Most conversations about money get stuck in the same place: treating money as a thing. A more accurate way to put it is that money is a claim on output. It is a voucher that decides who stands where in line for goods and services that actually exist. The number of vouchers can change; the pile of things they chase does not grow by a single item because of it.
The definition looks harmless. But if you keep walking in the direction it points, you pass through a series of increasingly uncomfortable conclusions. This essay is a record of walking that road to the end.
A society cannot save
I can save, of course. I can set money aside and spend it later. But humanity as a whole cannot store bread from 2060 in the present. What a society calls its savings is, at bottom, a legal claim on the output of future labor, not goods sitting in a warehouse. However a pension system is labeled on paper, funded or pay-as-you-go, at the physical level it is always pay-as-you-go: what retirees eat is always what the people still working that year produce.
Which is why the arithmetic of an aging society has no clean solution. The stock of claims keeps growing, as more people retire and hold assets, while the output that could honor those claims keeps shrinking. The claims must be written down somehow; the only question is the method. Inflation, asset repricing, taxation, or default. Historically the answer is almost always inflation, because it is the most diffuse, the slowest, and it never has to win a vote.
Money comes from a keyboard
A bank does not lend out its depositors’ money. You apply for a loan of a million, the bank approves it, types a million into your account, and books a million as an asset on its own balance sheet. That million did not exist before the keystroke, and no depositor’s balance went down because of it. The Bank of England published a paper in 2014, Money Creation in the Modern Economy, written specifically to correct the textbook story that deposits create loans. The symmetric half matters just as much: when a loan is repaid, the money is destroyed. The number moves out, the bank’s asset disappears with it. The money supply grows only when new lending outruns repayment, which is also why deleveraging is naturally deflationary. The money is not going somewhere else. It is ceasing to exist.
If money is created by lending, then how much was printed becomes a secondary question. The real question is where the loans sent the resources. The money created by a bad loan is real money, and it buys real things: cement, steel, years of labor, decades of land. Those resources are consumed for good, and what remains is a tower nobody lives in. The damage of a bad loan is not that there is more money around. It is that physical resources were steered to a place that produces no return, and money was merely the instrument that executed the steering.
The price effects come in two stages, which are easy to mix up. Stage one is asset inflation. Credit pours into real estate and pushes up house prices, not pork prices; the CPI can sit perfectly still while housing triples. The absence of inflation never proved the absence of monetary excess; it only meant the money went into the asset pool rather than the goods pool. Stage two is debt deflation. The loans stop performing, new lending dries up, repayment keeps destroying money, the net money stock starts to shrink, collateral loses value, banks tighten, and firms and households deleverage together. Japan performed this play for thirty years after 1990. It also resolves an apparent paradox: after all those years of printing, the problem now is deflation. There is no paradox. The previous two decades of inflation were hiding in house prices, and when house prices fall back, the deflation walks out into the open.
A demonstration in plain sight
America in 2008 supplies a necessary control case: a country with strong rule of law can run its credit underwriting into the ground just the same. What failed there was not law but incentives. Securitization meant the people writing the loans did not carry the default risk; package the loan, sell it on, and the losses are someone else’s problem. So the real variable was never a society of connections versus a society of laws. It is whether the people who lend bear the consequences. That condition can be broken under any system. China’s credit story over the past two decades is the same sentence, unfolded differently.
The interesting part is that China’s hole is not in household mortgages. With down payments starting at twenty or thirty percent and default rates below half a percent for years, mortgages are actually the healthiest corner of Chinese credit. The trouble lives in three other places.
Start with presales. You buy an unbuilt apartment for three million. Your down payment of nine hundred thousand goes to the developer, and the bank wires your two-point-one-million mortgage directly to the developer as well. Shortly after signing, the developer holds the full three million, while the apartment is still a hole in the ground, typically two or three years from delivery. In economic substance this is a loan: you hand three million to a company in exchange for a promise that pays off years later. It simply is not called a bond, and so none of the rules of any bond market apply to it. The loan pays no interest; the money sits with the developer for years while you pay mortgage interest over the same period. It has no collateral; the underlying asset is a set of construction drawings. And your recourse is weak: in the priority ladder, construction payables and the bank’s land mortgage both stand ahead of you, and worst of all, an abandoned project does not release you from your obligations to the bank. In the wave of mortgage strikes in 2022, many people discovered this only then: the tower would never be finished, and the payments would continue for thirty years. The first principle of credit is to place risk with the party best able to bear and price it. Presales invert that completely. The bank can read the developer’s books, has a credit department, spreads its exposure across thousands of loans, and can exit; the buyer can do none of those things, and stakes more than an entire net worth on a single exposure. Nor is the scale a figure of speech: unbuilt homes have long made up over eighty percent of new home sales in China, and deposits plus buyer mortgages have long provided around half of developers’ funding. The largest single source of financing for Chinese real estate is tens of millions of households, not the financial system. And the mechanism evolves into a Ponzi structure by necessity rather than by vice: proceeds from selling project A go to bid for land B, proceeds from selling B come back to build A, so solvency depends on new sales rather than on the projects’ own returns. That is the definition of a Ponzi structure, not a metaphor. Which is why, when sales stop, it is not a few projects that stumble but all of them that halt at once. For what it is worth, in most developed markets a presale deposit is five to ten percent of the price, held in mandatory escrow and released at delivery. Paying in full before the house exists is a Chinese particularity, not an industry norm.
Next, the local government financing vehicles. After the 1994 tax reform, local governments faced an impossible triangle: they had to spend, were forbidden to borrow, and did not collect enough revenue. The solution was to found a company. The government injects land into it, the company borrows against the land to build roads, the roads lift the value of the surrounding land, the government sells that land at the higher price, and the proceeds repay the debt. Look at the deadliest feature of this loop: the collateral is land, the revenue is land sales, and the repayment source is land sales again. Three things that are really one variable, and if land prices stop rising, all three legs break together. Legally, the government does not guarantee these companies’ debts. But everyone believes it will pay, because letting one vehicle default would instantly choke off funding for every vehicle in the province. So their bonds yield far less than their standalone credit would justify, capital flows toward political backing rather than economic return, lenders skip the diligence, and borrowers stop caring about project returns. The deepest layer of damage is that interest rates stop carrying information. The only reason a price system exists is to steer resources; once rates are bent by an implicit guarantee, the economy’s navigation system goes blind. This is not a matter of wasting some money. It is losing the ability to tell what is being wasted.
Third, shadow banking. The definition is simple: credit intermediation that happens off the balance sheets of regulated banks, and its sole reason for existing is regulatory arbitrage. The classic example is wealth management products, sold at bank branches, by bank employees, quoted with an expected yield that everyone treats as guaranteed. Economically they are deposits, legally they are not deposits, and to the regulator they do not exist, which makes the arrangement a bank with no deposit insurance and no lender of last resort. In fairness, one large function of shadow banking was precisely to fund the private firms that the formal financial system systematically shut out. It was a product of interest rate controls and credit discrimination, not their cause. Shadow banking is the symptom; financial repression is the disease. As long as rates are controlled, credit is allocated by ownership rather than by return, and local governments carry spending duties without legitimate funding channels, blocking one channel just makes the money grow a new one somewhere else. Over the past twenty years this has repeated at least three times.
Most of the damage from all of this required no one to break the law and no one to act in bad faith. It came from a large number of people making perfectly rational, perfectly compliant choices inside a badly built incentive structure. The most irreversible item is the permanent consumption of physical resources: the same point of growth now burns roughly twice the capital it did before 2008. Next is crowding out. A bidder who does not care about returns will systematically outbid one who does, so the famous difficulty private firms have in getting credit is mostly not discrimination; it is being priced out by a counterparty whose bids face no constraint. Economists studying Japan’s zombie loans in the nineties found that banks, unwilling to recognize losses, kept transfusing firms that could never repay, and that this directly depressed productivity growth across the whole economy. That is not wasting some money; that is breaking the engine of growth itself. And there is the hardest item to fix, the mismatch of horizons: officials serve terms of three to five years while the debts run for fifteen. The debt raised today is a successor’s problem, and the road built today is this year’s achievement. This requires no corruption at all, only every person rationally maximizing the metrics they are graded on.
The other side must be said too, or the account is dishonest. The financing vehicles built a great deal that genuinely pays: the high-speed rail network, the power grid, ports, urban metros, and in the 2000s these were genuinely scarce. The mismatch between duties and revenues after 1994 was real; local governments were not inventing reasons to borrow. The problem is at the margin, not in the total.
Watch the whole process end to end and it earns a name: a peaceful, legal, unvoted transfer of wealth. Financial repression holds deposit rates below inflation for decades, with hundreds of millions of household savers on the paying side. Presales move real estate’s largest single source of funding out of the financial system and onto families, assigning the most junior risk to the party with the least information, the least diversification, and no exit. Land finance has home-buying families carry a burden that is a tax in economics and not a tax in name. And the debt swaps convert the resulting hole into fiscal commitments that bind decades to come. The scale runs to tens of trillions. No step required anyone to break the law, no step was ever put to a vote, and the people bearing the cost mostly do not know what they have borne. Remember this template. It returns later.
Where growth comes from
Everything above is about allocation and distribution. But whether claims can ultimately be honored depends on output itself. Decompose output growth and part comes from adding capital, part from adding labor, and the residual that remains is called total factor productivity, TFP. It measures how much more the same people with the same machines manage to produce: pure efficiency. And only TFP can make output per person grow forever. Piling up capital runs into diminishing returns, and adding people raises the total without raising the average.
The data is not pretty. In the United States, TFP grew about 1.9 percent a year from 1920 to 1970, fell below one percent afterward, recovered briefly in the late nineties on the back of information technology, then fell back, and since 2005 has run at roughly half a percent. Every developed country shows the same shape. Why did the information revolution fail to turn it around? Three explanations persuade me most.
The first is narrow reach. What happened in the century from 1870 to 1970 was electricity, the internal combustion engine, running water and the flush toilet, fertilizer, antibiotics. Those changed how you eat, how you move, and how you avoid dying; infant mortality fell from one in five to a few per thousand, and housewives got tens of hours a week back. Information technology changed information and entertainment. However miraculous the thing in your hand, the home you live in, the food you eat, and the experience of a hospital visit are not far from 1990.
The second is Baumol’s cost disease, which I think is the most underrated item on the list. A string quartet took four people half an hour to play a piece in 1800, and it still does: zero productivity growth. But those four salaries must track the economy’s average wage, or the players change professions. The result is that sectors with stagnant productivity claim an ever-growing share of total spending. Health care, education, elder care, construction: precisely the sectors information technology cannot reach, and they soak up spending like sponges and drag the aggregate number down. The starkest data point: labor productivity in American construction has been flat or falling since 1970. Fifty years of zero progress, in the country that invented the internet and GPS. That is not a technology problem; it is a problem of permits, zoning, litigation, and how projects are organized.
The third is that ideas are getting harder to find. Research on research shows its productivity falling around five percent a year; keeping Moore’s law doubling at the same pace took more than ten times as many researchers in the 2010s as in the 1970s.
Can AI bend the curve back? The bull case is direct: if the bottleneck is a shortage of smart-person hours, and smart-person hours can now be manufactured for the first time, the curve breaks. AI is the first general-purpose technology aimed at cognition itself, and cognition is an input to every other kind of production. The bear case weighs more with me: Baumol eats everything anyway. Suppose AI makes software ten times cheaper to build; software is a few percent of GDP. Meanwhile houses still cannot be built quickly, someone still has to turn a bedridden patient, and clinical trials still take five years. If the gains stay locked in the symbolic layer, code, copy, presentations, aggregate productivity will barely move, and the binding constraint simply migrates from thinking of things to moving atoms. The test is clear enough: watch whether AI’s gains spill into the physical sectors. If in a few years the productivity numbers in construction, manufacturing, energy, and drug development start to move, the bulls are right. So far, the evidence sits mostly in the symbolic layer.
If growth stops
Several things are bets placed directly on aggregate growth. Asset prices, above all real estate and equity multiples, are priced on the assumption of an expanding customer base; commercial land in Japan’s six largest cities fell more than eighty percent from the 1991 peak while Japanese GDP per person did not collapse. Pensions and health care lose the denominator of the intergenerational transfer. Sovereign debt is measured against total GDP, not GDP per person. And there is a deeper point: ideas are non-rival, one person’s invention can be used by everyone, so more people means faster idea production. Under a shrinking population, not only does the total stagnate; the growth of income per person eventually goes to zero as well. The comforting thought that fewer people is fine because the average still rises does not survive the long run.
A long slide in productivity brings two further consequences. First, the debt dynamics flip. As long as the growth rate exceeds the interest rate, growth quietly dilutes the debt ratio and a government can roll its debt forever. Once the inequality reverses, the ratio climbs on its own and must be forced down by surpluses or by inflation. That is the hardest constraint facing developed countries over the next thirty years. Second, politics turns from positive-sum to zero-sum, which I find the most important and least discussed item. Growing societies lean tolerant, open, willing to extend rights to newcomers; stagnant societies turn to distributional fights, scapegoats, and protectionism. The reason is plain: in growth, your gain does not require my loss; in stagnation, it does. The broad turn in developed-world politics after 2008 looks to me mostly like this mechanism running, not the work of a few politicians.
And the ending is most likely not a collapse but financial repression: hold nominal rates below inflation for decades and slowly dilute the claims away. Britain and America ran this play from 1945 to 1975; that is how Britain worked its war debt down from two and a half times GDP, not through surpluses. It is not a Lehman moment. It is a slow bleed. You never see the system fall over; you just notice your savings buying a little less every year. So the answer to whether the system is sustainable comes in two parts: the monetary system itself is sustainable, and money works fine as a medium of exchange in a zero-growth world. What is not sustainable is the pile of promises built on top of it. Pensions, sovereign debt, house prices, valuation multiples: all of them are bets on growth.
Production solved, distribution not
Here the road forks, and the fork matters. The ceiling on productivity and the sustainability of monetary promises are two independent axes. Productivity can soar while the promises collapse anyway, and the logic runs clean: the output of the robots belongs to the owners of the capital, while the claims sit with the retired, and there is no automatic pipe connecting the two. Historically the pipe was called wages, and automation is dismantling exactly that pipe.
One common phrase needs correcting first. Society judges people by their productivity: not quite. What actually happens is that the market distributes income by marginal product, and our culture mistakes income for worth. The two can come apart, and through most of history they did. The idea that welds them together, that labor itself carries moral worth and idleness is shameful, is a rather recent invention. Industrialization needed masses of people willing to arrive on time and endure repetitive work, and it needed a moral story to match. That story is perhaps three hundred years old. It is not the human condition. Classical Athens held the exact inverse: not working was the precondition of a full human life, the Greek word for leisure is the root of the English word school, and a citizen’s worth came from participation in the city, not from output. It stood on slavery, which is a moral catastrophe; but as a logical structure it proves that equating human worth with productivity is not an anthropological constant. Feudal societies located worth in blood and land, decoupling income from contribution entirely, and they lasted a thousand years in remarkable stability. That stability is what should disturb you: a society in which most people do not matter to production can run steadily for a very long time. It is simply terrible for most people.
The most relevant modern model is the resource-rent state: Norway, Alaska, the Gulf, where citizens receive income by identity rather than by contribution. The divergence between Norway and the Gulf is the most important lesson I took from this whole line of thought. Norway’s sovereign fund is well governed and political power remains with the citizens. In several Gulf states, a pattern has been documented again and again: when the state does not need to tax its citizens, the citizens’ political leverage over the state withers. No taxation without representation runs in reverse just as well: when the ruler does not need your money, you do not need to be represented.
Push the thread one step further and you arrive somewhere uncomfortable. Universal suffrage and the welfare state were not the natural outcome of moral progress; to a large degree they were purchased by material conditions. The states of the late nineteenth and early twentieth centuries needed two things: mass labor that could be taxed, and mass infantry that could be conscripted. Both had to come from ordinary people, and that gave ordinary people real bargaining power. A strike could stop the economy; refusal to serve could lose a war. Historians have argued that mass-mobilization warfare was among the strongest equalizing forces in history: Britain’s male suffrage arrived nearly in step with the First World War, and women’s suffrage in many countries followed women’s mass entry into wartime factories. If automation eliminates the need for mass labor and mass soldiers at once, the material base of that bargaining power disappears. And the crux is that the process requires no malice. No conspiracy, no one deciding to discard anyone. It only requires the bargaining chips to drift quietly to zero, after which the existing arrangements get trimmed year by year under fiscal pressure, every single step backed by a perfectly reasonable technocratic argument. Which is exactly the template from earlier: peaceful, legal, unvoted, with the people bearing the cost mostly unaware of what they have borne. It needs no new machinery. It only needs the existing machinery to keep running.
Endgame
A few popular scenarios deserve a discount before anything else. Deliberate elimination is a low-probability script: it requires elites to coordinate tightly, and elites have torn at each other throughout history. Elite infighting has in fact been the main historical channel through which ordinary people gained power, because every faction needs supporters and therefore has to bid for them. The useless class story also deserves a discount: it assumes demand is fixed, while every past wave of technological displacement came with new wants nobody had imagined. In 1900, no one knew they would need programmers or therapists. But the rebuttal to that rebuttal is stronger still: in every previous shift, humans were moved to new tasks where humans still held the relative advantage, and a general-purpose cognitive technology leaves no reserved ground.
The real technical threat is not unemployment. It is the market-clearing wage falling below the cost of staying alive. Comparative advantage guarantees that people will always have something to do; even if AI is better at everything, differences in relative cost keep trade worthwhile. The theorem just never promised that the work would pay enough to live on. The horse population of the United States peaked around 1915 at over twenty million and fell to about three million by 1960. Horses never lost their comparative advantage; on certain terrain a horse beats a truck to this day. They were simply no longer bred wherever the feed cost more than the output was worth. But horses differ from people in two decisive ways: horses own no assets, and horses do not vote. People have both. So the decisive ground is not economics. It is whether people are owners of capital, and whether political power stays independent of economic necessity. The first can be engineered: sovereign wealth funds, broad share ownership, public ownership of compute, the Norwegian model rather than the Gulf model. The second has never truly been tested: we have never watched a democracy whose political power owed nothing to its citizens’ economic contribution, and so we do not know how long one lasts.
There is one more comforting argument that is weaker than it looks: if nobody has money, who buys the goods? It sounds like an automatic stabilizer and is not. Elite consumption plus capital accumulating into itself, robots building robots to build more robots, forms a perfectly self-consistent loop. Feudalism ran on exactly that loop for a thousand years: peasants in the manor economy barely touched money, while the luxury trade between nobles flourished. It does not end in economic collapse. It ends in an economy that gets smaller, with most people standing outside it.
As for the path by which all this converges, it is neither war nor plague. A population with no economic role, no political leverage, and no visible future for its children stops having children on its own. South Korea is at around 0.7 now, which means each generation shrinks by more than two thirds. This is already happening, in a country with no war, no plague, and a high standard of living. The most likely path is nothing dramatic at all: a voluntary, gradual population collapse assembled entirely out of individually rational choices. And to everyone living through it, it will not look like catastrophe. It will look like the wearying, twenty-year-old news story about young people not wanting children.
Closing
The whole essay compresses into one question: in a world that needs neither your labor nor your taxes, what still forces power to answer to you?
I can only find three candidates. Ownership: you hold assets, so your consent has a price. Legitimacy: rulers still need the ruled to believe they should rule, though history has replaced that belief with force too many times to count on it. Institutional inertia: rules keep running for a while after the material conditions that created them are gone, which is true, but it is a buffer, not a foundation. Whether there is a fourth, I do not know. What is clear is that only the first of the three can be worked on before the fact, and that the work has to be finished before the bargaining power is gone. Negotiating distribution after your leverage has vanished is sitting down at the table with nothing to bet.
Looking back, the road started from an unremarkable definition. Money is a claim on output. One step forward: a society cannot save, and every retirement promise is a claim on future labor. Another step: whether the claims get honored depends on growth, and the engine of growth is slowing. Another: even if the engine reignites, no automatic pipe connects production to distribution. And the last step: the pipe’s historical names were wages and taxes, which were also the entire source of ordinary people’s bargaining power. No link in this chain is sensational on its own. Together they point at something quite concrete: become an owner of capital while there is still time, and defend the institutions that force power to answer to ordinary people. The rest is a question our generation will live to see answered.