How Are Private AI Companies Valued?

A plain-English guide to where private AI valuations come from, why the headline numbers overstate what shares are worth, and what the evidence says about whether any of it is rational.

When a headline says a private AI company is "worth" $300 billion, the number did not come from a market. It came from a meeting. The company sold new shares to a group of investors at a negotiated price, and that price, applied to every share the company has issued, became the valuation you read about.

That mechanical difference explains most of what confuses people about private AI valuations. This primer walks through how the numbers get made, why they overstate what the shares are worth, and what the evidence says about whether any of it is rational. It condenses our full research piece, The Triumph of the Optimists, which works through the data in detail.

Where the number comes from

A public stock price is an average. It reflects everyone with an opinion, including the people betting against it. A private valuation is a maximum. It is the bid of the single most optimistic investor in the round, because in private markets nobody who disagrees has a way to act on their view. You cannot short a Series D. The skeptics decline the meeting, and the price is set by whoever did not.

The economist Edward Miller showed in 1977 that restricting short sales makes a stock price reflect the optimists rather than the average opinion. Private markets are that model with the dial turned all the way up. The more bidders compete for a round, and the more they disagree about the future, the further the winning bid rises above the average view. Competition pushes the hottest deals away from fair value, not toward it.

How investors actually model these companies

Serious investors triangulate rather than trust one tool. Revenue multiples anchor the price to comparable businesses, though a revenue dollar says little about profitability when serving costs are this high. Scenario-based discounted cash flow models run several complete futures through every line of the business and weight them by probability; small changes in long-run assumptions move the answer enormously. And venture underwriting asks a different question entirely: not "what is this company worth on average," but "what can I pay and still return the fund if this turns out to be the one?"

That last question explains more of AI-round pricing than any spreadsheet. The investor who wins the round is usually the one answering it.

Why the headline number overstates what shares are worth

The lead investor in a private round buys preferred stock wrapped in protections: liquidation preferences, ratchets, sometimes a veto over any IPO below the round price. The headline valuation takes the price of that most-protected share and applies it to every share in the company.

Researchers Will Gornall and Ilya Strebulaev repriced 135 US unicorns security by security, unwinding those protections, and found the average headline valuation 48 percent above fair value. Sixty-five of the 135 stopped being unicorns once the preference stack was priced properly. In the companion survey, 91 percent of venture investors said unicorns are overvalued. The people printing the numbers do not believe the numbers.

What happens when a real market shows up

Twice in recent memory, two-sided trading arrived to test private marks. When rate increases closed the primary market in 2022, the median company on the Forge secondary platform went from trading at a premium to its last round to 62 percent below it, then back toward par once primary activity resumed. And the IPO acts as a final exam: short sellers arrive, preferences collapse into common stock, and the 2021 vintage graded itself in public. The businesses were mostly fine. What collapsed was the price one very convinced buyer had once been willing to pay.

So are AI valuations rational?

Here is where the story turns. Venture-shaped returns follow a power law: a small number of outcomes carry nearly all the value. Across every AI-native company that first raised at a billion dollars or more between 2021 and 2024, fifty-one in all, three names account for close to eighty percent of the cohort's paper value creation.

For an investor holding what amounts to a call option on the rare enormous outcome, underwriting to the wildly bullish scenario is close to correct behavior. Which produces the strangest fact in private markets: the winning bid has been too high on almost every company and far too low on the few that mattered. The optimists set the price because the mechanism hands them the pen, and the power law occasionally makes them look conservative.

What this means for an investor

If nobody, including the winner of the round, can bid what the outliers turn out to be worth, then nobody can reliably pick them in advance either. The question with an answer is not "is this company cheap" but "do I own enough of the field that the two or three names that matter are in the book by construction." On the AI cohort's own marks, a basket of twenty-five names holds the winners nearly nine times in ten. A single position holds them about one time in sixteen.

The full analysis, including the data behind every figure here, is in The Triumph of the Optimists. EQUIAM invests in this market, and the research reflects how we underwrite it.