so “fastest-growing” is its characterisation rather than a number computed here. The September 2025 clearing prices are from Svenska kraftnät’s monthly FCR report for that month, which states them in euros per megawatt: 6.32 euros per megawatt per hour for FCR-D up, 31.86 for FCR-N and 8.00 for FCR-D down. These are capacity payments for standing ready, the distinction chapter 26 builds an argument on, and not energy prices. The 60 to 80 figure quoted for 2022 comes from a trade-press account that writes it as euros per megawatt-hour; on this product that can only mean per megawatt per hour, but it is a secondary source and the comparison should be read as an order of magnitude. Prices are on volume-weighted marginal pricing since February 2024 and pay-as-bid before it, which is a pricing-rule change rather than a change in the fleet and is part of why the series falls. Two cautions on the whole decomposition. Sweden’s share of a Nordic requirement can move without the Nordic requirement moving, since the allocation key is renegotiated. And a flat volume requirement is not the same as a flat cost: the price per megawatt can rise even when the megawatts do not, which is precisely what happened to FCR in 2021 and 2022 and then reversed.

27. German figures are the Regelleistungsvorhaltung line from the joint Bundesnetzagentur and Bundeskartellamt Monitoringbericht series — the 2011 report for 2009 and 2010, then the 2013, 2015, 2017, 2019, 2021 and 2025 reports. Each report states its own reporting year and the one before it in the text, and carries a four-year table of system-service costs, which is why the 2025 report supplies 2021 and 2022 as well as 2023 and 2024. The years are: 825, 697, 588, 417, 594, 437, 316, 198, 145.5, 123.3, 285.7, 152.4, 568.6, 628.6, 644 and 518 million euros for 2009 through 2024. The 2009 and 2010 reports, which would have carried 2007 and 2008, were not retrievable at any URL reachable here, so the series begins in 2009 rather than at this book’s usual 2006. The German figure covers the holding of capacity only; activated balancing energy in Germany is settled through imbalance prices paid by balance groups and never enters network charges, so the number is a floor on the total and not the whole of it. Swedish figures are Svenska kraftnät’s Årsredovisning for 2022, 2024 and 2025, which give gross ancillary-service costs of 3 628, 6 650, 6 934, 5 455 and 9 736 million kronor for 2021 to 2025, the mFRR and FCR components, and the explanations quoted. The per-product split drawn in the figure above is derived rather than quoted. Svenska kraftnät publishes two business segments, each with a net ancillary total and an mFRR figure, plus footnotes saying which products the segment covers, so FCR-D for Transmissionsnät and aFRR for Balansering are obtained by subtraction; only the segment totals, the two mFRR figures and the Balansering FCR figure are quoted. The three years reconcile against the published net totals of 5 855, 4 957 and 8 672 million kronor, and the build step fails if they do not. That check is narrower than it looks: it catches an error in either published segment total, because the sum reduces to those two, but it cannot catch an error in any of the three per-product inputs, each of which only moves value between products rather than changing the total. Swedish figures before 2019 are on a narrower definition — primärreglering, frequency containment alone — and run 298 million kronor in 2009, 453 in 2010, 629 in 2016, 675 in 2017 and 1 335 in 2018, from the 2010, 2016 and 2018 annual reports; the 2011 to 2015 and 2019 to 2020 reports are no longer served on svk.se or in the Internet Archive, so the Swedish series has a hole in it and its two halves are not continuous. The 2018 Swedish doubling is attributed by Svenska kraftnät to an opening cold spell, a violent spring flood, a dry summer with a hydrological deficit and high power prices, because at that time all Swedish primary regulation came from hydro; reactor closures are not in its explanation. Finnish figures are Fingrid’s financial statements bulletins for 2019 to 2023, which give 55.9, 63.5, 68.4, 186.9 and 185.6 million euros; note that Fingrid’s own 2021 bulletin states 83.7 million for that year against the 68.4 in the 2022 comparative, and the lower figure is used here so that 2021 and 2022 sit on one basis. British figures are aggregated by this edition from NESO’s daily Balancing Services Use of System cost data, by financial year, summing the frequency-control and reserve categories separately from constraints; they are narrower than NESO’s own headline total, which was 2.7 billion pounds for 2024/25 against the 2.1 billion these categories give. German congestion figures are the Bundesnetzagentur’s Netzengpassmanagement totals. Three cautions bear on the whole comparison. None of it is inflation-adjusted, and the 2021 to 2023 energy shock moves every series at once and by more than anything else visible here. The five definitions are not interchangeable, which is why no chart in this edition places them on one axis. And the attribution is contested rather than settled: Svenska kraftnät’s balancing-market chief has said the costs would have been much the same at comparable power prices had the reactors stayed open, while Tidningen Näringslivet, drawing on Svenska kraftnät’s own 2020 and 2021 reports and a PwC study it commissioned, argues that the connection to lost dispatchable plant is plain. What the published series will support is the narrower statement made above: the bill does not track the closures, and it does track the fuel price.

28. Prices are volume-weighted averages of what was actually paid for capacity over the four weeks from 1 to 28 September 2026, central European days, from ENTSO-E’s Transparency Platform under article 12.3.F of the electricity balancing guideline: one record per accepted bid, a quantity in megawatts and a price over a stated period, weighted by the megawatt-hours of capacity each bid held. The dataset has three traps and this figure walks into none of them, which is worth saying because the first two are invisible in the schema. The price is quoted per settlement period and not per hour, a factor of four on a fifteen-minute market, and both are labelled “MAW”; the reading used here is pinned against regelleistung.net, which publishes the same German and Austrian auction in euros per megawatt per hour, and the step that builds the figure refuses to write its file unless both countries and both directions come out at the price regelleistung prints — 2.30 and 0.68 euros for Germany, 2.28 and 0.78 for Austria, in the 00:00 to 04:00 block of 1 September, on volumes within half a percent. Four operators report in their own currency, which the document says and an unwary reader does not, so Switzerland, Hungary, Poland and Romania are converted at the European Central Bank’s reference rate for the day. And several operators publish one record per accepted bid at a shared position, which reads as a duration of zero unless the span is measured to the next distinct position instead of to the next record. Four systems are left out, each on a stated test rather than on judgement. Switzerland and Hungary buy 96% and 93% of their capacity on contracts longer than a day, where the published figure is the price of the contract and not of a period. Greece and Slovakia imply an aFRR capacity bill of 20.7 and 5.6 euros per megawatt-hour of their own demand, against 1.07 for Germany; the threshold is five euros, five times the whole German reserve bill, and Poland passes it at 4.67 with room to spare that is not large. Comparing the survivors still compares like products bought on unlike terms — Germany pays each accepted bid its own price where most of the others pay a single marginal price, the product is a four-hour block in some markets and a quarter-hour in others — which is why the figure carries the volume beside the price. Sweden is absent from the platform for every reserve product, so its row is Svenska kraftnät’s aFRR capacity market dataset from data.svk.se, CC BY 4.0, which covers the hours in which Sweden procured aFRR locally. Demand denominators are actual total load for the same areas and days from the same platform.

29. The fee figures are Svenska kraftnät’s Årsredovisning 2025, which reports that charge revenue from balance-responsible parties rose from 3 977 million kronor in 2023 to 6 408 million in 2025, “vilket i huvudsak förklaras av höjda avgifter”, with the base fee raised about 30% in 2024 and about 25% in 2025. The fee figure measures the fee, not the reserve. Svenska kraftnät does not itemise the balance-responsibility fee by product, so no part of it can be attributed to the cover for the reference incident, and the 2025 ancillary-service bill of 9 736 million kronor gross is larger than the whole fee revenue. What the published series shows is that the fee as a whole rose, on rises the annual report explains only as “höjda avgifter”; the cost-allocation claim above rests on how the charges are structured, not on that figure. Three limits on what is claimed here. This edition has not established that no Nordic mechanism differentiates a charge by the contingency a unit creates — the claim is the narrower one that the FRR dimensioning methodology contains no such mechanism. Article 6 sets the volume of FRR the block must hold, by requiring it to cover the block’s imbalances at least 99% of the time and to keep the frequency restoration control error inside its target ranges; Article 7 then requires the disturbance component to cover the control area’s reference incident and allows operators to share it. Neither says anything about who pays. And nothing above argues that charging the causer would be better policy: a single large unit is usually cheaper per megawatt-hour than several small ones, and a rule that priced its contingency to it would trade that efficiency for a different one. The point is only that the trade is currently made silently, in a tariff, rather than in the open.

30. Department for Energy Security and Net Zero, Quarterly Energy Prices, international industrial electricity price comparisons: of 28 IEA countries reporting for 2024, the UK had the highest industrial electricity prices including taxes and levies, with large users at about 25.33p/kWh against an EU-14 median of 11.25p/kWh. The 2024 price is compared here with the 2025 GB wholesale average, so the ratio is indicative rather than exact; and some energy-intensive users receive partial exemption from policy levies, so the figure for an individual plant may be lower. Sweden makes an instructive contrast: qualifying manufacturers there pay an electricity tax of 0.6 öre/kWh against a full rate of 43.9, a relief now under audit by Riksrevisionen.

31. The elmix cannibalization model, https://oluies.github.io/elmix/modell/system.html .