OpenAI would rather wait two years than let the market price its trillion dollars

The headline number this week is a trillion dollars, and the second number is 2027 — the year, The New York Times reported, that OpenAI now plans to go public, having quietly decided not to do it this year after all. The two numbers are usually read as one fact: OpenAI is worth a trillion and will list when conditions suit. Read them as a choice instead, because that is what they are. OpenAI's advisers, by the same account, laid out two options. Go public now, at a price the market would actually pay, which would be lower than a trillion. Or wait until 2027 and keep the trillion intact. Sam Altman, employees were told, treated the first option as a non-starter. That is not a scheduling decision. That is a decision about which kind of number the company is willing to live with.

There are two kinds of valuation, and the gap between them is the whole story here. A private valuation is a negotiated number. It is set by a handful of late-stage investors who agreed, in a room, on the figure that would appear on a term sheet — people who are already long the story and have every reason to want the mark high. A public valuation is a discovered number. It is whatever the marginal buyer, the one indifferent person who could just as easily own something else, will pay on a Tuesday. Most of the time the two roughly agree. The interesting moments are when a company that has only ever had the first kind goes looking for the second — and OpenAI has just decided it does not want to find out, not yet.

Why a lower mark is a "non-starter"

It is worth being precise about why the trillion is load-bearing, because "founder's pride" is the lazy answer and the wrong one. The mark is structural. It is the reference price for the equity that thousands of OpenAI employees hold, much of it as the reason they have not left for a competitor; reprice the company down by a third and you have repriced their compensation by a third overnight, which is how retention problems start. It is the figure that anchors the next private round, the one that bridges the company to whenever it does list. And it is the implicit collateral under a web of commitments — compute contracts, partner equity, the revenue-sharing arrangement with Microsoft — that all read more comfortably against a trillion-dollar counterparty than against a company the public market just marked at six or seven hundred billion. A lower public price does not stay politely inside the IPO. It travels backward through everything that was underwritten by the higher one.

So the instinct to protect the number is rational. The problem is that protecting it by not testing it has a cost of its own, and that cost is the part no one on the call wants to name. Waiting is a position. It is a bet that 2027 will offer a friendlier market, a fuller revenue line, and a calmer view of AI valuations than 2026 does. It might. But a company that defers price discovery because it fears the answer is telling you something about the answer.

The comp is sitting right there

OpenAI does not have to imagine what public price discovery would do to a mega-private mark, because it can watch it happen in real time. SpaceX listed on June 12 in the largest IPO ever recorded. It priced at $135, opened at $150, and ran — within a couple of weeks it touched roughly $225 a share, a level that briefly made it the fourth-most-valuable public company in America, ahead of Amazon and Microsoft. Then it reversed. A one-week slide of around 35% erased more than $600 billion of market value; the worst single day was down 16.4%. By this week the stock was back around $153, leaving it only modestly above its IPO price and well below the peak the early holders had already mentally banked. The biggest listing in history round-tripped most of its gains inside a fortnight.

The Times reporting makes the link explicit: part of why OpenAI wants to wait is precisely that SpaceX's shares have fallen since the debut. That is the comp doing its work. SpaceX is not some cautionary tale from another cycle; it is the freshest, largest data point on what happens when a colossal private valuation meets a public order book — and the answer it gave was that the marginal public buyer is far less convinced than the last private one. The mechanism is not mysterious. In the private market the buyer is a strategic investor who wants the number to hold because they are inside the trade. In the public market the buyer is a stranger who can sell tomorrow, and increasingly a passive fund that will own only what the index forces it to. The believers and the sellers stop being the same people. That is the repricing, and it has no brakes once it starts.

A private valuation is a number you negotiate. A public valuation is a number you discover. OpenAI has just decided it would rather keep negotiating.

Who is exposed to the trillion

Concentration is usually a story about an index. Here it is a story about a single private mark and everyone hanging off it. Employees hold paper wealth denominated in the trillion and cannot sell it. Late-stage investors — the funds that came in at the top of the last round — carry it on their books at cost, and their own returns to their own backers are reported against it. And then there is the wider AI-capital machine, the daisy chain of compute deals and vendor financing in which one company's valuation quietly serves as another's comfort. When the marks are all private, they can all support each other, because nobody is forced to print a price. The risk in this structure is not that a trillion is wrong. It is that a trillion is untested, and an entire layer of commitments has been built on the assumption that it is right.

Set the number against the operating reality, hedged as it should be. OpenAI's revenue has been reported climbing steeply — the often-cited ramp runs from a couple of billion toward a run-rate in the low tens of billions — against reported cash burn in the same enormous neighbourhood and a gross margin said to sit around a third. Those are figures a public filing would force into the open and a public market would price without sentiment. A trillion-dollar valuation on a business still spending roughly what it earns is not impossible to justify, but it is a valuation that asks the buyer to underwrite years of future growth as though it has already happened. Private investors are comfortable doing that. The public, on the evidence of the last fortnight, is in a more selective mood.

The new risk factor nobody had to disclose before

There is a second reason the timeline slipped, and it belongs in any honest read of the delay. The same week brought word that OpenAI's next flagship model, GPT-5.6, will not get a wide release on the company's own schedule. Its rollout is being staged through a limited preview for selected partners, with the government — the White House's Office of the National Cyber Director and the Office of Science and Technology Policy — approving access on a customer-by-customer basis during the preview. Whatever you make of the security rationale, consider what it does to a prospectus. The cadence of the company's most important product is now partly a matter of government discretion. That is a genuine risk factor, the kind that did not exist on the confidential S-1 the company filed earlier this month, and it is very hard to price a flagship AI business while the release schedule of its flagship is somebody else's decision to make.

Stack the two together and the delay reads less like patience and more like a company trying to hold a particular pose: almost-public, with a confidential filing already in and the optionality that comes with it, but not actually public, and so not actually subject to the discipline of a discovered price. It is a reasonable thing to want. It is also, in market terms, having it both ways — keeping the trillion-dollar mark as a fact while declining the one process that would tell you whether it is a fact or a hope.

The downside no one is pricing

Separate the company from the trade, the way you always should. OpenAI may well be the defining company of this era; it can be that and still be a dangerous price at a trillion dollars in a 2026 public market, and nothing about waiting until 2027 changes which of those questions is which. What waiting does is move the test, not pass it. If revenue grows into the number by then, the delay will look like discipline. If the AI-valuation mood that took $600 billion out of SpaceX in a week is still around — or worse — then all the company has bought is two more years of building commitments on a mark that has never met a real bid. The trillion is safe for exactly as long as no one is allowed to bid on it. That is a comfortable place to stand, right up until the moment you need the door, and the only price on the board is the one you spent two years refusing to discover.