The illusion of the finite world : why growth is our only way out

Aymeric Roucher is a graduate of École Polytechnique, one of France’s most selective engineering schools. He sat its entrance exam. He knows exactly what he is talking about. And in 2025, on the program Club Le Figaro Idées, he said it without any flourish, with an engineer’s dry precision: “In math, AI has already passed me. I tested it on the Polytechnique entrance exam, and it demolished it. For now, we see no ceiling to the scaling laws.” Not a prediction. An observation. Made by someone who cannot be suspected of naïveté.

This testimony should have shattered the French economic debate. It went almost unheard, drowned out by the prophets of decline, the experts in austerity, the apostles of the finite world. Because in France, in 2026, the dominant view of growth is still that of Jean-Marc Jancovici: you cannot have infinite growth in a finite world. A pithy formula. The appearance of rigor. And a fundamental category error.

To understand why that formula is false is to understand why we stand at the very beginning, and not the end, of human potential.

The Fundamental Error: Growth ≠ Matter

Let us state the problem plainly: the degrowth argument rests entirely on an implicit equation. Growth = consumption of physical resources. If that equation holds, the conclusion follows on its own. If it is false, the entire edifice collapses.

It is false.

GDP measures value added, not volume extracted. It measures utility created, problems solved, needs met. And value has no mass. It occupies no volume. It is not used up when consumed: it multiplies when shared. A medical-diagnostic program deployed across ten thousand hospitals does not consume ten thousand times more resources than a single copy. It creates ten thousand times more value.

The most immediate illustration fits in any high schooler’s pocket: two hundred grams of glass and aluminum that, within the span of ten years, rendered obsolete the camera, the camcorder, the GPS unit, the voice recorder, the encyclopedia, the music player, the calculator, the road map, and two dozen other physical objects. More value, less matter. Not a little less: radically less.

The economic history of the past half-century confirms what intuition already senses. Over fifty years, the Western economies have decoupled the creation of wealth from physical consumption. The material intensity of GDP (the quantity of resources needed to produce one dollar of value) has fallen steadily since the 1970s. Degrowth advocates object that absolute global consumption is still rising. True, and that is good news: billions of human beings are finally climbing out of poverty and reaching a decent standard of living. But the structural signal is unambiguous: each unit of value created consumes less and less matter, and that trend is accelerating rather than slowing.

The confusion between extensive growth (producing more by consuming more) and intensive growth (creating more value out of the same resources) lies at the heart of the misunderstanding. The first belongs to the nineteenth century. The second is the model of the twenty-first, and artificial intelligence is about to accelerate it at an unprecedented pace.

As for the classic rebound-effect objection (consume less per unit and you will consume more units), it misses the essential point. AI does not merely make things marginally more efficient. It makes possible things that were structurally out of reach. Operating a power grid built entirely on renewable energy (intermittent, distributed, unpredictable by nature) is a cognitively impossible task for a human operator at the scale of a country. An AI agent does it. This is not optimization. It is a change of paradigm.

A Resource Is Not a Stock, It Is a Form of Knowledge

There is one question the theorists of the finite world never ask: finite relative to what? Relative to our current knowledge? Certainly. Relative to the physical reality of the universe? The question makes no sense.

A resource is not an attribute of matter. It is a relationship between matter and the intelligence that knows what to do with it. Coal was a useless black rock until Watt grasped the power of steam. Silicon was sand until Shockley invented the transistor. And the rare earths we now fret over were nothing but geochemical curiosities fifty years ago. In every case, it was not the matter that changed. It was the knowledge that turned matter into a resource.

This dynamic has always held true, and it is accelerating. DeepMind’s GNoME project illustrates it with a statistical brutality that pessimists find hard to swallow: in a matter of months, an AI mapped 2.2 million stable crystalline materials, hundreds of thousands of which had never been synthesized or even theorized. The equivalent of eight centuries of crystallographic research, produced in a single project. The figure is striking, but what is truly revolutionary is the method: GNoME does not search blindly. It predicts the stability of structures before any physical synthesis, eliminating 99% of laboratory failures before they ever occur. This is pure efficiency applied to scientific discovery itself. Among the materials identified: serious candidates for room-temperature superconductors, revolutionary battery architectures, semiconductors with entirely new properties.

We are told that lithium is running out. GNoME identifies sodium-based materials (sea salt, endlessly abundant) capable of serving the same functions. We are told that rare earths are under Chinese control. New alloys with equivalent properties are emerging from models that did not exist five years ago. This is not optimism. It is materials physics applied at the scale of artificial intelligence. Perpetual substitution (every resource declared scarce is eventually engineered around) is not a theoretical promise. It is the engine of humanity’s technical history, now turbocharged.

When Intelligence Becomes the Primary Resource

Let us return to Roucher. What he describes is not an anecdote about an exam. It is the announcement of a civilizational shift. The fact that an AI now obliterates the Polytechnique entrance exam is not an isolated data point about the performance of LLMs: it is the signal that the ultimate constraint on human growth is giving way. For as long as humanity has existed, that constraint has always been the same: human cognitive capacity. We have never run short of matter: we have run short of minds capable of turning it into value. A Polytechnique engineer is worth what he is worth precisely because he is rare, and because training an equivalent takes twenty years. If that rarity becomes obsolete, the quantity of intelligence that can be brought to bear on resource problems (optimization, substitution, discovery) becomes effectively unlimited. That is the ceiling now giving way.

AI agents are not glorified chatbots. They are systems capable of acting in an autonomous loop, of spotting optimizations invisible to the human eye, of juggling thousands of interdependent variables at once. Deployed across global supply chains, they cut empty return trips, compress lead times, and eliminate stockouts. Deployed across irrigation networks, they meter water down to the individual plant, wiping out billions of liters of waste. The value created is real. The resources consumed are smaller. This is the very definition of intensive growth, and it is far from the hammer metaphor of a tool you pick up and set down.

What these models are capable of once they are properly aligned remains massively underestimated, including by the people who use them. The general public is testing versions that lag six to twelve months behind the true state of the art. Meanwhile, labs are solving olympiad-level mathematics problems, mapping proteins that human biology would not have approached for decades, and beginning to design scientific experiments on their own. This is not some bout of national gloom that will reverse itself: it is a reality advancing with or without our consent.

Producing the same value with 30% fewer resources is growth. Solving in six months a problem that would have taken twenty years is growth. Giving a Senegalese farmer access to the agronomic expertise of a Swiss university is growth. No degrowth has ever produced results like these. Neither has any imposed austerity.

Earth Is Not the World

There is, in the finite-world argument, a question-begging assumption we too often let slide: the world, supposedly, is the Earth. Closed. Finite. Exhaustible.

This confuses the address with the universe. To declare the world finite while looking only at the Earth is like announcing you are about to starve while staring at an empty cupboard, when in fact you are standing in a giant supermarket whose door handle you simply haven’t found yet. The Earth is one room. The solar system is a continent. And we have only just cracked open the first window.

Concretely: the Earth is a ball of rock weighing 6 × 10²⁴ kilograms, suspended in a solar system that contains trillions of tons of metals, ores, and rare elements we are only beginning to map. The asteroid belt between Mars and Jupiter holds a concentration of ferrous metals and platinum that would dwarf the entire output of terrestrial mining over the past two centuries. 16 Psyche, a single metallic asteroid, contains estimated quantities of iron and nickel greater than the world’s total reserves.

Does that mean we will go mine it tomorrow? No. It means the constraint is not matter (it has never been matter) but the cost of transport required to reach it. And it is precisely here that Jancovici’s argument turns against itself. His demonstration rests on EROI, the energy return on extraction: if the energy spent extracting a resource exceeds the energy it yields, the resource ceases to be viable. A valid line of reasoning, in a world where the cost of energy and transport stays what it is today. Yet that cost is collapsing: SpaceX cut the cost per kilogram to low Earth orbit fivefold with the Falcon 9. Starship promises a further tenfold reduction, a total improvement of roughly fiftyfold over the previous generation of launch vehicles. Once the cost of accessing extraterrestrial resources becomes trivial, the EROI of space extraction flips, and the finitude of the resource evaporates with it. Meanwhile, on the energy-production side, the Flamanville EPR delivers 1,675 net megawatts to the grid, and players like Airbus, Thales, and Leonardo are reviving a European space ambition that, if it materializes, will change the orbital-transport equation for the continent. Blue Origin, Rocket Lab, and a dozen other players are converging on the same trajectory. Moving heavy industry out of the biosphere (smelting, chemical synthesis, extraction) is no longer a science-fiction metaphor. It is an industrial timeline with funded milestones.

The Malthusian Trap, Revisited

Let us ask the question bluntly: who benefits from degrowth?

Not the billions of people who still have no reliable electricity. Not the economies of sub-Saharan Africa, which legitimately want the standard of living that Europe granted itself over a century of unbridled industrial growth. Not the Western middle classes, whose purchasing power is already eroding. Degrowth, formulated in lecture halls heated by people with guaranteed pensions, is a luxury of the well-off, who have consumed their share and now want to close the door behind them.

It is the Malthusian trap, in its contemporary form. Malthus, in 1798, had the numbers exactly right. He was wrong about the one variable he had not modeled: innovation. His calculations left no way out, and yet eight billion human beings now eat better than an aristocrat of his era. Degrowth reproduces precisely that error: it fixes the constraints, projects the trajectory, and forgets the intelligence. This is what I call cognitive nostalgia, the deep bias that pushes us to extrapolate the present in a straight line and to systematically underestimate the ruptures to come.

France, in particular, excels at this posture. Between its openly accepted industrial decline, the public budgets devoted to agencies that subsidize renunciation rather than innovation, and proposals for universal basic income as the answer to technological disruption (social anesthesia rather than adaptation), we have turned the precautionary principle into state doctrine. The result: while DeepMind maps matter, while SpaceX rewrites the physics of space transport, while precision-agriculture startups reinvent the food chain, we produce reports on digital sobriety.

And one last thing must be named: it is wealth, not poverty, that solves environmental problems. The cleanup of rivers, the restoration of forests, the rise of solar power, carbon capture, the protection of the oceans: every one of these advances was financed by growing economies, not by shrinking ones. A society that gives up on growth also gives up on the resources that would let it repair the damage it has done.

The Only Resource That Is Truly Finite

Let us return to where we began: Roucher and the AI that demolishes Polytechnique. That observation is not trivial. It signals that the cognitive constraint (the ultimate limit on our ability to turn matter into value) is receding at a speed no one had anticipated. Not an evolution. A regime change.

We are a civilization living on one planet when billions are accessible. Using a fraction of the materials available within its own solar system. Having just created an artificial-intelligence tool whose ceiling is not yet in sight. Only beginning to master nuclear fusion. To claim, in this context, that the pie can no longer grow is a matter of either ignorance or bad faith.

The only truly finite resource on this planet is not lithium, nor coltan, nor the rare earths. It is the imagination of those who decided, before they had even looked, that the door was closed. Pessimism is not wisdom. It is a self-fulfilling prophecy: if we stop looking for the keys, we do indeed stay locked in, not because the door does not exist, but because we have chosen to stop looking for it.

The pie is going to grow. In ways we cannot yet imagine. And the only thing capable of preventing it is precisely what the degrowth camp proposes.