The Ouroboros

---

On a Monday in June, in a blog post written in the calm register of a company announcing a milestone rather than crossing a Rubicon, OpenAI confirmed what Wall Street had spent weeks anticipating: it has filed, confidentially, to go public. Goldman Sachs and Morgan Stanley are steering. The targeted debut is the fall. The targeted valuation runs from $730 billion to as much as one trillion dollars — which would make the most famous company of the artificial-intelligence age also one of the most valuable ever to ask the public for money.

It is worth stating plainly what that public is being asked to buy, because the prospectus, when it comes, will be written to make the plain version hard to find.

The public is being asked to value at a trillion dollars a company that lost an estimated $14 billion last year and does not expect to turn a profit until around the end of the decade — a company whose own internal documents reportedly forecast operating losses approaching $74 billion in 2028 alone. A company whose cash burn is projected to climb from roughly $27 billion this year to near $63 billion the next. A company that, by its own chief executive's accounting, has signed compute and infrastructure commitments totaling around $1.4 trillion — a number so far beyond its means that the same executive admitted it was "significantly more than the company can currently afford," before the company quietly walked the figure back and began renting the data centers it had vowed to build. And a company whose shareholders, however many billions they put in, will own a thing controlled by a nonprofit that has reserved the legal right to put its "mission" above their money, and to veto the products they are paying for.

This is the most consequential offering of the AI era. It is also, read closely, one of the strangest financial structures ever floated to retail investors. Let us read it closely.

The number that doesn't add up

Every speculative era has its signature number — the one that, in hindsight, marked the moment ambition detached from arithmetic. OpenAI's is $1.4 trillion.

That is the figure CEO Sam Altman himself attached to the company's combined Stargate and cloud commitments: roughly $1.4 trillion of contracted future spending on the data centers, chips and electricity needed to train and run the models he believes will reshape civilization. Set that number beside the company's actual revenue — on the order of $20 billion by the end of 2025, growing fast but still nearly two orders of magnitude smaller than the promises — and the vertigo sets in. This is a company with the revenue of a mid-sized software firm writing checks, payable over the coming years, the size of a national infrastructure program.

Altman did not hide the gap; he more or less dared the market to look at it, conceding the commitments ran far beyond what OpenAI could currently afford. Then, as the concession echoed, the company began to manage it. It reset expectations, telling investors its compute target was "around $600 billion by 2030" — less than half the headline. It pulled back from data-center projects it had announced with fanfare, ceding one Stargate site to Microsoft. And it pivoted, tellingly, from building infrastructure to renting it — signing operating leases with Oracle, CoreWeave, Google Cloud and Microsoft's Azure, converting what had been pitched as nation-building capital investment into the corporate equivalent of a monthly bill. The trillion-dollar vow became a rental agreement. The market is being asked to fund the difference, and to applaud the discipline of the retreat.

The circle

Here is the part that should make a forensic reader set down the prospectus and reach for a diagram, because the money in this story moves in a circle, and the circle is the business. Critics have already given it a name borrowed from mythology — the ouroboros, the serpent that devours its own tail.

Consider Nvidia. It has agreed to invest up to $100 billion in OpenAI — an extraordinary sum, disbursed, by the deal's own terms, as OpenAI builds out and buys more of Nvidia's chips. Read that mechanism twice: the chipmaker funds its customer, and the funding is contingent on the customer spending it back on the chipmaker's products. Nvidia books the revenue; OpenAI books the compute; the same dollars make a loop and are counted, by each party, as growth.

Consider Oracle, which signed a five-year cloud deal with OpenAI reported to be worth as much as $300 billion — one of the largest cloud contracts ever struck — and which is separately buying tens of billions of dollars of Nvidia GPUs precisely in order to lease them back to OpenAI, taking the capital risk onto its own balance sheet so OpenAI can keep the spending off of its as an operating expense. Consider AMD, which agreed to supply chips for six gigawatts of capacity and, as part of the arrangement, handed OpenAI warrants to buy up to 160 million AMD shares — nearly 10% of the company — for one cent each, so that the buyer profits directly if the supplier's stock rises on news of the buyer's orders. Consider Microsoft, which owns roughly 27% of OpenAI, sells it the cloud capacity it runs on, and has separately capped the revenue OpenAI owes it at $38 billion through 2030. Investor, supplier, customer, landlord, warrant-holder — the same handful of names recur in every role, passing the same swelling sums among themselves, each transaction booked as validation of the next.

None of these arrangements is hidden. All of them are, individually, defensible. But step back and the aggregate is a closed financial ecosystem in which the leading indicator of "demand" is a small club of giants agreeing to finance one another's purchases of one another's products. It is the purest expression of a pattern visible across the whole AI boom — and OpenAI is not merely a node in that circle. It is the hub the serpent coils around.

The mission and the meter

Now to the structure itself, which is where the offering passes from aggressive into genuinely unusual.

OpenAI began as a nonprofit, became a "capped-profit" hybrid, and in October 2025 reorganized into a public benefit corporation — OpenAI Group PBC — sitting beneath the renamed OpenAI Foundation. The Foundation, the nonprofit, holds about 26% of the economics. Microsoft holds about 27%. Employees and other investors hold the remaining ~47%. But the economic split is a misdirection. Through special voting rights held by the Foundation alone, the nonprofit appoints every member of the for-profit's board and can replace any of them at any time. Whoever buys the stock, the nonprofit keeps the keys.

It goes further. As a condition of the restructuring, the attorneys general of Delaware and California extracted commitments that hard-wire the mission above the money: a safety and security commission that can veto the release of any new model, and an explicit requirement that, on safety questions, the for-profit board place the nonprofit's mission above for-profit motives. In plain terms, a public shareholder in OpenAI will own a claim on the profits of a company that is legally entitled — indeed obligated — to forgo those profits when its nonprofit parent decides the mission demands it, and whose flagship products can be held back from sale by a body the shareholder does not elect.

There may be excellent reasons, in a company building what OpenAI claims to be building, to bind commercial incentives this way. That is not the point. The point is that a valuation approaching a trillion dollars is being placed on an entity engineered so that the people providing the capital have neither control of the board nor an unencumbered claim on the earnings — and the prospectus will present this as governance, not as the extraordinary subordination of shareholder interest that it is.

The race for the exits

OpenAI is not filing into a vacuum. It is filing into a stampede.

Its closest rival, Anthropic, filed confidentially just over a week earlier, at a reported $965 billion valuation, after touching a trillion on the secondary markets where its shares change hands. SpaceX is queued for a debut near $1.75 trillion. Three of the most richly valued private companies in history are racing to the public markets within months of one another — a concentration of mega-offerings the market has not seen since the dot-com boom, and arriving, as those did, precisely as private valuations have run so far ahead of the fundamentals that the easy money has already been made in rooms the public never enters. The pattern is old and reliable: the richest growth stories stay private while the upside is steep, and are repackaged for everyone else only once the slope begins to flatten.

That OpenAI's own paper has appreciated a relatively modest 11% this year, even as Anthropic's ran up 123%, is its own quiet tell — a hint that, on the secondary markets where the most informed money trades, the appetite for the category leader is cooling even as the leader prepares to sell to the crowd.

The bag, and who holds it

None of this is a prediction that OpenAI fails. The revenue growth is genuinely without precedent — from roughly $2 billion annualized at the end of 2023 to $6 billion in 2024 to north of $20 billion two years later, faster than the companies that defined the internet and the smartphone, with the company telling investors it expects $30 billion this year and a path toward $200 billion by 2030. The technology is real, the demand is real, and it is entirely possible the models keep improving and the spending eventually finds its return. The bulls are not fools. They are betting that the most important company of the age grows into a valuation that today exceeds its revenue some fiftyfold.

But notice what the structure asks of the people at the bottom of it. The public is invited to supply the capital for $1.4 trillion of promises a $20-billion company cannot keep, in a business whose demand is substantially the sound of a few giants financing one another, inside a corporate shell engineered so that control stays with a nonprofit and the profits can be waved away in the mission's name — and to do so at up to a trillion dollars, in the same season that every comparable insider is racing to convert paper into cash.

When a company tells you, through its own chief executive, that it has promised to spend money that does not yet exist, and then asks you for some of yours, the honest thing is to take it at its word. The spending is real. The losses are real. The serpent, eating its own tail, is real. The only open question is who is left holding it when the circle finally breaks — and the structure has already answered that, in advance, in writing. It is not the nonprofit, and it is not the suppliers, and it is not the insiders.

It is you.