Abstract

What if everyone owned a little bit more of everything? Here, we argue that we should diversify the ownership of corporations to reduce risks from AI.

In Part 1, we highlight three benefits of diversification:

1. Power becomes decentralized, reducing risks from malevolent actors.
2. Actors developing TAI have fewer incentives to profit from negative externalities.
3. Greater global collaboration and less adversarialism in TAI development.

We believe diversification is tractable, as diversification both:

4. Hedges against risks to the domestic economy and technological unemployment

And:

5. Increases the expected returns of investors

In Part 2, we propose concrete policies that diversify the ownership of corporations.

All countries should:

1. Create sovereign wealth funds (SWFs) and pension funds that passively invest in the global market portfolio.
2. Create tax incentives to invest in the global market portfolio and reduce distortions and disincentives from doing so.
3. Improve corporate taxation through self-assessed taxes.

We close with a series of drawbacks, risks, and reservations of these policies.

Part I: The benefits of increasing common ownership

1. Decentralizing Power & Increased Global Collaboration

International tensions have put strains on collaboration within AI safety research,^[1] which could lead to coordination failures. We must cooperate to successfully reduce many different global risks,^[2] and rising national-security-based regulations on AI^[3] suggest that AI is not safe by default. Here we argue that more diversification of ownership of corporations can help reduce global risks, especially risks from transformative AI.

Currently, power over AI is very concentrated:

1. In a few countries: Especially in the US and China, with only some important suppliers of chips in Taiwan and the EU.^[4] 
2. In a few corporations: Frontier AI firms Google, Microsoft, Amazon, Anthropic, OpenAI, xAI (SpaceX), Facebook (Meta). Nvidia is the main producer of AI chips and now the world’s biggest corporation. Perhaps some Chinese corporations (e.g. Tencent). ^[5],^[6]
3. A few key people have outsized control (e.g. Google and FB’s founders have voting control through super-voting shares despite owning only ~10% of shares, Elon Musk owns roughly 60% of xAI).
4. A few players have outsized ownership: while Google, Meta, Nvidia, Amazon are all only ~⅔ commonly owned by institutional investors (e.g. BlackRock, Vanguard, etc.), Tencent is only half-owned by institutional investors.

And AI could be a winner-takes-all industry that might concentrate even further (cf. Big Tech has bought more AI startups recently, e.g. SpaceX has bought Cursor).^[7] These power concentration incentives trap AI firms in a zero-sum 'race to the bottom', sacrificing AI safety for speed.^[8]

By increasing diversification, we decrease the concentration of ownership. In turn, greater common ownership disincentivizes exploiting negative externalities (e.g. risks from AI uplift) for profit.

To illustrate, imagine a Big Tech firm with just one shareholder — call him Elon. Suppose Elon has total control and a substantial stake in the firm. The firm aims to maximize shareholder value, a lot of which is value-to-Elon. However, Elon's actions may affect Alice, who has no say in the firm's decisions and a much smaller stake. As the number of shareholders increases (say, to include both Elon, Alice and the rest of the world), the closer 'maximizing shareholder value' approaches 'maximizing value-to-everyone', rather than just 'value-to-Elon'. Both profits and negative externalities get shared more widely, and so diversification leads to fewer advantages to profit from untaxed externalities. In particular, diversification reduces the incentive to produce (global) public bads (e.g. risks from unsafe AGI).

As we diversify and the number of shareholders gets larger, i.e. as we converge on full common ownership, corporations become more and more incentivized to approximate acting in the interests of all. And so, the more we mutualize corporations, the more we reward safety and transparency,^[9] and decrease adversarialism between competing AI firms.

Diversified ownership increases collaboration, as up- and downsides of AI become more widely shared.^[10] The more we diversify, the larger our stakes in foreign firms, which incentivize us to collaborate and not support adversarial policies or arms races. If one company or country gets ahead, we are less incentivized to go to war. This is why trading partners go to war less often.^[11] Similar benefits from diversification apply to other public goods.

Put simply, generally, the more globally integrated our world economy becomes, the more incentive we have to coordinate to reduce global risk that would affect everyone vs. local risks that only affect us domestically.

Conventional Economic Benefits

Diversification is feasible and economically beneficial. Contrary to optimal portfolio theory, many investors show "home bias" by holding excessive domestic assets. ^[12] Home bias is a 'robust stylised fact'^[13], 'too large and widespread'^[14] across many countries to be due to chance.^[15] Though information effects may partially explain home bias^[16], evidence indicates it's also irrational,^[17] and reducing it will boost returns.^[18]

The more strongly markets of different countries correlate, the less beneficial it is to diversify.^[19] But even if one considers market correlations, domestic investment is inefficient.^[20],^[21] Further, local investment bias is to be expected given 'distance-based' models of investment decisions, which also explains a variety of international data in developed markets.^[22] US investor home bias is returning, indicated by decreasing international market correlations.^[23] Inefficient diversification exists despite minimal US economic home bias, because investors do not invest optimally across foreign countries, even if there is no obvious home-country-specific bias.^[24] Investors may even irrationally underestimate the most impactful tail risks, leading to yet greater inefficiency.^[25] In sum, inefficient diversification in many countries stems from home bias. 

Diversifying more leads to moral hazard: the less home bias, the less invested a country is in itself, and the less it is incentivized to produce more.^[26] (cf. the incentives of the CEO of Tesla, whose shares can increase by hundreds of billions a year, to those of Volkswagen's CEO, who earned a mere $10M—100,000x less because he owns a much smaller stake in VW).^[27] But the difference between current and optimal levels of diversification is stark and seemingly random: Empirically, in the US, efficiency loss is ~$112b/year^[28] or ~1% of GDP, ^[29],^[30], and in France, it’s ~$33b/year or ~2% of GDP.^[31] This has substantial implications for the macroeconomy.^[32]

Hedging against risks and technological unemployment

Global labor compensation is approximately $60T per year.^[33] Even if diversifying would not yield higher financial returns in expectation, hedging against risks from TAI, in particular labor displacements (‘technological unemployment’), might still increase expected utility. If AI does well, more people will profit, but if it doesn’t do well, current returns to labor will remain high.

Risks from events that have never happened before (like AGI) are underinsured.^[34] Diversification insures and hedges against such risks, particularly against technological unemployment, where AI has a transformative global impact before it can do all tasks humans can do.^[35] This could cause super-exponential economic growth,^[36] or be otherwise transformative (e.g. drastically increase income inequality^[37] or a large lead by one country or company could lead to a stable, perpetual global totalitarianism).

The more we diversify, the more we benefit from our competitors' success: either we get financially compensated,^[38] or we can use our increased resources to outcompete potentially malevolent (or even just self-interested) actors.^[39] 

Effects on differential technological development

Dilemma: The more directly we invest in AI startups, the more we speed up AI (ETFs of publicly traded AI firms vs. GMT) and the more governments are involved in picking winners. But the more diversified the portfolio, the less effective the hedge. If we were to invest directly into AI labs, then they would subsidize AI development, which might speed up AI.

And so we might want to invest in either just publicly traded AI companies or in the global market portfolio, which now has substantial exposure to the AI sector.

Current state of diversification

A large part of the AI sector is publicly traded via two channels:

1. Indirectly: Publicly traded companies either have their own frontier AI efforts or have substantial stakes in newer frontier AI firms (e.g. Microsoft owns a substantial stake in OpenAI,^[40] Google owns DeepMind and has a stake in Anthropic.^[41] Amazon and SK Telecom own substantial stakes in Anthropic. ^[42] Meta is publicly traded.)
2. Frontier AI companies: SpaceX has now gone public and owns xAI and Cursor. Anthropic and OpenAI might have their IPOs in 2026.

Such common ownership can have anti-competitive effects and reduce innovation.^[43] Recently, institutional investors have pressured Big Tech to spend less on 'moonshot R&D'.^[44] But when large investors own technologically related firms it mitigates this problem as their firms act in their interest and can profit by innovating via R&D as a 'public good to their competitors' that might otherwise be unprofitable.^[45] When technology spillovers are relatively large, an increase from the 25th to the 75th percentile of common ownership is associated with an increase of +13% in citation-weighted patents.^[46]