The AI underwriters are pricing hundred-fold multiples to extract your capital.
It is true, and the statistical literature is careful to state it, that base rates shift when the underlying architecture of an industry changes. A market observer in 1999 could not have used railway-era multiples to value a cloud infrastructure play, and a strict application of 1995 revenue multiples would have missed the compound growth the internet actually delivered. The trouble — and here it is worth being precise about what “the AI market” actually is, because the public discourse has the misleading habit of treating it as a monolith rather than a stack of mismatched capital expenditures and speculative derivatives — is that base-rate mathematics apply to the distribution of returns, not to the survival of one or two exceptions. You do not defeat a base-rate collapse by picking the one company that survives it.
The Morgan Stanley report that surfaced this morning, by Michael Mauboussin and Dan Callahan, applies exactly that distributional discipline to the revenue projections now being pitched to investors. OpenAI’s projection — the one the company floated ahead of its planned IPO — calls for revenue to hit $145 billion by 2029, a compound annual growth rate of 108 percent sustained for five years. Of the roughly nineteen thousand large companies the authors examined, not one has achieved that. Zero. The closest any company has come over a half-decade horizon is not in the same postal code. The authors themselves qualify the finding with the observation that base rates “can change as the world changes.” The qualifier does not erase the gravity of the null; it merely acknowledges that the null is historical, not prophetic. The discipline of formal verification — where you prove, mathematically, that a system does what its specification claims and not what its advertisers wish it did — has a clean analogue here. You cannot verify a system whose specification is mathematically impossible, and a trillion-dollar valuation built on a growth rate that has never been achieved is not an engineering specification. It is a financing document, structured specifically to move private capital out before the public window opens. As Main Street Independent has reported, the leading AI companies are pushing toward blockbuster IPOs even as Wall Street shovels record capital into the build-out. The architecture of this capital raise is not a market accident; it is the mechanism.
Cory Doctorow, extending J.K. Galbraith’s concept of “the bezzle” — the interval in the business cycle where the promoter holds the gain and the public has not yet felt the loss — notes that AI will leave a durable infrastructure residue, much like the telecom overbuild of the early 2000s, but that the speculative layer is the actual product being sold to public markets. The bezzle is sustained by what Doctorow and technology-studies scholars call “criti-hype,” a feedback loop where capability claims are repeated in the financial press without an audit of the unit economics. It is sustained because the capital flood pouring into data-center concrete and GPU clusters depends on revenue models — surveillance pricing, enterprise lock-in, utilization rates that, under current grid-capacity projections, face a hard physical ceiling before breakeven — that do not yet exist. The platform developers are not building a public utility; they are building the chokepoint.
The oddest part of the conversation is the speed with which this arithmetic is dismissed as philistine. The polite version of the objection — that base rates “can change as the world changes” — is technically true, in the way that the statement “machines can learn” is technically about function approximation. The room-temperature version is simpler: the skeptics “don’t get it,” which is what every party to a speculative mania says to the sucker who is still counting. The base-rate argument does not require getting it. The analogy in the report is well chosen: if you wanted to bet on a 100-degree day in Central Park in June, you would look up the climate record and note that it has happened three times in 157 years, or on 0.06 percent of June days. You would not, even with a generous nod to global warming, price the contract much above that number. The AI revenue projections ask the market to price a contract for something rarer than a heat wave in early summer, and the book-runners are calling it a “can’t-miss opportunity.”
Asset manager GMO has been tracking the ten-times-revenue threshold for four decades, and its finding is not ambiguous. A portfolio of companies trading above that multiple yields, after inflation, eighty percent less capital than a plain S&P 500 index fund bought and held. At the start of this week, nearly seventy U.S.-listed companies — Palantir, Broadcom, Nvidia, Intel, Alphabet among them — fetched a multiple of at least ten times trailing revenue. That is a larger share of the market than sat above the same threshold at the peak of the dot-com bubble. Ten times revenue is the cheap end. SpaceX, the next IPO up, has already priced its private shares at ninety-two times sales. At ninety-two times, the revenue has to be not merely real but enormous, durable, and untaxed by a price war between two companies that are burning cash faster than a hyperscaler builds a data center — a possibility the markets acknowledged, nervously, in the same session that produced this report. Big Tech’s own capital expenditure — Oracle’s larger-than-forecast data-center spend, Super Micro’s $7 billion equity raise to feed the server appetite — is the infrastructure that makes the bezzle look like progress. When the bezzle collapses, as all bezzles do, the durable residue will be a thicket of GPUs and underwater lease obligations and a few well-capitalized incumbents who bought the wreckage at a discount. The employees? The retail investors? They are the sucker’s side of the trade.
What is being sold is not a stake in future cash flows; it is a stake in the bezzle. The GMO data shows that leaving the extraction layer unchecked destroys public returns. The remedy is not to avoid the sector but to impose the interoperability mandates that prevent a handful of platforms from capturing the entire value chain, and to require any company projecting nine-figure revenues five years out to supply, alongside the forward-looking statement, the historical base rate for achieving a growth arc of that shape. It is not a radical proposal. It is the kind of line item an institutional investor would demand before buying a minority interest in a privately held widget factory, and it is merely being proposed for the largest concentration of capital in recent industrial history.
In 1930, NYSE president Richard Whitney tried to restore public confidence in the market by placing a bid — with his own money — for a 60,000-share block of U.S. Steel at $160. The stock kept sinking and bottomed two years later at $21. The AI bankers are putting up your money to bid on something rarer than a heat wave, at ninety-two times the revenue it has not yet earned. The confidence they are restoring is their own.