The market is not a machine for producing wealth; it is a machine for allocating scarce resources: land, labor, capital, and attention. Every economic model we teach begins from that premise, and the premise has held because the inputs to production have always been scarce. Artificial general intelligence is the first technology that attacks the scarcity at the root: the capacity to set a goal, reason about it, and act on it, the thing we have always paid humans for, becomes a capital good that can be replicated at near-zero marginal cost. The post-AGI economy is not a faster version of the current one. It is an economy with a different binding constraint, and many of our intuitions about prices, wages, and value will not survive the transition.

1. The assumption that breaks

Economics begins with scarcity, and scarcity has always meant, at bottom, scarce intelligence. Land was scarce because deciding how to use it took judgment; capital was scarce because deciding where to put it took judgment; labor was scarce because the ability to reason, coordinate, and execute was attached to slow, mortal, expensive human bodies. Every market we have ever observed priced the fact that somebody had to think.

AGI breaks that link. It does not remove scarcity; thermodynamics sees to that. But it changes the character of the scarce thing. Thinking stops being a service performed by people and becomes a manufactured good. The difference between AGI and every earlier technology is a difference of degree raised to a difference of kind. The industrial revolution replaced muscle and left judgment to humans; the information revolution replaced routine cognition and left the hard parts to humans. AGI is the first technology that aims at judgment itself: not a tool for one task but a capacity that can be pointed at almost any task. That is why the standard reassurances, that every previous technology created more jobs than it destroyed, are at best incomplete. We will return to that question in a moment.

2. The price of intelligence falls toward zero

Software is the most deflationary good ever produced: high fixed cost, near-zero marginal cost. Once written, a program can be copied at negligible expense, and competition drives the price of each additional copy toward zero. AGI extends that property from the product to the capability. Once a model is trained, every additional act of reasoning, whether an analysis, draft, design, diagnosis, or line of code, costs almost nothing to produce. The McKinsey Global Institute's estimate that generative AI could add between $2.6 trillion and $4.4 trillion to annual global output ^[1] captures the upside; the deflationary side is quieter. The new value will not accrue to wages, because wages are the price of scarce human labor, and the labor is precisely what the technology replaces.

The consequences appear first in prices. Intelligence-intensive services that were once expensive, such as translation, drafting, basic analysis, routine programming, and customer support, now fall toward the cost of electricity and compute. This is Baumol's cost disease in reverse. Baumol observed that labor-intensive services grow relatively more expensive as productivity rises elsewhere; in the post-AGI economy, the sectors that resist automation become the expensive ones. In-person care, the exercise of taste, legal accountability, and physical presence are not easily substituted by models. The result is a two-speed economy: near-free, superhuman intelligence on one side; stubbornly expensive, irreplaceably human attention on the other.

3. What remains scarce

If the price of intelligence falls toward zero, what is left to price? The list is shorter than we are used to, but it is not empty.

Notice the structure of the list. The first three items are physical: they sit on the supply side and cannot be conjured by intelligence, however general. The last four are social: they exist because humans are finite and because markets require trust. Rents do not disappear in the post-AGI economy; they migrate upstream, away from who can do the work and toward who controls the physical substrate of the work, and who is trusted to do it. That migration, more than any technology, will determine the shape of the economy.

4. Labor and the missing frontier

The historical record is genuinely reassuring, and it should be taken seriously. In 1900 roughly forty percent of American workers farmed; today the figure is below two percent. Manufacturing employed about a quarter of the workforce in the middle of the twentieth century and employs roughly eight percent today. Yet total employment rose, because new technologies did not simply replace tasks; they created new ones, and, as David Autor has argued, the new tasks repeatedly required new human skills that machines could not yet perform.^[2]

The unsettling discontinuity is this: every previous wave left humans a comparative advantage somewhere. There was always a frontier of new tasks demanding judgment, dexterity, or social intelligence, where the machine was still weaker than the person. The frontier is exactly what AGI erases. If a general intelligence can perform the new tasks better than any human, then the famous race between education and technology^[3] becomes a race in which one competitor writes both sides of the equation. The wage for a task is set at the margin: the least-productive person still doing it sets the price. When a machine can do the marginal task for near-zero cost, the floor under wages erodes.

This is not a prediction of zero employment. Humans will continue to work for meaning, for status, and for structure, and labor markets will persist in attenuated form, especially in the trust-and-taste sectors. But the economic function of work changes. The labor share of national income in the United States has already fallen from roughly 65 percent in the 1970s to under 60 percent today, in an era of comparatively weak automation.^[4] The post-AGI economy is not necessarily an economy without workers. It is an economy in which work is no longer the primary mechanism by which output is distributed to people.

5. Compute and energy: the new oil

Every act of intelligence in the post-AGI economy has a physical cost, and the cost is denominated in electricity. Training a frontier model consumes on the order of tens of gigawatt-hours; serving it to billions of users multiplies that demand indefinitely, because inference scales with usage rather than with production. The International Energy Agency estimated that data centers consumed about one and a half percent of global electricity in 2022, and every projection since has pointed sharply upward.^[5] Meanwhile the compute itself has been doubling roughly every six months in recent years,^[6] and doubling is a statement about a physical supply chain of chips, fabrication plants, and power grids, with lead times measured in years.

The strategic consequence is that energy security becomes intelligence security. The jurisdictions that control cheap, abundant, reliable electricity, and the industrial capacity to turn it into compute, will host the means of production of the era, the way coal fields and ports hosted the industrial one. This is a concentration dynamic: a handful of firms and regions already own the frontier models, and the economics of scale favor them. Compute-hours will also become a kind of money; contracts are already denominated in compute. But it is a peculiar currency: it depreciates quickly, is useless without energy, and concentrates wherever the infrastructure is. It prices the scarce thing well, but it does not store value.

6. The ownership economy

If wages are the price of scarce human labor, and the labor becomes cheap, then the distribution of output shifts to the other factors of production: capital, energy, compute, and the social permissions to deploy them. The post-AGI economy is an ownership economy. Output becomes a function of capital held rather than hours worked, and the critical variable is not the technology but the distribution of that capital.

Consider three scenarios. In the first, ownership is narrow: a few firms own the models, the chips, and the energy contracts; rents rise, wages fall, and the gilded-age pattern of the late nineteenth century repeats with faster dynamics. In the second, ownership is broad: public or widely held institutions own the substrate of intelligence, including sovereign compute funds, universal dividends from energy and model revenues, pension and sovereign wealth funds invested in the new infrastructure. The abundance that used to flow out as wages is paid out as distributed capital income. The Alaska Permanent Fund is the small-scale precedent for this kind of institution. In the third, and most likely, scenario, both things happen at once: the models are concentrated, but taxes, dividends, and public compute partially redistribute the surplus. Nothing in the technology forces any of these outcomes; the technology only determines that distribution happens through ownership rather than through the wage bargain.

The uncomfortable novelty is that the means of production in this economy are unusually easy to own and unusually hard to contest. A factory requires workers; a model requires capital and energy. The people who build and operate the machines will be few relative to the population they serve. Who owns the weights is the twenty-first-century version of who owns the land, and the answer, whatever it is, will be written into law, taxation, and corporate governance long before it is noticed in the headlines.

7. Measuring the impossible

GDP is a measure of market production, not of welfare, and the gap between the two is about to become enormous. When a service that once cost a hundred dollars becomes free, GDP records a loss: the paid transaction disappears, while welfare rises. The post-AGI economy will be spectacularly deflationary in this sense: vast quantities of value will be produced at the margin of almost nothing, and most of it will not appear in the national accounts at all. Policy made on GDP will be flying blind.

The macroeconomics of the transition are unfamiliar. Productivity growth of the kind AGI promises is normally a blessing, but here it arrives with a distributional wound: output per worker rises while the price of the worker falls. Deflationary pressure on goods and services will sit awkwardly beside inflationary pressure on the truly scarce assets: energy, compute capacity, land near infrastructure, and the few remaining goods that certify trust. Monetary policy, calibrated to a labor-wage economy, will be pushed toward responses it was not designed for. New metrics are needed: measures of consumer surplus, capability-adjusted output, and energy and compute throughput as proxies for the real production of the era. Until we build them, the economy will be larger and more unequal than the numbers say, simultaneously more abundant and more fragile than it appears.

8. Policy levers

The policy problem is not that the tools do not exist; it is that the shock arrives faster than institutions adapt. A few levers are available, and each trades one risk for another.

9. Conclusion: An economy is a choice

Markets do not distribute scarcity; they price it, and the post-AGI economy will price what remains scarce: energy, compute, attention, trust, and the willingness to be accountable. That is not the end of economics; it is the beginning of a different one, and the difference is a matter of choice rather than of destiny.

Keynes, writing in 1930, predicted that within a century technological progress would give us a fifteen-hour workweek.^[7] He was wrong about the timetable and wrong about the politics, but the underlying claim, that the constraint on human flourishing is not production but distribution and meaning, has never looked stronger than it does now. The institutions of the transition, who owns the models, who is paid for attention, who is trusted with liability, who receives the dividend, will be decided by policy, law, and bargaining, and they will be decided while the transition is still underway. The post-AGI economy will be the first economy in history in which the binding constraint is not the scarcity of things but the scarcity of choices well made. The machinery will be built either way. Whether the abundance is shared is up to us.

Notes

1. McKinsey Global Institute, "The Economic Potential of Generative AI: The Next Productivity Frontier" (2023). ↩
2. David Autor, "Why Are There Still So Many Jobs? The History and Future of Workplace Automation," Journal of Economic Perspectives 29, no. 3 (2015). ↩
3. Claudia Goldin and Lawrence F. Katz, The Race between Education and Technology (Harvard University Press, 2008). ↩
4. Daron Acemoglu and Pascual Restrepo, "Robots and Jobs: Evidence from US Labor Markets," Journal of Political Economy 128, no. 6 (2020). ↩
5. International Energy Agency, Electricity 2024 (2024). ↩
6. Epoch AI, "Trends in Training Compute" (2024). ↩
7. John Maynard Keynes, "Economic Possibilities for our Grandchildren," in Essays in Persuasion (1930). ↩