In late 2025, OpenAI's leadership said publicly that an initial public offering was not the priority. The stated sequence was to reach a $1 trillion valuation first. Reported figures place the company near a $500 billion private mark after an employee share sale in the second half of 2025 — up from roughly $157 billion in October 2024 and about $300 billion in the spring of 2025. The valuation curve is a straight line. The cash flow curve is not. That divergence is the whole story, and it is a story anyone who has read a token distribution schedule has already seen before.
In 2017, I audited an ICO that raised $12 million. The whitepaper described a utility token, but the emission schedule told a different story: 40% of supply unlocked to insiders in month nine, no vesting cliff, no burn tied to usage. I wrote the flaw out line by line and published it. The community hated it. Six months later the token was trading below issue. The lesson was not that I was clever. The lesson was that a capital structure which cannot be verified by the people funding it is not a capital structure. It is a promise wearing a spreadsheet.
OpenAI is not a token. But the mechanics rhyme. A private company raising on a valuation none of its eventual counterparties can independently audit has taken on a different class of obligation than a public company. Understanding why the IPO is being deferred requires reading the capital structure the way an auditor reads a balance sheet — not as a set of aspirations, but as a set of claims against future cash.
Start with the revenue. According to public reporting, OpenAI's 2024 revenue was roughly $3.7 billion. The 2025 expectation sat near $12.7–13 billion. The 2026 expectation sits near $29–30 billion. Those are extraordinary growth rates. They are also, at a $1 trillion valuation, insufficient. A $1 trillion mark against $30 billion of revenue is a price-to-sales multiple of about 33. Against a more conservative multiple of 10, the same valuation demands roughly $100 billion in annual revenue — three to ten times what 2026 guidance suggests. The gap between the valuation and the revenue is not a rounding error. It is the product being sold.
Now look at the obligations stacked on the other side. The Stargate project, announced in January 2025, carried an initial figure near $100 billion with a stated target of $500 billion. Multi-year cloud commitments to Oracle run into the hundreds of billions across their terms. Semiconductor procurement agreements with NVIDIA, AMD, and Broadcom lock in demand before a single unit of revenue is invoiced. The aggregate of these commitments has been reported at figures approaching the trillion-dollar scale.
Here is where my audit training overrides my optimism. A take-or-pay obligation is a debt instrument with a friendlier name. If a contract requires minimum purchase volumes regardless of demand, it behaves like a fixed liability on the cash flow statement, even if it never appears on the balance sheet as borrowed principal. In traditional credit analysis, we would model it as a lease-like obligation and stress-test the covenant. In the private AI market, it is being modeled as growth.
The circularity sharpens the concern. NVIDIA invests in OpenAI; OpenAI commits to purchasing NVIDIA silicon. Oracle signs multi-year hosting contracts with OpenAI; OpenAI's projected revenue supports Oracle's equity story; Oracle commits to more data-center capacity. Semiconductor vendors take warrants in exchange for purchase commitments. This is not fraud. Every leg is a legitimate contract. But it is a closed loop, and a closed loop can be sustained by its own momentum until the day it cannot. I have watched this exact architecture run in DeFi — the Terra/Anchor reserve, the reflexive collateral loops of 2021. The mechanism is the same. The only difference is that on-chain, the loop was visible to anyone who ran an archive node. Here, it is visible only to the parties inside it.
The governance structure compounds the opacity. OpenAI reorganized into a public benefit corporation in 2025, with a nonprofit foundation holding a minority equity stake while retaining control through special voting rights. This is a defensible structure. It is also a structure that a public-market due-diligence committee will interrogate for months. IPO disclosure requires a clean, unambiguous cap table and a stable control narrative. A nonprofit foundation with special voting rights over a for-profit subsidiary, layered against a $10 billion-plus strategic relationship with Microsoft, is not clean. It is a legal argument waiting to be tested in Delaware and California.
The copyright litigation adds another contingency. Multiple suits from publishers, authors, and rights holders remain unresolved. For a private company, an unresolved suit is a footnote. For an IPO, it is a line item in the risk factors section with an estimated range of loss attached. Every month that litigation stays unresolved is a month the listing window stays partially closed.
So why defer? Because the private side still works. Sovereign funds, SoftBank, and strategic industrial investors have both the capital and the structural incentive to mark the company higher than a public market would. SoftBank's staged investment is the load-bearing wall here; a $1 trillion valuation is, with high probability, not a personal wish but a contractual milestone attached to the tranches. When the buyer and the seller are the same counterparty family, the price stops being a discovery mechanism and becomes a negotiation.
The crypto industry knows this pattern intimately. We spent 2017 and 2018 learning that a valuation set by insiders, for insiders, with no secondary market clearing, is a valuation that has not been tested. Verify everything, trust nothing — not because counterparties are dishonest, but because untested claims fail in the dark.
Here is the contrarian angle, and it cuts against the comfortable pessimism of my own field. The reflexive instinct among crypto commentators is to call this an AI bubble and wait for the pop. That read is too simple. A deferred IPO is not a failure signal; it is an optionality purchase. By staying private, OpenAI keeps its cost of capital below what a public market would charge a company with these loss ratios, and it keeps the valuation narrative under the control of the people who set it. If the 2026–2027 revenue curve delivers 2x growth, the $1 trillion mark becomes defensible and the IPO prices into strength. If it delivers 1.4x, the mark gets renegotiated quietly in private rather than punished loudly in public. The maneuver is rational. It is simply not the strategy of a company confident that the public market would agree with its peers.
The bear market teaches one thing above all others: survival is a function of what you can verify about your counterparties, not what they tell you. The AI capital cycle and the crypto capital cycle are converging on the same structural problem — enormous commitments made against unverifiable future demand, financed by parties who are also beneficiaries. In crypto, we solved part of this by making state public. Every commitment, every unlock, every transfer sits on a ledger any auditor can read. Code is the only law that holds, precisely because it does not negotiate with its own narrative.
The next eighteen months will produce a test. Track the tender-offer pricing on employee share sales; it is the closest thing to a public mark this company has. Track the earnings calls of the cloud and semiconductor counterparties; delayed or restructured commitments will surface there first. Track the copyright rulings. If all three stay quiet, the private round absorbs the risk and the loop keeps turning. If any one breaks, the $1 trillion marker stops being a target and becomes a ceiling.
The interesting question is not whether OpenAI reaches $1 trillion. It is who is holding the last tranche when the verification finally happens.
