Where Scarcity Goes When Thought Is Cheap

Time, trust and money in an age of machines.

The people building machine intelligence increasingly say where they think it leads: to a world in which robots and software produce more than anyone can consume. Their argument is that scarcity then ends, and that prices, money and the need to economise end with it. In this article I argue that the strong form of that argument fails, and that abundance does not remove the reasons for prices and money. I do not claim that every abundant society must use them. First I ask what stays scarce when machines do most of the work. Then I ask how competing claims on those scarce things can be settled, and why exchange and money remain useful for it. Finally I turn to trust: why most money is a promise, what a base that is nobody’s promise adds, and what the people who hold one actually do with it.

1. A world without money

Iain M. Banks published the Culture books between 1987 and 2012. The Culture is a civilisation that has left scarcity behind. Its machines make whatever anyone asks for, and nobody has to work. Artificial Minds of almost unlimited ability run its ships and habitats. With today’s pace of development, that scenario seems to be approaching ever faster. It is also a system without money. The first novel calls money a clumsy way of rationing, made pointless once production outran every demand.

The books are favourites of the people now building such machines. SpaceX’s rocket-recovery ships carry the names of Culture spacecraft, a meeting room at DeepMind is named after Banks, and he has been called one of the technology industry’s favourite writers (The Economist). For at least one of them the novels are more than entertainment. He has called them his best picture of the future that intelligent machines could bring (Fortune). The forecast taken from Banks runs as follows. If machines and software agents can produce more goods and services than anyone can consume, nothing is left to economise. Prices would fall toward zero, work would become a hobby, and money would be a relic.

Two claims are folded into that forecast, and they should be kept apart. The strong one is that trade-offs disappear altogether. The weaker one is that money comes to matter much less in ordinary life. The next sections take on the first. The second is harder to settle, and the article returns to it at the end.

Take the strong claim literally. A world in which everything is given is a world in which nothing has to be given up. People could still like one thing more than another, but the ranking would never bind, because having the first would never cost them the second.

Economics starts from the opposite observation. Robbins (1932) defined the subject as the study of behaviour where ends are ranked and the means to reach them are scarce and have alternative uses. Mises opened Human Action (1949) with the same thought in its simplest form: to act is to choose, to prefer one state of affairs and set another aside. Wherever someone prefers one thing to another and cannot have both at once, something is scarce.

So the strong claim holds only if nothing people want ever excludes anything else they want. The first task is to see whether that can be so.

2. Scarcity is a relation: the land of Cockaigne

Start with what scarcity is. It is a relation between wants and means, not a property of things, so abundance moves it without ending it. Menger (1871) defined an economic good as one whose available quantity falls short of requirements. Machines can push many goods across that line. They cannot abolish the line.

The oldest picture of the other side of the line is the land of Cockaigne. In the medieval poems its rivers run with wine, its houses are roofed with cakes, and roast fowl fly into open mouths. Bruegel painted it in 1567: a clerk, a peasant and a soldier asleep on the ground beneath a laden table. It was a peasant’s dream, and what it abolished was labour.

Detail of a painting showing a clerk, a peasant and a soldier asleep on the ground beneath a round table laden with food
Pieter Bruegel the Elder, The Land of Cockaigne (Das Schlaraffenland), 1567. Alte Pinakothek, Munich.

Mises (1949) used it as a test case and found what the dream leaves standing. Even in a land of Cockaigne, where every material want is met, a person must still decide what to do first. Some satisfactions exclude each other and cannot be enjoyed in the same hour.

Life is finite and satisfactions are sequenced, so time is economised whatever else is free. The only exit Mises allows is immortality with eternal youth, and even then the ordering of incompatible pleasures remains. One could argue that a digital identity may last forever. Two things have to be kept apart, though. Lasting forever is not the same as doing everything at once: a mind that never dies still acts in sequence, and every hour it runs uses energy and hardware that could have served something else. And the wants the machines serve are still those of people, whose hours are numbered.

Finite time forces choices between now and later. Time preference is a further premise: other things equal, people rank a satisfaction now above the same satisfaction later. It is an assumption about people, not a theorem, and anticipation can cut the other way. Where it holds, present goods trade at a premium over future ones, and that premium is interest. It would exist in Cockaigne too, as soon as anyone there promised anything to anyone. Bruegel’s sleepers cannot feast while they sleep.

Machine intelligence is the nearest thing to Cockaigne yet built or anticipated, and only for one class of goods: those made of thought. Writing, analysis, design, diagnosis and proof used to need a trained person’s hours. Machines now do more and more of this work for the price of the electricity and hardware they use. That is what it means to say that thought is becoming cheap, and the question is where scarcity goes when it does.

3. Where scarcity goes

When one factor of production becomes cheap to reproduce, income moves to the factors that complement it and cannot be reproduced. Three older literatures describe the same movement from different sides.

Hirsch (1976) separated material goods, which technology can multiply, from positional goods, whose value lies in relative standing. Baumol (1967) showed that services which need a person’s hour grow dearer as everything else grows cheaper. George (1879) argued that land rent absorbs the gains of everyone else’s productivity, because land is fixed.

All three describe value retreating into what cannot be reproduced: rank, another person’s hour, a location. Rank is scarce by definition, which is why status competition survives any level of output. A house with a view is preferred to one without, however many houses are built. The argument is ordinal. It needs only that people rank outcomes, never that anyone measures them.

Böhm-Bawerk (1889) supplies the classification. Land and labour are original factors. Capital goods are produced, so their prices tend toward cost. If machines make substitutes for labour reproducible, rents shift to the factors that stay original.

The evidence predates the machines. The rise in the net capital share of advanced economies since 1950 is largely housing (Rognlie, 2015), and the long rise in house prices is largely land (Knoll, Schularick and Steger, 2017). Software became abundant, and the returns went to whoever owned its scarce complement.

So the list of what stays scarce is short. It holds time and attention, land and location, energy at a given place and hour, and rank. Two newer items join it. One is proof of origin: when text, images and voices can be synthesised without limit, evidence that a thing existed before a certain moment, or came from a certain person, becomes a good in its own right. The other is a unit whose quantity is fixed by a rule that no issuer can change on his own. Like land, more of it cannot be made. Land is limited by nature and the unit by a rule.

4. What creates value, and who decides

The abundance forecast is at bottom an argument from cost: what costs nothing to make will be worth nothing. For goods that can be reproduced, that is right. For the scarce things just listed it is not, because their value does not come from cost.

Value runs from ends to means. A machine, a mine or an hour of work has value because the thing it helps to produce is wanted. Economists call this the law of imputation. Costs do not escape it, since the cost of a factor is the value of what it could have produced elsewhere. So when the cost of making things falls, value does not vanish. It attaches to whatever is still wanted and still limited.

That does not yet show that money is needed. Scarce things can be allocated by queues, by lottery, by shared rules or by one person’s decision, and inside a household or a firm they usually are. The hard case is a different one: many independent parties, with different aims and different information, competing for the same rival resources. Even if most manufactured goods became nearly free, people and their agents would still compete for a particular location, for computing capacity, for electricity at a particular hour. Where those claims are settled through exchange, prices express what users are willing to give up, and money lets those decisions be compared and carried through time.

The Culture settles them another way. Its Minds are intelligent enough to coordinate everything directly, so prices are not needed and money withers. Four different difficulties confront that ambition. The knowledge an economy runs on is dispersed, local and often tacit (Hayek, 1945). Individual rankings cannot always be combined into one consistent social ranking; Arrow (1951) proved this under specified conditions, and how demanding those conditions are is still debated. People’s values change with the experiences they choose, so the target moves (Paul, 2014). And without exchange between owners there are no prices for capital goods, which makes it hard to compare one plan with another (Mises, 1920).

None of these settles the matter alone, and they answer different questions. The last bears hardest on machines, because the problem it names is the absence of the numbers, not the difficulty of processing them. More intelligence processes given data faster. It does not create the bids of rival owners risking their own property, and those bids are what a price is. A machine could simulate a market, but a simulated bid risks nothing, and it is the risk of loss that makes a bid informative. Giving the machine a fixed objective only moves the question to who sets it.

A world of software agents sharpens the point. Billions of agents will compete for electricity and computing capacity, both of which have other uses. Deciding which task gets the next kilowatt-hour is the coordination problem just described, and prices are the tested way of handling it. That is the reason to expect markets in such a world, and probably more prices than a human economy needs, not fewer.

5. Whose promise is money?

Prices have to be stated in something, and that something is money. Money is the most saleable good: the one thing everyone accepts because everyone else does. In barter, a trader who cannot find a direct match accepts a good that is easier to sell on than his own. The more people do this with one good, the easier that good becomes to sell, until it is wanted for that reason alone.

The abundance forecast sees only one of money’s jobs. It assumes money exists to buy produced goods, and that job does shrink as goods get cheap. But money also allocates whatever stays scarce, and it carries decisions through time. It is easy to forget that money is meant to be a store of value and not only a medium of exchange. Neither of those jobs depends on goods being dear.

The second of those jobs is the one Section 2 prepared. Present goods trade against future goods through money, so it is the medium in which time preference is expressed. A person who holds money is keeping a decision open.

Holding money through time raises the question this section is named for, because most of what people call money is a promise. A bank deposit is the bank’s promise to pay, and it passes as money for as long as nobody doubts it. Holmström (2015) calls this debt that nobody examines. A monetary system is a hierarchy of such promises, and ideally it has at its base an asset that is nobody’s liability. The promises are convenient and usually sound. They are also where trust enters, and where it can fail.

In digital form the promises go all the way down. A balance at a bank, a central bank’s digital currency and a dollar token from a private issuer are all entries that someone keeps and could change. The move to stablecoins is a move to private issuance, and it is still issuance: a claim on whoever stands behind the token. When anything digital can be copied, money is the one digital thing that must not be, and whoever is trusted to keep it scarce has discretion over other people’s savings. That is a narrower power than a planner’s. It raises the same question of whom one has to trust.

None of this shows that a base outside any issuer’s discretion is necessary. It shows why some people want one. Savings are decisions carried through time, and they are safest in something whose supply nobody can change on his own. Gold was such a base. It is scarce by nature and nobody’s promise, but it is heavy and hard to test, so people held claims on it instead. A digital base has to get its scarcity from a rule, and the difficulty is to have a rule without a ruler.

The construction that solved it can be stated briefly. Transactions are put in a single public order by chaining each batch of records to the one before, so that changing an old entry would mean changing every later one (Haber and Stornetta, 1991). Adding a batch requires a computation that is costly to perform and trivial to check, so rewriting the past would mean redoing that work faster than everyone else combined. New units are issued with each batch on a schedule fixed in advance and capped at twenty-one million, and anyone can verify the whole record for himself. A holder who keeps his own keys then has no counterparty.

The work is paid for in energy, and this is where the earlier point about value and cost becomes concrete. New units go to whoever wins the computation, so competitors enter until the expected cost of competing equals the expected reward. In a stylised equilibrium, leaving out transaction fees and holding everything else fixed, annual energy use is:

E* = φ × N × R × P / c    (kWh per year)

Here N is the number of awards per year, R the units per award, P the price of a unit, c the price of electricity and φ the share of power in marginal cost. Machine efficiency does not appear, and energy use moves with revenue. The energy spent does not give the unit its value. The value of the unit decides how much energy is spent. Calling the unit “stored energy” gets this backwards. It is a cost theory of value with joules in place of labour hours, in the line of Ricardo, Marx and Henry Ford’s energy dollar of 1921. It is not Menger’s.

What the energy does buy is specific. It sets the cost of rewriting the record, it makes new units costly to obtain, and it shows that real resources were committed to this history and no other. It does not govern supply. Ford’s unit was a claim on energy, so abundant energy would have meant abundant money. Here cheaper energy raises the work done per batch and leaves the schedule alone.

“Trustless” needs a careful reading. The design does not remove assumptions. It moves trust from persons and institutions to three other things: that certain mathematical problems are hard, that no single party controls most of the work, and that users keep running the rule. A rule announced in advance is not a command. But rules age, and changing one is a collective decision that no code can make.

6. Scarce is not yet valuable

Building a scarce base does not make it valuable. Scarcity is necessary for value and not sufficient. Menger required four things of a good: a human need, a causal link to satisfying it, knowledge of that link, and command over the thing. Holding the keys gives command. The other three depend on people wanting to hold the unit, and a cap on supply cannot provide that.

The pricing rule is simple. When supply is fixed, the price settles where the existing stock equals the total demand to hold it (Rothbard, 1962). The supply curve of a rule-fixed unit is vertical by construction, so its price is demand and nothing else.

Demand for money looks circular: people hold it because it has purchasing power, and it has purchasing power because people hold it. Mises (1912) broke the circle with history. Today’s demand rests on yesterday’s purchasing power, and the chain runs back to a commodity once valued for its use. This is the regression theorem. Modern monetary theory meets the same circle and finds many possible outcomes, including a price of zero (Garratt and Wallace, 2018; Biais et al., 2023). Read side by side, both say that nothing in the asset fixes its price and that history selects among the possibilities. That is convention settling on a money, not the market finding the best one.

The new unit tested the strict version of the theorem, since it was never a commodity. It acquired a price anyway. Davidson and Block (2015) reply that the commodity requirement binds only when money first emerges from barter. Luther (2018) concludes that the theorem is narrower than claimed. One detail supports a middle reading: the first published exchange rate, in October 2009, was computed from the cost of electricity. Cost gave the first trade a focal point, and demand took over afterwards.

The consequence for this article is simple. A rule can secure the supply. It says nothing about demand, so theory provides no floor under the price and no ceiling over it.

Hayek predicted in 1976 that if people could choose their money, they would choose stable purchasing power. The market has run the experiment. Dollar-pegged tokens stood at roughly $285–290 billion in mid-November 2026 (DefiLlama). They exist to be spent, not held for gain. So the demand for a stable private medium was met by a tokenised dollar, and the fixed-supply unit kept the demand for savings. Its design is a fixed stock indifferent to the price level, not Hayek’s managed private money. And each token is a promise by an issuer to redeem, which returns the argument to trust.

7. Trust returns with every promise

Only the base asset held with one’s own keys has no counterparty. Everything built on it is a claim, enforced by law, and the market price of each claim is a running measure of how far it is trusted. A share in a company that holds the asset, a bond issued against it, a loan secured by it, a fund unit and a balance at a custodian are all promises of this kind, and each rests on a court and a balance sheet. Trust in the base is bought with time, by waiting for settlement to become final. Trust in a claim is bought with yield.

That is why law came back. Dai (1998) hoped that contracts among pseudonyms could be enforced by posted bonds and reputation. Reputation works inside small closed groups, and impersonal exchange needs a third party who can compel (Greif, 2006). Code settles. It does not adjudicate.

So what do status-seeking, credit-using people do with a base that cannot be inflated? They do three things, and together they show that a scarce base removes neither credit nor state money nor rivalry. What it changes is who bears the losses.

They lend against it. The demand for credit comes from differences in patience, not from the monetary base, and a technology that permits full reserves does not make anyone choose them. Within fifteen years a lending market had formed on the new base, and in 2022 it had its first crisis. Several large lenders had taken customers’ holdings into custody and lent them on, and when the loans went bad they could not give the holdings back. What happened next is the instructive part. Nobody could create new units to rescue them. The lenders went bankrupt, their creditors took the losses, and those who had kept custody of their own units still had them. In that episode a hard base did not prevent a credit cycle. It changed who paid for it.

They keep their accounts in something else. Credit on the new base is secured by it and denominated in state money. Few borrowers will promise to repay an asset they expect to appreciate, and people quote prices in the unit their wages and debts are already in (Doepke and Schneider, 2017). The base therefore works as savings and as collateral inside the existing price system. It is not a unit of account. That sets the limit of what it protects. It shelters savings from whoever manages the money, which today means the central bank. It does not shelter the price system.

They compete with it. The abundance forecast is not new. Keynes made it in 1930, expecting the economic problem to be solved within a century and the working week to fall to fifteen hours. He was roughly right about productive capacity and wrong about the hours. As material goods got cheaper, people spent a larger share of income bidding for what cannot be multiplied: the flat near the good school, the seat, the view. A scarce asset is natural collateral for that contest, because it lets the owner bid without selling.

New credit reaches some hands before others, a point as old as Cantillon (1755), and in collateralised lending those hands belong to the owners of collateral. That holds whether the credit comes from a central bank or from a private lender. Borrowing against a trustless base does not abolish the channel. It adds the holders of the base to it.

8. Machines, work and who owns the capital

The remaining question is who ends up owning the scarce things. If machines make labour reproducible, income shifts to whoever owns what machines cannot make.

For a task a machine can do, competition pushes the price toward the cost of the electricity and hardware time it uses, and that cost becomes a ceiling on what a person can charge for the same task (Hanson, 2016; Korinek and Suh, 2024). This is not the end of work. Comparative advantage still holds: even if machines are better at everything, people keep a role while machine capacity is scarce and has better uses, and it is scarce while energy and computing power are. The tasks that stay outside the equation are those where the buyer is paying for a person, such as care, teaching, craft, performance and accountability (Baumol, 1967). But nothing in the theory keeps any particular wage above any particular level, or makes the adjustment quick. British output per worker rose for about forty years from 1800 while real wages stayed nearly flat (Allen, 2009). Forty years is a working life.

The abundance forecast skips the question of who owns the machines, and it is the old question of who owns the means of production. Money stops mattering to consumers only if the output reaches them. It can reach them through shareholding, which rewards those who saved and bore the risk and is defensible only while entry stays open. It can reach them through redistribution by the state. A cash transfer leaves people free to choose in markets, so it is not planning, but its size and terms are set by whoever controls or taxes the machines. Or it can follow collateral, so that whoever can borrow against scarce assets acquires the machines. The politics of the coming decades will be fought between these answers, and a bearer asset outside every issuer’s control is the hedge for those who doubt whichever planner wins.

9. Where this leaves us

This week supplied a small example. On 20 November 2026 Levent Alpöge, a mathematician at Anthropic, released a proof of the Riemann hypothesis, open since 1859, that he had found working with an AI model. It runs to about 17 million lines of formal code. Within a day the Lean and Mathlib communities had certified it with independent checking programs (joint statement), and no person has yet read all of it (Al Jazeera). The proof was costly to produce and cheap to check, so its correctness could be checked without taking its discoverer’s word for it. Understanding it is another matter, and that is rationed by human hours.

What follows may matter more than the proof. The next results will be built by machines on what machines have proved, checked by machines, and handed to us afterwards. The pace is then no longer set by how fast people can read.

A piece of speculative fiction that circulated online in the summer imagined that sequence and guessed this week for its first step. Its last item is not a result at all but a song. The image deserves a moment, as speculation and no more. Music, stories and advertising already shape what people want. A system that knows what moves people better than they know it themselves could do so with far more precision, and what it would compete for is the scarcest thing in this article, the hour. Unlike the proof, nothing in such a work could be checked. The old defence is a commitment made in advance. Odysseus had himself tied to the mast before the Sirens sang (Elster, 1979). A monetary rule that no issuer can alter is one small commitment of that kind. It guards savings against someone else’s discretion. It does not guard anyone’s attention or judgment.

Elon Musk, who often cites the Culture novels, said in 2026 that money will not matter by 2036 (Fortune). His reasoning is that people want money for goods and services, and that robots will supply more of both than anyone can use. He also grants that electricity and mass will stay constrained (Business Today).

One part of that forecast this article has argued against. Trade-offs do not disappear, because time, location, energy and rank stay limited. The other part, that money will matter much less in daily life, is open. It depends on choices still to be made: whether competing claims on what stays scarce are settled by exchange or by an authority, what money prices are stated in and whose promise it rests on, and how widely ownership of the machines is spread.

Whatever the answers, the day will still have twenty-four hours, and each of us will still be deciding what to do first.

References

The linked web sources were opened on 21 November 2026. The works below are cited from the standard literature; editions and page numbers should be checked before formal citation.