The Unfalsifiable Decade

Three companies are doing something no three companies have done before. Within a single year, SpaceX, OpenAI, and Anthropic are coming public at a combined valuation north of three and a half trillion dollars - a figure that would seat them, as a trio, above every listed company in the world but a literal handful. SpaceX went first, listing last week at a $1.77 trillion valuation on a record $75 billion raise - the largest initial public offering in history - before closing its debut session up nearly twenty percent, worth more than two trillion dollars. OpenAI and Anthropic follow close behind, the latter reportedly preparing to list as early as this autumn at something approaching a trillion dollars. The instinct is to greet this as either a triumph of American capital formation or the topping signal of an AI bubble. Both readings are available, and I believe both are a little lazy. The more interesting fact is quieter and harder to look at directly: these companies did not arrive late to the public markets because the markets were closed to them. They arrived late because lateness paid, and the bill for that lateness is about to be presented to people who never agreed to pay it.

Start with the alibi, because it is the part most often repeated and least often examined. The standard explanation for why the great unicorns stayed private so long is that the public markets were inhospitable - volatile, short-sighted, punitive in their disclosure demands, unwilling to underwrite a long-horizon vision. I think there is truth in the complaint. Being public is genuinely expensive and genuinely constraining. But as an account of these companies’ behavior, the story collapses on contact with the evidence. The public markets were not closed to them. In 2020 and 2021, the IPO and SPAC windows appeared to be as wide open as any in a generation, and the firms that had the option to walk through them - SpaceX already worth a hundred billion dollars at the time - declined. They are choosing to come public now, into a market that is, by their own bankers’ account, eager to have them, at the largest scale ever attempted. A firm that sits out the hospitable window and lists only once it has grown to systemic size has told us, through its conduct rather than its press releases, what drove the decision. It was never access. It was scale. Fine.

This is what economists call revealed preference, and it is worth pausing on the moral texture of it. When the owner of a generational asset explains a decade of private tenure by pointing to the deficiencies of the public market, he is relabeling a choice as a constraint - describing his own agency as someone else’s failure. I think the story flatters everyone. It lets the founder pose as a victim of quarterly tyranny while doing the single most value-maximizing thing available to him. And what made the choice available was not heroism but plumbing. Private capital has swollen from under ten trillion dollars in 2012 to roughly twenty-two trillion today; the median company now waits about sixteen years to go public, a third longer than it did a decade ago; the old regulatory tripwire that once forced companies public once they crossed a few hundred shareholders has been quietly defanged. So rational financial actors act rational. Deep private market did not make anyone go private out of necessity. It removed the necessity of ever leaving, and then rational actors optimized. “The IPO market was bad” is the narration a system supplies after the fact to make a strategy sound like a circumstance.

What, precisely, did the strategy optimize? The honest answer has two parts, and the temptation is to mention only the one that flatters the critic. The damning part first: the steepest, most convex stretch of a company’s value creation - the part where a few thousand dollars becomes a few million - now happens almost entirely behind a wall that only the already-wealthy are permitted to stand behind. By the time the public is invited in, the convexity has been harvested. Uber is the textbook case; public investors boarded, as one fund manager put it, on the top floor. The private holders rode the elevator up and sold the view. That is real, and it is a wealth-concentration engine of considerable power, because the public buyers on the other side of the trade are overwhelmingly us - pension funds, index funds, the retirement accounts of people who will never be accredited investors and will therefore only ever meet these companies at their maturity, priced for perfection.

But the part that flatters no narrative is that staying private is not only extractive. For a SpaceX, the long opacity was arguably the precondition for the achievement: a decade of reusable-rocket and satellite-network capital expenditure that a quarterly market would very likely have strangled in the cradle. Insulation from short-term feedback can be the thing that lets a hard, slow, capital-hungry bet compound. So, the uncomfortable truth is that the same mechanism does both jobs at once. It shelters the company from the market’s worst myopia, and it lets the owners capture the asymmetric upside before the public arrives. These are not two strategies. They are one coin, and any argument that only shows you one of its faces is selling you something. And mind you, this is fine by any rational standard. Private capital taking heightened risks to drive new companies and new products should get paid for their locked-up investment and risks.

Now to the part, though, that has been undersold in almost every account of this moment, including, until recently, my own. The cost that staying private imposes on the rest of us is not the disclosure itself. Disclosure is a public good; an equitable market runs on it. The cost is the timing structure of disclosure that a long private tenure produces - and here the two senses of “scale” finally fuse into a single problem.

Consider how a company that goes public early actually behaves. It discloses continuously as it grows. Each earnings print, each guidance revision, each contract win and lost customer and amended risk factor is a small increment of information that the market prices in close to real time. The volatility of discovery is therefore distributed - smeared across years, and across a market footprint that begins small enough that no one else’s portfolio depends on the answer. A firm that stays private for a decade and a half does the precise opposite. It accumulates an enormous, untested stock of material information - the true shape of its adoption curve, its real unit economics, its customer and contract concentration, its compute and energy dependencies, its competitive position - while simultaneously growing to a size at which it anchors an entire ecosystem. Then it discloses all of it at once. The first genuine, adversarial price discovery on fifteen years of accumulated reality arrives compressed into a roadshow, and it arrives when the company is no longer too small to matter but too large to ignore. The volatility that an early-listing peer would have metabolized in digestible pieces, while it was beneath the market’s notice, instead detonates in a single event, at systemic scale.

And the detonation does not stay contained to the issuer - which is the real heart of it, and the reason these are unlike any IPO’s that have come before. These three are not merely large companies that happen to be entering the index. They are the central nodes of the very economy the rest of the market has been pricing. OpenAI and Anthropic are, between them, among the largest customers on earth for the products that have powered the entire market’s ascent: Nvidia’s silicon, the hyperscalers’ cloud capacity, the power and data-center buildout that has dragged utilities and industrials along in its wake. The relationship runs in a circle that ought to make a careful person uneasy. Nvidia invests in the labs; the labs commit to buy Nvidia’s chips. The hyperscalers invest in the labs; the labs commit to rent the hyperscalers’ compute. Capital flows out of the suppliers, into the labs, and back to the suppliers as revenue - and the public, until now, has had no instrument with which to test how much of the great AI revenue boom is genuine exogenous demand and how much is capital recycling among a closed set of mutually invested players.

That is the disclosure that will reprice everything. When a nearly-trillion-dollar Anthropic and an eight-hundred-billion-dollar OpenAI put audited financials into the public record - their real gross margins, their churn, their cost of inference, the cadence and conditionality of their multibillion-dollar compute commitments, and in OpenAI’s case a projected loss on the order of fourteen billion dollars this year with profitability not expected until the end of the decade - those numbers do not merely set the issuers’ prices. They recalibrate the priors underwriting Nvidia, Microsoft, Amazon, Alphabet, Broadcom, the data-center REITs, the independent power producers, and the long tail of application-layer software companies whose entire cost structure is a function of what the labs charge for a token. For three years the market has priced the picks and shovels off a story about demand from companies whose actual economics no one outside a boardroom could see. The suppliers’ valuations validated the labs’ private rounds; the labs’ private rounds validated the suppliers’ valuations; and the whole edifice marked itself against itself, with no exogenous, falsifiable anchor anywhere in the loop. The AI trade has been, in the most literal sense, an unfalsifiable proposition - a thesis that could not be wrong because nothing in the public record was permitted to contradict it.

IPO’s end that. The S-1 is the first hard surface the narrative has been made to strike. And here is what makes the resulting volatility genuinely outsized rather than merely large: the second-order effects have no precedent and therefore no calibration. The market has a rough working model of how a new consumer-staples listing affects its sector. It has no model whatsoever for how the simultaneous public debut of the three most central firms in the AI economy reprices their suppliers, their customers, and the macro bet that a meaningful fraction of the entire index now expresses. The error bars on the knock-on effects are enormous, in both directions. If the disclosed economics are worse than the priors - if inference is less profitable, if the capex intensity is more punishing, if the demand is thinner or more circular than hoped - the correction does not land only on the issuers. It forces a wholesale re-underwriting of every asset that was priced on the assumption that these labs would consume ever more compute, profitably, forever. If the economics are better than feared, you get the mirror image: a melt-up that pulls the same chain in the other direction. Either way, the move is transmitted through the counterparty web, not merely through the issuers’ own index weights, and it falls hardest on holders who never bought a share of the IPO and never shared in the private decade’s gains.

This is the externality, stated plainly. The moment that is optimal for the private owner to sell - peak narrative, peak bid, maximum accumulated value - is also the moment of maximum valuation across every adjacent asset, and therefore the moment of maximum systemic fragility. Private optimization and public vulnerability are scheduled, structurally, for the same date. The owners chose when to reveal; the rest of the market, having been priced for years off a thesis it was never given the tools to falsify, absorbs the repricing whenever the revelation comes. And because the interim was a one-sided story - sustained by funding rounds and secondary marks that set prices without ever submitting them to adversarial discipline - the surprises, when they come, skew toward the downside. It is easy to inflate an estimate no one is allowed to audit.

None of this is a case against disclosure, and it is not quite a case against private capital, which has done real work compounding hard things slowly. It is an observation about what how our unchecked systems afford this choice, and a market chooses when it permits its own central thesis to remain unexamined for a decade and a half. It chooses comfort over discipline, and it agrees, tacitly, to settle the difference all at once. We have spent years admiring a wager whose terms we were not allowed to read. The reading is about to begin. The question worth sitting with is not whether these are great companies - they, and several others plainly are - but whether a system/market that lets its most important questions go unasked for this long has any right to be surprised by the answer.

_{Disclosures: The securities identified and described do not represent all of the securities purchased, sold or recommended for client accounts. The reader should not assume that an investment in the securities identified was or will be profitable. This article contains opinions which are subject to change without notice. The reader should not assume these are recommendations to purchase or sell any securities discussed herein nor is this investment or financial advice. The views expressed herein are those of Peter Lupoff and do not necessarily reflect those of Beatrice Advisors.}