You Can't Index Into a Private Company

Most people think they participate in economic growth through their portfolio.

The logic is simple, and for most of the last seventy years it held. You own a 401k or an index fund, the economy grows, companies get more valuable, and your balance climbs. You never have to pick a winner, because the market picks for you. That worked on an assumption nobody bothered to say out loud: the most valuable companies were publicly traded, so once a business became important, it eventually listed, and anyone with a brokerage account could own a piece.

Three assumptions sit under that promise: the most valuable companies are publicly traded, the index captures the winners, and the gains flow through to public shareholders. All three are breaking down at once, and they break down most completely in AI.

The companies most likely to capture the gains from artificial intelligence are, right now, almost entirely private. Most people holding index funds have not yet processed what that means, treating it as a minor detail or a passing phase. It is neither. It is a structural problem.

Consider who builds the frontier models. Anthropic is private, backed by Amazon and Google. OpenAI is private, wrapped in a capped-profit structure few outside investors fully understand. xAI is private and Musk-controlled. Google DeepMind is a subsidiary of Alphabet, not a company you can buy on its own. The organizations training the most capable models in the world are, with few exceptions, closed to retail investors.

The owners are venture firms, strategic corporate investors, and sovereign wealth funds, alongside a few hundred early employees sitting on equity that stands to become generational wealth. Add it all up, and the population with meaningful direct ownership of frontier AI is measured in the thousands, not the millions.

The obvious objection is Nvidia: public, accessible, and responsible for some of the best returns in market history. All true. But Nvidia sells the hardware and earns the hardware margin, not the margin on intelligence itself. Being long Nvidia is not the same as being long AI: one is hardware margin, the other is intelligence margin, and the index hands you only the first.

None of this would matter if these companies followed the old road to public markets. They are not going to.

The IPO timeline has changed completely. In the 1990s, companies went public early, frequently before they turned a profit. Amazon listed in 1997 at a market capitalization around $438 million, and the growth that made it a giant happened in full view of public shareholders. Through the 2000s and 2010s, that timeline stretched to seven or ten years, and in the 2020s the biggest technology companies stay private more or less indefinitely. Private capital is abundant, staying private lets founders keep control, and the public offering is no longer required to fund growth.

Watch what that does to wealth. When a company stays private, the gains from its early growth accrue entirely to insiders, so by the time an ordinary investor can buy in, most of the value already exists. Facebook went public in 2012 at roughly $100 billion, and the foundational growth had already happened in private hands. For the leading AI companies, that foundational growth is likely to happen entirely in private. The public offering, if it comes, is the liquidity event for insiders, not the entry point for everyone else.

A legal structure seals this off, and it was built with good intentions. To invest in private companies you need to be an accredited investor: $200,000 or more in income, or $1 million or more in net worth. The rule protects people who cannot afford to lose money from risky, illiquid bets. Its unintended consequence is that it walls ordinary investors off from the best-returning asset class of the past decade, the one that has beaten the public index it was supposed to protect them into. The people who most need the upside from AI are the ones least allowed to reach it.

Your S&P 500 fund is not nothing. It holds Nvidia, Microsoft, Alphabet, real AI exposure. But it is not frontier exposure, and it holds none of the pure-play model companies where value compounds fastest. An index is a rearview mirror, capturing winners only after they have already won.

Here's the arithmetic that should keep an index investor up at night. If AI value compounds mostly inside private companies for the next decade, and those companies list only after their steepest growth is behind them, then public investors are buying into the maintenance of value, not its creation. That's a fundamentally different proposition than what the postwar model promised.

Pension funds are a partial answer, and only partial. Some allocate to private equity and venture, but regulation, liquidity rules, and fee structures cap how much they can hold. And here is the part worth sitting with: the workers most exposed to AI automation are the least likely to hold a pension with any real private-market exposure, so the same people absorb the displacement and miss the gains. There is a real cruelty in that alignment.

Most proposals to fix AI wealth concentration aim at the wrong target.

Universal basic income skips the ownership question entirely. Nearly every version taxes income or corporate profits and redistributes the proceeds, but AI wealth held as private equity is not income. It is unrealized. Nobody sells it and nobody distributes it, so an income tax never touches it. You cannot tax what nobody has cashed out.

Universal basic capital has the right instinct: hand people an ownership stake, not a check. The execution is brutally hard. Which companies go into the fund, chosen by whom, and updated how often? Give everyone shares of BlackBerry in 2007 and you have impoverished them by 2013. Sovereign wealth fund models, like Norway's, are the most convincing version, and they require the political and institutional infrastructure most countries do not have.

Timing matters more than any single mechanism. Redistributing wealth after it concentrates is a far heavier political lift than building the channels to spread it beforehand, and the will required now is a fraction of what it will take once the ownership structure hardens. The window is open now, while it is still cheap.

A few levers actually address ownership instead of income. Mandatory equity pools come first: require AI companies above a set revenue threshold to contribute a slice of equity to a broadly held public fund, so the public owns a piece before the liquidity event, not after. Then recalibrate access, because the accredited-investor thresholds were never designed for an economy where the most important wealth creation is private. Ring-fence a portion of AI revenue taxes for a public investment vehicle that buys equity ahead of the IPO, and in countries without one, build a sovereign wealth fund for exactly this purpose.

Then there is the scenario that fixes the problem with no policy at all. If AI models commoditize, competition grinds margins down and no single company holds durable pricing power, so value flows to users and to the applications built on top, not to the model providers. Think of electricity: ConEd earns a utility margin, not a monopoly rent. Open-source AI is the strongest force pushing toward that outcome, which makes it the most powerful solution on the table. It is also the one nobody controls, because it is a market outcome, not a policy lever.

Strip everything else away and two questions remain, and people keep collapsing them into one. The jobs debate asks whether you will have work to do; the ownership debate asks whether you will share in the value you help create. Those are not the same question. You can have full employment and radical wealth inequality in the same economy, because a steady paycheck does not entitle you to a piece of the thing you are running.

The Industrial Revolution generated staggering wealth, and it took roughly a century to spread that wealth broadly. We do not have a century. The ownership architecture for the AI economy is being set right now, in rooms and deals most people are not watching. The index fund worked because the winners went public, and nothing guarantees that architecture repeats.

Does your organization have a position on whether the equity upside from AI adoption flows broadly or concentrates narrowly? Most don't. That absence of a position is itself a decision, and it will outlast the people making it.