In his 2013 Capital in the Twenty-first Century, the socialist economist Thomas Piketty argued that, absent strong redistribution, economic inequality tends to increase indefinitely through the generations—at least until shocks, like large wars or prodigal sons, reset the clock. This is because the rich tend to save more than the poor and because they can get higher returns on their investments.
As many noted at the time, this is probably an incorrect account of the past. Labor and capital complement each other. Wealthy people can keep accumulating capital, but hammers grow less valuable when there aren’t enough hands to use all of them, and hands grow more valuable when hammers are plentiful. Capital accumulation thus lowers interest rates (aka income per unit of capital) and raises wages (income per unit of labor). This effect has tended to be strong enough that, though inequality may have grown for other reasons, inequality from capital accumulation alone has been self-correcting.
But in a world of advanced robotics and AI, this correction mechanism is likely to break. That is, though Piketty was wrong about the past, he will probably be right about the future.
Indeed, in some ways, he may well be more right than he knew. A lot of AI wealth is being generated in private markets, which only large and sophisticated investors have access to. You can’t get direct exposure to xAI from your 401k, but the sultan of Oman can. This trend toward the “privatization of returns”, already ongoing (1, 2, 3) and especially pronounced in the AI startup world, could well continue indefinitely.
Furthermore, with full automation,1 the main source of catch-up growth for developing countries goes away; namely, that by importing capital and know-how, they rapidly make their underutilized labor more productive.
If AI is used to lock in a more stable world, or at least one in which ancestors can more fully control the wealth they leave to their descendants (let alone one in which they never die), the clock-resetting shocks could disappear. Assuming the rich do not become unprecedentedly philanthropic, a global and highly progressive tax on capital (or at least capital income) will then indeed be essentially the only way to prevent inequality from growing extreme.2 Without one, once AI renders capital a true substitute for labor, approximately everything will eventually belong to those who are wealthiest when the transition occurs, or their heirs. Or more precisely, it will belong to the subset of this group who save most and most invest with a view to maximizing long-run returns.
The rest of this post is about the lessons we can draw by “re-reading Piketty with our AGI hats on”. We won’t try to integrate these lessons with every thought one might have about the implications of AGI, though some of that integration will appear now and then. And despite the title, we’re making no claims about when AGI will arrive or how long growth will subsequently last. Our aim is just to explore what follows if we accept “Piketty’s model” of inequality in a world of capital-driven growth and believe that AI will eventually bring such a world about.
For those familiar with the basics of growth theory, this section can be summarized briefly.
The accumulation of capital by the rich only produces a divergence in income inequality if capital and labor are gross substitutes.
Piketty argues that they have been, historically. But the standard view, which we follow, is that they have been gross complements.
With advanced enough AI, though, they will become gross substitutes after all.
We think it’s worth emphasizing the arguments for thinking that Piketty was wrong about the past, even though they’re somewhat tangential, because the wronger you think he was about the past, the more you think will change once his assumption comes true.
We produce the things we value using labor and capital. Capital just refers to everything productive besides labor: the chair and scissors in a barber shop, say, or the equipment in a factory.
Piketty’s claim about the historical forces driving inequality begins innocuously. As noted above, he points out that the capital of the rich tends to grow more quickly than that of the poor (a) because they tend to have higher saving rates3 but more importantly (b) they can earn higher interest rates on their capital (because of economies of scale in wealth management). For example, Piketty documents that, from 1980 to 2010, America’s largest university endowments earned outsized rates of return, which themselves increased in the size of the initial endowment. He argues that this is because larger funds can hire better money managers and access a wider range of investment opportunities.
While this persists, global capital belongs ever more to the heirs rather than the entrepreneurs, and capital grows ever more concentrated.
Concentrated ownership of some good only drives income inequality, however, if that good is generating a large share of overall income. The fact that two private firms produce 60-80% of America’s carrots contributes very little to aggregate income inequality. To be sure, if capital ownership were highly concentrated, this would come with a worrisome concentration of political and military power, unlike in the carrot case. But it would not directly produce a lot of income inequality unless the “capital share”—the share of national income coming in the form of interest on investments, as opposed to wages—were high. The capital share in the developed world has historically fluctuated between 20 and 40 percent.4 A key question for Piketty, therefore, is whether this tends to rise or fall as capital-owners, in effect, start accumulating their capital ever more quickly. He argues that it tends to rise.
When an individual capital-owner decides to save more, or invest in a way that allows her capital to grow more quickly, she increases the growth rate of the aggregate capital stock. But while labor remains an important input to production, each extra unit of capital (per worker) will tend to increase our production capacity by less than the last. The third hammer isn’t as useful as the first two while people only have two hands. Our saver thus slightly lowers the marginal product of capital, which should roughly equal the interest rate: the income one can earn each year off of each unit of capital. That is, though she may raise her own future income by accumulating more capital, she slightly lowers the incomes of other capital-owners. The question is, by how much?
If capital is eventually near-useless without the labor to make use of it, then accumulating more capital causes its marginal product to fall quickly. Our saver lowers the incomes of other capital-owners by more than she raises her own.5 She lowers total capital income. (Also, because each hour of work is more valuable when there is more capital for each worker to make use of, her capital accumulation also raises wages.) So the capital share falls. In short, if the rich all try accumulating capital more quickly, this does not lead to a spiral of income inequality, but the reverse.
If capital can keep adding ever more value as it accumulates (fixing the population and everything else), on the other hand, our saver lowers the incomes of other capital-owners only slightly. Capital incomes can rise indefinitely, outstripping labor incomes ever more.
To illustrate the first scenario, water is extremely valuable when it is scarce, but important uses for it vanish quickly when it grows abundant. If a small group of people own all the oases in a desert, they can earn high incomes by selling the water or renting out time at the wells. If any individual oasis-owner digs a new oasis, he raises his income, but starts to lower the price of water. If they all follow suit, until the desert has become a swamp, the water they all have to sell has grown many times over, but the price has fallen even more. Their combined incomes have fallen. The benefits of the water have accrued instead to the rest of the water-using society.
Likewise, if AI is just a historically unprecedented productivity driver, leaving us still bottlenecked by labor for some important goods and services, then the demand for labor will go off the charts and the share of our budgets we spend on AI services will eventually plateau or fall. The labor share of income could actually increase. (This would be what is sometimes called the “Baumol effect”.)
But once robots and computers are capable enough that labor is no longer a bottleneck, we will be in the second scenario. The robots will stay useful even as they multiply, and the share of total income paid to robot-owners will rise to 1. (This would be the “Jevons paradox”.)
Piketty argues that capital has been highly substitutable for labor, in the “Jevons” sense, for hundreds of years. In his view, capital accumulation by the rich—well-building by the well-owners—has essentially always been a force exacerbating income inequality, and it will be enough to widen inequality ever further even if we don’t advance AI at all.
This is an extremely non-standard interpretation of history and of the modern world. It was roundly criticized when the book was first published, for many good reasons.
It implies that it is only by coincidence that the capital share has been roughly constant for centuries. When labor and capital are both necessary for production, it is no coincidence that expenditures on each have grown in tandem (and thus the capital share has been about fixed): expenditures on left shoes and right shoes have stayed proportional too.
It is contradicted by essentially every other way of estimating the substitutability of capital for labor. At the micro level, it is contradicted by direct estimates at the firm- or industry-level that the marginal product of capital tends to fall quickly when more capital becomes available (e.g. Oberfield and Raval, 2021). At the macro level, it is contradicted by the observation that across countries all at the technological frontier at a given time, the countries with more capital per worker tend to have smaller capital shares (e.g. Bentolila and Saint-Paul, 2003). A literature review of 2,419 estimates from 77 studies from 1961 to 2017 finds that, at least for the US economy, the result that capital is not highly substitutable for labor is very robust (Knoblach et al., 2019).
Innovations are predominantly designed to save labor rather than capital.6 This strongly suggests that labor really is a bottleneck (Acemoglu, 2003). If labor were not a bottleneck, it would be more valuable to free up capital, since it is more abundant, or can quickly become so. If you have a hundred workers and a thousand equivalent robots, increasing the efficiency of the robots by 1% is ten times more valuable than increasing the efficiency of the workers by the same proportion.
To take the point above to its conclusion: in the Jevons world, technological development would not merely sustain economic growth. Rather, each technological advance could permanently raise the growth rate.7 This is because, if labor were already not a bottleneck to production, then capital accumulation could sustain growth on its own: we would already be, more or less, in a world of self-replicating robot factories (even if the supply chains involved to build each factory part were long enough that no single factory could literally replicate itself on site). Better technology could then speed growth simply by raising the rate at which capital self-replicates.8 But frontier economies have seen roughly steady growth for centuries, not rapid acceleration.
Remarkably, Piketty’s only evidence for the “Jevons” assumption underpinning his central thesis is that in the mid 20th century, Britain and France had a lower capital stock (per unit of output) than they had before or have now, but the capital share of income did not increase during this period, as the Baumol scenario would predict: instead it dipped.
But this anomaly may be explained in many ways which would not overturn our entire understanding of economic history, technological development, and common sense.
In some contexts, capital might be able to accumulate up to a point without its marginal product falling much, even if this would not be possible indefinitely. That is, it might be that the first doubling of the capital stock makes capital’s usefulness fall only slightly, since there is an overhang of labor or technology allowing useful capital to be built to take advantage of it, but further doublings would make it fall by more than half, since we will have already built all the machines the current state of technology lets workers make good use of.9
As Piketty elsewhere emphasizes, capital was unusually highly regulated in Britain and France in the middle of the 20th century.
From generation to generation, as people get richer or tastes or demographics change, consumption might shift somewhat from less- to more-capital intensive industries or vice-versa, producing fluctuations in the capital share which sometimes offset those driven by capital accumulation alone.
Rognlie (2014) argues that if one accounts properly for capital depreciation and fluctuations in the prices of capital goods relative to consumption goods, and if one excludes housing (an important kind of capital which should not be expected to complement labor in production), the capital stock and capital share actually moved in opposite directions in Britain and France during the 20th century after all.
Nevertheless, given how valuable automating all work would be, it seems hard to deny that Jevons’s world is coming. If the forecasters and AI developers are right, it is likely coming soon.10 The world Piketty describes may not have existed in the past, but we will wake up in it one day; and he has thought more than most about how it might unfold, and how to tax and regulate it so that inequality stays at least somewhat contained.
The US Gini coefficient for income before redistribution is currently 0.42. For wealth, it’s 0.75.11 About 25% Americans have no wealth at all, or negative wealth, having borrowed against their future wage income. As Piketty documents, the US is not an outlier in this respect: though it is more unequal than other developed countries, and more unequal than it was in the past, wealth (or capital) has been much more concentrated than income in every time and place on record.
Capital takes many forms: hammers, microchips, housing in Kansas City, land in Montana. If every kind of capital were on track to become equally productive in an automated world, or if everyone’s wealth were composed equally of every kind (i.e. if we all just owned some number of shares in an index fund that owned all the capital in the world), then as the capital share rose to 1, the US income Gini coefficient would (“just”) rise to around 0.75.
But the lower you are in the wealth distribution, the larger a fraction of your wealth probably consists of real estate. For the poorest half of Americans, the fraction is about half: typically, from people owning (or partially owning) their own homes. And a cheap house is a form of capital almost uniquely ill-suited to taking advantage of a leap in automation. It plays no part in the development, production, operation, or transportation of computer equipment, robotic equipment, training data, or energy. Nor is it one of (or does it contribute to the production of) the luxury consumption goods that will be in high demand in a rich, unequal future. Quadrillionaires might bid up the price of the world’s most beautiful beach, for example; they will want no part in the million-odd single-family homes around Kansas City. Land might eventually grow much more valuable, if we start filling the world with solar panels and robot factories. But you can buy over 400 acres of pasture land in Montana for the price of the median US house.12 As this illustrates, most of the value of a house usually comes not from the value of land itself but from either (a) the physical “improvements”—the all-too-slowly-depreciating hunk of capital generating a stream of a single good: mediocre outdated housing—or, in an urban environment, (b) the proximity to other human beings. These will both be ever less significant sources of value.
What will grow more productive is the capital held by firms, especially firms in the most AI-exposed industries. Without taking a stand on which firms or industries these are, it seems safe to say that stocks as a whole will grow more valuable. (That is, the dividends produced by an index fund will rise. Stock prices may fall, for reasons explained by Chow et al. (2025), but that is not the relevant comparison here.) Public stock ownership is even more unequally distributed than wealth as a whole, and by far. Its Gini coefficient appears to be well over 0.9.13 Though the share of Americans with no stock holdings at all has been falling, as of 2022 it is still as high as 42%.
And what will likely grow more productive still are the startups, those begun in the last few years and those still to come, which have been built from the ground up to be links in an AI-centered supply chain. While these firms remain private, their ownership of course remains even more unequally distributed than that of public companies; and typically, it is even more concentrated among the wealthy.14
If, after the transition to full automation, everyone
faced the same tax rate,
suffered no wealth shocks,
chose the same saving rate, and
earned the same interest rate,
income inequality would stabilize at some high level: at a Gini coefficient of, say, 0.95. This would already be insanely high. The highest income Gini coefficient in any country-year on record is 0.63, observed in South Africa in 2014.
But the wealthiest will probably tend to save more and earn higher interest rates than average. That is, fixing (1) and (2), the failure of (3) and (4) could drive income inequality ever higher.
As for (3): historically, the rich have tended to save more than the poor (Dynan et al., 2004; Saez and Zucman, 2016). A complete analysis of the future of income inequality would try to explain why this is and whether this pattern will persist. But for our purposes, this is unnecessary. In a world where almost all income comes from interest on investments, as long as some people persistently save more than others for whatever reason—and all investment decisions are made by the best AI investment advisors—the high-savers will eventually be among those with the highest incomes, and their incomes will pull ever further away from the rest.15 As discussed above, this process will no longer be corrected by the dynamic whereby capital accumulation lowers interest rates and raises wages.
As for (4): in general, the rich can earn higher interest rates than the poor because some investments come with fixed costs. For arguably the most significant example, consider that a significant proportion of each year’s growth comes from the “Luttmer rocket” phenomenon, in which small new firms grow very rapidly, largely by taking market share from incumbents (see e.g. Jones and Kim, 2018). Since the firms are typically private during this growth phase, most investors cannot participate in this growth: only (i) the firm’s founders and early employees and (ii) its private equity investors. The exclusion of the public is mitigated, but only partially, by the fact that public firms (like Microsoft) can themselves own equity in private ones (like OpenAI), and by the fact that private firms can borrow from banks, which aggregate the funds held in millions of low-risk, low-return saving accounts.
Though the concentration of firm ownership among founders and early employees during its fast growth phase drives a lot of income inequality within each generation (as noted above), it can mitigate the compounding of inequality across generations, since it prevents the biggest new fortunes from accruing entirely to the biggest investors of the past.16 But this source of “wealth churn” will disappear when the entrepreneurs are pieces of capital, owned and passed down. What will remain is (ii): the fact that some investments, and perhaps disproportionately the fastest-growing investments, are open only to large, private investors. It is therefore important to understand why this is and whether we should expect it to persist after automation is complete.
In fact, the fraction of corporate capital held by private firms has not only persisted but risen steeply in recent decades: in the US, from 8% in 2000 to 19% today.17 This “privatization of returns” is primarily attributed to the growing importance of intangibles (see e.g. Doidge et al., 2018). When much of a firm’s value comes not from how much equipment it has but from how promising its proprietary technology is, its value is harder for outsiders to assess; and naturally, publicizing the details of the technology can erode its value. Because of this information asymmetry, small investors on a public market, who do not have the means to do independent due diligence, are willing to pay less for a share in such a firm than the large private investors who can. So if intangibles form an ever larger share of firm values as technology advances, as we think they probably will, then all else equal, we should expect the privatization of returns to continue.
Furthermore, when firms eventually do go public, they are willing to pay the associated costs only for the benefits that come with being available to a much larger pool of investment funding: the new investors can bid up one’s shares and make them more liquid. In a world of extreme inequality, these benefits will shrink. If essentially all wealth belonged to a few people, not many companies would pay the costs of going public to get access to everyone else’s nickels.
AI could reverse this trend by making it easier to price intangibles or aggregate small pools of savings, or by lessening any of the other frictions that keep capital from obeying the “law of one price”. Whether the gap between the returns of the rich and the returns of the poor grows on balance will depend on the outcome of a race between the financial services technology and the economic trends discussed above. But it seems unlikely to vanish entirely, let alone for the poor ever to face higher returns than the rich.
For the past 75 years or so, the world’s poorest countries have, on average, grown more quickly than the world’s richest countries. This is because poor countries are rich in an important resource that is being poorly utilized: human labor. Whereas growth in the richest countries requires technological development, growth elsewhere “only” requires the local population to import the latest equipment and adopt the latest production techniques (not to deny the difficulty of achieving the institutional, educational, and cultural conditions needed for the import to be successful). As Korinek and Stiglitz (2021) point out, when capital can well substitute for labor, this powerful force for catch-up growth will end. To a first approximation, countries’ relative incomes per person will remain proportional to their initial stocks of capital per person.
For countries rich in the natural resources that will be most important in an automated future, some catch-up may remain possible. American firms are building data centers in the UAE and other Gulf states not because of the quantity or quality of the local labor but because of the quantity of the oil. But globally, the natural resource share—the share of income received in exchange for the use of natural resources—is on the order of 5%, much less than it was in the distant past, and shows few signs of rebounding (Eden and Kuruc, 2024). In other words, even if all the world’s natural resources belonged to the world’s poorest countries, it would probably be a long time before their share of global income exceeded 5%, if ever. And as Piketty emphasizes, the developing world’s natural resources are largely foreign-owned: about 50% in South Africa, for instance.
Currently, the primary way wealth is “transferred” across generations is what economists sometimes call the OLG [overlapping generations] mechanism: the young invest some of their wage income so that they can consume it in retirement, and when they are old, the capital they have accumulated augments the work and raises the wages of the next generation. This “intergenerational catch-up” is closely analogous to the international catch-up discussed in the previous subsection, and, in an automated world, it will fail for the same reason (see e.g. Sachs and Kotlikoff, 2012).
One implication is that, unless parents start explicitly transferring larger fractions of wealth to their children well before they die, intergenerational inequality will greatly increase. (Or perhaps: we should expect parents to start transferring larger fractions of wealth to their children well before they die, because otherwise intergenerational inequality would greatly increase.) Especially if lifespans increase more quickly than the average age of parents when their children are born, this increase could in principle grow to comical proportions, with each person orders of magnitude richer in her last two or three decades than throughout the vast majority of her life.
Another implication is that the (eventual) wealth distribution in a given cohort of children will be governed by essentially two things, in equal measure: the parents’ wealth per child, and the fraction of this wealth they pass on. The importance of intergenerational transmission via imparting knowledge, culture, and genes will all fade away. Though we disagree with Piketty that the income distribution throughout the past would have concentrated indefinitely in the absence of shocks, it is certainly true that one important limitation on the longevity and ultimate relative size of family fortunes has been that the fortunes are periodically splurged. In an automated world, it will be the only limitation; and preventing it will be the only path to avoid (at least relative) poverty. A splurge will be essentially impossible for the splurger or his heirs ever to recover from.
Anticipating this, we should probably expect the shift to automation to coincide with increased investment in “commitment technology”. That is, condition (b) sustaining high income inequality should come closer to reality. As has long been recognized, one important way in which AI agents may differ from human ones is that, for better or worse, it may be much easier to get AIs to commit to a given policy. But even apart from this, as Piketty discusses, it is already somewhat easier than it once was for people to establish complex family trusts that leave wealth to their descendants without giving them the right to spend it quickly.18 We should expect this trend to continue, and for an ever larger share of income to accrue to those who participate in it.
Finally, by the same token, we should expect an ever larger share of income to accrue to foundations and charitable trusts, especially those that commit to spending slowly. Though the law differs from place to place, foundations, unlike trusts, are typically legally required to spend some percentage of their assets each year (but are allowed more flexibility as to their mission): in the US, the minimum is 5%. But this can be largely circumvented in various ways; and in any case, since interest rates in a fully automated world will probably be many times higher than 5%, the minimum will be close to meaningless if it is not raised.
The final variable determining the long-term distribution of capital is of course how it gets invested. Idiosyncratic investing genius will presumably not matter once the AIs are doing the investing—though the AI outputs may be somewhat stochastic, in essence everyone with a given budget to spend on investment advice will equally get the best AI investment advice that that budget can buy.1920 But the risk-tolerance with which a given pot of capital is invested can greatly affect its long-term growth rate.
The Kelly rule is a well-known formula describing how one should invest, under stylized conditions, if one wants to maximize one’s expected log(wealth). It is not in any further sense the “optimal” way to invest; one may be more or less risk-tolerant than a logarithmic utility function would demand. But putting aside some technicalities, it turns out that, if everyone adopts the same saving rate and has access to the same set of investment opportunities, and if at least some people invest in an “approximately Kelly-like” way, then for any f<1, the probability that at least fraction f of the world belongs to these people tends to 1 as the time horizon tends to infinity. If the rich have access to investment opportunities offering higher returns, the set of people among whom the world will be divided is further restricted to those who start out richest when the regime of capital-driven growth begins.
In short, if capital rather than labor will eventually be the primary source of income, the earth will not be inherited by those with high birthrates, contra e.g. Kaufmann (2010). It will be inherited by those who
start out richest;
are most patient (or act that way by saving a lot);
invest in the way that is, in the relevant sense, most Kelly-like; and
are most willing and able to commit to the saving and investment behavior above.
If you want to inherit (or want your heirs or favored philanthropic causes to inherit) a non-negligible share of the pie—that is, if you are concerned with your place in the future distribution, rather than your absolute future budget—the broad implications of this section are straightforward. While you are not among the world’s richest actors, start accumulating capital as early as possible. Get into (or start) those private firms with the AI intangibles. Invest unusually risk-tolerantly: you will probably go bust, but it is your only hope of catching up. Especially among illiquid investments, prioritize those in which the capital is mobile and not quickly bottlenecked by natural resources, if there is a risk that some shock or tax will hit the country where it is located. And if you do make it to the top, save almost everything; invest it in the precise way that balances the need to keep growing with the need to avoid going bust; and try hard to tie your hands and the hands of your inheritors.
An old and widespread worry is that, even if we begin in a democracy, inequality can entrench itself and grow arbitrarily extreme, as the rich push policy or steer technological development in the direction that best serves them. Relatedly—or as an extreme case of the same phenomenon—revolutions may be much easier to suppress when a state or ruling elite has a robotic military that obeys its commands without the need for any popular support.
But as long as democracy lasts, the shift to capital should make redistribution easier, not harder, for at least two reasons. First, and most importantly, it will no longer be necessary to encourage hard work and entrepreneurship by letting people keep a large share of what they earn. Second, as Piketty emphasizes, income inequality due to capital, especially inherited capital, is seen as less just than income inequality due to labor; so perhaps capital income can be radically equalized without setting precedents that raise the risk that the state will violate personal rights more generally.
A massive wrinkle is that taxing capital effectively requires better international coordination, since capital is generally more mobile than labor (at least in the sense that investors can reallocate new investment overseas). We discuss this further in the next section. But this is in some sense a political constraint, which the countries of the world will all have a strong interest in relieving, not a technological one.
And it should be feasible to retain democratic control of the state even after labor has grown comparatively useless.
Some have argued that this would be unstable. What caused the spread of democracy, the argument goes, was that industrialization spread “real economic power” to a large mass of workers who could easily organize (Acemoglu and Robinson, 2009). Once the masses lose this economic power, due to AI, the political power will follow (Garfinkel, 2021). The only hope for democracy would be an ever more extreme—in our view, hopeless—program of “directed technical change” which somehow permanently preserves labor’s place as a tight bottleneck to production.21
But real power may consist in being necessary for production or in being sufficient for destruction. The latter may be widely distributed even when the former is not. Indeed, an important worry about AI is of course that it will make it possible for ever more people to wreak ever more havoc, e.g. by engineering biological weapons. It would currently be illegal for a disgruntled worker to blackmail the rich by threatening to release a virus unless he is transferred, say, a hundred thousand dollars a year; but that is again a political question, not a technological one. So, though we should not stumble blindly into a world of full automation assuming that democracy will be preserved by default, it should be possible with some imagination to maintain a system in which “real power”—hopefully not enough to bring the world to its knees, but enough to demand a say in how laws are written and resources are allocated—is widely shared. Physical fitness too used to be a byproduct of work, and the transition away from physical work came with a rise in obesity. The solution is not to return to physical work, or give up on fitness, but to start going to the gym.
[Added 1/3/26, for the economists. — In an AK economy, there is a tax that does not introduce any inefficiencies: a flat tax on consumption, distributed as a lump sum. This doesn’t discourage how people spend, and it doesn’t discourage saving (i.e. when people spend) either, since it equally penalizes consumption at every period.
When there are economies of scale to investment, however, concentration of capital ownership is efficient. If widespread capital ownership persists at all, it is presumably due to a market failure: people would want to give all their capital to a few in exchange for a promise of a share of the returns, proportional to their contribution, if they could frictionlessly borrow against this income stream and if this transaction incurred no other costs. By slowing the process whereby capital grows more concentrated, the flat tax/transfer policy exacerbates this inefficiency, as does all progressive redistribution.
But in the limit, it still leads to complete capital concentration. This puts it in an awkward place. If we care only about efficiency, we should just centralize capital ownership altogether, as discussed in footnote 19 and the preceding discussion. If we dislike wealth inequality for reasons other than its implications for income inequality—e.g. for political economy reasons—we can’t rely (entirely) on a flat tax/transfer. We need a tax code in which the present value of your future tax bill, as a proportion of your wealth, increases with your wealth.]
To keep income and “real power” alike widely distributed, the obvious place to start is with Piketty’s headline recommendation: to highly and progressively tax capital.
This will be inefficient, for the same reason that taxing capital is inefficient today: because it discourages future consumption (via saving) relative to present consumption. This is inefficient even if the saving rate in the absence of capital taxation is optimal, but especially unfortunate if one believes that people tend to save too little, out of impatience or lack of concern for future generations. And in fact it will be much more inefficient than today, because in a regime of capital-driven growth, a decrease in the saving rate decreases the growth rate, as it does not when the main engine of growth is the advancement of labor-saving technology.2223
It will be worth doing anyway, if one wants to limit inequality, because it will be essentially the only way to do so. Progressively taxing either capital income or consumption would amount to roughly the same thing—every dollar invested will eventually be consumed—but comes with a risk: if the state cannot commit to it, the rich may just somewhat shift their consumption to low-tax periods. And only redistributing capital itself will keep the means of production (and “real power”) themselves widely distributed, which is presumably the main reason to propose maintaining private property in an automated future at all. Otherwise we might as well go fully communist, centralizing all capital in the hands of the state, reaping the increasing returns that result, and sharing its fruits equally.24
A particularly promising way to redistribute capital will be to tax big inheritances and subsidize small inheritances, perhaps from a positive “baby bonus” baseline.
The motivation for treating inheritances differently is not the traditional meritocratic one. In the world we are considering, income is already almost completely divorced from any notion of “merit”. The motivation is just that people on average seem to have imperfect intergenerational altruism: they care less about the money they leave to their heirs than about the money they themselves will get to consume in the future, and in practice around half of the former is just wealth left over when a person doesn’t live as long as they were planning to (Kopczuk and Lupton, 2007). As a result, taxing big bequests or subsidizing small ones will probably discourage saving less than taxing the capital or consumption of the living.
But even if maintaining democracy proves easy, taxing capital will be hard. People care about where they live (and about who lives in the same country) much more than they care about where their capital is located (or what capital is in the same country). As a result, capital is much more mobile than labor: taxing even capital income in some jurisdiction, let alone capital itself, motivates its owners to rapidly shift their investment elsewhere, much more rapidly than labor income taxes produce emigration. (Also, clever accounting by a multinational can relocate capital income; but it’s not clear whether this will be more or less of an issue in an automated future, and we will put it aside.) Capital’s mobility is probably the main reason why capital income is already taxed at significantly lower rates than labor income.25 And full automation will probably increase capital mobility, and thus (as long as jurisdictions are competing for capital) render the revenue-maximizing tax rates on capital or capital income even lower than today, for three reasons.
First, in most cases, equipment is not literally mobile, but because most of it depreciates every decade or two, shifting new investment amounts to roughly the same thing. This will be even more true when the capital stock is growing rapidly. Suppose an investor starts out with the same amount of capital in two countries, each of which is taxing capital at the same rate. Then suppose one country raises its tax rate, and in response, the investor shifts all new investment to the other country. If the interest rate is only 5% per year, then even if all of this interest is being reinvested, around a decade has to pass before twice as much capital is located in the lower-tax country (given that the capital in the higher-tax country is not depreciating quickly). If the interest rate is 100%, twice as much capital is located in the lower-tax country by the end of one year.
And shifting capital by shifting new investment will be easier still if the depreciation rate rises. This seems likely to happen for two reasons. First: old capital grows obsolete more quickly when technology is advancing more rapidly. Second: as anyone with, say, a computer and a footstool knows, as capital has grown more complex, its depreciation rate has risen.26
Second, to return to the recurring theme, the primary bottleneck to putting capital to good use is currently the scarcity of labor—and, even scarcer, labor with the skills needed to use the capital in question. When this bottleneck is gone, the robot factories can easily go anywhere, even to the most unpopulated or uneducated corners of the world. (At least until natural resources become a significant bottleneck; but as discussed briefly in the next section, this is probably a long way off.)
Third, new technologies may allow large amounts of capital to reside productively in the places currently outside any country’s jurisdiction: international waters and outer space.
If Piketty’s headline message is that a high and progressive capital tax will be necessary to prevent inequality from spiraling, his headline corollary is that strong international coordination will be necessary to tax capital. The discussion above suggests that this too is even truer than he reports.
Unfortunately (if we value equality), it seems likely that international coordination on taxing capital will grow even more difficult. This is, again, because under full automation, capital accumulation can let inequality grow indefinitely. Today, tax havens’ real economies remain small because the economic capacity of a country is essentially determined by its population and, if it is at the technological frontier, the quality of the latest labor-saving technology. The people of the world and their governments can thus, at least to some extent, sanction tax havens at a small cost to themselves and at a large cost to the tax havens.27 In a world of rapid, capital-driven growth, the country that attracts the most capital might soon far outgrow the rest of the world combined (Davidson, 2025), to the point of economically or militarily threatening any would-be sanctioners.
Against this we can only set the tenuous hope that AI may prove especially good at facilitating coordination in general, through superhuman monitoring (so that defections are more quickly punished) and commitment (so that punishments are credible even when irrational). But the stakes will be high—with our Piketty hats on, it seems the problem of devising a global capital tax that sticks will one day be the world’s most pressing problem—so the prospect that all our and our AIs’ ingenuity will manage it is perhaps not entirely hopeless. If we can put a man on the moon, as the cliche goes, maybe we can keep some capital on the earth.
Finally: this section has considered the case that, due to capital flight, a progressive global capital tax might be necessary to prevent income inequality within a country from growing indefinitely. But as discussed earlier, full automation bodes especially badly for inequality between countries: international catch-up will slow or end. International redistribution will thus become even more valuable than today, though of course this hope would be even more utopian.
So far we have discussed taxing capital as if all capital were accumulable capital, as almost all of it currently is. In this context, we have discussed two large costs to capital taxation: that it slows growth and that it drives capital abroad. As any Georgist will tell you, taxing non-accumulable capital—land and other natural resources—avoids both costs. Land reclamation efforts aside, the world’s acreage will not grow more slowly if we tax landlords, and the landlords will not move the land overseas.
These are good reasons to tax natural resources more than accumulable capital. But if the goal is to put a lid on inequality, rather than just to maintain some minimal tax base, taxing natural resources will not be sufficient.
A standard difficulty with taxing a natural resource in practice is that it is often difficult to distinguish its value from the value of the improvements that have been put into it. A piece of farmland is not priced separately from, say, its irrigation system. But a more important issue with relying heavily on natural resource taxes is that they cannot exceed the natural resource share of income without driving the price of natural resources below zero. If the marginal product of an acre of land—the most a landlord can rent it for—is $x per year, no one will want to own it if the property tax is higher. As noted above, the natural resource share is currently no more than 5%, and historically flat at best.28 Though it will presumably rise to 100% if we approach technological maturity and fill the world with solar panels and robot factories, it may stay low for a long time.
There are a few ways beyond direct redistribution that the state could slow or halt a “Pikettian inequality spiral” in an automated economy.
First, by making it easier for small investors to pool their resources until their returns match those of large investors. For instance, we could deregulate how banks, or at least some banks, can invest their savings deposits, even if this implicitly means that the banks are subsidized with stronger deposit insurance.
Second, and relatedly, by making it easier for firms—or at least the kinds of firms most likely to exhibit outsized growth—to go public (or harder for them to stay private). This could be done by loosening the regulatory requirements on public firms (or tightening them on private firms), or simply by taxing them at different rates.29
Third, by imposing on individuals the same regulation already imposed on foundations to prevent them from growing too quickly: a spending requirement. A minimum spending rate—either on an annual basis or, by capping inheritances, over a lifetime—would prevent those inclined to adopt high saving rates from outgrowing the rest.30 Functionally, assuming that any unspent income below the requirement is confiscated, a spending minimum is just a somewhat ham-fisted 100% tax on saving above some bar. But if enforced, it would certainly limit income divergence; and its simplicity, and the fact that it need not be framed as a tax, may be valuable. Likewise, though a maximum spending rate is essentially unheard of today (except in the sense that people cannot withdraw their social security while they are still young), in the absence of other redistributive measures it might be necessary to prevent the bottom end of the income distribution from collapsing, via people permanently spending down the assets on which they and their heirs will have to rely forever.
If the policymakers of the future succeed in capping income inequality—putting a permanent upper bound on, say, the ratio of world’s highest income to the median—then the future could belong to those with the highest birthrates after all, as opposed to the people with the attributes listed at the end of Section 3. In fact, the fertile would inherit the earth even more straightforwardly than today. Today, having more children leaves each with a smaller inheritance and thus lower capital income. More significantly, while most income comes from labor, parents face what is sometimes bluntly called a “quality-quantity tradeoff”: more children means each child gets less investment in human capital, and thus lower labor income.
Whether this would be a desirable or undesirable development is up for debate, and well beyond the scope of this already rambling essay. A widely celebrated byproduct (and cause) of the Industrial Revolution was the shift in wealth and ultimately political power away from a hyper-conservative, decadent landowning aristocracy to a bourgeois middle class that prized experimentation, thrift, and hard work. For better or worse, once the robots are doing all the work, wealth and power will shift again. Whether the optimal distribution of capital in the 22nd century is equal or otherwise, we can agree with Piketty that it may not emerge by default.