Who Owns the AI Economy?

As AI shifts income from labor to capital, we need to rethink taxation and redistribution. A levy on large firms, paid in their own non-voting shares to a public benefit trust and enforced through market access, sidesteps the failures of past wealth taxes. The proceeds could be used to pay every citizen a dividend that scales with AI's advance. The international club that enforces the levy could also give AI governance a mechanism to temper the AI arms race.

Introduction

A PDF version of this post is available for printing.

When industrial machinery displaces an assembly line worker, or software displaces a call center worker, a part of their salary becomes income for the capital owner. Historically, this shift was tempered by two counterforces. Automation created new jobs that machines could not do, and human work became more specialized and better paid . Even so, labor’s share of income has been slipping for four decades (Figure 1). If increasingly capable AI comes to compete for most work, those counterforces may lose their strength, leaving workers with an ever-diminishing portion of the pie .

As this trend continues, two problems arise. The immediate one is distributive: capital concentrates and compounds in the hands of the few.Three mechanisms contribute to the centralization of capital: (a) compounding: capital income gets reinvested and grows faster than labor income ; (b) economies of scale: large firms accumulate advantages of scale in distribution, brand, network effects, data, and talent; and (c) near-zero marginal cost: intangibles and automated services cost almost nothing to replicate, so marginal revenues in the digital economy flow almost entirely to capital. At the individual level, wealthy households tend to earn higher rates of return on their investments , so their wealth compounds faster than everyone else's. Already, in the U.S., the top 1% of households hold roughly half of all corporate equities . The less obvious problem is fiscal. Roughly seven-tenths of U.S. federal revenue is derived from labor ,Payroll taxes supply about a third of federal revenue and fall entirely on labor. The individual income tax supplies about half. Wages and salaries account for roughly 66% of reported income and retirement-related income another 13%, so the figure allocates roughly 80% of individual income-tax receipts to labor-derived income . Combining that share with payroll taxes puts the labor-derived share of federal revenue at roughly 70%. This split is an estimate, as CBO reports revenue by tax, not by labor and capital. so as labor’s share of income shrinks, the tax system’s reach over the economy shrinks with it. The gap in Figure 1 tells the story: labor earns a falling share of total income, but still pays its historical share of the total tax.The fiscal version of this problem, a tax base that erodes as labor income shrinks, is set out in Korinek and Lockwood's public-finance framework for AI .

The AI-safety literature has reached the distributive problem from its own direction: the "windfall clause" would have leading labs pre-commit a share of extreme profits to the public , and others include a public ownership stake in highly automated firms among the measures for keeping power dispersed .

Figure 1. Labor's share of national income has fallen from roughly 63% in 1980 to roughly 55%, while the estimated labor-derived share of federal taxes has remained near or above 70%, rising from roughly 68% to roughly 74%. The decline appears across most countries and industries . The labor share is the BLS nonfarm-business measure, anchored at 56.6% in 2017 . The tax share is calculated from federal receipts as payroll taxes plus roughly 80% of individual income taxes, allocating the latter in proportion to the reported-income share from wages and retirement distributions .

The proposal below starts with a 1% floor aimed at the concentration and fiscal imbalance already present, and the rate rises only if labor’s share falls further. The fund would take decades to mature, so workers displaced early would still need separately financed transition support.

The tax system’s focus on labor income is not without reason: taxing capital correctly is hard. While taxes on income and realized gains from capital are administrable, the ability of taxpayers to defer realization, reinvest, and move capital across borders limits how much these taxes can raise. A natural next step is a wealth tax, which taxes the value of the capital directly, regardless of when it produces income or where that income is booked.

But three problems stand in the way . The first is valuation: income from capital is forward-looking and uncertain, making fair valuation of capital difficult (and easy to game). The second is liquidity: taxing unrealized capital gains may force early sales, which are economically inefficient, or even infeasible. The third is capital flight: because tax rates vary by jurisdiction, capital owners can change their tax domicile or move capital offshore, avoiding the tax entirely. Together, these issues have contributed to the repeal of wealth taxes in most of the countries that have tried them.Of the twelve European countries that levied a recurrent net wealth tax on individuals in 1990, nine had repealed it by 2020, leaving only Norway, Spain, and Switzerland . France's ISF, the most prominent casualty, was replaced in 2018 by a tax on real estate alone, after years of wealthy taxpayers relocating abroad. The surviving regimes run at rates low enough (Switzerland's cantonal rates range from roughly 0.1% to 0.9%) to function closer to property taxes than to the redistributive instrument proposed here.

Recent U.S. proposals face the same problems. Warren's and Sanders's individual wealth taxes require the IRS to value every private business, piece of real estate, and art collection owned by anyone above the threshold, annually and under adversarial conditions. A cash obligation on illiquid wealth forces asset sales, and wealthy individuals can change their domicile (e.g., Facebook's co-founder Eduardo Saverin renounced U.S. citizenship before the company's IPO). Mark-to-market proposals (Wyden's Billionaires Income Tax, Biden's minimum tax on unrealized gains) are only slightly less impractical. While they avoid the deferral problem for publicly traded assets by taxing gains as they accrue, they still demand cash for paper gains, still struggle with assets that do not trade (valuing them each year or falling back to deferral charges), and still target individuals who can relocate.

A liquid wealth tax

What if, instead of a cash tax on wealth, we required large corporations to pay a tax denominated in newly issued shares? On its face, such a tax would amount to a naturally progressive, valuation-free, indirect tax on individual wealth that defers the liquidity issue until later.The tax is progressive because equity ownership rises more than proportionally with wealth. But note that much of the half of corporate equity held outside the top 1% sits in pension and retirement accounts, which would be diluted at the same rate. The tax can be designed to offset this by compensating retirement accounts proportionally, as discussed under Open design questions. I will call this a liquid wealth tax.

The idea is natural enough that multiple people I’ve spoken with have reinvented it with minimal prompting: forced sales for a cash tax would only crash the market, so tax the shares themselves. In the literature, Saez and Zucman worked out the economics of such a tax in a 2021 proposal , and many, including Sam Altman, have floated similar ideas in the past.Related proposals. The broad direction (distributed capital ownership and a citizens' dividend) is well populated. Meade proposed a state-held capital fund paying a social dividend as early as 1964, partly in anticipation of automation , and Atkinson revived the idea . In the tax-law literature, Galle, Gamage and Shanske reach the same share-denominated logic, with governments taking notional equity interests in lieu of cash . The closest policy cousins are public funds that hold corporate equity and pay citizens, several funded partly by a "scrip levy" (a tax paid in shares): Lansley, McCann and Schifferes's 0.5% annual share issue , the IPPR's scrip tax of up to 3%, which pays a one-off capital dividend to young adults , and Bruenig's U.S. social wealth fund, which combines a market-cap levy payable in cash or shares with leveraged purchases . Other cousins include Dean Baker's plan to swap the corporate income tax for a non-voting public equity stake , Varoufakis's universal dividend funded from a share of public offerings , the UK Labour "inclusive ownership funds," which would have had large firms transfer 1% of equity a year into a worker-controlled, voting fund , and Stout, Belinfanti and Gramitto's donation-funded Universal Fund, which gives citizens shares and a voice in how the fund votes .

In the AI debate, Sam Altman's "American Equity Fund" is the closest of all to this post: a levy on large companies paid in their own shares into a fund that pays every citizen a dividend. Altman also considers avoidance, the treatment of private firms, dividend sizing, and a constitutional cap on the rate . David Shapiro's "post-labor economics" similarly emphasizes broad ownership . Marinescu pairs displacement insurance with a small universal "digital dividend" funded by taxes on the digital sector, built to scale up if job scarcity becomes permanent . The discussion has continued while this post was being written. Senator Sanders's initial 50% equity tax on the largest AI firms and Anthropic's unemployment-indexed policy framework are taken up under Other proposals below.^,What this post adds. Besides the synthesis of past work, this post presents my version of the idea in a concise and self-contained form. Saez and Zucman treat an international agreement as the primary route, noting that a country acting alone could give others an incentive to follow. This post goes further, arguing that the United States has enough market power to effectively "force" others to join the club, overcoming the delays of negotiation and benefiting everyone. It also offers commentary on constitutionality, on the ways firms might try to avoid the charge, and on the governance design that keeps a growing public stake from becoming control of the firms it holds. Finally, I analyze the capital requirements and timelines for funding a sufficiently large universal dividend, propose indexing the levy's rate to labor's share of income so that the tax keeps pace with the problem it is meant to address, and discuss how the club structure the wealth tax creates could carry over to the enforcement of international AI governance.

A share tax largely sidesteps the valuation issue: 1% is 1% regardless of the company’s value."1% non-voting equity" still needs implementing rules for dual-class shares, preferred stock, outstanding options, and partially owned subsidiaries, and covered firms may need to amend their charters to authorize the new non-voting class. Valuation re-enters only at the edges: constitutional considerations in the U.S. may require a cash alternative, private companies lack a daily price, and foreign private groups may only be able to satisfy the obligation in cash.In the U.S., the Constitution requires "direct taxes" to be apportioned among states by population, which would make a uniform-rate wealth tax unworkable if it were deemed a direct tax. In Pollock v. Farmers' Loan & Trust Co. (1895), the Supreme Court treated the 1894 tax on income from property as a direct tax and struck down the broader income-tax scheme on these grounds. Congress responded with the Sixteenth Amendment (1913), which authorized income taxes without apportionment but did not address wealth taxes. This is the gap that recent individual proposals try to work through. The mark-to-market plans (Wyden, Biden) attempt to reach unrealized gains through the income tax, by marking publicly traded assets to market each year and recognizing the gains. Warren's and Sanders's taxes on net worth face the direct-tax question more squarely, and attempt to make the argument that a tax on wealth is not a "direct tax" in the narrow sense the Constitution intends.

The most plausible path for a liquid wealth tax is to structure it as a corporate excise on the privilege of doing business in corporate form, following Flint v. Stone Tracy Co. (1911), which upheld such an excise measured by income. Measuring it by equity value rather than income is an extension Flint does not directly address, but the underlying logic (taxing a corporate privilege, not property itself) is the same. Most recently, Moore v. United States (2024) upheld a tax on a foreign corporation's realized but undistributed earnings, attributed to U.S. shareholders, while declining to rule on broader taxes on wealth or appreciation, leaving the constitutional question open.

Conditioning market access on payment of a share tax could also face a takings challenge, especially if the stock transfer is viewed by courts as a compelled transfer of property rather than a tax payable in cash. In Horne v. Department of Agriculture (2015) the Supreme Court found a taking even where growers kept a contingent claim on the proceeds, and conditioning a government benefit on surrendering property raises an unconstitutional-conditions question of its own. Both are reasons to preserve a cash alternative.

The path to constitutionality is plausible, but a challenge is likely. If the courts ultimately classify the liquid wealth tax as a direct tax, Congress would likely need to enact a constitutional amendment, the same route it took after Pollock to enable the income tax. These pricing concerns shrink when the tax is focused on very large corporations (e.g., over $1 billion enterprise valueTo avoid perverse incentives, the tax should phase in over some range, rather than activate based on a hard cliff. For example, from zero at $1 billion of enterprise value to the full rate at $10 billion. For simplicity, the figures in this post use a $1 billion cutoff, which overstates receipts by the discount owed to firms inside the band (on the order of a few percent of the estimated tax).): public firms have a daily price, large private ones are priced by recent funding rounds, and the administrative burden is concentrated on a manageable number of entities.

The liquidity issue is likewise deferred. A covered firm does not have to raise cash or sell assets to issue new non-voting shares, so a force-selling spiral is avoided. The cash question comes later, when the public fund that receives the shares needs cash to pay a dividend or finance public spending. Much of the share-to-cash conversion could run through machinery that already exists: dividends and buybacks by the share issuers. Over 2020–2024, the largest U.S. firms returned roughly $1 trillion or more a year to shareholders through buybacks and dividends combined , well more than the roughly $500 billion a 1% tax would collect on the roughly $50 trillion of covered equity before any repricing (a conservative base, revisited below).

Although the annual dilution is small, markets would price a permanent levy all at once, marking covered shares down today by the value of all the future dilution. The markdown would be on the order of 15 to 25% at a 1% rate, borne by whoever holds covered equity when the tax becomes a credible prospect.A permanent annual dilution of τ reduces per-share value by τ/(r − g + τ), where r is the required return and g is the growth rate. At τ = 1% with r − g between 3% and 6%, this gives a repricing of roughly 15% to 25%. The repricing would begin as soon as the tax becomes a serious legislative possibility and adjust as the probability and terms become clearer. The loss would fall on all holders, including the roughly half of equity held outside the top 1%, much of it in pension and retirement accounts. The Open design questions section returns to this, including how retirement accounts can be held harmless. After that adjustment, a new buyer would face the same market-required return on each remaining share as before . What changes is who holds the capital: each year the fund accumulates more equity, so the public’s share of the economy’s wealth rises even as the per-share return is unchanged.The total economic cost may go beyond the repricing if the tax distorts behavior: firms may invest less, leverage more, restructure to avoid coverage, or pass part of the cost through prices and wages. The broader the tax base, the smaller the distortions from restructuring, but the capital allocation decision (reinvest vs. return capital to shareholders) for covered firms would be distorted regardless. This distortion limits the practical size of the share tax (a bit more on this below).

Figure 2. Each year, large corporations issue new non-voting shares (e.g., 1% of equity) to a public fund. Firms may pay the tax in cash, or buy the shares back if they prefer not to dilute their holders. Proceeds could be used to fund a public dividend or other public spending.

The serious objection to any capital tax is that it lowers the expected return on new investments and so less gets built. But the tax would reach only firms above a high threshold, and by the time a firm arrives there, the investment that built it has already succeeded.Founders and early investors still price expected exit values, so some effect on new company formation would remain. Large firms also tend to earn economic rent, i.e., the return that flows from scale, network effects, and market position rather than from marginal new investment . Rent can be taxed with less effect on investment, since it is by definition the return in excess of what the investment required.A standard result in optimal tax theory (Chamley-Judd) holds that the long-run tax on the return to capital should be zero. It rests on a model where the government can still raise revenue from labor or consumption, and it applies to the return on new investment rather than to the rents described above. As AI erodes the labor base, capital becomes harder to leave untaxed, and the part that is rent may be taxed with less effect on the new investment the result is meant to protect. For a formal treatment showing the zero-tax conclusion is less general than often assumed, see Straub and Werning . A related concern in open-economy tax analysis is that a charge on corporate capital is partly shifted onto wages, because capital can escape the charge by relocating or respond to it by investing less. Market-access enforcement limits the first channel: a base that cannot leave cannot shift its burden by leaving. The second channel remains, including the incentive to pay earnings out rather than reinvest them, which is taken up below.

Another incentive distortion would occur inside the covered firms themselves. Value retained in a covered firm bears the dilution every year, while value paid out to shareholders escapes the base, so the levy leans on large firms to pay their earnings out instead of reinvesting them. This is the effect I would watch most closely.

That leaves capital flight, the hardest of the three problems. By taxing wealth through corporations rather than individuals, we can overcome capital flight by making the tax conditional on market access. While a firm can move its capital, or even its founder, it cannot move its customers.

From market access to a global club

Enforcement through market access inverts the asymmetry that doomed earlier wealth taxes. The U.S. alone accounts for nearly a third of global household consumption . That gives it leverage over any firm with material U.S. sales, as walking away from the market would give up too much value.An obvious avoidance play is to split the company in two: hold the valuable part (the AI model and its intellectual property) in a privately held shell offshore that makes no direct sales, while a thin-margin local subsidiary does the selling. The charge reaches the subsidiary, which is worth almost nothing, while the valuable parent sits beyond it. The policy response is to treat the corporate group as a single entity, following OECD Pillar One , the 2021 framework for reallocating taxing rights over the largest multinationals' profits to the markets where their customers are.

This concentration of consumption creates an opportunity for one large market, the U.S., to go first in a way that incentivizes the entire world to follow, without waiting for a treaty or a negotiated global agreement. As first mover, the U.S. would have the unilateral ability to set the amount of the tax. How it uses that power decides whether the rest of the world follows or fights. Taxing every firm that sells to Americans the maximum amount it can without having them exit the U.S. entirely, keeping all of the money, and paying only Americans would invite a trade war. The version that works is an offer, which extends to every country the same claim the first mover makes for itself. The U.S. collects the share that matches sales in the U.S., France the share that matches sales in France, and India the share that matches sales in India. A foreign company selling to Americans is treated the same as an American company selling to Americans, and an American company selling abroad is treated like any other company selling there.

Since going first is not without its costs,The mover's own firms are repriced first, so its shareholders bear the markdown years before the dividend arrives to repay them, and it draws the first retaliation, before any ally shares the burden. and the goals of the wealth tax are best served if all countries cooperate, it makes sense for the first movers to reach beyond their own share, into the shares of countries that have not yet followed suit, up to a cap.If uncapped, firms with marginal club sales would face a charge on their whole equity, and might rationally exit the club's markets instead. This additional reach can be used to repay the movers for going first, as an added incentive for other countries to join the club, or both.

Thus, a country that joins collects its share, but a country that holds back does not shield its firms from the tax. So long as those firms keep selling into the club’s markets, they must pay the share matching their home-market sales as well (up to the cap).Saez and Zucman proposed domestic apportionment as a fallback, with the market country taxing a foreign company in proportion to its local activity when the company's home country charges no equivalent tax . Rather than a fallback, the design proposed here always taxes corporations based on local sales, and further, proposes to tax them on foreign sales as well if their home country does not. What becomes of these non-member shares is a policy choice: they could be escrowed for the absent countries and released when they join, or they could be claimed by the members (perhaps after a grace period).The OECD's global minimum tax, agreed by more than 135 countries in 2021, was built to work this way through its undertaxed-profits rule. If a multinational's home country does not tax its profits up to the agreed floor, the countries where it operates may collect the shortfall, which gives the home country a reason to levy the tax itself and keep the revenue . The European Union's carbon border charge runs on the same logic in goods, crediting a qualifying carbon price a producer has already paid at home . The precedent is two-sided, though. When the undertaxed-profits rule threatened to reach U.S. multinationals, the United States threatened retaliatory taxes on the countries applying it, and in January 2026 the OECD adopted a "side-by-side" arrangement that keeps the rule from applying to U.S.-headquartered groups (though domestic minimum top-up taxes still reach their local operations) . The collect-or-be-collected mechanism works, but it cannot survive defection by the largest market. That cuts in favor of the design proposed here, where the largest market is the club's founder rather than its holdout, but it is also a measure of how much depends on who goes first. Nor is a hostile counter-coalition the threat it first appears. A coalition that could block the club would need the European Union at its center, and the club offers the European Union what it has sought for a decade, a way to tax the largest (mostly American) firms with American blessing rather than American retaliation. Under either arrangement, the holdout’s firms are charged while it waits, and joining determines when the country receives the proceeds and whether the existing members retain any of them. The result is an incentivized club (Figure 3Assumptions behind Figure 3. The figure applies a 1% levy to the end-2024 market value of publicly traded companies. It uses each market's total listed value as a simplifying proxy for the value above the $1 billion cutoff. U.S. index data indicate that companies below the cutoff make up only a few percent of listed value, although comparable evidence is not available for every market. The model groups listed value into six market buckets and assigns each a constructed sales pattern based on published revenue data and company disclosures. Many disclosures report only broad regions, so judgment is required to divide them and the country figures are approximate. The cap is applied to each group's sales pattern. Full sources, data notes, and code are on GitHub.) that countries are free to join but firms cannot escape.

Figure 3. Who would collect under the club, shown at the levy's 1% floor using EOY 2024 market values. Each member collects the share of every large firm's annual charge that matches the firm's sales into that member's market. Shares that match holdout markets are collected by the club, up to a cap, and divided among the members proportionally. The policy caps collection from any one firm at three times its member-market sales slice. This regional illustration applies that cap to the constructed sales pattern for each group, which is why part of China's charge would stay uncollected until it joins. Modeling assumptions are summarized in the preceding footnote , and the simulation code is available on GitHub.

Explore the club →

While much of the dilution would be borne by the shareholders of foreign firms, that asymmetry cuts both ways, with foreign investors holding close to a fifth of U.S. corporate equity . This introduces some political risk for the U.S. as first mover. If foreign shareholders bear part of the dilution while only U.S. citizens receive the dividend (before anyone else joins the club), other countries may treat the levy as discriminatory and push back, similar to how the United States opened trade investigations and threatened tariffs when European countries taxed American technology companies in 2019.When France and several others taxed the revenues of the largest (mostly American) technology firms with digital services taxes beginning in 2019, the United States opened Section 301 trade investigations and threatened retaliatory tariffs of up to 100% on French goods . The suspended duties were terminated in 2021 after an agreement tied to the OECD's multilateral negotiations . The precedent points in both directions here, showing how much force market access carries and previewing the reaction a club founder should expect from countries whose investors bear the charge. It also shows why the founder matters. Retaliation has real bite against a midsize economy, but much less against the market with nearly a third of world consumption. To work, the design must weather a (hypothetical) trade fight as well as the domestic political one.Retaliation cannot directly shield a foreign firm from levy, so the point of retaliating is to change the founder's politics. The natural pressure points are the founder's own companies. The American firms with the deepest revenues in China are also among the loudest voices in Washington, and licensing delays, security reviews, and quiet boycotts of foreign firms are a practiced playbook there. The dangerous window is the one between credibility and enactment, when the repricing has arrived, the dividend has not, and the threats are loudest. Retaliation of this kind is costly for the retaliator, which is why it is used sparingly. Once others join the club, however, the discrimination argument and asymmetric treatment are largely defused (see Figure 3).

One asymmetry survives full membership. American portfolios hold roughly half of the world’s covered equity and would therefore bear about half of the initial worldwide markdown, while the sales matrix gives the U.S. fund about 37% of collections. In distributional terms, this makes the levy resemble a source-based tax more than a strictly domestic individual wealth tax: collections follow customer markets rather than investor residence. This asymmetry does suggest, however, that a politically viable bargain may have to give the United States more than its sales share, e.g., through a temporary founder’s credit, a modest ownership-based component, or some combination of the two.An ownership-based component remains conceptually possible without changing how the levy is enforced: firms could still owe the same charge under the sales and market-access rule, while the member funds divide some of the proceeds according to where the firms' investors live. That would track the initial markdown more closely, but beneficial-ownership data are not consistently available at the required global scale, holdings can be obscured through funds and custodians, and residence can move. It would also reward the countries that already own the most capital rather than the markets that make the club work. A cleaner solution might take the form of a temporary founder's credit, funded from the residual shares collected on holdout markets. The credit could run for a fixed period or until it reaches an agreed present value.

Charging for market access is not new. A tariff does exactly that, but it raises the price of goods at the border, so much of the cost lands on the taxing country’s own consumers. Further, because it targets goods by where they come from, it invites retaliation.Studies of the 2018 to 2019 U.S. tariffs found their cost passed through almost entirely to American importers and consumers rather than absorbed by foreign exporters , and the retaliation that followed fell heavily on U.S. exporters, especially agriculture. A charge applied equally to domestic and foreign sellers has no border price wedge and gives trading partners no origin-based discrimination to retaliate against. That neutrality is analogous to the logic of broadly adopted destination-based taxes . The levy avoids both problems. Nothing happens at the border to raise prices, and because domestic and foreign firms are charged alike, no country’s products are singled out.The United States is currently using market access in the opposite spirit, with unilateral tariffs and a carve-out of its own firms from the global minimum tax. The design asks it to make market access an offer where it has lately made it a weapon, and the club is a chance to recapture some of the political capital those moves have cost.

A closer precedent than tariffs is a law that used U.S. market access to change the behavior of firms worldwide. In 2010, the U.S. Foreign Account Tax Compliance Act (FATCA) required essentially every bank in the world to report on its American account holders. The alternative was a 30% tax withheld from their U.S. payments. Since access to the U.S. financial system mattered more than the burden of reporting, the banks moved to comply. Where home-country secrecy laws stood in the way, intergovernmental agreements struck after the fact cleared the path.

Even if the club never forms, the United States would still collect from its own covered firms, and the dividend arithmetic in Figure 4 uses that domestic base alone. Failure would sacrifice the global reach and AI-governance extension, but the mechanism would retain its standalone value as a wealth tax.

Extending the club to AI governance

The tax gives members a continuing reason to sustain the club, and the same machinery might support a second bargain, this time over AI governance. International cooperation on AI is stuck on a familiar problem. Commitments are voluntary, and every proposal to set standards or slow down runs into the same objection, that restraint by one side hands the lead to the other. The existing tools do not answer it. Export controls and sanctions punish, but a punished state’s best answer is to build its own supply, so the stick weakens each time it is used. Voluntary commitments ask for restraint and offer goodwill in return. By contrast, by paying countries their share of the tax, club membership could be used as a direct incentive for cooperation.

To use the club for AI governance, members would impose and enforce a common set of conditions on every firm selling into the club, wherever the firm is headquartered."Selling into the club" would need a broad definition, covering API access, cloud deployment, and app distribution, not just goods crossing a border. The conditions might include safety standards for the most capable systems, reporting on the largest training runs, and limits on letting AI systems act on their own in high-stakes settings. Conditioning market access on compliance has already been proposed for AI governance on its own . The tax brings the machinery and the motive, since a club built for taxation already tracks which firms sell where and what they must do to keep selling, and its revenue gives members a continuing reason to stay in.

How much governance the club buys depends on whether the relevant states join, and joining is binary. Market access reaches commercial deployment more readily than frontier training, military systems, or state programs that need not sell anything into the club. For a state that stays out, the club’s conditions reach only what its firms sell into member markets, and, depending on the club that forms, that may not be much. China’s frontier labs sell little into the West today (their fast-growing export revenue is, by most accounts, concentrated in Southeast Asia and the Middle East). So a club without China might buy little of the governance that matters, and the race would continue. The club’s own conditions would be written with that in mind, since members would not handicap their firms against a rival that accepted no constraints.

For a state that joins, the club offers that state’s share of the levy in exchange for its cooperation on enforcing governance on its own firms. For China, the fact that the same governance conditions would bind American frontier labs could be a substantial benefit in itself. The economic benefit also compounds with time, since the levy’s rate, as proposed below, rises with AI’s advance, so the payment for membership grows alongside the capabilities that make the race dangerous. If the payment and reciprocal restraint clear each state’s strategic bar, the members can stop racing each other and start watching each other for cheating, a different and older problem. While the club would not solve it, every member would be on the side of wanting it solved.

Coupling the wealth tax with AI governance is not without costs. Bundling gives an opponent of either piece a reason to fight both. The tax revenue arrives as soon as a country joins, while the strategic benefit of reciprocal restraint is less certain, so countries are incentivized to seek the revenue without accepting the conditions. The conditions would therefore have to be a prerequisite of joining, and a coupled club would grow more slowly than a tax-only club. Since nothing in the tax case depends on the coupling, the choice can be made after the club exists.

Public ownership without control

A public fund taking 1% of corporate equity each year raises a fear that it would slowly nationalize the economy. But even at a 1% rate, the fund would hold around 18% of covered equity after twenty years and 39% after fifty years, and these are upper bounds that assume the fund never parts with a share. In practice, buybacks by the issuing firms, occasional sales to fund the dividend, and new firms entering the market would pull the stake lower.

For perspective, labor currently earns around 55% of income (Figure 1). If AI shifts a large part of that income to capital, a fund holding a minority stake after decades of accumulation would claim only a fraction of the returns that used to flow to wages.

If we want to fund a sufficiently large dividend, significant public ownership cannot be avoided. Even a fund that passes much of the levy through to citizens as it arrives would come to hold a substantial share of covered equity along the way. In the simulation behind Figure 4 (below), the fund holds 14 to 22% of covered equity after fifty years, depending on the scenario.

For ownership on that scale to be politically and economically sustainable, the fund’s shares must carry no votes, and thus no direct control.Where a share class cannot legally be stripped of votes, the fund can mirror-vote by casting its votes in the same proportion as other shareholders so it minimizes its effect on voting outcomes. Economists worry that when one investor holds stakes in every firm in an industry, competition can soften, since the investor profits from all the rivals at once . That effect depends on the investor being able to influence management, through votes, board seats, pressure on managers, or the threat of selling a large stake. A non-voting public fund that holds its shares in trust has none of those levers (cf. on pure passivity and mirror voting). In economic terms, it is closer to the corporate tax, which already claims a share of every firm's profits, than to an institutional investor like BlackRock or Vanguard. Even so, residual channels remain (litigation and disclosure rights, class votes on some corporate actions, political pressure on managers), and the design has to keep them closed too. While the design cannot make public ownership at this scale apolitical, it can remove the formal channels that would turn the fund into an industrial-policy board. The fund’s assets would be held for citizens and kept outside the ordinary budget, with protections placed at the constitutional or charter level and optionally reinforced through the club, making it harder for a later majority to redirect the money.Sweden's wage-earner-fund debate provides a cautionary precedent. Economist Rudolf Meidner proposed that large firms issue voting shares to union-controlled funds, which would over time gain ownership control of the firms. Businesses treated the project as a slow-motion takeover, which led to the funds becoming politically toxic. Even the much weaker, tax-financed version of the plan that was enacted in 1983 was abolished within a decade .

A public dividend

Early on, the fund’s accumulated holdings would start small and compound slowly, so its natural first use is insurance. As AI displaces particular workers and industries, the fund can cushion the transition, paying those most directly displaced while displacement is still identifiable . The first years might also benefit from a second, temporary displacement tax. Firms whose headcount falls while their AI spending or sales grow, for example, might pay it directly to the displaced, or into the fund itself. Either way, the tax would retire as the fund grows large enough to carry the insurance on its own.That displacement tax has its own design problems (what counts as AI displacement, how to keep it from being gamed). An alternative bridge could involve fund borrowing against future equity receipts, which arrive according to a statutory formula. This introduces leverage and political risk, but avoids the administrative complexity of identifying displacement.

Displacement insurance has a shelf life. After enough displacement, it becomes impossible to say who was displaced. Early on you can point to the call center worker and the truck driver. But as AI diffuses into everything, we will have graduates whose entry-level job never existed, businesses that never formed because software did the work, and wages lower than they would have been. Insurance needs a victim you can name, and past some point there are no nameable victims left, just an economy producing more with less labor.

Instead, once the fund is large enough, it makes sense to pay a per-capita dividend. The dividend would be the same amount for every citizen, with no test and no claim to adjudicate. While a means test or variable dividend might still be administrable, it would give up the simplicity and the universal constituency that protects the fund politically (see Protecting the fund below). To keep the net transfer progressive, the dividend could be paid as ordinary taxable income, since the income tax claws back the most from those who need it least.

For the dividend to reach meaningful size, the fund should retain most of what it collects. Even as buybacks and dividends provide a stream of cash receipts, reinvesting that cash in the broader market allows the fund to compound while it is small. The dividend is then funded according to a preset schedule from the fund’s flows, the yield on its accumulated holdings plus a capped fraction of each year’s receipts.Even at maturity the scale stays manageable. The fund's net receipts in the Figure 4 scenarios remain below two-fifths of what covered firms pay out each year through dividends and buybacks, so the mature dividend can ride the payout machinery that exists without forcing net sales of stock. As the labor tax base erodes, the same flows could fund public spending alongside the dividend, restoring part of the reach the tax system is losing. The principal itself would be locked in trust, so a legislature short of cash could spend from the fund’s income but could not sell down the holdings that produce it.

Growing the dividend

How large could such a dividend get? It depends on how high the levy can go. As noted above , the levy’s cost arrives up front, through a one-time markdown of roughly 15 to 25% for a permanent 1% rate and deeper at higher rates. The market-required return on equity is the same after the markdown, so the ceiling is set partly by economics, since avoidance and distortion grow with the rate, and partly by politics, since today’s holders will not accept an excessive markdown.

Wealth holders will fight an outright loss, but parting with a share of an AI-supplied windfall is a different matter. If AI displaces enough work to move a large share of income from labor to capital, the gains to shareholders would be far larger than the markdown, so a much higher rate could go through with holders still better off than they are today. And the arithmetic is not the only consideration, since the worlds that hand shareholders the windfall are also the worlds where an economy of displaced workers puts every asset at risk.

This discussion suggests the rate should be indexed rather than fixed, rising automatically as labor’s share of income falls.The direction has grounding in the economics literature: the U.S. tax code already taxes labor much more heavily than capital, which subsidizes automation , and in calibrated models of automation, simply maintaining government revenue requires the capital tax rate to rise as labor's share falls . On the optimal-tax side, Atesagaoglu and Yazici find that a falling labor share justifies rising taxes on capital income only when the decline reflects a rising share of pure profits , which is the mechanism the rent argument above describes. The tax then scales with the problem it exists to solve. Because the higher rates apply only in futures where shareholders capture more income, markets would price less than the maximum levy into shares today.The index would need a precise legal definition. Labor's share is published, but its historical path has changed when statistical agencies changed how they treat items such as intellectual-property investment . The statute should name the series and responsible agency, use a multi-year average, and calculate each year's rate from the data then available. Once set, that year's rate would not be reopened by later revisions. If the agency changes its methodology, the trigger would be rebased prospectively so that the change in accounting convention is not itself treated as a shift from labor to capital. An international club would also need a common rule for the index because national labor shares follow different paths and statistical conventions. Members could use a harmonized club-wide series or agree on a common method for translating national series into the rates applied in each market.

The levy already varies with outcomes in one sense. Even at a fixed percentage, its dollar value rises and falls with covered firms' market capitalization. Tying the percentage to labor's share adds a second adjustment. Shareholders give up a larger percentage when more income has shifted from labor to capital and a smaller percentage when the shift is limited. This should narrow the range of their after-tax outcomes and reduce the enactment markdown relative to a fixed levy expected to collect the same amount. The dividend provides similar insurance to households by growing most when wages fall furthest. A simulation of the fund’s accumulation shows the shape of the path (Figure 4).Assumptions behind Figure 4. All amounts are real 2026 dollars. Covered-firm payouts begin at $1.5 trillion, equal to 4.69% of a $32 trillion economy. GDP grows by 1.8%, 3%, or 4% across the three scenarios. Labor's share moves from 55% toward 45%, 35%, or 30%, and half the resulting income shift from wages becomes covered-firm payouts. The model discounts projected payouts at 5.5%, 6.25%, or 7% to estimate market value, with permanent future share issuance included in the valuation. The annual levy starts at 1% and rises with the labor-share decline, subject to a 4% cap. Avoidance reduces issuance and dilution together by 12% to 18%. Retirement protection routes part of each year's issuance to recognized funds and fades as enactment-date holdings run off. Opening holders receive each year's corporate payout before that year's shares are issued. The public fund initially retains most of what it receives and gradually shifts toward paying through its corporate income and new share receipts. A simpler cross-check holds listed market value at its current share of GDP and produces about $44,000 per person in strong AI at year fifty, compared with about $52,000 in the model shown. Full equations, an alternative model, and the simulation code are available on GitHub.

Figure 4. Illustrative annual public dividend per U.S. person under a tax indexed to labor's decline, with all amounts expressed in real 2026 dollars. The model discounts future covered-firm payouts into market value and includes the dilution from future share issuance. Under strong AI the dividend reaches about $52,000 by year fifty and crosses the $35,000 reference line around year 41. The dots mark the fund's stake in covered equity, about 17% after twenty-five years and 22% after fifty. The interactive version also shows the same world with the tax fixed at 1%, which produces about $27,000 at year fifty, and lets readers vary the assumptions. Full equations, an alternative model, and simulation code are on GitHub.

Explore the assumptions →

Protecting the fund

A fund this large is a standing temptation for a government short of cash. Defending it is the second reason, beyond the impossibility of singling out the displaced, to pay everyone the same dividend. Alaska’s Permanent Fund pays every resident a dividend (with no reduction in employment, though a small shift toward part-time work ), and for decades that dividend was the most untouchable line in the state’s budget, because every voter received it and every voter would notice its loss. That is also an argument for starting with a small dividend early on. The scenarios in Figure 4 pass a quarter of each year’s receipts through from the first year. While the first checks would be small, a few hundred dollars a person, they would create the constituency during the decades when the fund is most vulnerable and least able to defend itself.

But a constituency defends the fund better than it defends the rules around it. When oil revenue fell after 2016, Alaska legislators cut the dividend through the appropriation process anyway, and the state’s supreme court held that the statutory formula could not bind the legislature’s power. A hundred million people who want their dividend check larger are also the coalition that would vote to raise the rate, spend down the principal, or restore the voting rights of their non-voting shares. So the voting limits and the lock on the principal need deeper entrenchment than a statutory formula. The proposed design supplies some of this on its own. The share tax collects shares of a non-voting class, so the ban on control is written into the securities themselves. And if the fund is a genuine trust whose beneficial interests vest in individual citizens, the accumulated principal might be considered their property, so a government that later reached for it may run into the Takings Clause.At the federal level a charter is itself a statute, and one Congress cannot prevent a later Congress from changing it through ordinary legislation. In Flemming v. Nestor (1960), the Supreme Court held that Social Security benefits were not accrued property rights, which is why the trust would need to create a much clearer vested interest. Clear drafting, an independent trustee, and amendment rules could add legal and political friction, but the constitutional protection would remain uncertain.

Vesting creates a boundary problem of its own. The class of citizens changes through births, deaths, migration, and naturalization. Property rights strong enough to protect today's beneficiaries may also make it harder to admit tomorrow's on equal terms. The trust would need to reconcile individual vesting with a beneficiary class that changes forever. Taken together, these provisions offer practical and political protection for what has accumulated, while the levy itself remains exposed because no one holds a vested right in taxes not yet collected. Protection of the tax going forward would come from the global club.

Other proposals

While this post was being finished, versions of the idea arrived from three directions at once. OpenAI reportedly discussed giving 5% of its equity, once, to a prospective U.S. sovereign wealth fund, with Altman suggesting other leading AI firms do the same.CNBC reported the discussions in July 2026, citing the Financial Times . The talks were preliminary, and the fund does not yet exist. Altman and other OpenAI executives have suggested that America's leading AI companies allot 5% of their equity to a vehicle modeled on the Alaska Permanent Fund, discussed above. At OpenAI's reported valuation of roughly $850 billion, a one-time 5% is about $43 billion; a 3% to 5% annual payout would be $1 to $2 billion, or a few dollars per American. Senator Sanders introduced a bill that would initially take half the equity of the largest AI companies as voting shares managed by a new public commission, then tax later issuances to preserve the fund’s stake.The American AI Sovereign Wealth Fund Act (S. 4825), introduced in June 2026, levies an initial 50% tax on the equity of AI companies designated systemically important, plus a tax on later issuances to maintain that stake, paid in stock to a sovereign wealth fund managed by a seven-member Independent Commission for Democratic AI, which would vote the shares in the public interest . At the valuations the sponsors cite, the fund would start near $7 trillion and pay out about 5% of its value a year, more than $1,000 per person. And Anthropic published an economic policy framework whose measures escalate with the unemployment rate, beginning with citizen capital accounts that could hold AI-company equity and ending with sovereign-wealth and basic-income models.Anthropic's Economic Policy Framework (June 2026) keys its tiers to unemployment thresholds, roughly at 5%, at 10%, and beyond historical peaks, moving from wage insurance, retraining support, and individually held capital accounts that could be funded with equity in AI companies toward expanded unemployment insurance and, in the last tier, a menu that includes basic income, AI sovereign wealth funds, and levies on AI use measured by tokens, compute, or revenue . The framework begins from the same fiscal premise as this post, that the tax system leans on labor while income moves to capital, and its argument for the capital accounts, that they compound and must be seeded before disruption is visible, is an argument for not leaving collective ownership to the end.

Each of these contains a piece of the design and omits the rest. OpenAI’s one-time 5% stake is a snapshot of today’s winners. It collects nothing from firms not yet founded, and it shrinks relative to an economy that keeps growing past it. The Sanders proposal avoids dilution from later equity issuances, but remains limited to a designated set of AI companies. The arithmetic is also not close. A one-time 5% of one firm funds a dividend of a few dollars per person per year, the Sanders fund projects about a thousand dollars a year, and even the dashed line in Figure 4, which collects 1% of all covered equity every year for fifty years, does not reach a livable income. The Sanders version also shows how an initial taking can be too aggressive while its sectoral base remains too narrow.Half of a company seized in a single stroke would be fought as expropriation, drawing the takings challenge and the united business opposition that a modest recurring levy is designed to avoid. Its governance repeats an old mistake as well, since voting shares managed by a public commission would have the fund govern the companies it holds, the feature that turned business against Sweden's wage-earner funds and that non-voting shares exist to prevent. The other design choices are drawn too narrowly, in scope and in timing. A base defined as AI companies invites a fight over what counts as one, while a size threshold draws no sectoral line. And an unemployment trigger arrives late, since wages and labor's share can fall before full automation . Anthropic’s framework does escalate as displacement deepens, but what escalates is relief rather than ownership. Its early tiers are wage insurance and retraining, its collective-ownership measures sit in a final tier that is explicitly a research agenda, and the levies on that menu tax AI use, by tokens, compute, or revenue, stopping at the sector’s edge. It is a framework for cushioning displacement that postpones the question of who owns what the displacement builds. And none of the three has an answer to capital flight, which is what market-access enforcement is for.

The instinct behind all of them is right. Anthropic writes that AI companies whose returns prove transformative have an obligation to share them broadly, and says it is ready to pay its fair share. But the gains will not stay with the companies that build the models. The leading labs are losing money today while the value shows up around them, in the chipmakers and the platforms already, and eventually in every large firm that replaces labor with software. If open models end up carrying much of the work, the builders may capture the least of it. In this way, a tax on the AI industry would miss the AI economy. This post is one attempt to specify the public’s share of that economy and the practical machinery for claiming it at the scale widespread job displacement may require.

Open design questions

Despite this post’s length, the design space is far larger than I alone can fully analyze. I’ve tried to hint at certain gaps or shortcomings in the footnotes and below, but there are likely more, and probably some errors and oversights. However, I believe the central narrative is sound, and my intent is for this proposal to serve as a starting point for discussion rather than a finished blueprint. At the very least, the following points remain underspecified by the proposal.

The levy is meant to sit alongside the corporate income tax. Swapping one for the other would raise nothing new, and the two reach value at different times. The most valuable AI companies have enormous valuations but minimal profits. For example, at OpenAI’s reported valuation, a 1% levy would require a firm that so far has no profits to tax to transfer shares worth about $8.5 billion that year. But keeping both is a significant increase in corporate taxation. At a 1% rate, the levy would collect roughly as much as the entire corporate income tax does today.Consumption taxes belong in the same conversation. A VAT is administrable, hard to avoid, and reaches capital income when it is spent. It may be part of the fiscal answer. It creates no public claim on capital and bears proportionally hardest on the households whose wages are being displaced.

The tax base has unresolved edges. Phasing the charge in according to enterprise value softens the cliff, though firms near the band still gain from staying below it. The charge itself remains measured on equity value, which also gives firms an incentive to borrow to fund payouts and shrink the base, adding a debt bias to the incentive to distribute earnings. Private firms pose a harder valuation problem because funding rounds are infrequent and headline values can be inflated by preference terms. Coverage also has to follow size and economic function closely enough that a firm cannot escape by becoming a partnership, staying private, or delisting.

The cash payment option creates another problem. A firm that thinks its stock is overpriced will pay in shares, while one that thinks its stock is cheap will pay cash. Requiring shares would remove that choice, but the constitutional analysis above cuts the other way. The problem is worse for private firms, whose valuations are contestable, though public firms would still gain a timing advantage if the obligation were priced on a single day.

The fund’s influence over the firms it holds remains a separate risk. Non-voting shares, no board seats, and no engagement with management greatly reduce the concerns about common ownership discussed above . Residual legal rights and political pressure on managers remain, and the safeguards would have to survive decades of changing governments.

Crisis behavior is also untested. In normal times, issuer buybacks would supply much of the fund’s cash without requiring the market to absorb large stock sales. In a crisis those buybacks can stop just when the fund’s payouts are most needed. The club makes the problem larger because many national funds could sell at once. A shared rule should cap the pace of sales, route them through issuer buybacks first, and suspend them during market-wide stress.Pooling the stakes in a common vehicle would allow much of the flow to be netted before it reached the market. Charter-level limits on resale could provide another backstop.

The enforcement rule has an aggressive edge of its own. While the club is incomplete, the residual lets members collect part of an out-of-club firm’s charge on sales into other countries that have not joined. This reach makes holding out costly, but it also stretches the destination principle beyond the member markets supplying the enforcement power. Governments outside the club may retaliate against it. The grace-period escrow softens the claim by holding those shares for the absent countries, at the cost of delaying part of the reward for going first. Its treatment under the WTO and investment treaties would still require separate analysis, especially where the club reaches value attributable to sales outside its markets.

The hardest problem is the politics of a loss taken now for gains that arrive over decades. Past wealth taxes lost support after capital fled, valuations were contested, and revenue disappointed. This design tries to avoid those failures, but the repricing would still arrive when the tax becomes credible and the dividend would still take decades to matter. Indexing helps because the higher rates arrive in the futures where shareholders receive the largest gains, and holders in those scenarios can remain wealthier after the repricing than they are today.Markets would price an indexed levy across the possible paths. The markdown ranges from roughly 15 to 25% at a permanent 1% rate to roughly 50% in the strong-AI scenario. The high rate remains contingent on AI producing the income shift that triggers it, so today's markdown should sit below the strong-scenario figure. If AI disappoints, the rate stays near its floor and the part of the markdown associated with the higher-rate paths unwinds.

Without explicit protection, retirement accounts would bear part of the initial loss. To remedy this, part of each firm’s levy issuance could be routed to recognized retirement funds, leaving total issuance unchanged and netting out their dilution. Limiting the offset to holdings that exist at enactment closes the obvious loophole and lets the protection run off as those savings are drawn down. Figure 4 uses this grandfathered design.Recognized funds would mean broad-based retirement savings, with sovereign wealth funds excluded. Custodians already certify qualified accounts, and pooled funds could certify the retirement share of their assets subject to audit. Retirement accounts hold roughly a third of U.S. equities , so a permanent version would cut the strong-AI dividend at year fifty from about $52,000 to about $38,000. Grandfathering preserves most of the fund's long-run receipts because the protected holdings run off while it is still compounding. New contributions buy at prices that already reflect the levy and need no offset.

Further design questions appear in the footnotes, including the constitutional path for the levy , the division of collections among member countries , financing transition support before the fund matures , the legal definition of the labor-share index , and the durability of the trust as its beneficiary class changes .

Closing

AI threatens to move income from labor to capital faster than any technology before it. Perhaps the best response is to give citizens a direct and durable claim on the capital whose returns are replacing their wages. To do this, the cleanest instrument is a levy on large firms, paid in their own non-voting shares, enforced through market access, with the proceeds returned to every citizen as an owner’s dividend. Current capital owners have their own stake in the outcome. Asset values rest on the stability of the institutions that enforce them, and an economy where citizens own nothing is not one where property stays secure for long. Part of what the levy buys, for the owners who pay it, is the durability of the arrangement that makes their assets worth anything at all.

The United States is in a unique position, with the market power to implement such a levy and the leverage to bring the rest of the world along. The same global club could give international AI rules an enforcement mechanism and a continuing incentive for cooperation. Both possibilities should be given serious consideration now, before the transition to an AI workforce accelerates. Figure 4 shows why the work must begin early: even the ambitious path takes nearly twenty-five years to pay an annual five-figure dividend per person. But while first dividends would be small, the fund’s holdings would compound from the day it was created. And waiting compounds too, as every year without a public claim is a year the private one grows.

FAQs

Isn't this all AI hype?

The policy proposal has merit even if we froze AI's capabilities today. Labor's share has already fallen, equity is already concentrated, and the tax system already leans heavily on wages. The 1% floor is meant to address the economy we already have, and indexing the tax to labor's share of income adjusts it if the shift toward capital continues (see Growing the dividend above ).

Won't firms just raise prices and pass the cost to consumers?

Unlike a tariff, issuing shares does not directly increase the marginal cost of making or selling another unit, so the levy creates no immediate reason to raise prices. A firm that could already increase its profits by raising prices would presumably have done so. The immediate cost would therefore fall mainly on capital owners, through the markdown in the value of existing shares . Over time, however, the recurring levy could alter investment, financing, market entry, and corporate structure in ways that eventually affect prices and wages. The clearest distortion may be the incentive to distribute earnings that would otherwise be reinvested, which is discussed under A liquid wealth tax above.

Could companies escape by relocating?

Changing a company's legal home would not escape the levy because the charge follows the corporate group's sales into member markets. A covered firm could avoid it only by abandoning those customers, and for a firm with material American sales, that would generally cost much more than moving its headquarters. The possibility of withdrawal still places a practical limit on how far a country or a small club can reach (see From market access to a global club above).

Isn't this confiscation?

The levy's real cost is a markdown of covered share prices, on the order of 15 to 25% at a permanent 1% rate, relative to the value those shares would have had without the levy. Existing holders keep their shares but own a smaller fraction of the firm after it issues new stock, using the same mechanism firms already use to compensate executives and finance acquisitions. In the scenarios that drive the indexed rate higher, a holder who bears the full repricing still ends up far wealthier than they are today because the deeper markdown arrives only when AI has produced much larger gains (see Open design questions above ).

Is the levy constitutional?

The answer is unsettled because courts have never considered a levy of this form. The strongest path would be to structure it as a corporate excise tied to the privilege of doing business, with a cash payment option that reduces the risk that issuing shares would be treated as a compelled transfer of property, but a court that classified the levy as a direct tax on property could make a constitutional amendment necessary. The legal case and its limits are discussed above .

Isn't this nationalization?

The design gives the public an economic stake while withholding the powers that would let the fund direct the firms it holds. After fifty years in the strong-AI path of Figure 4, the fund owns about 22% of covered equity, but its shares carry no votes and the formal channels that could turn it into an industrial-policy board are closed (see Public ownership without control above ). In this important respect, the arrangement resembles the corporate income tax, which already gives the government a claim on part of every firm's profits without giving it control of the firm.

Why would this fare better than earlier wealth taxes?

This design is built around the valuation, liquidity, and capital-flight problems that led many European wealth taxes to be repealed (see A liquid wealth tax above). Publicly traded firms have observable market prices, payment in shares allows the firm to satisfy the charge without raising cash, and enforcement through customer markets makes a change of domicile ineffective. Private firms and the eventual conversion of the fund's holdings into cash still present difficult edges, but the central case no longer requires the government to value every asset a wealthy person owns or to prevent that person from moving abroad.

Would the levy reduce investment and innovation?

It probably would at the margin, although the size of the effect is uncertain because the levy would begin at 1%, phase in across a high threshold on the order of a billion dollars, and reach firms only after the investment that built them had already succeeded. Founders and early investors still price expected exits, so some reduction in new company formation could remain, although the higher rates would arrive in futures where AI had already multiplied the value of successful firms (see Growing the dividend above). Within covered firms, the more immediate concern is whether the levy encourages managers to distribute earnings that they would otherwise reinvest, a distortion discussed under A liquid wealth tax above.

Won't competition pass AI's gains to consumers, leaving nothing to tax?

If competition passes most of AI's gains to consumers, covered equity values will grow less and the levy will collect less. If wages also hold up, the indexed rate will remain near its floor, leaving little need for the larger public claim modeled in the stronger scenarios. The levy becomes more consequential as gains accumulate in equity values and labor's share falls, with the indexed rate rising only as far as those changes actually occur.

Why not just raise income taxes, or add a VAT?

Higher taxes on corporate profits, realized capital gains, and consumption could all supply revenue, and a VAT may well be part of the fiscal answer . Taxes on income and gains reach value only after it appears as taxable income or an investor realizes a gain, which leaves room for profits to be booked elsewhere and gains to be deferred. A VAT reaches capital income when it is eventually spent, but places a relatively heavy burden on households that consume most of what they earn. None of these policies gives citizens an ownership claim on the capital producing the income, which is the additional distributive purpose of the share levy.

Why not just tax AI companies directly?

AI's gains are unlikely to remain inside the sector that builds the models, because retailers, banks, manufacturers, and other firms will capture some of the value when they use the technology. Restricting the base to AI companies would miss those gains and invite a continuing fight over which firms qualify, especially if open models allow the builders to capture less value than their users (see Other proposals above). A size threshold reaches large firms across the economy, while indexing adjusts the rate according to the shift in labor's share that the policy is meant to address (see Growing the dividend above).

What about wealth the levy never touches, like real estate?

The levy reaches large incorporated firms, leaving wealth in real estate, small businesses, art, and other uncovered assets outside it. Much of the wealth at the very top would still be reached, since the top 1% of households hold roughly half of all corporate equities and the largest fortunes consist mainly of stakes in large companies. Extending the levy downward would reintroduce the annual valuation problems that the share-payment mechanism avoids, while housing would remain the sharpest gap because land and rent absorb a large share of any livable budget. Land and housing policy would therefore be a natural companion to this design.

Won't retirees and pension savers bear the cost?

Without protection, retirees and pension savers would bear part of the initial markdown. Figure 4 therefore protects savings already held in recognized retirement accounts when the levy is enacted by routing enough of each firm's new shares to those accounts to offset their dilution, while the public fund receives the balance. This slows the public fund's growth, but the protection phases out as the grandfathered savings are withdrawn, and new contributions need no offset because they buy shares at prices that already reflect the levy .

Could a future government raid the fund?

No ordinary statute can remove that risk permanently, because one Congress can revise the work of an earlier one. The proposed trust would vest beneficial interests in individual citizens, and the fund's non-voting shares would carry their limits inside the securities themselves, making both the principal and the ban on corporate control harder to reach. These protections would create legal and political barriers rather than an absolute lock, and the unresolved problem of preserving them as the beneficiary population changes is discussed above .

Won't other countries retaliate?

Other countries may retaliate, especially while the club is small and the first mover is collecting shares whose ultimate owners live abroad. The design limits that risk by treating domestic and foreign firms alike and offering every country the same right to collect on sales into its own market, so a government that fought the system would also be declining revenue available to it through membership (see From market access to a global club above). Retaliation could still place serious pressure on the founder's own firms during the period between the levy's becoming credible and the club's becoming large, and the precedents and limits are discussed in the footnotes there , .

Would American investors bear more than the United States collects?

Under full membership, the current estimates say yes, because American portfolios hold roughly half of the world's covered equity and would bear about half of the initial worldwide markdown, while the sales matrix gives the American fund about 37% of collections. Some asymmetry is consistent with ordinary source-based taxation, under which countries tax economic activity within their markets rather than matching receipts to the residence of investors. It still creates a real negotiating problem for the first mover, and a temporary credit funded from the residual shares collected while the club remains partial could narrow the gap without changing the market-access rule or permanently directing most collections toward the countries that already own the most capital .

Why pay everyone instead of only the people actually displaced?

Payments can remain targeted so long as particular workers and industries can still be identified as displaced, which is why the design begins with insurance for those directly affected. As AI diffuses across the economy, the causal line between one person's job loss and a particular system will become harder to draw even if the aggregate wage share continues to fall, and a narrowly administered program would miss much of that broader loss (see A public dividend above). A universal dividend extends compensation across the economy-wide shift and creates the broad constituency that would help protect the fund over time (see Protecting the fund above).

Isn't the dividend too small to matter?

The dividend would be far too small to replace lost wages during its early years, so workers displaced before the fund matures would need separately financed transition support . That delay strengthens the case for beginning early, since even the ambitious indexed path needs nearly twenty-five years to reach an annual five-figure dividend per person (Figure 4). A recurring levy also keeps acquiring shares in firms that become valuable later, allowing the public's claim to grow with the economy instead of leaving it with a one-time stake in today's winners (see Other proposals above).

What will people do without work?

Nothing in this proposal answers that question, which remains important because work supplies structure, identity, and community along with income, while a dividend replaces the income alone. If AI ends work as the default organizer of adult life, people will need other sources of purpose, such as hobbies, sports, and learning, along with communities and social institutions that organize those pursuits as schools, professions, and workplaces once did. A tax design cannot build those institutions, although the dividend could provide the material security people need while they search for purpose and community elsewhere.