Sustainable Energy — Without the Hot Air

2026 revised edition · work in progress

Author

David J. C. MacKay — with revisions by Örjan Lundberg

About this edition

This is a work-in-progress 2026 revised edition of David MacKay’s Sustainable Energy — Without the Hot Air (2008), rebuilt with Quarto from the markdown source preserved by the Life Itself climate project.

MacKay’s text is reproduced unchanged. Most of his figures have been replaced by versions carrying the same series to the present; where a figure of his is retained, or replaced, the caption says so. New material added in this revision is marked in the chapter where it appears, and several chapters are new in their entirety. The table below is the authoritative list of them.

Latest changes

Last updated 22 September 2026. The complete history is in the repository.

Date Chapter What changed
22 September 2026 11a

The Commission adopted a rating scheme on 21 September: a label for every data centre over 500 kW, showing power and water efficiency, low-carbon energy, waste-heat readiness, grid services and whether it has paid for new clean generation. The report published beside it is the first look at what the 2024 reporting duty produced: only 36% of obliged sites answered, and six member states reported none. Water averages 0.58 L/kWh, well below the industry figure. Heat reuse is where the press release and the report part company — half the waste heat could warm four million homes, and 1.8% of it actually does. The 86% renewable share rests largely on certificates and contracts the regulator itself says are not tied to new plant. The chapter’s table gains an EU row: 68 TWh, 2.5% of the Union’s electricity, and 0.41 kWh/d per person — MacKay’s figure exactly.
22 September 2026 20, 13

Chapter 20 gains the road half of MacKay’s freight priority. An electric 40-tonne truck uses about 103 kWh per 100 km against roughly 300 for a diesel, the same factor of three as cars; the problem is the working day, not the kilometre. China has written its answer into an eleven-department plan of 12 June 2026: 40% of heavy-truck sales by 2030, 1.6 million trucks and about 3000 charging and swapping stations — a target the market already touched in December 2025, at 54%. In August CATL bought a quarter of the state-owned Qiyuan network, so the battery maker now holds much of the swap network on both competing formats. Europe standardised the charger and is now hedging, with Swaptopus in Britain and a CATL–DHL memorandum. Swapping wins where routes are fixed and machines never stop, which leads chapter 13 into the forest: a Swedish RISE project built an automatic swap station for a forestry machine and a farm machine in 2021–23, so the book’s case for swapping off-road is no longer only its own inference.
20 September 2026 20

MacKay counted the carbon pollution of the cars in one British showroom in 2006 off a car-buying website; every member state now keeps the same register by law, so his figure 20.9 is redrawn from the 10.8 million new cars registered in the EU, Norway and Iceland in 2025. The histogram has marched left — his UK average of 168 g CO2 per km against Europe’s 97 — and grown a spike at zero holding a fifth of all registrations. That spike is an accounting convention: those cars emit about 45 g/km on the European grid, the plug-in hybrids beside them are typed at 32 and metered at 145, and the petrol and diesel cars are metered at 169 against a type approval of 132. Corrected, the 97 becomes 143. His own proposed ceilings can now be marked against the register: the first has arrived without being legislated, the second is half-arrived on paper and not at all on the road, and the third would still ban seven cars in ten.
20 September 2026 20

His figure 20.21 plotted nineteen recharges of one G-Wiz with guide lines at 16, 21 and 33 kWh per 100 km. Rather than print it beside the 2026 version, the remade figure carries all three of his numbers — the 21 as a line, the 16 and the 33 as the edges of a grey wedge — and the two figures become one. All three numbers are quoted from his own text, so nothing is read off his axes; what is lost is where within the wedge his nineteen points fell. The wedge turns out to be the most useful thing on the chart: one car, one driver and one city produced a factor of 2.1 between the best recharge and the worst, which is about the spread between the best and worst vehicles drawn on it.
20 September 2026 20

Germany sold 220 300 cargo bikes in 2025, 84% of them electric, and has 17.2 million e-bikes in use against 2.03 million battery- electric cars — the electrification of German transport happened on two wheels, more than eight to one. The Swiss survey of 696 owners is the best evidence that the thing replaces a car, and the most equivocal: 31% gave up one, but the mode it substituted most was public transport, and 31% now walk less. Babboe’s 22 000-frame recall is what it looks like when a bicycle is really a vehicle. On regeneration, the chapter’s rule is sharpened: what blocks it is the freewheel and not the motor’s position, which is why Valeo’s gearbox mid-drive can recover where a derailleur cannot, and the arithmetic is redone for 250 kg — three fifths of the descent reaches the brakes rather than two, and the case for fitting it is the brake heat rather than the eight hundred metres of range. Figure 20.23b gains the most efficient car on sale and a VW ID.7 as an ordinary family one.
14 September 2026 1

Figure 1.3 shows that Britain’s feared generating gap never appeared. What it does not show is what the surviving fleet burns — gas — or where that gas comes from, so figure 1.2a’s question is now asked of the other fuel. Britain’s gas surplus lasted nine years: 1995 to 2003, peaking at 5.6 kWh/d per person in 2000, which is about a fifth of what its oil managed in oil’s own best year. The country went from importer to exporter and back inside a decade, and in 2025 imported a net 12 kWh/d per person of gas on top of 15 of oil. Self- sufficiency in the fuel that keeps the lights on was not an era; it was an episode. The other three lines end differently, which is why all four are drawn. Norway’s never crosses zero: it peaked at 612 kWh/d per person in 2017, has wandered since without trend, was 571 in 2025, and is now the largest single supplier of the gas the others import. Denmark’s dips below zero for four years and comes back: not depletion but the Tyra hub being rebuilt, the only interruption in either figure that is engineering rather than geology. And the Netherlands has the longest run of all, fifty-three unbroken years to 2017, ending not because the gas ran out but because Groningen was shut in over the earthquakes it caused — the one line in these two figures that comes down for a reason the geology had no say in. The task checks the nine-against-twenty-four comparison and the one-fifth ratio across both figures, reading figure 1.2a’s own committed table.
14 September 2026 1

Figure 1.2 plots what the three North Sea countries pump, which is what MacKay’s own figure plotted. But production is not what a country has to sell — what it has to sell is production minus what it burns itself, and on that measure the three stop being one story at three scales. Britain’s surplus lasted twenty-four years: it began in 1981, peaked at 29 kWh/d per person in 1999, ended in 2004, and Britain now imports a net 15 kWh/d per person, more energy in oil than it produces. Denmark’s lasted seventeen years, 1998 to 2014, and it now imports 16. Norway’s is in its fifty-first year and has never been interrupted: 1074 kWh/d per person at its peak in 2000, about eight and a half times MacKay’s figure for average British consumption, and still 494 in 2025. One sea, three decades, opposite outcomes — and what differed is how much each country burned at home. Norway consumes about a tenth of what it pumps, so nine-tenths is still there to sell. The data task checks each surplus is a single unbroken run rather than a line flickering across zero, and that this figure and figure 1.2 agree on the peak year, 2000, despite being in different units.
14 September 2026 M

Chapter M gains its first figure, and it is the one that makes the chapter’s whole argument visible: Hall, Balogh and Murphy’s oil cascade, drawn as a Sankey diagram. Of 100 units of crude in the ground, 10 go on extraction, 10 on running the refinery, 17 leave as the part of the barrel that never becomes motor fuel, 3 move it to where it burns, and 24 build and maintain the roads it drives on — leaving 36 still doing work, which is where their much-quoted “minimum EROI of 3:1” comes from. Three nested brackets mark the boundaries: about ten for one at the wellhead, six at the petrol tank, three and a half where the work happens — the paper scales its wellhead ratio by the fraction of the barrel that survives each stage, which is a different operation from reading the ratio off the picture, and the chapter says so. Nothing about the oil changes between those answers; only the question does. Chapter Q’s passage on why primary energy falls faster than anyone uses less now points at it, since these are the same losses seen from the other end. Drawn from the 2009 paper’s own percentages rather than from the widely reproduced 2014 chart of them, which charges 37.5 units to roads and leaves 20.5: this edition could not reconcile that with its cited source, since 37.5 is 64.7% of the flow before it — the source’s own 64% total applied a second time — and it would imply five units of crude per unit of service where the same authors say three. The note says so, and the task checks the arithmetic against the conclusion the paper draws from it.
13 September 2026 Q

Chapter Q gains a passage on a trap almost every summary of these pathways falls into. Europe’s primary energy falls 19% by 2050 and its final energy 27%, and neither is mostly a story about Europeans doing less. The chimneys and the engines close: a coal station wastes 60% of its fuel and a petrol engine three-quarters of its, both losses counted as primary energy, and neither counted in what replaces them since the report counts a kilowatt-hour of wind or solar as a kilowatt-hour of primary energy. Europe doubles its electricity generation while the primary energy each unit costs falls from about 1.6 to about 1.0; across the system 40 EJ a year of fossil primary leaves and 26 of renewables and biomass arrives, and the service does not fall by the difference. European transport gives up 11.1 EJ of liquid fuel for 3.5 of electricity, a ratio of 3.2 to one, which is appendix A’s arithmetic arriving unannounced in a global model. And something pulls the other way — the primary-to-final ratio gets worse, 1.26 to 1.39, as chimney losses are replaced by hydrogen made from electricity and liquids made from biomass. That third term gets scepticism: the pathways assume a European hydrogen industry sixty times today’s, and chapter O records the sixty major projects cancelled during 2025. If hydrogen underdelivers, the service is either unmet or plugged in, and plugging in is the more efficient of the two — so the overhead is the models’ pessimism.
13 September 2026 Q

The scenario database holds the demand side as well as the supply side. A European in the median 1.5 °C pathway uses 71 kWh/d of final energy in 2020 and 50 in 2050, with the electricity inside that shrinking total rising from 15 to 29 — from a fifth of final energy to three-fifths. Transport falls hardest, from 22 to 14, which is appendix A’s arithmetic arriving in a global model rather than a story about driving less. Between the categories, final energy per European falls 30% for 1.5 °C and 15% for 2 °C: the tighter target asks twice the reduction. Set side by side in 2050 itself, a European in the 1.5 °C pathways uses 17% less energy than one in the 2 °C pathways while having a quarter more wind capacity and a third more solar — both levers move and neither dominates. What stands out instead is how little the 2 °C pathways ask of demand at all: their buildings go from 27 kWh/d per person to 26 in thirty years. As a rate, the 1.5 °C median asks Europe for about 31 GW of wind a year now and 35 after 2030, and 61 GW of solar falling to 38; Europe built 19.1 GW of wind and the Union 65.1 GW of solar in 2025. Solar is ahead of the pathways, wind at two-thirds, and neither figure counts replacement.
13 September 2026 Q

The regional files of the scenario database arrived, so chapter Q can ask its question of a continent, and Europe answers differently from the world. In the median 1.5 °C pathway Europe goes from 90 to 69 kWh/d per person of primary energy by 2050 against the world’s 55 to 43; coal falls 98%, oil 81% and gas 68%; solar is multiplied by nine and wind by seven, against the world’s 32 and 16, because Europe starts higher. Nuclear is the one carrier where the two point opposite ways: it roughly doubles for the world and falls 58% for Europe — 91% on the tighter European Union cut of the same file. The spread there runs from 0.1 to 13.8 exajoules, and the chapter says why a number assembled before 2022 should be read as a dated input rather than a finding. For Britain and Sweden the answer is that the database has nothing: it holds 48 scenarios for them and none were sorted into a temperature category.
13 September 2026 Q

The question the chapter had to leave open — how much solar, how much nuclear — is answered from the scenario database itself. In the 1.5 °C category the median pathway multiplies solar by 32 and wind by 16 between 2020 and 2050, roughly doubles both nuclear and biomass, grows hydro by about four-fifths, and cuts coal by 95%, oil by 62% and gas by 42%; in the 2 °C category, solar by 21 and wind by 13 with coal down 82%. In this book’s units the median 1.5 °C world of 2050 runs on a median 43 kWh/d per person of primary energy — a third less than Britain uses now — with medians of 9.3 kWh/d for biomass, 8.5 for solar, 6.9 for wind, 2.4 for hydro and 1.6 for nuclear, which are separate medians over partly different sets of pathways and are not a decomposition of the total. The spread is the point: across those pathways solar in 2050 runs from 31 to 200 exajoules and nuclear from 4 to 70, so the models agree that coal ends and disagree about what replaces it, which is why the report prints the fossil declines as numbers and the rest as words. The extraction checks itself against those published declines and fails if any of the six drifts more than eight points.
13 September 2026 Q

The AR6 Scenarios Database’s metadata file arrived, so chapter Q’s pathways are computed from the ensemble the report is built on — 1202 vetted scenarios — instead of hand-entered from the printed Table SPM.2. The medians come back as the table prints them, to within half a point: 43.3, 68.9 and 84.2% for C1 against 43, 69 and 84, and 20.7, 46.3 and 64.5 for C3 against 21, 46 and 64, with the category counts matching too. The chapter gains the whole eight-rung ladder, C1 to C8, each rung’s median 2050 cut divided by the people then alive: 0.90 tonnes per person for 1.5 °C, 2.02 for 2 °C, 4.05 for 2.5 °C, and 8.31 for C8 at the bottom, which is more per person in 2050 than the world emits today. The step from two degrees to one and a half costs about 12 kWh/d per person of burning, roughly what an average European spends on driving. What the report does not give is a solar or nuclear number per category: the mix is per pathway rather than per temperature outcome, and the chapter says why.
13 September 2026 Q

A new chapter converting the Sixth Assessment Report’s headline scenarios into this book’s units. Its two lead categories cut global greenhouse gases by 43, 69 and 84% by 2030, 2040 and 2050 for 1.5 °C, and by 21, 46 and 64% for 2 °C, measured against 55 GtCO₂- equivalent in 2019. Divided by the people who have to live on them — 7.8 billion then, 9.7 billion by 2050 — that is 7.0 tonnes per person today falling to 0.93 or 2.07, and the per-person cut is deeper than the global one the report prints: 87% rather than 84%, because the denominator grows. Run through the carbon chart’s own 250 g per kWh, today’s 7 tonnes is 77 kWh/d of burning and C1’s 2050 budget is 10 — the whole greenhouse allowance, farming and cement included, against a British 69 kWh/d of primary energy now. The chapter also collects what the report says about energy itself: coal down 95% by 2050 and gas only 45%, almost all electricity zero- carbon, costs down 85% for solar and batteries in a decade, half the 2030 job available under $100 a tonne, demand-side measures worth 40 to 70%, and investment needing to run three to six times higher.
13 September 2026 Charts

The interactive charts gain marks for the present, and the scales are extended to hold them: MacKay’s power chart stopped at 125 kWh/d per person, which fitted the Britain of 2004, and his carbon chart at 11 tonnes of CO₂ a year, which fitted the Britain of 1990. An American today uses 206 kWh/d and emits 14.2 tonnes, so the scales now run to 210 and 15. The marks are computed rather than typed: primary energy per person for 2025 comes from the same Energy Institute series as chapter L’s explorer, and CO₂ per person for 2024 from the Our World in Data file the emissions figures use, both written to a small JSON the page reads — so refreshing the data moves the marks. Pressing United States, 2025 reads 206 kWh/d per person, 514 GW, 4,503 TWh/y and 387 Mtoe/y for a UK-sized population. One bug fixed on the way: a browser gives a background tab no animation frames, so a mark pressed in one used to glide nowhere; the value now arrives whether the animation runs or not.

13 September 2026

13 September 2026

Bibliography

Charts

MacKay’s bibliography is his and is untouched. Under it there is now a second list: the works and data sources the 2026 material actually draws on, in three groups — books, papers and reviews; reports and official statistics; and the live data, registers and instruments the figures are computed from. Forty-one entries, each saying what the source is and where the edition leans on it: Murphy’s finite-planet textbook, the KEITO scenarios and the Finnish data-centre roadmap, Lucy Yu’s grid review, ORNL’s Transportation Energy Data Book, the Dutch vehicle register, the Cambridge rooftop and Cairngorm summit weather stations, Elexon, ENTSO-E, NESO, Fingrid, Svenska kraftnät, Open-Meteo and the rest. The precise citation for any particular number stays in the chapter’s own note, which is where it was checked, and the list says so. ::: {.callout-note collapse=“true” appearance=“minimal” icon=false title=“The two translation charts are interactive”} MacKay’s power and carbon translation charts are nomograms — parallel scales, all linear in one underlying quantity, so a single line laid across them reads the same amount in every unit. On paper you use a ruler. They are now rebuilt as SVG you can drag: move the line, or type a number into any scale’s box, and every other scale follows. The power chart carries kWh/d per person, GW, TWh/y and Mtoe/y per UK; the carbon chart carries chemical and electrical kWh/d per person, tCO₂ per person, MtCO₂ for a UK, and GtCO₂ and GtC for a world. MacKay’s own marks are buttons that glide the line into place — UK electricity in 2004 lands on 401 TWh/y where his printed chart marks 400, and the 60, 80 and 90% targets are measured on the UK’s 1990 emissions as his chart measures them. His printed originals are kept underneath each one. Plain JavaScript in the page, following the pattern chapter L already uses; Scala.js would have meant adding a compile step to a Quarto book for 150 lines of arithmetic.
13 September 2026 P

A new chapter, and a short one: it points at Thomas Murphy’s Energy and Human Ambitions on a Finite Planet, free from the University of California and licensed for reuse, and reproduces the argument of its first chapter. Take the 2.3% a year that industrial energy use has actually managed — a factor of ten a century — and extrapolate: in 400 years humanity uses every scrap of sunlight the planet receives, in 1,400 the entire output of the Sun, in 2,500 the galaxy, in 5,000 the visible universe. The point is not the timescales but that the growth rate we treat as normal runs out of universe. Figure P.1 is the limit that arrives first, recomputed here from the equilibrium chapter 1 already uses: today’s 18 TW is 0.14 W/m² against 961 of absorbed sunlight, but grown by ten a century it boils the oceans in 417 years, whatever the energy source, because heat is the end of every use. The recompute lands on four of Murphy’s five published rows and disagrees with the fifth, which the note sets out.
13 September 2026 B

Figure B.1, the Beaufort scale in miles per hour, km/h and m/s, had been flattened by the 2008 conversion into a column of 57 loose numbers with a caption under it. It is a table again, with MacKay’s own values — including the one row that carries no Beaufort force, which his own online edition has too. Figure B.7, the two standard wind-shear formulas, was a 219-pixel bitmap; it is recomputed from the formulas printed on the same page and redrawn, and it gives 6 m/s at 10 m as his does.
13 September 2026 C

The same conversion damage in appendix C: table C.6, the jumbo jet against the albatross, had become a column of loose lines and is a table again. Figure C.5, the thrust a plane needs against its speed, is recomputed from the formula above it and the numbers its own caption prints — 319 tonnes, wingspan 64.4 m, drag coefficient 0.03, frontal area 180 m², air at 0.41 kg/m³ — with the area of the sausage of air the wings throw down taken as the square of the wingspan, which figure C.7’s caption gives. The minimum lands at 226 m/s and 113 kN against the 220 m/s his caption states, and the note records that the 130 kN the chapter derives later in the appendix uses table C.6’s fully-laden 363 tonnes instead. And figure C.7, the frontal view of a 747, had a yellow field baked in behind it where the drawing’s transparency used to be; the background is white again.
13 September 2026 A

Both of MacKay’s sources for figures A.12 and A.13 had decayed, and better ones exist. For A.12a, ORNL’s Transportation Energy Data Book collects dynamometer measurements of fuel economy at steady speeds from four studies — 13, 15, 9 and 74 cars, from 1973 to 2012 — plus modelled 2016 vehicles: 111 cars against his two. They say what he said, with more authority: every curve has a minimum, and from 80 to 113 km/h the 74-car study costs 32% more fuel where a square law demands 96%. For A.13a, the Dutch vehicle register publishes a type- approval maximum design speed, which almost no other register does, alongside maximum net power: on 665 models with a hundred registrations or more since 2023, combustion cars still follow Tennekes’ cube law at an exponent of 2.9, while electric cars sit far above the line at 2.3 — a 504 kW Tesla Model S declares 263 km/h where a 478 kW Porsche 911 Turbo S declares 327. And 41 models declare exactly 250 km/h, the German limiter. Pooled, the sample fits 1.4, which looks like a refutation of the cube law and is really a policy replacing a physical limit.
13 September 2026 A

The four model plots in appendix A were low-resolution bitmaps from the 2008 EPUB, and a reader could not take a value off them. They are recomputed from the assumptions printed in their own captions and redrawn as vector figures: the car, the bicycle and the train at a steady speed, and the electric-car range curves for lead-acid at 40 Wh/kg and lithium at 120. Nothing is added and no curve moves — the captions say so, and the 2025 vehicles and batteries stay in figures A.9a and A.14a. The other figures in the appendix are photographs, cartoons, or data this edition cannot re-read, and are left as printed.
12 September 2026 4

Figure 4.2a carries MacKay’s other wind figure forward on its own station, Heriot-Watt’s on the summit of Cairn Gorm. His caption says six months of 2006, and the archive says why: that year’s anemometer failed in July and again at the end of November, so January to June is what 2006 has. Those six months averaged 11.2 m/s with 133 of 179 days at or above 6; the same months of 2026 averaged 6.5 m/s over 92 of 175 days. No trend is claimed — summit years vary enormously and several archive years are instrument failures rather than calm weather — but even the quieter half-year is more than twice the Cambridge rooftop in its windier one, which is the chapter’s case for putting windmills on hills.
12 September 2026 A

Figure A.9a redraws figures A.9, A.10 and A.11 from their own captions and puts 2025 vehicles beside MacKay’s. The check is his petrol car at 110 km/h: 78 kWh per 100 km against the 80 this book is built on. A 2025 electric car uses 17.7 at the same speed, a factor of 4.4 where chapter 20 measures 3.8 on the road; an electric sport-utility of the same drivetrain uses 25.5, which is 44% more for nothing but size. An electric bicycle costs about 1.1 kWh per 100 km of electricity against 3.1 of food for his cyclist. And his train at 40% of its seats costs 3.9 kWh per 100 passenger-km instead of 1.6, which is a fact about occupancy rather than about trains. Figures A.12 and A.13 are left as printed, and the text says why: the Prius source behind A.12 is a 404 and the BMW page publishes plots rather than numbers, while A.13 is Tennekes’ scatter and would need a new sample of cars this edition cannot cite.
12 September 2026 4

Figures 4.1a and 4.6a carry MacKay’s Cambridge wind year forward on the same instrument at the same address: the Computer Laboratory rooftop station, still publishing the half-hourly file he used. The recompute finds his own result first — 26 days with a daily mean of 6 m/s or more in 2006, against the about 30 in his text — which is the check that the undocumented wind column is being read in tenths of a knot, as he must have read it. The update comes with a warning: the anemometer failed, 2024 is 97% zeros and 2025 and 2026 entirely so, which leaves 2023 as the last usable year, and its share of dead-calm readings had already doubled since 2006. The fall from 2.7 to 1.9 m/s is therefore the instrument as much as the weather, and only the direction survives — which is enough, because the bias runs downwards and the chapter’s 6 m/s is an upper bound. The cube is the quantity that matters: averaged over 2006’s half-hours it is 86 m³/s³ against 216 for a steady 6 m/s, so the rooftop in the windier year saw about 40% of the wind power the assumption implies.
12 September 2026 A

Figure A.14a recomputes MacKay’s electric-car range model from his own assumptions — it returns the numbers he printed, 179 km where his text says 180 and 538 km for 500 kg of 2008 lithium — and then moves the two inputs that have changed since. A 2025 pack at 160 Wh/kg carries 717 km in his 740 kg car and needs 170 kg rather than 237 for the 300 km he guessed people would settle for, so his verdict that 120 Wh/kg was already good enough has been walked past rather than overturned. But 740 kg for car and occupants is a 2008 abstraction: a car of that class now weighs about 1500 kg before the pack goes in, and the same pack in that car still gives 500 km while the transport cost rises from 13 to 19 kWh per 100 km — a tenth below the 21 that chapter 20 measures across 342 real electric cars. The extra density bought permission to build a heavier car more than it bought range.
11 September 2026 11a

Figure 11a.2: what Finland has actually committed, against what its own scenarios assumed it would. Thirty-three data centres and 285 MW were operating in September 2025; 1300 MW had an investment decision at the end of that August; Google’s four sites are perhaps another 1300 MW, which makes 2885 MW committed — past AFRY’s strong-growth case for 2030 and half as much again past the assumption the national KEITO scenarios carry for 2050, with 2500 MW more in planning that nobody has counted. Finnish tax records say the built fleet draws about half its nameplate, so the committed capacity meters at 13 TWh a year and 6.4 kWh/d per Finn on today’s habits and 21.5 TWh and 10.5 kWh/d if it runs as an AI campus does — either way above Ireland’s 3.6, for the same number of people. It is also 19% of the 15 553 MW peak Finland set in January 2026, against the 9% figure 11a.1 forecast for 2031. And the deal above looks different from here: the 80% coverage is an artefact of a utilisation assumption, and AFRY finds that building wind to match the annual energy still raises the average price by more than 10%.
10 September 2026 11a

Britain’s answer to the connection queue, which turns out not to be about the queue. Lucy Yu’s independent review for DESNZ asks whether the grid needs the headroom its own rules insist on: deterministic standards insure against a fixed worst case whatever the weather is doing, and “hardcode higher than necessary costs into how the grid is run”. The share of British half-hours with weather-dependent renewables above 30% went from about 2% in 2015 to 64% in 2025, and balancing costs from £1.2bn in 2018/19 to £2.7bn in 2024/25, forecast at £3.4–4.9bn a year in 2031–35 even after the 2030 network build — £39 a head now, £50 to £72 then. The review’s own quantified savings from putting AI into the grid add to about £90m a year against that £2700m bill, some thirty times smaller than the problem, while British data-centre demand quadruples by 2030 on a source that is not the one this chapter already cites.
9 September 2026 11a

What the hyperscalers’ European power deals actually buy. Google’s €13bn in Finland comes with a 22-year offtake of half of Loviisa, 629 MW of new onshore wind and a 94 MW battery: about 6 TWh a year, 680 MW continuous, 7% of Finnish electricity and 2.9 kWh/d per Finn — twice what the table at the head of the chapter gives an American for the whole American fleet, contracted by one company in one country. The battery is eleven minutes of a gigawatt. And the money buys certainty rather than plant: the licence to run Loviisa to 2050 has existed since February 2023, so the 2030 shutdown the deal averts was a commercial prospect and not a legal deadline. Ireland has since made the same thing a rule — 80% of annual demand in new Irish generation — against announced deals covering 12% and 3%; Britain had the Homer City site over again at Cottam and answered with SMRs at 230 W/m² instead of gas at 290.
9 September 2026 25

Figure 25.9a: what one degree of temperature buys, computed from ENTSO-E load (NESO for Great Britain) and ERA5 reanalysis for six countries rather than quoted from anyone. France rises 2.14 GW per degree colder against Italy’s 0.42, because France heats with electricity; Italy rises 1.01 GW per degree hotter, because it heats with gas and cools with air conditioning. The summer arms add to about 3 GW per degree across five countries and they arrive together, which is the arithmetic behind July’s export restriction — and every heat pump chapter 7 recommends moves a country from Italy’s shape towards France’s.
9 September 2026 11a

Figure 11a.1: data centres measured against national peak load rather than against a grid’s kilowatt-hours, which is the quantity a system operator rations. Ireland is already 29% of its peak and forecast at 46% by 2031; Sweden goes from 2.5% to 7.0%, 2.8-fold over six years, against Svenskt Näringsliv’s 3-to-10% over five by an entirely different route — the same order rather than the same number. Installed capacity is not consumption, so the chart is a ceiling and the chapter’s table is the meter.
8 September 2026 28

Germany’s capital plan for 2035, read from the source rather than the coverage. The wires cost about what the machines cost — €350–400bn of grid against €300–450bn of generation — and table 28.3 has two lines for that, totalling £2bn, and none at all for distribution. The index’s own scoreboard is quoted with the formula that produces it, because four of its fifteen indicators score above 100% and one of those is a job count that has fallen by 46 000.
7 September 2026 28

The term MacKay leaves out: the price of the money. His £870bn spread over forty years is an interest-free mortgage; borrow the repriced plan — £1620bn — at 5% and it is £94bn a year rather than the £40bn an interest-free reading gives. Britain has run the experiment — Hinkley borrowed at about 9% and Sizewell C at about 4.7% for the same reactor, and 9% money costs about 77% more per year than 4% money for identical concrete.
7 September 2026 28

Table 28.3 is a table again — it had decayed into 190 lines of loose fragments, one per cell — and now carries a 2025 price beside each of MacKay’s. Three lines fell, two held and ten rose: solar by a factor of nine downwards, nuclear sixfold up, heat pumps threefold. Repriced line by line the plan costs about 13% more than his own total in the same money, having been read for eighteen years as though the total were the durable part.
1 September 2026 26

Figure 26.16a gains Britain, and its price side is now computed from the sources rather than typed in. Britain’s day-ahead spread is wider than SE3’s and earns less, because it has almost no cheap hours; its reserve markets pay about half what Sweden’s do, which is what Sweden’s will look like once the batteries have arrived.
31 August 2026 26

The Swedish case: day-ahead arbitrage does not cover a battery’s capital even with 17% of hours under €10, while the reserve markets do — and they pay per megawatt of availability, so nothing in them rewards a fifth hour of storage.
31 August 2026 1

Figure 1.6 replaces MacKay’s two population graphs. Britain multiplied six-fold while its coal did, then grew 69% in the 125 years since; the world went the other way.
31 August 2026 1

MacKay’s four rectangle diagrams are numbered 1.8 to 1.11, so his own construction can be referred to and it is visible which new figure updates which.
31 August 2026 1

Figure 1.13 adds emissions per person over time. China passed the world average in 2006, the EU in 2013 and the UK in 2014, while every wealthy bloc roughly halved and the world average went from 4.3 to 4.7.
31 August 2026 1

Figure 1.14 adds cumulative emissions. MacKay’s claim that Britain is second only to the United States among historical emitters per person is still true — 1 158 tonnes a head against 1 259 — and is the one thing in the chapter that has not moved.
31 August 2026 1

Figure 1.12 redraws MacKay’s rectangle construction for 2000 and 2023. China goes from the wide, low block he describes to the largest area on the chart, taller than Japan; the United States falls from 26 tonnes a head to 18.
31 August 2026 1

Rebuilt: MacKay’s superseded figures are replaced rather than duplicated, and figures 1.16 and 1.17 are new — the world is 1.4% above its 2007 emissions per person on paths that required a steep fall, and a third of all greenhouse gases now come from electricity and heat.
31 August 2026 1

Figures 1.3, 1.4 and 1.5. The energy gap closed more by demand falling a quarter than by wind rising; CO₂ is 426 ppm against the 383 where MacKay’s chart ends; and the early coal history redrawn legibly.
31 August 2026 1

Figure 1.7 carries the coal chart to 2025. British coal peaked in 1913 and produced 1 TWh last year; world production is five and a half times its 1913 level and set its record in the 2020s.
31 August 2026 1

Figure 1.2 carries MacKay’s North Sea chart forward to 2025. Production is down 57% from its 2000 peak, Britain’s own share down 77%, while the real price went above anything on his original.
31 August 2026 1

Picks up the security-of-supply motivation MacKay drops: three-fifths of world oil comes from autocracies, Sweden buys 96% of its crude from democracies, and fungibility means that does not help. Chapter N contrasts it with uranium.
30 August 2026 26

A cheaper cell arrives with a tax advantage behind it, and changes nothing about duration — with the sodium cost claim checked against the lithium price it is measured against.
28 August 2026 31

The carbon cost of a war — four years of the war in Ukraine counted in tonnes, what removing them would cost, and the peacetime military emissions no inventory records; chapter 17 points to it.
21 August 2026 6 · L · 31

A check on solar power per unit area from Chinese utility-scale data, which reaches Cleve Hill’s figure by an independent route; chapter L the land totals that go with it; chapter 31 a daily carbon dioxide series behind the “nine seconds” comparison.
21 August 2026 11a

The electricity cost of a single AI answer, the land a data-centre campus stands on, who pays for its winning bid, and what generates its power.
20 August 2026 28a

Two claims about flexibility narrowed to what the study behind them supports.
19 August 2026 28a

The German flexibility study’s own figures moved into the text.
17 August 2026 28a

Argues the agency problem from the mechanism rather than a political slogan. Figure scripts declare their dependencies and run under uv.
16–17 August 2026 L

A third way of counting emissions and what it does to Sweden, and corrections to the carbon explorer instructions.
15–16 August 2026 31

Drained peat, and why filling the ditches in does not undo it.

New chapters

Each of these says so at the top, and none of them is MacKay’s writing.

Chapter What it is for Added
11a The machines behind the screen Data centres, from MacKay’s 0.4 kWh/d footnote about servers to a load that outbids industry and housing for a grid connection. 3 August 2026
28a The value of renewable energy as it scales What happens to the economic value of renewable electricity as it grows to the scale the balance sheet calls for. 30 July 2026
L The world in 2025 MacKay’s stock-taking redone for the most recent complete year. 1 August 2026
M Energy return on investment The question chapter 3 asks and then declines to answer properly. 2 August 2026
N Peak oil, peak gas, peak uranium Chapter 23’s Jevons calculation carried forward to the fuels it now matters for. 2 August 2026
O The hydrogen ladder Which uses of hydrogen are worth the electricity they cost, ranked. 9 August 2026
P A book that starts where this one stops Where MacKay’s method runs out: growth, and the waste heat that stops it. Points to Tom Murphy’s free textbook. 13 September 2026
Q The IPCC’s scenarios, in this book’s units What the Sixth Assessment Report’s pathways ask of each person, in kWh/d rather than gigatonnes. 13 September 2026

Licence

The original book is © 2008 David J. C. MacKay and licensed CC BY-NC-SA 2.0 UK. New and revised material in this edition is © 2026 Örjan Lundberg and is released under that same licence, as its ShareAlike term requires. Figures that carry third-party rights — the Private Eye cartoons, named-photographer photos, and the Ordnance Survey Crown Copyright maps — are omitted here and marked in place, because MacKay’s licence did not extend to them.