Chapter Twenty: Who Will Buy the Cars?

In January of 1914, Henry Ford did something that the other industrialists of his age regarded as either madness or socialism: he doubled his workers' wages overnight, to the famous five dollars a day. The standard explanation, the one Ford himself sometimes encouraged, was a piece of homespun economic wisdom — that he wanted to pay his men enough that they could afford to buy the very cars they were building.

Historians will tell you, correctly, that Ford's actual motives were more tangled than the legend — he was bleeding workers to crushing turnover and brutal monotony, and the high wage bought him a stable, disciplined workforce. But strip away the question of what was in Ford's head and look at the structural truth the legend captured, because the legend captured something real and permanent about how an economy works. A factory is worthless without customers. And customers are not an abstraction; they are workers with money in their pockets. The mass-production economy Ford was inventing could not exist without a mass-consumption economy to absorb its output, and the two turned out to be the same people, seen twice: once as the producers who made the cars, and once as the consumers who bought them. Production and consumption are not two separate things. They are two ends of a single circuit, and the circuit runs through human beings who both earn and spend. Break the circuit at either end, and the most magnificent factory in the world grinds to a halt for want of anyone to sell to.

I begin with Ford because the largest investment boom in the history of the world is, right now, running his insight precisely in reverse. We are building the most powerful production machine ever conceived — and we are simultaneously dismantling the incomes, and shrinking the population, that would have to consume what it produces. We are pouring trillions into the factory while quietly firing the customers — and the people placing the bet have not noticed that it cannot work.

The One Thing Every Investment Assumes

Begin with the most basic logic of capital investment, because everything in this chapter follows from it, and because it is so foundational that the people deploying the capital have stopped consciously seeing it.

When you make a long-horizon capital investment — when you build a factory, lay a railroad, raise a data center — you are not building for today's demand. You are building for tomorrow's, and specifically for a tomorrow in which demand is larger than it is now. That is where the return lives. The factory pays off because in ten years it will sell more than it sells today; the railroad pays off because the territory it crosses will fill with people and freight; the whole exercise rests on the proposition that the future market will be bigger than the present one. Take that proposition away — imagine you somehow knew, with certainty, that your market would be smaller in ten years than today — and almost no long-horizon capital investment would ever be made. The growth is the reason.

And now ask what a "market" actually is, at its root, beneath the jargon. A market is people who want to buy things and have the money to do it. Its size is simply the number of those people multiplied by their purchasing power. Which means the foundational assumption beneath all capital investment — that the future market will be larger — is, when you strip it down, a demographic assumption. It is a bet that there will be more buyers tomorrow, and richer ones, than there are today. Every factory, every railroad, every data center ever financed has been, secretly, a wager on population and prosperity rising together. For two hundred years that was the safest wager on Earth. It is the wager this entire book has been arguing is no longer safe.

The Largest Bet Against the Surest Trend

Look, then, at what is actually being built. The artificial-intelligence build-out is the largest deployment of capital in human history — hundreds of billions of dollars a year and climbing, data centers rising across the deserts and drawing the electrical power of small nations, the most valuable companies on the planet committing sums that dwarf the Apollo program and the interstate highway system combined, and stock valuations that only make sense if one assumes years of explosive, compounding growth in demand for what these machines produce. Every dollar of it is, by the logic we just established, a bet that the future market for the output will be vastly larger than today's.

And by every measure this book has assembled, the future market is going to be smaller. Fewer people, as the population crests and turns down. Poorer people, in the knowledge classes whose incomes the same technology is deflating toward the cost of compute. The single largest capital bet ever placed is being staked, at this very moment, against the single most certain trend of the coming century. The capital is wagering on growth, into the teeth of decline.

The Robot That Buys Nothing

But there is a problem here deeper than "the market will be a bit smaller," and it is the one that makes AI different in kind from every investment boom that came before. To see it, return to Ford's circuit — production and consumption joined through human beings who both earn and spend — and watch what AI does to it.

Every previous labor-saving technology displaced workers on the production side while leaving them intact on the consumption side. The combine put the farmer out of the field, but the farmer (or his children) moved to the city, took a factory job, and kept buying things — kept being a customer. The displaced worker remained a consumer; the circuit stayed closed; demand was preserved even as the particular job was destroyed. This is, quietly, a large part of why the historical transitions worked: the technology changed what people produced, but it did not remove people from the economy, and people were always both halves of the circuit.

Artificial intelligence, carried to its logical end, does something no prior technology did: it removes the human from the loop entirely — and the human was not only the producer. The human was also the consumer. A robot that builds a house does not then need to buy a house. A model that writes the code does not pay rent, or buy groceries, or take a vacation, or raise children who will grow up to be customers in their turn. The machine produces, and consumes nothing. It earns, in a sense, but it does not spend, because it wants nothing. And so for the first time the labor-saving technology is not just shifting workers from one part of the circuit to another — it is cutting the consumption end of the circuit out altogether. You can build production capacity without limit. But you cannot sell the limitless output to a machine that needs nothing and to a human population that is shrinking and whose incomes you are simultaneously deflating. For a machine economy to sustain itself, the machines would have to be both productive and consumptive — to generate demand for their own output, to want and buy and need. They are only ever the first. A civilization of perfect producers that consume nothing is not an economy. It is a warehouse, filling up with goods that no one is left to buy.

Why the Old Booms Were Rescued

The believer has a powerful rejoinder, and it is the same one from the last chapter, wearing investment clothes: every great technology boom looked like a reckless bubble at the time, and every one was ultimately vindicated. The railroads of the nineteenth century were wildly overbuilt — speculative manias that ruined thousands of investors when they burst — and yet the rails remained, and a continent was bound together, and the investment, in the long run, paid for itself many times over. The internet boom of the late 1990s poured fortunes into fiber-optic cable and server capacity far beyond what the moment could use, and crashed spectacularly — and then the world grew into that capacity and more. The smartphone, the cloud, each the same story: overbuilt, corrected, vindicated. Why should AI be any different? Why is this bubble not simply the next boom that the future will justify?

And the answer is the one this book keeps returning to, because it is the hidden variable under everything: each of those booms was overbuilt for a population that was about to expand into it. The railroads were laid across a continent that was filling with tens of millions of new people and the freight they would generate. The fiber of the dotcom era was lit for the billions of human beings who would come online over the following two decades. The smartphone was a device manufactured for a species of nearly eight billion, almost all of whom would acquire one. In every case, the overbuilding was rescued by an incoming tide of new people pouring into the empty capacity and filling it. The boom was a bet on growth, and growth showed up, because there were always more people coming.

AI is the first great investment cycle of the modern era to be built precisely as that tide goes out. It is the first boom in two centuries with no expanding human population coming to fill the capacity it is creating — the first to place the old reckless bet on growth at the exact historical moment that growth reverses. The pattern that vindicated every previous overbuilding is, this one time, missing. The believers are pattern-matching to railroads and fiber. They have not noticed that the thing which actually saved the railroads and the fiber was not the technology. It was the people who were coming. And this time, no one is coming.

The Momentum Trade

If the premise is this shaky, why is the smartest money on Earth pouring in regardless? Because capital, in large quantities, does not actually test premises. It follows other capital. What we are watching has a precise structure, and financiers have a name for it: a momentum trade. A genuine breakthrough generates real and justified excitement; capital rushes in; capacity is built; valuations rise; and then the rising valuations themselves become the argument for the next wave of capital, which points to the last wave as its proof. Round and round, each dollar justified by the dollar before it. And nowhere in that self-reinforcing loop does anyone stop to test the buried assumption — that there will be a large and growing market of human beings to buy all this cheap output — because testing it is not what keeps the loop spinning. The loop spins on the arrival of new money, and new money keeps arriving as long as new money keeps arriving.

This is the structure of every great bubble in financial history, from the South Sea Company to the railway mania to the dotcoms: investment justified primarily by the fact that investment is happening. And the tell is always identical. There is one assumption that the whole edifice rests on — in the language of finance, the terminal assumption, the bet on perpetual growth in the final market — and it is precisely the assumption that no one examines, because to examine it honestly would be to stop the music. In the AI boom, the terminal assumption is a growing population of prosperous human customers. It is false. And it is the one thing the trillions are not allowed to look at directly.

What the Correction Looks Like

Let me be careful here, because it would be easy to overreach, and the overreach would discredit the real point. I am not predicting a single dramatic crash on a particular morning. The technology is genuine; the machines work; much of the infrastructure will prove useful. The correction I am describing is slower and grimmer than a crash, and more certain. It is a long, grinding disappointment: hundreds of billions of dollars of AI infrastructure that cannot generate returns remotely commensurate with its staggering cost, because the markets it was built to serve are shrinking even as the output it produces deflates in price. Not zero value — just nothing like the returns the valuations require. The capital is not incinerated in a night; it bleeds away over a decade, as the promised growth fails, quarter after quarter, to arrive, and the gap between what was spent and what comes back slowly becomes impossible to ignore. Like the railroads, the infrastructure will survive and find uses. Unlike the railroads, there will be no tide of new people coming to make it pay. We will come, later, to what that decade of disappointment actually feels like to live through.

There is, however, one form of the automation bet that escapes the trap of this chapter — not because it solves the problem of vanishing consumers, but because it changes the board entirely. Everything here has concerned AI as a producer of information, selling into a thinning market of human minds. But the same automation is about to climb out of the screen and into the physical world, in the form of generally capable robots — machines that do not merely draft the memo but lay the brick, pick the fruit, tend the body, run the line. And at the moment that human labor itself becomes a machine you can switch on anywhere, the deepest organizing principle of the global economy — the differential in the cost of human labor between one country and another — simply evaporates. When a robot costs the same to run in Ohio as in Guangzhou, the entire reason to ship the world's production across the oceans disappears with it. And the thing that rises to take labor's place as the source of national advantage is one almost no one is pricing today: geography itself.

That inversion is where we turn next.