
...
By Gabriel Snyder
Executive Editor, Enterprise
0
Share
Newsweek is a Trust Project member
See more of our trusted coverage when you search.Prefer Newsweek on Googleto see more of our trusted coverage when you search.
Whenever doubts are raised about the economics of the AI boom, ardent believers have two magic words at the ready to wave them away: “Jevons paradox.” The incantation got its first major test in January 2025 after a little-known Chinese lab called DeepSeek released a reasoning model that rivaled the best from Anthropic and OpenAI but reportedly cost only $5.6 million to train compared to the hundred-million-dollar-plus price tags for frontier models—and, critically for customers, cost 95 percent less to use.
If near-frontier AI could be built that cheaply, what were trillions of dollars in chips and data centers for? The news led to a rout for AI stocks, wiping out more than $1 trillion in value. But the night before markets opened, Microsoft CEO Satya Nadella had his answer ready: “Jevons paradox strikes again!”
Read More on AI
He explained (and helpfully linked to the Jevons paradox Wikipedia page for those unfamiliar) that the 19th-century economic theory had predicted that as AI got cheaper, “we will see its use skyrocket.” To anyone who thought crashing AI token prices were a threat to the buildout, Nadella argued that it only added to the case for it.
“Jevons paradox,” named after the English economics and philosophy professor William Stanley Jevons, has remained a near-constant refrain in the AI discourse ever since, even when not cited by name. In October, before he resigned as Nvidia’s CEO, Jensen Huang told the BG2 podcast, “We want intelligence costs to decrease by a factor of two or 20 every single year, because the consumption of it is going to more than make up for it.”
It does not seem that many in the AI world have read the 1865 book, The Coal Question, that Jevons paradox is drawn from.
Econ 101 in a Top Hat
The paradox he is now famous for was something of an aside to his main argument—as the book’s subtitle put it, “The Probable Exhaustion of Our Coal Mines”—laid out in a single, slender chapter entitled “Of the Economy of Fuel.” Its aim was to rebut any who believed technological innovations and efficient ways of using coal would forestall a time when Britain’s cheap coal became scarce.
“It is wholly a confusion of ideas to suppose that the economical use of fuel is equivalent to a diminished consumption. The very contrary is the truth,” Jevons wrote. If a furnace required less coal, then it would make that furnace less costly to operate, resulting in more furnaces being used and more coal being burned. “It is the very economy of its use which leads to its extensive consumption. It has been so in the past, and it will be so in the future. Nor is it difficult to see how this paradox arises.”
No, it is not. Because stripped of its Victorian phrasing, Jevons’s insight is the same elastic demand curve taught in every intro to economics: When the price of something goes down, sales go up.
Jevons paradox was largely forgotten for a century until it started getting notice in the wake of the 1970s energy crises. Economists Daniel Khazzoom and Leonard Brookes argued that, counterintuitively, more efficient appliances, cars and factories would not cut overall energy use and might well increase it. In 1992 Harry Saunders christened the Khazzoom-Brookes postulate. It was a skeptic’s argument, aimed at those selling efficiency as a free lunch. It would take three more decades, and an AI boom, to turn Jevons paradox into a growth strategy.
What Jevons Was Actually Worried About
At the time Jevons wrote, Britain was near the height of its economic and military power, and undergirding it all was cheap coal. By his own tally, Britain mined 80 million of the 136.5 million tons produced worldwide around 1860, about three-fifths of the planet’s coal. And the rate of mining was only growing with British coal output climbing to 98 million tons in 1865, roughly 50 percent growth in a decade.
“Coal in truth stands not beside but entirely above all other commodities,” he wrote. He called his era “the Age of Coal,” and Londoners could see it in the air: the city’s famous fog was in large part coal smoke.
But Jevons was not celebrating demand. He was worried about price. Britain did not get rich selling coal (it exported less than a tenth of its output) but rather by burning it. Cheap, accessible coal at home allowed it to produce the iron, cloth and machinery it sold to the world, as well as to power the steamers that brought its manufactured goods to them. This abundance of cheap coal underwrote what Jevons called Britain’s “manufacturing and maritime supremacy.”

...
His question was never whether Britain would literally run out of coal. “We may some day have to pay dear for fuel,” he wrote, but “it will never be positively wanting.” His primary concern was how much more coal would cost in the future and “the duration of our present cheap supplies of coal,” because on them “England’s manufacturing and commercial greatness, at least, is at stake.”
That was what made his arithmetic so alarming. Jevons cited an estimate that the total accessible coal remaining in Britain was 83.5 billion tons. Consumption was growing about 3.5 percent a year, and, he wrote, “if our consumption of coal continue to multiply for 110 years at the same rate as hitherto, the total amount of coal consumed in the interval will be one hundred thousand millions of tons.” (That’s 100 billion for those of us in the 21st century.)
As the easy seams were worked out, remaining British coal would be deeper, more difficult and costlier to mine, while countries with vast untouched coalfields of their own would be able to catch up with Britain’s lead. His country, Jevons predicted, “must witness the coal produce of other countries approximating to our own, and ultimately passing it.”
From Warning to Rallying Call
This is also why Jevons didn’t believe efficiency could save Britain. Rebutting his critics in the preface of a second edition, he noted that they forget “that other nations can use improved engines as well as ourselves, so that our comparative position will not be much improved.” Better engines wouldn’t protect Britain’s lead because rivals would copy them. It is a strange portion of the argument to forget on the night a Chinese lab had just built a more efficient engine.
Which makes it ironic that modern Jevons admirers now invoke his name in the face of collapsing prices. For Jevons, cheap coal was crucial because Britain was the one burning it. The hyperscalers are on the other side of the trade: They sell tokens for others to burn. When prices collapse, the windfall goes to their customers.
Consider Google’s AI token stats: At Google I/O 2026, CEO Sundar Pichai disclosed that Google now processes more than 3.2 quadrillion tokens a month, up from 9.7 trillion two years earlier. That roughly 330-fold expansion is presented as proof that Jevons was right about AI: Demand is exploding!
But here is the other half of that story: The cost of AI has been collapsing. In early 2023, months after ChatGPT was released, industry-leading intelligence cost $30 per million input tokens. Since then, open-weight rivals like Meta’s Llama, DeepSeek and Moonshot’s Kimi have flooded the market, and the price users actually pay for LLMs has fallen far below that.
In 2025, when Silicon Data introduced its index measuring what the market pays per million AI tokens across hundreds of models, weighted by use, the figure stood at about $1.09 per million tokens, peaking at $2.07 this past May. But with “tokenmaxxing” out of fashion in the C-suite, it has since dropped to $0.76.
He Was Right
Jevons correctly foresaw that Britain's coal boom would end and was right about how: not with empty mines, but with cheaper competition. He even spotted the competitor. A footnote in the first edition records Pennsylvania’s anthracite output climbing from 3.2 million tons in 1850 to 8 million in 1860. In 1899 the United States, thanks to Pennsylvania coal miners, passed Britain as the world’s largest producer.
British output peaked in 1913 and then the economics Jevons feared arrived. British coal grew costlier to mine and too expensive for its overseas markets. In 1921, the government ended the control of the mines it had maintained long after World War I, and the owners immediately went after wages. The result was a lockout that year and another in 1926, a nine-day General Strike and a lockout of more than a million miners. In 1947 the U.K. nationalized the industry.
Demand never disappeared; Britain kept burning coal for more than a century. What disappeared were the margins. The money in the coal economy went to those who used cheap coal to make things higher up the value chain, not to the people who dug it.
That is the part of the story the AI boosters skip. The hyperscalers are spending close to $1 trillion this year on data centers whose chips will be outdated in a few years, often with borrowed money. Jevons had a line for that, too: “nothing so presses upon trade as the necessity for a large capital expenditure; it is so much more risked, so much more to pay interest on.”
Brief Greatness
The deepest misreading, though, is about duration. At root, Jevons’s real subject was how long a boom can last. “The question here treated,” he wrote in the preface to the second edition, “regards the length of time that we may go on rising, and the height of prosperity and wealth to which we may attain.”
Looking at the Britain of his time, with its “accumulating capital, with rising profits and interest,” he called it “a union of happy conditions which hardly any country has before enjoyed, and which no country can long expect to enjoy.”
The AI industry has borrowed his paradox as proof that the good times can only go on. The book it comes from argues the opposite. Jevons’s final sentence is one which people namedropping him might read twice: “We have to make the momentous choice between brief greatness and longer continued mediocrity.”
View 0 Comments
Submit Correction
View Editorial & AI Guidelines