In January 2025, the Chinese AI lab DeepSeek released a model that appeared to match the performance of far more expensive American competitors — at a fraction of the cost. Tech stocks briefly tumbled. Commentators rushed to declare that the age of massive AI investment was over: if you could do the same work with less, surely we would need fewer chips, less energy, smaller data centers. Within hours, Satya Nadella, CEO of Microsoft, posted a single corrective on social media: "Jevons paradox strikes again!" [C1]
He was invoking a Victorian economist who had been dead for nearly a century and a half. And he was exactly right.
William Stanley Jevons published The Coal Question in 1865, when he was thirty years old and still scrambling for professional recognition. [C2] The book posed a deceptively simple question: as Britain's steam engines grew more efficient, burning less coal per unit of output, would the nation's coal reserves last longer? The intuitive answer — yes, obviously — was wrong. Jevons demonstrated that every improvement in the efficiency of coal use had historically been followed by an increase in total coal consumption. The better you got at using a resource, the more of it you used, because efficiency lowered the effective price, which expanded demand into new applications and new markets. The improved steam engine of James Watt did not conserve coal. It made coal the foundation of industrial civilization. [C3]
This is the Jevons Paradox, and it remains one of the most counterintuitive and durable ideas in economics. It is also, in our present moment, one of the most relevant. When DeepSeek showed that powerful AI could be run cheaply, the correct inference was not that the world would spend less on AI. It was that AI would become so affordable that it would be embedded in everything — every appliance, every workflow, every throwaway consumer app. Cheaper AI means more AI, not less. Nadella understood this because the logic is the same logic Jevons laid out about coal in the age of Palmerston.
But the most sweeping empirical confirmation of the Jevons Paradox was assembled not in the nineteenth century but in the twentieth, by the economist William Arthur Nordhaus, and it concerned not coal but something even more elemental: light.
The Economics of Light: A Four-Thousand-Year Experiment
William Nordhaus traced the true cost of illumination from ancient lamps to modern bulbs, revealing how efficiency multiplies consumption.
Nordhaus wanted to understand the true history of economic progress — not as measured by GDP or industrial output, but by the actual cost of something human beings have needed since the first campfire. He chose illumination. [C4] His project was almost absurdly ambitious in its temporal scope: he traced the cost of producing light from Babylonian sesame oil lamps around 1750 BC through to the compact fluorescent bulbs of the 1990s, a span of nearly four thousand years. The findings were staggering.
A Babylonian burning sesame oil in a clay lamp got roughly 24 minutes of light for a day's labor. A Roman with a tallow candle did somewhat better. A Victorian with a kerosene lamp did better still. But the gains were gradual and grinding for most of human history. Then came the revolutions: gas lighting, the incandescent bulb, the fluorescent tube. By the late twentieth century, an hour's labor at the average American wage could purchase roughly 20,000 lumen-hours of illumination. The efficiency gain over the full four millennia was on the order of 500,000-fold. [C5]
Now here is the question that Jevons would have asked: did we use less energy on lighting? Did the plummeting cost of a lumen-hour mean that humanity consumed fewer resources to see in the dark?
The answer, of course, is the opposite. The Babylonian lit a single lamp in a single room for a few precious minutes each evening. The medieval peasant might have a single rushlight. The Georgian household had candles, but used them sparingly — they were expensive, they smelled, they dripped. As late as the early nineteenth century, most human beings lived in something close to darkness after sunset. Then gas lamps arrived and cities began to glow. Then Edison's bulb, and the interior of the home was transformed. Then neon, and the commercial streetscape became a theater of light. Today we have Times Square, which pumps out roughly 161 million lumens across its billboards and screens, burning continuously, and nobody considers this particularly remarkable. [C6] We have office buildings that stay fully illuminated through the night with no one inside. We have highways ribboned with sodium vapor for hundreds of miles through empty desert.
The 500,000-fold improvement in the efficiency of producing light did not reduce humanity's energy expenditure on illumination. It increased it by orders of magnitude. We went from 24 minutes of dim, flickering sesame-oil light per day to a civilization that is visible from space. This is the Jevons Paradox operating across the longest time horizon anyone has ever measured. And it tells us something profound about efficiency, demand, and human appetite: making a good thing cheaper does not sate the desire for it. It reveals how much desire was always there, suppressed by cost. The Babylonian did not want only 24 minutes of light. He wanted what we have. He simply could not afford it.
Two Revolutions Hidden Inside One
Jevons's marginalism came packaged with empiricism and averageism, sowing the seeds of a theoretical tension.
Jevons is remembered today, when he is remembered at all, primarily for this paradox and for his role in the marginal revolution — the near-simultaneous discovery, by Jevons in England, Carl Menger in Vienna, and Léon Walras in Lausanne, that economic value is determined not by total utility but by marginal utility, the value of the last unit consumed. [C7] But Tyler Cowen has argued, persuasively, that Jevons was really leading two revolutions at once, and that understanding the second one — what we might call the "average revolution" — is essential to understanding why his version of marginalism ultimately prevailed.
To see what this means, we need to step back to the peculiar figure of Adolphe Quetelet.
Quetelet was a Belgian astronomer and statistician who, in the 1830s and 1840s, hit upon an idea that would reshape the social sciences. Astronomers had long understood that when you take many observations of a star's position, the measurements cluster around a central value — the true position — in a pattern we now call the normal distribution, or the bell curve. Quetelet's great move was to apply the same logic to human beings. He collected vast quantities of data on human height, weight, chest circumference, crime rates, marriage rates, and mortality, and found that these too clustered around central values in approximately normal distributions. From this he derived the concept of l'homme moyen — the "average man" — a statistical fiction who nonetheless, Quetelet believed, represented the deep tendency of a society. [C8]
The idea was enormously influential. It suggested that social life was governed by regularities as lawlike as those of astronomy. It gave reformers and governments a tool — the average — with which to describe and compare populations. And it shaped the entire statistical enterprise of the nineteenth century. When people collected data and summarized it, they computed means. When they looked for patterns, they looked at how averages changed over time. The intellectual world Jevons inherited was saturated with averageism, with the conviction that the average was the deep truth lurking behind the noise of individual variation.
Jevons absorbed this thoroughly. His work on index numbers — the attempt to construct a single measure of the "average" price level from many individual prices — was, in the judgment of later economists, "as careful and intelligent a piece of empirical work as can be found" in the economics of the period. [C9] Jevons did not merely compute arithmetic means of prices. He explored the properties of geometric means, considered weighting schemes, and thought hard about what an index number was actually measuring. This was not a sideline to his theoretical work. It was an expression of the same intellectual temperament: the desire to find patterns in data, to reduce complexity to manageable summaries, to discipline speculation with measurement.
From his index number work came a practical proposal that was ahead of its time by roughly a century: the tabular standard. Jevons suggested that long-term contracts — debts, wages, rents — should be indexed to the price level, so that inflation and deflation would not arbitrarily redistribute wealth between creditors and debtors. [C10] The idea was elegant, technically sophisticated, and almost entirely ignored in his lifetime. It would not be widely adopted until the inflation crises of the 1970s forced governments to take indexation seriously. But it shows Jevons operating in a mode quite different from the abstract theorizing of marginal utility. He was in the data, building tools, trying to make economics useful in a direct and practical sense.
And then there were the sunspots.
Jevons became convinced, late in his career, that business cycles were driven by periodic variations in solar activity — specifically, by the roughly eleven-year sunspot cycle, which he believed affected harvests, which in turn affected commercial activity and credit conditions. [C11] He was wrong. The correlation was spurious, or at best vastly overstated. But the enterprise was revealing. Jevons was trying to do something that no economist before him had seriously attempted: to find the physical cause of an economic phenomenon, to link the rhythms of commerce to the rhythms of nature. He wanted economics to be a science in the same sense that astronomy was a science — quantitative, predictive, grounded in observable regularities. He sought to unify the social and natural sciences at a time when the dominant tradition, represented by John Stuart Mill and John Elliott Cairnes, held that economics was a fundamentally different kind of inquiry from physics, because it dealt with human purposes and choices rather than with blind mechanical forces.
This is what Cowen means when he says that The Theory of Political Economy was really a plea for two distinct revolutions. The first revolution — marginal utility — was about the theory of value. The second revolution — what we are calling the average revolution — was about method. Jevons wanted economics to be empirical, statistical, data-driven, modeled on the natural sciences. He wanted it to traffic in index numbers and correlations and testable hypotheses. He wanted, in short, something recognizable to us as modern quantitative economics.
When Method Devours Theory
Empiricism's rise turned marginalism inward, replacing radical subjectivism with aggregate averages.
Here is the irony that Cowen draws out with characteristic relish: the two revolutions were in tension with each other, and eventually the second would undermine the first.
Marginalism is, at its core, about the individual. It says that value arises from the subjective experience of the last unit consumed, and that this experience differs from person to person, moment to moment. My marginal utility for a glass of water depends on how thirsty I am, what else I could spend my money on, my particular tastes and history. There is nothing "average" about it. The marginal revolution was a revolution precisely because it moved economics away from aggregate, objective measures of value — like the labor theory of value, which said that a good was worth the average labor time embodied in it — and toward the irreducibly individual and subjective.
But the average revolution pulled in the opposite direction. If your method is to collect data, compute means, build index numbers, and look for statistical regularities, you are necessarily averaging over individuals. You are doing exactly what Quetelet did: treating the variation as noise and the central tendency as signal. And the more successful the average revolution became — the more economists invested in national income statistics, price indices, macroeconomic aggregates — the less room there was for the radical subjectivism that marginalism, in its purest form, implied.
This is what Cowen means by the arresting claim that "the Jevons strand of marginalism contained the seeds of its own destruction." [C12] Jevons won the race — his version of marginalism gained traction faster than Menger's, spread more widely, became more influential in the English-speaking world — precisely because it was packaged with empiricism, with data, with the tools of the average revolution. Economists could adopt Jevons's marginal utility theory while simultaneously doing the kind of statistical, aggregate, average-focused work that felt like real science. But in the long run, the averageism swallowed the marginalism. The discipline moved toward macroeconomics, toward representative agents, toward models in which individual subjective experience was averaged away into aggregate demand curves and national accounts. Jevons's marginalism was the Trojan horse that brought averageism inside the gates of economic theory, and averageism, once inside, took over the city.
Menger, by contrast, was a purer marginalist. His economics was built more firmly on individual choice, on subjective valuation, on the logical structure of human action. He did not traffic in index numbers or sunspot correlations. He did not seek to unify economics with physics. His tradition — carried forward by Eugen von Böhm-Bawerk, Friedrich Wieser, Ludwig von Mises, and Friedrich Hayek — maintained the radical individualism of the marginal insight more faithfully. [C13] But precisely because it lacked the empiricist packaging, it was slower to spread and easier to marginalize (no pun intended) as the discipline professionalized around data and measurement in the twentieth century.
The Systematizer's Advantage
Jevons won not because he invented marginalism, but because he integrated it into a complete program.
Jevons did not arrive at his theoretical innovations in isolation. He had precursors — thinkers who had glimpsed pieces of the marginal idea before he assembled them into a system. Richard Jennings, an obscure English writer, had discussed the diminishing intensity of wants in the 1850s. Henry Dunning Macleod had played with related ideas in his idiosyncratic writings on credit and banking. Dionysius Lardner, a prolific popularizer of science, had discussed demand curves in the context of railway pricing. And Fleeming Jenkin, a Scottish engineer better known for his work on telegraphic cables, had independently developed supply-and-demand diagrams that anticipated much of what Jevons would formalize. [C14]
What Jevons brought that these precursors lacked was not any single idea but a vision — a sense of where the whole picture was headed. He saw that marginal utility, mathematical formalization, statistical method, and empirical investigation were all parts of a single program for remaking economics as a rigorous science. He was, in Cowen's word, a systematizer. He could take scattered intuitions and partial insights and arrange them into a coherent intellectual architecture. This is a rarer and more important talent than originality in the narrow sense. Many people have good ideas. Very few can see how the ideas fit together and what they imply for a field as a whole.
This systematizing vision is why Jevons prevailed — not because his version of marginal utility theory was more correct than Menger's or Walras's, but because it came embedded in a larger program that the economics profession found irresistible. It offered theory and measurement, abstraction and data, the elegance of mathematics and the respectability of empirical science. It was, in marketing terms, the full package. And like many full packages, it involved compromises that would only become apparent with time.
The Paradox Applies Everywhere
Efficiency gains in thought itself follow the same pattern: cheaper theory means more theory, more application, more everything.
Return, for a moment, to the Jevons Paradox and the history of light. There is a deep structural similarity between what happened with illumination and what happened with marginalism itself.
Jevons made economic theory more efficient — more compact, more rigorous, more powerful per unit of intellectual effort. You might have expected this to mean that less theorizing would be needed, that economics would settle into a stable body of established truths. Instead, cheaper theory meant more theory. The efficiency of Jevons's mathematical approach opened up vast new domains of application. Economics expanded into every corner of social life — into law, politics, family structure, crime, religion — precisely because the marginal framework made it cheap to generate economic analyses of almost anything. [C15] The Jevons Paradox applies not only to coal and light and computing power. It applies to ideas themselves. Make a mode of thought more efficient, and you do not get less thinking. You get Times Square.
Satya Nadella was right to invoke Jevons in January 2025. But the paradox runs even deeper than he suggested. It is not merely that cheaper AI will produce more AI. It is that every great efficiency gain in history — in energy, in illumination, in computation, in economic reasoning itself — has followed the same pattern. We do not economize on what becomes cheap. We feast. The Babylonian with his sesame lamp could not have imagined Times Square. Jevons with his coal tables could not have imagined a world saturated with marginal analysis. But the logic was already there, waiting, in the simple observation that when something becomes easier, we do not do less of it. We discover, at last, how much of it we always wanted to do.