Energy after fossil fuels

Foundations

The forecasts were all wrong in the same direction

Start with a fact that should have changed how everyone talks about this.

For about twenty years, official projections of solar photovoltaic costs and deployment were published by serious institutions with serious modelling teams. And they were wrong. Not scattered around the truth in the way honest uncertainty produces, but wrong in the same direction, year after year, always too pessimistic, often by a lot. The scenarios had to be rewritten repeatedly because reality kept arriving early.

When every error points the same way, the problem is not noise. It is the method.

The models mostly assumed costs would flatten out. What actually happened is that solar followed a pattern seen across many manufactured technologies: cost falls by a roughly constant percentage for every doubling of cumulative production. Not with time — with volume produced. This is sometimes called Wright’s law, and it is an empirical regularity rather than a law of nature, but for solar it has held across several orders of magnitude of cost decline.

Batteries have done something similar. A careful re-analysis using a much larger dataset found the historical decline in lithium-ion costs was faster than earlier estimates had suggested.

Which means the sacrifice frame is probably wrong

The standard framing of the energy transition is that it costs money and we should do it anyway because the alternative is worse. Sacrifice now, safety later.

Way and colleagues asked whether that is actually true, using forecasts built from observed technology cost trends rather than from expert judgement, and running them probabilistically.

Their result: a fast transition to solar, wind, batteries and electrolysers is likely to cost less than a slow transition, and less than continuing with a fossil-dominated system. Not less once you count avoided climate damage. Less before you count it.

Two things to say about this immediately.

It is a forecast, and forecasts are wrong. The paper’s own contribution is a method for being wrong less badly — using data on how technologies have actually behaved rather than assumptions about how they should behave. That is a real improvement and it is not a guarantee.

And it applies to the parts of the system these technologies serve. Which brings us to the rest of it.

How solar actually got cheap, since it matters

It was not one thing, and the answer is politically inconvenient for everybody.

A decomposition of photovoltaic cost reduction found the decline was driven principally by research and development and by economies of scale, acting through larger plants, more efficient modules, and cheaper materials.

The share shifted over time. Early on, publicly funded research did most of the work: the efficiency gains that made the technology viable at all. Later, scale did most of it: bigger factories, better processes, cheaper silicon, learning by doing.

So the answer to “was it government research or market deployment?” is: both, in that order, and neither alone would have worked. Research made a technology that was worth deploying. Deployment policy created the volume that made Wright’s law apply. Remove either and the curve does not happen.

This is the single most useful policy lesson available in energy, and it is routinely flattened into a slogan by whichever side is arguing.

Now the hard part, at length

If the lesson stopped here it would be cheerleading, and cheerleading is how people end up shocked later. So: the parts that are genuinely difficult.

Storage is cheap for hours and expensive for weeks.

This is a physical distinction that most coverage misses. Power is a rate, measured in watts. Energy is an amount, measured in watt-hours. A battery’s cost is dominated by the energy it stores, not the power it can deliver.

Shifting solar output from midday to evening needs a few hours of storage. That is a solved and increasingly cheap problem. Covering a still, dark week across a whole continent — the Germans call it Dunkelflaute — needs storage on a completely different scale, and the cost does not shrink proportionally.

Seasonal storage is harder again. In much of the world, solar output in winter is a fraction of summer, and no battery technology is remotely economical for storing summer energy until January.

Which is why firm resources are argued about.

Modelling of deeply decarbonised electricity systems finds that including firm low-carbon resources — nuclear, geothermal, hydro with storage, or combustion with carbon capture — lowers total system cost compared with variable renewables and storage alone. The intuition is that the last few per cent of a variable system is very expensive, because you are building enormous over-capacity to cover rare bad weeks.

That conclusion is sensitive to assumptions. It depends on how cheap storage becomes, how much long-distance transmission gets built, and how flexible demand turns out to be. Cheap enough storage or a big enough grid weakens the case; either could happen.

This lesson does not pick a winner, because the field has not. What it does say is that “just build more solar” is not an answer to the last ten per cent, and anyone claiming otherwise is skipping the hard part.

And a quarter of the problem is not electricity at all.

Aviation needs energy dense enough to carry itself. Long-distance shipping has the same problem at a different scale. Steelmaking uses coal as a chemical reducing agent, not merely as heat. High-temperature industrial processes are hard to electrify.

Cement is the sharpest case, and the most under-appreciated. Making cement means heating limestone until it decomposes, and that reaction releases carbon dioxide from the rock itself. Even with a perfectly clean furnace, the chemistry emits carbon. There is no fuel switch that fixes it.

Together these hard-to-abate uses are roughly a quarter of emissions. They have candidate solutions — hydrogen, synthetic fuels, novel cement chemistries, carbon capture — and none is yet cheap or proven at scale.

The equity problem, stated properly

Now the part that a course written in India has no business softening.

Committed emissions. Existing power plants, factories and vehicles will emit over their operating lives unless retired early. Tong and colleagues estimated that infrastructure already existing in 2018 committed emissions comparable to the entire remaining budget for 1.5 degrees.

And the geography of that commitment is the whole argument. Coal plants in Europe and North America are old, near the end of their design lives, and retiring them is mostly a matter of not replacing them. Coal plants in Asia — India very much included — are young. Built recently, financed on thirty- or forty-year lifetimes, with debt outstanding and workforces attached.

Meeting temperature goals would require operating those plants for a fraction of the life they were financed for. That is not a technical problem. It is a balance-sheet problem, and somebody has to absorb it.

Energy poverty. Per capita energy consumption in India remains far below the global average. A large share of humanity consumes too little energy for health, education and productive work — which is the condition every industrialised country escaped by burning fossil fuels, as lesson 8.3 described.

So the transition has to do two things at once that pull against each other: raise energy consumption for a third of humanity, and cut emissions globally. Any framing that treats reduced energy use as the goal has quietly assumed that the people currently without energy stay that way.

Both halves of this are why the fairness argument in lesson 8.4 is not a debating point. Whoever pays for early retirement of Asian coal, and whoever funds the alternative build-out, is the actual content of climate policy. The physics is the easy part.

One thing that has never been done

Past energy transitions — biomass to coal, coal to oil — took roughly fifty to a hundred years to reshape the primary energy mix, and none of them was deliberate. They happened because the new source was better and cheaper for the user.

This one has two features with no precedent. It is on a deadline set by physics rather than by economics. And it is a transition to sources that are, for most uses, now genuinely cheaper — which is the one thing working in its favour, and the reason the sacrifice frame deserves retiring.

Fast enough is the open question. Not whether.

Deeper

What Wright’s law is and is not

The empirical claim is that unit cost falls as a power law in cumulative production, with an exponent that is stable over long periods for many manufactured goods. It is descriptive: it does not explain why, and the underlying mechanisms — learning by doing, scale economies, induced innovation, input cost declines — differ by technology. It fails or reverses where a technology is dominated by site-specific construction rather than manufacturing, which is the standard explanation for why nuclear fission costs have risen in some countries while solar has fallen. Whether that failure is intrinsic to the technology or a consequence of regulatory and construction practice is contested, and a technology built repeatedly in factories rather than bespoke on site would be the test.

Why the last ten per cent is expensive

In a system supplied mainly by wind and solar, the residual load — demand minus renewable generation — has a long tail. Most of the year it is small; occasionally it is large and persists for days. Meeting that tail with generation means building capacity that runs for a very small number of hours a year, which spreads its capital cost over almost no output. Meeting it with storage means sizing the store for the worst event on record, which is enormous. Meeting it with transmission means betting that the weather is uncorrelated across the connected area, which for large continental systems it often is not. The cost curve of a decarbonised grid is therefore convex: cheap to reach eighty per cent, steep thereafter, which is exactly where firm resources and demand flexibility earn their place in the models.

The cement chemistry

Portland cement clinker is produced by heating limestone, calcium carbonate, to around 1450 degrees. The reaction is calcium carbonate decomposing to calcium oxide and carbon dioxide. Roughly half to two thirds of the emissions from cement production come from this reaction rather than from the fuel used to drive it. Options are therefore capture at the kiln, substitution of clinker with supplementary materials such as slag or calcined clay, or alternative binder chemistries that do not start from carbonate. Each is being pursued; none currently matches Portland cement on cost, availability and a century of engineering experience simultaneously.

How committed emissions are estimated

The method inventories existing infrastructure by type, age, capacity and region, assumes typical operating lifetimes and utilisation rates, and integrates expected emissions forward. Its sensitivity is mostly to the assumed lifetime and capacity factor, so the estimates are ranges rather than points. The regional asymmetry is robust to these choices, because it follows from the age distribution of the fleet, which is directly observed rather than assumed.

Frontier

Long-duration storage

Approaches under development include flow batteries, thermal storage, compressed and liquefied air, gravitational storage and hydrogen produced by electrolysis and burned or fuel-celled back. Hydrogen is chemically the most flexible and round-trip efficiency is its weakness. Which of these, if any, becomes cheap enough to change the firm-resource argument is genuinely open.

Whether solar’s learning rate continues

Extrapolating a learning curve assumes the mechanisms behind it continue. For solar, module cost is now a minority of installed system cost in many markets, so further module cost declines matter less than reductions in installation, land, grid connection and financing. Whether those non-module costs follow their own learning curves is much less well established.

Transcript and shot list

Target runtime about 10 minutes. Built to overturn the sacrifice frame with data, then spend the second half on the genuinely hard parts so nobody leaves complacent. Shot types: ANIM is rendered from code with Manim, AI is a generated shot from Higgsfield with the shared style reference, REAL is agency imagery with credit, CAM is the narrator.

[00:00] S01 · ANIM · A stack of official solar cost and deployment projections from successive years, each drawn flat from its publication date. Then the actual cost curve plunging through all of them.

For about twenty years, serious institutions with serious modelling teams published projections of solar costs and deployment. And they were wrong. Not scattered around the truth the way honest uncertainty looks. Wrong in the same direction, year after year, always too pessimistic, often by a lot. The scenarios kept having to be rewritten because reality kept arriving early.

[00:27] S02 · CAM · Narrator direct to camera.

And when every error points the same way, the problem is not noise. It is the method.

[00:35] S03 · ANIM · Cost against cumulative production on log axes, tracing a straight line across several orders of magnitude — a constant percentage fall per doubling of volume, not per year. Then the same pattern drawn for lithium-ion batteries.

The models mostly assumed costs would flatten. What actually happened is that solar followed a pattern seen across a lot of manufactured technologies. Cost falls by a roughly constant percentage for every doubling of cumulative production. Not with time. With volume produced. It is sometimes called Wright’s law, and it is an empirical regularity rather than a law of nature, but for solar it has held across several orders of magnitude. Batteries did something similar, and a careful re-analysis with a much bigger dataset found the historical decline was faster than earlier estimates had said.

[01:18] S04 · ANIM · The sacrifice frame drawn as a trade-off curve, then replaced by the Way et al. result: fast transition costing less than slow, and less than continuing, before climate damages are counted.

Which matters, because the standard framing of this whole subject is that the transition costs money and we should do it anyway. Sacrifice now, safety later. Way and colleagues asked whether that is actually true, using forecasts built from observed technology cost trends rather than expert judgement, run probabilistically. Their result is that a fast transition to solar, wind, batteries and electrolysers is likely to cost less than a slow one, and less than carrying on with a fossil-dominated system. Not less once you count avoided climate damage. Less before you count it.

[02:01] S05 · CAM · Narrator direct to camera.

Two things immediately. It is a forecast, and forecasts are wrong. What that paper contributes is a method for being wrong less badly, by using data on how technologies have actually behaved instead of assumptions about how they ought to. That is a real improvement, not a guarantee. And it applies to the parts of the system these technologies serve, which is not all of it.

[02:32] S06 · ANIM · Cost decline decomposed: research and development, and economies of scale, acting through plant size, module efficiency and materials. Then the share shifting over time — research dominant early, scale dominant later.

So how did solar get cheap? Not one thing, and the answer is politically inconvenient for everybody. Decomposing the cost decline, it was driven principally by research and development and by economies of scale, working through bigger plants, more efficient modules and cheaper materials. And the share shifted. Early on, publicly funded research did most of the work — the efficiency gains that made the thing viable at all. Later, scale did most of it: bigger factories, better processes, cheaper silicon, learning by doing.

[03:11] S07 · CAM · Narrator direct to camera.

So: government research, or market deployment? Both, in that order, and neither alone would have worked. Research produced a technology worth deploying. Deployment policy created the volume that made the learning curve apply. Take away either one and the curve does not happen. That is the most useful policy lesson available in energy, and it gets flattened into a slogan by whichever side is arguing.

[03:41] S08 · CAM · Narrator direct to camera.

If I stopped there this would be cheerleading, and cheerleading is how people end up shocked later. So here are the parts that are genuinely hard.

[03:53] S09 · ANIM · Power as a rate in watts, energy as an amount in watt-hours; battery cost scaling with stored energy. Midday-to-evening shifting shown as a few hours and cheap. Then a Dunkelflaute — a still, dark week across a continent — and the storage requirement growing enormously. Then winter-to-summer seasonal storage drawn far off scale.

Storage is cheap for hours and expensive for weeks, and that is a physical distinction most coverage misses. Power is a rate, in watts. Energy is an amount, in watt-hours. A battery’s cost is dominated by the energy it stores, not the power it can deliver. Shifting solar from midday to evening needs a few hours of storage, and that is a solved and increasingly cheap problem. Covering a still, dark week across an entire continent — the Germans call it Dunkelflaute — needs storage on a completely different scale, and the cost does not shrink proportionally. Seasonal is harder again. In much of the world winter solar output is a fraction of summer, and no battery is remotely economical for holding summer energy until January.

[04:51] S10 · ANIM · The residual load curve with its long tail; capacity built for a handful of hours a year; the system cost curve drawn convex — cheap to eighty per cent, steep after. Then firm resources added and the curve flattening, with the assumption dependencies flagged.

Which is why firm resources get argued about. Modelling of deeply decarbonised grids finds that including firm low-carbon resources — nuclear, geothermal, hydro with storage, combustion with carbon capture — lowers total system cost compared with variable renewables and storage alone. The intuition is that the last few per cent of a variable system is very expensive, because you are building enormous over-capacity to cover rare bad weeks. But that conclusion is sensitive to assumptions. It depends on how cheap storage gets, how much long-distance transmission is built, and how flexible demand turns out to be. Cheap enough storage, or a big enough grid, weakens the case, and either could happen. I am not going to pick a winner here, because the field has not. What I will say is that just build more solar is not an answer to the last ten per cent.

[05:58] S11 · ANIM · The hard-to-abate quarter: aviation needing energy dense enough to lift itself, shipping, steel using coal as a chemical reducing agent, high-temperature heat. Then cement: limestone heated to 1450 degrees, the carbonate decomposing, CO₂ released from the rock itself with a perfectly clean furnace shown alongside.

And about a quarter of the problem is not electricity at all. Aviation needs fuel dense enough to carry itself. Shipping has the same problem at a different scale. Steelmaking uses coal as a chemical reducing agent, not just as heat. High-temperature industry is hard to electrify. Cement is the sharpest case and the most under-appreciated. Making cement means heating limestone until it decomposes, and that reaction releases carbon dioxide out of the rock. Even with a perfectly clean furnace, the chemistry emits carbon. There is no fuel switch that fixes it. These uses have candidate solutions — hydrogen, synthetic fuels, new cement chemistries, capture — and none of them is yet cheap or proven at scale.

[06:52] S12 · ANIM · Committed emissions: existing infrastructure integrated forward, totalling comparably to the remaining 1.5-degree budget. Then the age distribution of coal fleets — old in Europe and North America, young across Asia — with financed lifetimes, outstanding debt and workforces attached.

Now the part a course written in India has no business softening. Existing plants, factories and vehicles will emit over their operating lives unless somebody retires them early. Infrastructure already standing in 2018 was estimated to commit emissions comparable to the entire remaining budget for 1.5 degrees. And the geography of that commitment is the whole argument. Coal plants in Europe and North America are old, near the end of their design lives, and retiring them mostly means not replacing them. Coal plants across Asia, India very much included, are young. Recently built, financed over thirty or forty years, with debt outstanding and workforces attached. Meeting temperature goals means running them for a fraction of the life they were financed for. That is not a technical problem. It is a balance-sheet problem, and somebody has to absorb it.

[07:55] S13 · ANIM · Per capita energy consumption by country, India far below the global average; a third of humanity below the threshold for health, education and productive work. Then two arrows pulling opposite ways: raise consumption for a third of humanity, cut emissions globally.

And energy poverty. Per capita energy use in India is far below the global average, and a large share of humanity consumes too little energy for health, education and productive work — which is precisely the condition every industrialised country escaped by burning fossil fuels, as lesson 8.3 described. So this transition has to do two things that pull against each other. Raise energy consumption for a third of humanity, and cut emissions globally. Any framing that treats using less energy as the goal has quietly assumed the people who currently have none stay that way.

[08:40] S14 · CAM · Narrator direct to camera.

Which is why the fairness argument from lesson 8.4 is not a debating point. Who pays for the early retirement of Asian coal, and who funds the alternative build-out, is the actual content of climate policy. The physics is the easy part.

[08:59] S15 · ANIM · Past transitions — biomass to coal, coal to oil — drawn over 50 to 100 years, each happening because the new source was better for the user. Then this one, marked with a deadline set by physics, and with the new sources already cheaper. Calendar returns.

Past energy transitions, biomass to coal and coal to oil, took roughly fifty to a hundred years to reshape the primary energy mix, and not one of them was deliberate. They happened because the new source was better and cheaper for the person using it. This one has two features with no precedent. It is on a deadline set by physics rather than economics. And it is a transition to sources that are, for most uses, now genuinely cheaper — which is the single thing working in its favour, and the reason the sacrifice framing deserves retiring. Fast enough is the open question. Not whether. Next: the technology that is currently eating a noticeable share of that electricity, and whether it counts as a new kind of learning.

Shot list

Zero AI shots, ten code-rendered shots, five narrator shots. The narrator carries the argument in this one; the animation carries the numbers.