The blog · Scarcity mechanics
The Estate Sale That Never Comes: Money, Resources, and Property When Nobody Dies and the Robots Do the Work
A thought experiment · two assumptions removed · what is left standing
Every price, pension, and property deed quietly assumes that owners die and that people work. Take both away and money still matters. What changes is who ends up holding it.
Sources and arithmetic in Exhibits B, C, and G
In 1726 Jonathan Swift sent Gulliver to the island of Luggnagg, where a few children in every generation are born unable to die. Gulliver is delighted. He tells his hosts what he would do with forever, and the first item on his list is money: by thrift and good management he expects to be the richest man in the kingdom in about two hundred years. Wisdom can come afterward.
His hosts find this funny. Their immortals, the Struldbrugs, go on aging, and the kingdom has had a long time to work out what to do about them. At eighty each one is, in Swift's phrase, "dead in law." The heirs take the estate, a small allowance is set aside, and the immortal may never again buy land or take a lease. Swift gives the reason without embarrassment. The old hold on to what they have, people who never die never let go, and left alone the immortals would end up owning the whole country and running it badly.
He meant it as satire. Read as economics, it is the cleanest statement we know of a problem nobody has yet had to solve, and it was written three centuries early. We came back to it this autumn for a practical reason. Two things that used to belong to fiction now have budgets. One is machines that can do most kinds of work. The other, further off and much less certain, is medicine that slows aging itself. We do not know when either arrives in full, or whether the second arrives at all. But a newsletter named for compounding ought to be able to say what happens to compounding when the people doing it stop dying, and what happens to the businesses we own when the people who work in them are no longer needed.
So this piece is a thought experiment, run the way we would run one on a business: change two assumptions and see what breaks. The assumptions are that owners die and that most people have to work. Nearly every price, pension, tax, and property deed leans on one or the other. We take both away and follow the consequences through money, then resources, then property rights. After that we bring it home to three things of our own: the debt plan we published in September, and the two model portfolios with the most riding on the answer.
A warning before we start. Nothing below is a forecast, and none of it has a date attached. Where we cite a number, it describes the world as it stood in November 2026. Where we do arithmetic, we list the inputs so you can redo it. And where we choose between two defensible answers, which we do in one section, we say that we are choosing and give the other side its strongest form.
| The argument on one screen | Our short answer | Where |
|---|---|---|
| Does money still matter? | Yes. It stops being something most people earn and becomes something they own or are sent | 03–05 |
| What gets cheap, and what gets dear? | Anything a machine can make gets cheap. Land, energy, legal rights, and rank absorb the wealth | 04, 09 |
| Can everyone compound forever? | No. Returns are a share of output, and given enough time ruin beats return | 07 |
| What happens to property rights? | They survive, with a clock or a meter attached | 10–13 |
| Which settlement would we choose? | Broad ownership through a permanent fund, paid for by a small yearly charge on holding | 14 |
| What does it do to our debt plan? | The consumption tax survives. The taxes on paychecks and on dying do not | 15 |
| And the portfolios? | Own the bottlenecks, not the labor. Spend income the way an endowment would | 16–17 |
Two assumptions nobody wrote down
The first assumption is that owners die. Every asset on earth is repriced, divided, taxed, or sold about once in a human lifetime. Nobody designed this and nobody voted for it, but it does a great deal of work. A fortune assembled by one person is split among several heirs, and some of them are bad with money. The tax collector takes a share. Land that has sat in one pair of hands for fifty years comes to market. The estate sale is the economy's reset button, and by the United Nations' count it is pressed about 62 million times a year.
Once you look for the assumption, you find it built into institutions that seem to have nothing to do with death.
| Institution | How it leans on death | If nobody dies |
|---|---|---|
| Inheritance | A fortune is divided among heirs every generation | One owner, undivided, indefinitely |
| Estate tax | 40% federal tax above a $15 million exemption in 2026, collected at death | Never collected |
| Capital gains | Heirs take assets at market value, so the gain built up over a lifetime is never taxed | "Buy, borrow, die" loses its last step, and the gain is deferred without end |
| Retirement income | Pensions, annuities, and Social Security pay for a remaining life that actuaries can estimate | A payment for life becomes a payment forever, which nobody can fund |
| Copyright | Lasts for the author's life plus 70 years | Never expires |
| Trusts | The rule against perpetuities sets its limit in lives plus 21 years | The clock never starts |
| Offices | Federal judges serve for life, and so in practice do many founders and chairmen | A seat held for centuries |
The second assumption is that people work. Wages are the main pipe through which income reaches households, and that pipe has been narrowing for a long time. Labor's share of output in the U.S. nonfarm business sector was 52.8 percent in the second quarter of 2026, the lowest reading in a series that begins in 1947. For most of the postwar period it sat between 60 and 66 percent.
We make no claim about how much of that decline is artificial intelligence. Globalization, the growing weight of a few very profitable firms, and the way statisticians treat housing all have their advocates, and the slide began long before a chatbot could write a memo. The reason to show the chart is narrower. The direction was set before the machines arrived, and the thought experiment simply runs the same trend much further.
The rules, and how long forever is
A thought experiment needs rules, or it turns into a wish list. Ours has three.
- Aging is curable; accidents are not. "Forever" means you stop getting frailer, not that you become indestructible. Swift's version, aging without death, is a nightmare nobody is trying to build.
- Machines can do almost any task for less than a person can. People may still work, and some will want to. Few need to be hired.
- People and their governments still own and run things. This is the largest assumption of the three, and we come back to it in Who owns the robots.
Neither change would arrive overnight, and most of what follows describes the destination, not the road. Where the road matters, as it does for resources and for lenders, we say so.
The first rule raises an obvious question: if a body stopped aging, how long would it last? The Social Security Administration's period life table gives a starting point. It puts a 25-year-old's chance of dying in a given year at about 0.16 percent for men and 0.06 percent for women. Most of that is accident, violence, and misfortune, not age. Hold the risk at that level forever and the arithmetic is simple: the average lifespan is one divided by the yearly risk, which comes to roughly 900 years.
At a 25-year-old's risk, half of any group is gone within about 630 years, and a third is still alive after a thousand. People with nine centuries to lose would presumably drive more carefully than 25-year-olds do, so the true figure could be higher. It could not be infinite. The estate sale still comes. It comes about every nine centuries instead of every eight decades.
It helps to put a face on that interval. An estate settled this year, on a nine-century cycle, would last have changed hands when Henry I sat on the English throne. For compounding, that changes everything, and a later section works through the numbers. For the law, which was built around turnover every generation, nine centuries is the same as never.
Does money still matter?
The most common reaction to this premise is that money would stop mattering. If machines make everything and nobody has to work, what is there left to buy, and what is a dollar for? It was the first thing we asked ourselves, and it deserves a direct answer before anything else, because the rest of the piece depends on it.
Money is a claim on things that are scarce. It prices them, it pays for them, and it stores the right to buy them later. The thought experiment does not end scarcity. It moves it. A machine can make a second car as easily as the first, but it cannot make a second lot on the same block, and it cannot make the hours in a day more numerous. As long as two people can want the same thing at the same time, something has to decide who gets it, and for all its faults money is the least bad decider anyone has found. There are only two ways it could stop mattering. One is that nothing is scarce, which is not this world. The other is that people stop trusting it, which is a failure of government and has nothing to do with abundance.
What does change is what money is for. John Maynard Keynes saw this in 1930, in an essay that predicted his grandchildren would work fifteen hours a week. He divided human wants in two: absolute needs, which can be satisfied, and relative ones, which exist only by comparison with other people and can never be. He was right about the wealth and wrong about the workweek, and the second category is why. In a world of machine labor the first category gets very cheap. Food, clothing, devices, and most services stop separating a modest income from a large one. The difference between the two becomes position: where you live, what is yours alone, and how much say you have.
So our answer is that money matters less for living and more for ranking. That is not a small change. It means the things worth owning are the things that confer position or collect a toll, and it means a great deal of what we now call the cost of living turns out to have been the cost of labor.
One more point, which matters for the portfolio sections later. Shares are not money. A share is a claim on what a business produces, whatever currency it is counted in. If dollars ever did lose their importance, ownership of the things that produce would gain it. The people who should worry about money mattering less are the ones who hold only money.
What gets cheap and what gets dear
When labor costs almost nothing, the price of anything a machine can make falls toward the cost of what goes into it: energy, raw material, land, and whatever the owner of the design is able to charge. Anything that cannot be copied moves the other way, at least in relative terms. Economists call the second effect Baumol's cost disease, after William Baumol's observation that a string quartet takes the same four players, and about the same time on stage, as it did in 1800. When productivity soars in one part of the economy, the parts it cannot reach get more expensive by comparison.
| What | Direction | Why |
|---|---|---|
| Factory goods, routine services, software | Down, toward the cost of energy and materials | Labor was most of the cost |
| A house, meaning the structure | Down | Machines build it |
| The lot under the house | Up | Nobody is making more of it |
| Energy and raw materials | The new floor under every price | Robots need them too |
| Anything valued because a person did it | Up, relative to everything else | It cannot be automated by definition |
| Rank and exclusivity | Up | There is one front row and one original |
The first row is already happening, to the product doing the automating. Epoch AI, a research group that tracks the field, finds that the price of reaching a fixed level of performance from a language model has been falling by a factor of somewhere between 9 and 900 a year, depending on the task. We know of no other product whose price has fallen that fast. It is what "toward the cost of the inputs" looks like while it is happening, and it will come up again when we ask who keeps the money.
The house is the row to remember. A home is two assets sold as one. The structure is a manufactured good that happens to be assembled on site, and in this world it gets cheap. The lot is a position, and it does not. Anyone who has priced a teardown in a good school district already knows which of the two the money is paying for.
How money reaches people
Today a little over half of what American business produces is paid out as compensation. If machines do the work, that half goes to whoever owns the machines. Everything in the second half of this piece follows from that sentence, so it is worth being careful about how sure of it anyone can be.
Economists disagree about whether wages collapse or merely shrink as a share. Anton Korinek and Donghyun Suh model the question as a race between automation and the build-up of capital. If automation moves slowly enough, there is always more for people to do and wages can keep rising. If every task can be automated, wages collapse. They also find that wages can fall well before the end, if automation runs ahead of the investment needed to put displaced workers to use. Our second rule puts us in their full-automation case. Readers who think that rule is too strong should read the rest as a description of the direction, not the destination.
Some paid work survives even on our premise. A buyer may want a person, as with the string quartet. The law may require one, as with a judge or a signature on an audit. And a great many jobs exist because someone has to be answerable when things go wrong, which a machine cannot yet be. We would not want to guess how many jobs that is. We are confident it is not half of national income.
So household income has to come from somewhere else, and there are only two places. One is ownership: dividends, rent, and interest. The other is transfers: a check from the state, or from family. Which of the two carries the load is the central political question of that world. We take it up in Who owns the robots and answer it for ourselves in the section after.
What money earns
Two forces pull on interest rates in this world, and they pull in opposite directions.
Growth pulls rates up. If machines can build machines, the economy could grow far faster than it ever has. People who expect to be much richer later save less today, so borrowers have to pay more to attract what saving there is. Trevor Chow, Basil Halperin, and Zachary Mazlish ran the textbook formula and found that going from 2 percent growth to 30 percent would lift the real interest rate from 3 percent to 31. Critics have argued the effect could be weaker, or even run the other way, so read the figure as an illustration of the pull and not as a forecast.
Long lives pull rates down. A person planning for nine centuries has every reason to save more, and nobody ever spends down a nest egg because nobody ever reaches the end. More savings chasing the same opportunities means lower returns.
We do not know which force wins, and we distrust anyone who says they do. What we can say is that over an endless horizon the rate does almost all the work. Every asset held forever is a perpetuity, and a perpetuity has the simplest valuation in finance:
A dollar a year forever is worth $25 at 4 percent, $33 at 3 percent, and $50 at 2 percent. A single point on the rate moved the price by a third, and then by half. Owners who never die, holding assets that never mature, should expect small changes in rates to produce very large changes in prices. That is true whichever way rates end up going.
Compounding forever, and the three things that stop it
This newsletter is named for compounding, so allow us a moment. Warren Buffett had fun with the idea in his partnership letters of the 1960s. Had Queen Isabella skipped Columbus and put her roughly $30,000 out at 4 percent, he reckoned, Spain would have had about $2 trillion by 1962. His moral was that such sums reward one of two things: a very long life or a decent rate of return. Our thought experiment hands the first one to everybody, which is Gulliver's plan exactly.
| Years | At 3% | At 5% | At 7% |
|---|---|---|---|
| 50 | $44,000 | $115,000 | $295,000 |
| 100 | $192,000 | $1.3 million | $8.7 million |
| 200 | $3.7 million | $173 million | $7.5 billion |
| 500 | $26 billion | $393 trillion | $4.9 quintillion |
The first time we filled in that table we checked the bottom row twice, because it looks like an error. It is not. It is the tell. Those are sums no economy could pay to one saver, let alone to every saver at once, and Gulliver's plan fails for the same reason a chain letter does. Three things stop it.
First, returns are a share of output, not a law of nature. A return is somebody else's payment: a customer's, a tenant's, a borrower's. If everyone's claims grow at 5 percent a year and the economy grows at 3, the claims outrun the things being claimed, and then something gives. Returns fall, or prices fall, or the claims are cut by default, inflation, or tax. In total, wealth cannot outgrow the economy forever unless the owners spend the difference. An individual can beat that arithmetic for a long time. Everyone cannot.
Second, reality intrudes. Benjamin Franklin ran the experiment for us. In 1790 he left £1,000 each to Boston and Philadelphia, to be lent to young tradesmen and left to grow for two hundred years. He projected £131,000 per city after the first century, which works out to 5 percent a year, and £4,061,000 after the second.
In 1990 Boston's fund held almost $5 million and Philadelphia's about $2 million. Those sound like large numbers until you set them beside the plan.
Boston, the better steward of the two, ended with about a quarter of what Franklin had penciled in. Philadelphia ended with about a ninth. Nothing dramatic happened to either fund. Borrowers defaulted, cash sat idle, politicians took an interest, and fees were paid, year after year for two hundred years. That is what happens to a careful man's 5 percent.
Third, ruin. Over an ordinary lifetime an investor can live with a small yearly chance of disaster. Over nine centuries nobody can. The arithmetic is the same as the mortality arithmetic in Exhibit C, applied to a fortune instead of a body.
A 1 percent yearly risk, which most investors would shrug at, leaves four fortunes in every hundred thousand standing after a millennium. Even one-tenth of that risk loses nearly two fortunes in three. For the investor who never dies, the first question about an asset is not what it earns. It is what can kill it: debt that forces a sale, a single bet that can end the game, fraud, confiscation. That has always been the first rule. Given enough time it becomes the only one.
What breaks outright
Some institutions bend in this world. A few simply stop working, and it is worth naming them, because each is something readers hold or will depend on.
Retirement. A pension that starts at 67 and runs for eight more centuries is not a pension. The whole idea rests on a working life followed by a shorter one of rest. Careers would turn into a series of chapters with long breaks between them, each break paid for by what a person owns.
Annuities and life insurance. Both are bets on a mortality table. Flatten the table and a life annuity is a perpetuity, priced by the formula in the last section, while life insurance becomes accident insurance.
Pay-as-you-go programs. Social Security taxes wages to pay people who have aged out of work. With few wages to tax and nobody aging out, it has to be rebuilt on some other base. That one is our own problem, since our debt plan leans on it, and we face it in its own section.
Debt, as a contract. A thirty-year mortgage is sized to a working life. Lenders facing borrowers with nine centuries ahead of them would write far longer loans against land, and the price of land would rise to meet the credit. Unsecured debt changes more. Today a debt nobody can collect dies with the debtor. Without death, the only release is bankruptcy, and a society of immortals would need to decide how often a person may start over. The oldest answer on record is a calendar: the law of ancient Israel released debts every seventh year and returned land to its original families every fiftieth.
Inheritance, as a plan. Charles III became king at 73, after the longest wait of any heir to the British throne. That is what being an heir looks like when the previous holder lives a long time. Stretch it tenfold and inheritance stops being something anyone can plan around. Parents who want to help their children would have to give while living, which most tax codes treat very differently from giving at death.
Swift, characteristically, thought of one more. In Luggnagg a marriage between two immortals is dissolved by law once the younger turns eighty, as a mercy. We leave that one to the reader.
Resources: abundance has a bottleneck
A world of machine labor is rich in everything that can be manufactured and poor in everything that cannot. Wealth collects at the bottlenecks, and it helps to list them.
- Land and location. More floor space, yes. More coastline or more Manhattan, no.
- Energy. Every robot and every data center is a new mouth for electricity. Supply can grow, but it grows on a lag, and the lag is where the money is made.
- Raw materials. Copper, lithium, and the rest are atoms. Machines can dig faster. They cannot dig what is not there.
- Rights granted by rule. Permits, rights-of-way, radio spectrum, water rights, and orbital slots are scarce because someone decided they would be.
- Attention and rank. Each person still has twenty-four hours a day, and some things are valuable only because other people cannot have them.
Korinek and Suh make the same point inside their model. Scarce inputs that cannot be reproduced can choke growth and make the outcome for wages worse, because the gains from automation flow to whoever holds the thing in short supply.
Land has been rehearsing this for seventy years. We went looking for evidence that a fixed factor really does soak up the gains from everyone else's productivity, and the best we found is a study of house prices in fourteen rich countries since 1870, by Katharina Knoll, Moritz Schularick, and Thomas Steger. Real house prices were roughly flat from the nineteenth century to the middle of the twentieth and then climbed steeply. When the authors split the price into the structure and the land beneath it, more than 80 percent of the rise since 1950 turned out to be land. Building got more efficient. The lot took the winnings. Make building nearly free and we see no reason for that to reverse.
Energy is the bottleneck of the road, if not of the destination. This is the one place where the transition matters more than the end state. The machines have to be built before they can do anything, and building them is the largest capital project under way anywhere. Meta alone has guided to $115 to $135 billion of capital spending for 2026. Its Hyperion campus in rural Louisiana is designed to draw as much as five gigawatts, which is the output of several large power stations, for one customer on one site. Chips can be ordered. Substations, transformers, and transmission lines take years and permits. For as long as that gap lasts, the people who can deliver electricity to a particular piece of ground are collecting a toll.
More people, same planet. The world recorded about 132 million births and 62 million deaths in 2024, which nets to 70 million more people. Take deaths to zero and the same births add 132 million a year, nearly twice today's pace. Falling birth rates soften this: the world average is 2.3 births per woman, and more than half of all countries are already below the 2.1 needed to hold a population steady. The direction still holds. More people, and the same amount of shoreline.
The auction that never ends. Put a fixed supply together with bidders who never die. The best lot on the lake is bought once and does not come back to market for nine centuries. Markets for things that cannot be replaced get thinner, with fewer sales at higher prices. We would expect far more renting of the scarce things, because owners never need to sell and everyone else cannot save fast enough to buy. A society can live that way. Most of them did for most of history, and the word for it is feudalism.
A stress test for toll booths. Our own method is to look for businesses that sit at a chokepoint, something everyone must pass through and nobody can easily copy. The thought experiment is a hard test of that idea, and we think the idea mostly survives. When labor is abundant, every edge built on labor fades: the skilled workforce, the service culture, the sales force. What remains are edges built on things that cannot be copied, such as a location, a legal right, a physical network, or a deposit in the ground.
The part that does not survive is permanence. Technology breaks bottlenecks as well as making them. Energy is scarce today and may not be after a few more decades of engineering. A bottleneck is worth owning only for as long as it is one, and nine centuries is a long time to be right about that.
Property: the dead hand was the easy problem
Property law has spent seven centuries fighting owners who are gone but still in control. Lawyers have a name for that grip, the dead hand, and a long shelf of rules for prying it loose. An owner who never leaves is a harder case, because most of those rules are keyed to a death that no longer arrives.
Death is the reset nobody legislated. Every generation, fortunes are divided among heirs, taxed, mismanaged, or sold. Even so, wealth is stickier than most people assume. Guglielmo Barone and Sauro Mocetti of the Bank of Italy matched Florence's tax rolls of 1427 against those of 2011 and found the surnames at the top six centuries ago still near the top. Descending from a family at the 90th percentile in 1427, instead of the 10th, was worth about 5 percent more income in 2011.
Five percent does not sound like much until you remember what lies between the two dates: twenty generations of deaths and divisions, the fall of the Medici, plague, two world wars, and the invention of the income tax. The advantage survived all of it. Remove the deaths and nothing shuffles the deck at all.
We have met owners who never die. The medieval Church was one. In England a lord collected his dues when land changed hands at death. Land given to the Church never died, never came of age, and never lacked an heir, so the dues stopped. The lawyers called it mortmain, which is where the dead hand gets its name.
Edward I answered in 1279 with the Statute of Mortmain, which barred such gifts on pain of forfeiture unless the king agreed. Lawyers found ways around it for the next two hundred and fifty years. Henry VIII ended the argument in the 1530s by dissolving the monasteries and taking the land. The lesson is blunt. When owners stop dying, the state loses a tax event, and states do not take that quietly. The first response is a rule. The last is confiscation.
The will that compounded. Peter Thellusson, a London merchant, died in 1797 with more than £600,000. His will ordered the income to pile up through the lives of every descendant then living and to go only to those alive when the last of them died. Estimates of the eventual sum ran to £14 million or more, an enormous sum for the time. Parliament was alarmed enough to pass the Accumulations Act of 1800, which capped how long income could be left to pile up.
The ending is better than the law. When the estate was finally divided in the 1850s, the lawyers had eaten the growth, and the heirs received little more than the original sum. Franklin's fund and Thellusson's will were started within a decade of each other, and they teach the same thing from opposite directions.
The rule against perpetuities. The common law limits control from the grave with a deadline measured in lives plus twenty-one years. Notice the yardstick. If lives do not end, the rule never bites.
American states have been dropping the rule anyway, to attract trust business. South Dakota abolished it in 1983, and by the end of 2005 twenty-one states allowed trusts that last forever. The law has been moving toward perpetual ownership well before any owner became perpetual.
Jefferson's nineteen years. In 1789 Thomas Jefferson wrote to James Madison that "the earth belongs in usufruct to the living" and that the dead have no rights over it. From mortality tables he worked out that a generation turns over in about nineteen years, and he argued that no debt and no law should bind for longer. Madison replied, politely, that the idea would not survive contact with practice.
What strikes us, reading the exchange now, is that both men took generations for granted. Their disagreement was about how much the departed may bind the arriving. With immortals nobody departs. The generation that wrote the rules is still in the room, and it still holds the deeds.
Seats, and the other grim reaper
The same problem applies to positions as to property. Pierre Azoulay and two co-authors tracked 452 eminent life scientists who died early. After a star died, papers from that star's collaborators fell by about 40 percent. Papers from outsiders rose 8.6 percent, and the outsiders' work was unusually likely to be highly cited. The authors titled the paper after a remark of Max Planck's, that science advances one funeral at a time, and their data suggest he was not joking.
Fields renew when incumbents leave. Now picture board seats, tenured chairs, founders' voting shares, and Senate seats held for centuries. We would expect term limits on seats as well as on assets, and we would expect the fights over them to be worse, because a seat cannot be divided.
There is a more hopeful observation to set beside that one. We already live among immortal owners, and they die all the time. A corporation has no natural lifespan. Yet the average company's stay in the S&P 500 fell from 33 years in 1965 to 20 in 1990, and the consultancy that tracks it expected 14 by this year. Nothing in the law removes these firms. Competitors do.
Competition is the other grim reaper. It does to companies what mortality does to people, and it would go on working on immortal people's fortunes for as long as a newcomer can still take a customer from an incumbent. Whether machine intelligence makes that easier or harder is, to us, the most important open question in this whole exercise. If anyone can rent the same intelligence by the hour, entry gets cheaper and fortunes churn faster. If the advantage lies in scale, in owning the compute, the data, and the land under both, entry gets harder and the Florentine result gets worse. That is why we would count antitrust and open entry as part of the property regime in this world, not as a separate subject.
The menu
Societies already use a handful of tools on owners who do not die: on corporations, churches, trusts, and governments themselves. None of them is new, which we find reassuring. The question is which ones scale.
| Tool | Where it exists now | What it does | The catch |
|---|---|---|---|
| Ownership with an end date | Singapore has sold residential land on leases of no more than 99 years since 1967; China grants residential land-use rights for up to 70 | Land returns to the pool on a schedule | Value decays as the end nears, and holders press for free renewal |
| A yearly charge for holding | Property tax; the land value tax Henry George proposed in 1879 | Makes idle ownership costly, and needs no death to collect | Needs honest valuations; hard on owners with assets and no cash |
| A tax on income from capital | Corporate and capital-gains taxes everywhere | Takes a share of the yield and leaves the title alone | Capital moves, so high rates need many countries to agree |
| A cap on accumulation | The Accumulations Act; the rule against perpetuities | Stops fortunes from compounding untouched | Lawyers, and states that compete to repeal it |
| Legal death at a set age | Luggnagg | Forces the estate sale | We would rather not |
| A citizen's dividend | The Alaska Permanent Fund paid each resident $1,200 in 2026, a $1,000 dividend plus $200 of energy relief | Gives everyone an owner's income from a shared asset | Politicians set the amount each year |
Who owns the robots
When machines earn the income, the map of who owns the machines becomes the map of who has income. That is the sentence we flagged in section five, and this is where it leads.
The case for a runaway. The sharpest version of the worry is Capital in the 22nd Century, an essay published last December by the economist Philip Trammell and the podcaster Dwarkesh Patel. Their argument starts with a brake that has always worked. Capital needs workers. When the rich save more, capital piles up faster than the workforce, so each new dollar of it earns a little less and each worker a little more. Inequality that comes from saving has tended to correct itself.
Robots remove the brake. If machines can replace workers outright, more capital no longer needs more people. Returns stop falling as capital accumulates, and whoever saves the most pulls away without limit. The authors conclude that only a broad, progressive tax on capital, coordinated across countries, would stop it, and they do not claim to know how to get one.
Now add our first rule. Their essay worries about heirs. In our world there are no heirs. The original saver simply keeps going, and everything that used to wear a fortune down is switched off: the split among children, the spendthrift grandson, the estate tax.
The case against. Critics answered within days. The writer Zvi Mowshowitz, in a long reply, accepts the arithmetic and attacks the premises. The essay takes for granted that property claims go on being honored at any scale, however lopsided they become, and that people, not the machines themselves, remain the owners. We think the first objection is the heart of the matter.
Property is a settlement, not a law of physics. A deed is a promise by everyone else to leave you alone, enforced by courts and police that everyone else supports. The promise holds when enough people have a stake in it holding. Henry VIII did not need an economic theory to take the monasteries. He needed a great deal of land to sit in hands that could not defend it.
That leaves three ways it can go. Two can last and one cannot.
| Outcome | Who owns the machines | How everyone else gets income | Does it last? |
|---|---|---|---|
| Broad ownership | Most households, directly or through a common fund | Dividends | Yes. Most voters are owners |
| Narrow ownership, heavy sharing | A few | Taxes on capital income, paid out as transfers | Yes, with a permanent fight over the rate |
| Narrow ownership, light sharing | A few | They do not | No |
The assumption we promised to come back to. All of this takes our third rule as given: people still own and run things. If AI systems come to hold property or make the decisions themselves, the question stops being which humans own what. We are not qualified to answer that one, and we are suspicious of people who say they are. We note it and stay inside our premise.
The version we would choose
An earlier draft of this piece stopped at Exhibit I and declined to pick between the first two rows. That was a dodge, and we would not accept it from anyone else. So here is our answer, followed by the best case against it.
We would choose the first row, broad ownership, and we would build it on purpose instead of hoping for it. The design is a permanent fund in which every citizen holds an equal share that cannot be sold, filled by a small yearly charge on land and on large holdings of capital, and paying out all of its income as a dividend. It is the Alaska model at national scale, with Alaska's one flaw repaired.
Reason one: a title outlasts a promise. A transfer is a line in a budget, and every legislature writes its own budget. Ownership is a title, and titles are hard to take. Alaska has run the comparison inside a single institution for fifty years. The fund's principal is protected by the state constitution and cannot be spent without amending it. It never has been. The dividend rested only on a statute, a formula written in 1982. In 2016 the governor vetoed about half of it, cutting a check of roughly $2,052 to $1,022, and the state's supreme court ruled that the dividend has to be appropriated each year like anything else. It has been an annual negotiation ever since: $3,284 in 2022, $1,000 in 2025, $1,200 this year. The part that was property held. The part that was a promise became a fight.
Franklin Roosevelt understood this about his own program. Asked why Social Security was financed by a payroll tax that his advisers thought bad economics, he said the contributions were there to give people a legal, moral, and political right to their benefits, so that "no damn politician can ever scrap my social security program." The most durable transfer in American history is the one that was designed to feel like property.
Reason two: it keeps property rights popular. This is the argument from the last section turned into a design. People defend rules they benefit from. A country in which every adult receives a dividend from the national capital stock has no constituency for confiscating it.
Reason three: it avoids the hardest thing in tax policy. The second row needs high tax rates on income that can move, held steady across many countries at once. Trammell and Patel, who see no other route, admit they do not see how to get there. A charge collected in kind on assets that cannot leave, such as land, or shares in companies that want access to the country's courts and customers, asks much less of the tax collector.
How much difference would one percent make? We built the simplest model we could to find out. Two groups of equal size start with equal wealth. Capital earns 5 percent a year after inflation and, this being the robot world, goes on earning it however much piles up. One group reinvests 40 percent of its income and the other 10 percent. Then we add the charge, paid into a fund that both groups own equally and that pays out everything it earns.
Two things stand out. The first is that the runaway is real, even between two groups whose only difference is a habit: no charge, and the thriftier half owns nineteen-twentieths of everything inside two centuries. The second is how little it takes to stop. One percent a year does not reverse the savers' lead. It caps it at about three-quarters, permanently, while leaving them every incentive to keep saving.
None of this is exotic any more. Versions of the idea have surfaced on both sides of American politics in the past eighteen months, and we mention them as evidence of direction, not as endorsements. The tax law passed in 2025 created Trump Accounts, which launched this July: the Treasury puts $1,000 into an index fund for each American child born from 2025 through 2028 whose parents open an account. Altman has reportedly offered 5 percent of OpenAI to a public fund, and Senator Bernie Sanders has proposed a one-time 50 percent levy on the stock of AI companies for the same purpose. The federal government has owned a tenth of Intel since August 2025. These differ enormously in size and in spirit. They agree on the instrument, which is shares.
What the other side gets right. Readers who prefer the second row have real arguments, and we would be poor analysts if we hid them.
- The charge is a wealth tax under another name. In our debt piece we passed on a wealth tax ourselves: it faces a serious constitutional challenge, it depends on valuing assets nobody has sold, and most of the European countries that tried one repealed it. Payment in kind solves valuation for listed shares and for nothing else. A family business, a painting, and a private fund are as hard to value as they ever were. We come back to our own inconsistency in the next section.
- A fund that owns a third of the capital stock is a political prize. Our own exhibit says the fund ends up there. Norway's fund invests only outside Norway and stays a small minority holder everywhere. An American fund holding American companies would face a standing temptation to vote its shares for whoever holds power.
- Transfers can be aimed, and dividends cannot. An equal dividend sends the same check to those who need it and those who do not. Dollar for dollar, a transfer targeted at hardship relieves more of it. Dividends also fall in bad years, which is when need rises.
- A share nobody may sell is paternalism. It tells adults they cannot be trusted with their own property. We think nine centuries of temptation justify it. A reasonable person can disagree.
- It taxes the saving that built the machines. If capital earns 5 or 6 percent after inflation, a 1 percent charge takes something like a fifth of the return. Whether the robot economy still needs that incentive at full strength is an open question, and the transition certainly does.
Most rich democracies today are closer to the second row than the first: they leave ownership where it lies, tax the income, and pay transfers. That is a respectable arrangement with a long record, and a reader who weighs the five objections more heavily than we do should prefer it. What tips us the other way is Alaska. Given fifty years and one fund, the part that people owned survived every budget crisis, and the part they were merely promised did not.
What it does to our debt plan
In September we published a plan to stop the federal debt from outgrowing the economy: seven steps and about $10 trillion of deficit reduction over ten years. It was written for the world as it is, which is the only world a budget can be written for. Still, a plan meant to hold for thirty years ought to be asked how it fares if those thirty turn out stranger than expected. So we reread it with this piece's two rules in hand. About a quarter of it leans on wages or on death. The load-bearing wall does not.
Start with what the Treasury taxes. Federal receipts were $5.40 trillion in fiscal 2026. By the Congressional Budget Office's final tally, individual income taxes brought in about $2.85 trillion of that and payroll taxes $1.81 trillion, against $0.38 trillion from corporations. The first two are 86 percent of everything the government collects. Through August, about 70 cents of every revenue dollar had been withheld from a paycheck before the worker ever saw it. We knew the federal government was financed by wages. We had not appreciated how completely until we added it up.
That is the vulnerability. In the thought experiment, output soars while the share paid as wages shrinks, and GDP is not revenue. A country can get much richer while the particular kind of income its tax code is aimed at dries up.
| Piece of the plan | What it rests on | Ten years | In this piece's world |
|---|---|---|---|
| Step 1 · Payroll tax on all wages, plus half a point | Wages | ≈ $2.2T | Shrinks with the wage share |
| Steps 1 and 2 · Retirement at 70, Medicare at 68 | A lifespan of about eighty years | ≈ $0 | Meaningless at nine centuries |
| Step 2 · What Medicare pays hospitals, insurers, and drug makers | The illnesses of old age | ≈ $1.7T | Unknown. A cure for aging is either the largest saving in the budget or its largest new cost |
| Step 3 · Gains taxed at death; the estate tax restored | Death | ≈ $0.6T | Collects nothing |
| Step 3 · Corporate rate, gains rates, the top bracket | Profits and gains | ≈ $1.7T | Grows. Profits are where the income went |
| Step 4 · The VAT and the carbon tax | Spending and energy | ≈ $3.6T | Holds. Machines make things and people buy them |
| Step 4 · Transactions tax | Trading | ≈ $0.8T | Holds |
| Step 4 · Cap on the health exclusion | Employer health insurance | ≈ $0.4T | Shrinks with employment |
| Step 5 · Appropriations freeze | The cost of agencies and defense | ≈ $1.8T | Easier, if government lets machines do the work |
| Step 7 · The debt brake | The VAT | not counted | Still works, and matters more |
The pattern is easy to state. The pieces that tax what people earn at work, or what they leave behind, fade. The pieces that tax what people spend and what businesses earn hold or grow. We called the VAT the load-bearing wall of the plan in September, and it is the one wall the thought experiment leaves standing. That was luck more than foresight. We will take it.
The debt brake comes out better still. It raises the VAT automatically whenever debt stops falling as a share of the economy, and it does not ask why. If the wage base erodes and payroll receipts fall short, the brake fires and moves the burden onto consumption, a point at a time. We designed it as insurance against high interest rates and our own arithmetic. It turns out to insure against this too.
Would growth make the problem go away? The equation at the center of the September piece says the debt ratio moves with the gap between the interest rate the Treasury pays and the growth rate of the economy. We refused to count on growth then, even though CBO's own scenarios put half a point of extra productivity growth at roughly 40 points of debt-to-GDP over thirty years. A machine economy would be the largest growth surprise in history, and it is tempting to say the debt simply melts.
It might, for a while, because much of the debt is locked in at yesterday's rates and takes years to roll over. It would not melt forever. The force that raises growth also raises the interest rate, as we saw in What money earns, and in the textbook version the rate rises at least as fast. A plan that depends on the gap is still a hope and not a plan. That sentence survives the thought experiment intact.
Social Security becomes something else. Our plan raises the retirement age to 70, on the argument that life expectancy at 65 is five to six years longer than when benefits began. At nine centuries the adjustment is absurd. No retirement age makes a lifetime annuity affordable when the lifetime does not end, and a program triggered by age stops making sense when age stops meaning frailty.
There are two things it could turn into. One is insurance against need, paid to people without means. The other is a payment that does not ask about age at all, which is the dividend from the fund we just described. We would choose the second, for Roosevelt's reason. Social Security has lasted ninety years because people believe it is theirs. A dividend from a fund they own would be theirs in fact.
Medicare is the coin flip again. In September we called the medical cost curve the biggest unknown in the plan. Here it is larger. Aging is most of what Medicare insures against, and a therapy that slowed aging would attack that spending at the root. It would also be something every beneficiary wanted in the same year. CBO has found that covering weight-loss drugs for obesity would raise federal spending, because the savings from healthier patients arrive slowly and fall short of the cost of the drugs. A cure can be both the best news in the history of medicine and a line item nobody can pay for.
The inconsistency we promised to own. In the debt piece we passed on a wealth tax, for the three reasons quoted above, and argued that taxing gains at death and restoring the estate tax would reach the same fortunes through taxes that already exist. In this piece we propose a 1 percent yearly charge on large holdings, which is a wealth tax with better manners.
We do not think the September reasoning was wrong for the world we live in. It does not survive our own first rule. If nobody dies, the two taxes we relied on collect nothing, and the choice is between a charge on holding and no tax on accumulated fortunes at all. Given that choice we would take the charge, at a low rate, paid in kind. The September argument was never about principle. It was about the cheapest way to reach the same dollars, and the cheapest way changes when the funerals stop.
And the dollar itself. The debt is a promise of dollars, so it is fair to ask what a dollar would be worth. If machine-made goods got cheaper every year, would the currency gain value and make the debt heavier? Not while the Federal Reserve targets inflation. A central bank aiming at 2 percent offsets falling prices for made things with rising prices elsewhere, and in this world elsewhere means land, energy, and the few services still performed by people. Incomes in dollars would then grow at least as fast as output. The debt would not get heavier by itself. It would simply be owed, as now, in a currency that buys less and less of the things that are scarce.
There is one piece of good news for the Treasury in all of this. Savers who need safe income forever are natural buyers of very long bonds. A government facing immortal lenders could borrow for a century at a time, or with no maturity at all, as Britain did for more than two centuries.
What we would change. Three amendments, none of which we would have thought of in September. Tie the retirement and Medicare ages to measured life expectancy by formula, instead of legislating a number that medicine can make obsolete. Move the payroll tax, over time, from wages to income of every kind, so that the program's base cannot shrink out from under it. And treat the taxes at death as placeholders, to be replaced by a charge on holding in whatever year the actuaries report that Americans have stopped dying on schedule.
What it means for a growth portfolio
Our Capital Appreciation portfolio is concentrated, on purpose, in the companies building the physical plant of artificial intelligence, alongside a few holdings chosen for reasons that have nothing to do with it. So the question we kept returning to while writing this piece is the selfish one. If the machines do arrive, who keeps the money? The chip makers? The labs that train the models? Nobody, because money stops mattering? Or the tax collector?
We answered the third already. Money goes on mattering, and shares are not money in any case. The other three take more work.
Start with the base rate, which is humbling. The economist William Nordhaus once measured how much of the value created by new technology ends up with the companies that create it. Looking at American business from 1948 to 2001, he estimated that innovators captured about 2.2 percent of it. The rest went to customers, as lower prices and better products.
Buffett made the same point with history in 1999, when everyone around him was sure the internet would make its builders rich. He counted at least 2,000 makes of American car, of which three companies survived. He listed 129 airlines that had gone bankrupt in twenty years and noted that, as of 1992, the whole industry had earned nothing since the Wright brothers. What matters, he concluded, is not how much an industry changes the world but each company's competitive advantage and, "above all, the durability of that advantage."
So the importance of machine intelligence tells us nothing about who profits from it. The useful question is narrower: at which layer of the business can competition not get in?
The labs deserve their own paragraph, because one of them is about to be a public company. Anthropic, which is scheduled to price its initial public offering on November 18, reported $33.2 billion of revenue for the first nine months of 2026, against $4.6 billion in all of 2025, with an operating loss of $1.3 billion. OpenAI, expected to follow next year, took in about $10 billion in the third quarter and lost about $10.9 billion at the operating level on its own adjusted measure, according to documents reviewed by the Wall Street Journal. The two companies define their costs differently, so the losses do not compare line for line.
We know of no precedent for revenue growing like that. We also know of no precedent for the price trend in section four. A product whose price falls ninefold or more every year is one of two things. If a handful of firms can keep the frontier to themselves, it is the best business ever built. If they cannot, it is the airline industry with better margins for a few years. Which of those it is turns entirely on durability, and durability is the one thing a prospectus cannot show. Anyone who wants the mechanics of what an offering does to existing and new owners will find them in our piece on IPOs.
The middle of the stack is where we are most comfortable, and we know that should make us careful. The facts in layer three are remarkable. They are also facts about the present. A chokepoint in manufacturing lasts as long as the frontier keeps moving and the leader keeps leading. Machine intelligence could lengthen that lead, since the leader has the most data about its own process, or shorten it, since a follower can rent the same intelligence. We do not know which, and we size positions as people who do not know.
Does it simply get taxed away? Some of it. Under the design we favor, a 1 percent yearly charge on large holdings takes something like a fifth of a 5 or 6 percent real return. Under the alternative, taxes on profits would be heavier than they are now. An owner can live with either.
The larger risk is older and blunter. When a toll booth becomes essential, it gets regulated. The railroads got the Interstate Commerce Commission in 1887. Standard Oil was broken up in 1911. Power and telephone monopolies spent the twentieth century earning whatever a commission allowed them. The tolls that last are the ones that stay small next to the value passing through, collected by businesses whose customers could in principle go elsewhere. We would sooner own ten modest tolls than one that makes headlines.
Our working answer, in order of confidence. We are most confident that the things nobody can make more of hold their value, and least confident about when. We are confident in the manufacturing chokepoints for this decade and agnostic beyond it. We think the plumbing of ownership and the businesses built on people's time are steady and do not depend on who wins the race. And we think the labs have the widest range of outcomes of anything in the stack, which is a reason for interest and for caution in equal parts. Two things would change our minds: evidence that the frontier is turning into a commodity faster than the price data already suggest, or the first serious move to regulate a chokepoint as a utility.
What it means for an income portfolio
We describe our Cash Flow portfolio as built for life. In this piece's world that phrase needs a second look, because the life in question just got ten times longer.
Much of the portfolio's income comes from business development companies, the publicly traded lenders we took apart in an earlier piece. A BDC lends to mid-sized private businesses, mostly at floating rates. It borrows against its own loan book to do so. And it must pay out at least 90 percent of its taxable income to keep its tax status. Each of those features meets the thought experiment differently.
First, what "for life" requires. A retiree with thirty years ahead can spend income and, late in the day, some principal. Someone with nine centuries ahead cannot spend a dollar of principal, ever, and has to keep it whole against inflation besides. That is an endowment's problem, and endowments solve it with a rule:
A lender that must distribute nearly everything it earns cannot do that reinvesting on its holders' behalf. The holder has to do it. Over thirty years, the gap between spending the whole yield and spending the sustainable part shows up as a smaller estate. Over nine centuries it shows up as Exhibit G. We made a related point about funds whose yield is partly the investor's own money handed back in our first piece.
This is the one lesson in the section that needs no robot and no cure. A retirement can already run forty years. A yield is not a spending rate, and the longer the horizon, the more the difference costs.
Second, what happens to the borrowers. Here the thought experiment has already begun, in miniature. The Bank for International Settlements counts about $115 billion of BDC loans to software companies, roughly a fifth of all BDC lending. When investors began to fear that AI would undercut those companies, software shares fell almost 30 percent between October 2025 and February 2026, and BDC shares fell about 10 percent with them. In July, BIS researchers found that lenders had not yet priced software loans any differently from other loans, while the cushion of extra yield had been getting thinner.
Software was the ideal borrower for a decade: recurring revenue, few physical assets, customers who rarely leave. It is also labor in a box. It is code that people wrote, sold to replace work that people did. If machines write the code, the moat is shallower than the loan documents assumed. The general rule follows. A loan underwritten on cash flow is a bet that the cash keeps flowing. A loan secured by something that cannot be reproduced is not. In a world of nearly free labor, the question to ask of any borrower is what part of its revenue survives a competitor with an unlimited workforce.
The structure of these funds helps here. Most of their loans are senior and secured, they reprice with short-term rates, and they are repaid within a few years. A lender that gets its money back that often can change its mind that often. When nobody can see far, that is the right way to lend.
Third, what happens to lending as a business. On the road to the machine economy, the demand for credit booms. Somebody has to pay for the build-out. Meta's Louisiana campus was financed with about $27 billion of private-credit loans in a vehicle that is 80 percent owned by an asset manager, the largest private capital deal on record. Morgan Stanley's researchers have estimated that private credit will supply about $800 billion of a $2.9 trillion build through 2028.
At the destination, two things squeeze the lender. There are more savers, because immortals save, and more capital chasing the same loans means thinner spreads. And there are better underwriters. A private lender's edge is knowing borrowers that banks will not take the time to know. Reading every filing and every covenant is exactly the kind of work machines do first. We would expect the extra yield in private lending to shrink toward what the risk alone deserves. Floating rates, meanwhile, cut whichever way short-term rates go, and we have already admitted we cannot call that.
Fourth, falling prices are hard on debtors. A borrower whose product gets cheaper every year owes the same dollars every year. That is the oldest way lenders lose money, and a world of collapsing manufacturing costs has a great deal of it. The defense is the same as before: lend against what cannot be reproduced, such as land, contracted power, or a license.
Our reading. Of everything in the Cash Flow portfolio, the lenders are what this world would test hardest. They are leveraged, they must pay out what they earn, and many of their borrowers' advantages are built on labor. They are also short-dated, senior, and unusually transparent, which is why we can watch the test in the filings quarter by quarter instead of discovering the result afterward. The holdings that own assets outright, or that collect a fee on other people's capital, sit on the other side of the ledger. Where capital's share of income rises, a claim on capital rises with it.
So "built for life" survives, with a correction. What lasts is a habit, not a yield: spend less than the statement says, put the difference back, and keep asking what could kill each source of income. That was the right habit for thirty years. It is the only one for nine hundred.
Six guesses, and what to do with them today
If we had to reduce all of this to a list, it would be this one. We hold every item loosely.
- Money stays and the paycheck fades. Most income arrives as a dividend, a rent, or a transfer.
- Prices split in two. Made things fall toward the cost of energy and materials. Land, legal rights, and rank take a growing share of all wealth.
- Interest rates are the wild card. Faster growth pushes them up and longer lives push them down. Either way, asset prices become far more sensitive to them.
- Ownership gets a clock or a meter. Leases replace some freeholds, and a yearly charge on holding replaces the estate tax. The tax event moves from the funeral to the calendar.
- Seats get term limits. Offices, like assets, have to turn over without waiting for a funeral.
- Broad ownership is what keeps property rights alive. The settlement holds where most people are owners. Elsewhere it gets renegotiated, politely or not.
None of this needs the science-fiction version to matter. Labor's share of output is already at a record low, and lives are already longer than the systems built around them assumed. The direction is enough to act on, and what it suggests is mostly old advice with a new reason behind it.
Be an owner. In every version of this future, the claim on output moves from work to ownership. The easiest time to become an owner is while there is still a paycheck to buy with.
Favor what cannot be copied. Ask of any business what it has that a well-funded rival with unlimited free labor could not rebuild.
Put survival ahead of return. Exhibit G is the whole argument. Avoid debt that can force a sale and any single bet that can end the game. Time is the edge, and ruin forfeits it.
Distrust long extrapolations. The bottom row of Exhibit E never arrives. Franklin, a careful man, got a quarter of his plan in his better city. Assume modest returns and be pleasantly surprised.
Expect the rules to move. Taxes on holding look more likely than taxes on dying. Any asset whose value rests on a legal right carries political risk, and that risk rises as ownership narrows.
Gulliver's plan for eternity was to spend two centuries getting rich and then turn to wisdom. Luggnagg made sure nobody could try it. We would aim somewhere between the two: own things that last, in a country where enough other people own things too that nobody needs to declare us dead at eighty.
Sources and a note on the numbers
The literature of the thought experiment:
- Jonathan Swift, Gulliver's Travels, Part III, Chapter 10 (1726).
- John Maynard Keynes, Economic Possibilities for our Grandchildren (1930).
- Anton Korinek and Donghyun Suh, Scenarios for the Transition to AGI (2024).
- Trevor Chow, Basil Halperin, and J. Zachary Mazlish, AGI and the EMH (2023).
- Philip Trammell and Dwarkesh Patel, Capital in the 22nd Century (December 2025), and Zvi Mowshowitz's reply, Dos Capital.
- William Nordhaus's estimate is from the NBER Digest's summary of his 2004 paper. Buffett's is from "Mr. Buffett on the Stock Market" (Fortune, November 1999) and, for the partnership letters, a Morningstar summary.
Numbers about today:
- Death probabilities and survivors by age are from the Social Security Administration's 2023 period life table, as used in the 2026 Trustees Report.
- Labor's share is from the Bureau of Labor Statistics and Fortune.
- Federal receipts are from CBO's summary for fiscal year 2026; the withholding share is from its review for August, covering eleven months.
- Estate-tax figures are from Kiplinger and the New Jersey Society of CPAs.
- Births, deaths, and fertility are United Nations figures, from StatisticsTimes and the UN's key messages for World Population Prospects 2024.
- The price of machine intelligence is from Epoch AI.
History and law:
- Franklin's bequest: the Philanthropy Roundtable, When and Where in Boston, and Billy Penn.
- Florence: Guglielmo Barone and Sauro Mocetti's own summary of their study.
- Mortmain and Thellusson: Statutes of Mortmain, Encyclopedia.com, and Thellusson v Woodford.
- Perpetual trusts: 24/7 Wall St. on South Dakota, and Max Schanzenbach and Robert Sitkoff, Perpetuities or Taxes? (2006).
- Jefferson and Madison: Founders Online, September 6, 1789 and February 4, 1790.
- Science and funerals: Pierre Azoulay and Joshua Graff Zivin in ProMarket; the paper is NBER Working Paper 21788.
- Corporate lifespans: Innosight's 2016 study, as reported in Inc. magazine. House prices and land: Knoll, Schularick, and Steger, No Price Like Home (American Economic Review, 2017).
- Leaseholds: the Government of Singapore and, for China, the National Immigration Administration.
- Roosevelt's remark is from the Social Security Administration's history archive.
Funds and proposals:
- Alaska: the Permanent Fund Dividend Division, the state supreme court's opinion in Wielechowski v. State, and the Anchorage Daily News.
- Norway: Norges Bank Investment Management.
- Sam Altman, Moore's Law for Everything (2021); MIT Technology Review and Newsweek on this year's proposals; Axios on Trump Accounts.
Markets and companies:
- The AI labs: Yahoo Finance on Anthropic's amended prospectus and Forbes on OpenAI.
- Chips: MarketScreener.
- The build-out: CoStar, Bloomberg via Business Standard, World Finance, and a Columbia Business School deck citing Morgan Stanley Research.
- Lenders and software: the Bank for International Settlements' March 2026 Quarterly Review and Bulletin 128 (July 2026).
Exhibits C, E, F, G, K, and N are our own arithmetic. They are illustrations, not forecasts, and anyone with a spreadsheet can reproduce them from the inputs below.
- Exhibit C. Average the SSA table's male and female death probabilities at ages 10, 25, and 30: 0.000113, 0.001102, and 0.001482. The share still alive after n years is (1 − probability) raised to n. Average lifespan is 1 ÷ probability.
- Exhibit E. $10,000 × (1 + rate) raised to the number of years.
- Exhibit F. Boston: $5 million ÷ $4,444. Philadelphia: $2 million ÷ $4,444. Franklin's plan: 1.05 raised to 100 is about 131; keep 31 of the 131 and grow it another hundred years to get 4,061.
- Exhibit G. (1 − yearly chance of ruin) raised to the number of years.
- Exhibit K. Two groups start with 50 each, and a fund starts at zero. Each year: the fund pays its whole 5 percent return to the two groups in equal halves; each group's income is 5 percent of its own wealth plus its half of that payout; the savers add 40 percent of their income to their wealth and the spenders 10 percent; each group then hands the fund 0, 1, or 2 percent of its wealth. The plotted line is the savers' wealth plus half the fund, divided by the total.
Disclosure: our model portfolios hold companies and funds in several of the categories discussed here, including semiconductor companies and business development companies; subscribers receive every position and every change as it happens. Nothing here is individualized investment advice, and nothing here is a forecast. Our full Disclosure Policy applies.
Tools we use — see our full toolkit (some links earn us a commission).
This is how we reason.
The memos apply the same standard of proof to three live model portfolios — every position change explained.