The valuation table is the tell. October 2024: $157 billion. Spring 2025: $300 billion. Late-2025 secondary sales: $500 billion. Target before any listing: $1 trillion. That is a 6.4x re-rating in roughly eighteen months, set against revenue that public reporting places near $3.7 billion in 2024 and a projected $12โ13 billion for 2025. The code whispered secrets the audit missed; here the ledger is doing the whispering, and almost no one is reading it.
OpenAI has reportedly told staff it will not prioritize an IPO until it clears a $1 trillion valuation. I have audited systems that made the same promise with different nouns. In 2022 I spent six weeks reverse-engineering UST's depegging mechanism, and the shape was identical: an asset whose mark was sustained not by cash flow but by a loop of fresh commitments that held only while new capital kept arriving. That loop had an arithmetic end. Every structure with that shape does.
To be fair to the substance, the underlying business is real. OpenAI operates a layered model โ API, subscriptions, enterprise contracts, ecosystem revenue shares โ and the products are used. The distinction that matters for anyone holding exposure, on-chain or off, is between a real business and a valuation that has outrun it. The two can coexist for years. They cannot coexist forever at a fixed multiple.
The capital-commitment side is where the mismatch bites. Stargate's announced scale, multi-year cloud agreements, and silicon purchase plans collectively describe obligations that run ahead of the revenue that would service them. This is the definition of a maturity mismatch: long-dated commitments funded by equity rounds that must keep clearing at the same or higher price.
The reorganization into a public benefit corporation, with a nonprofit retaining minority equity and special voting control, is the institutional backdrop. It is also, read coldly, an IPO obstacle. Public diligence requires a clean capitalization table and a settled control structure. Reports indicate the current arrangement is contested in several jurisdictions and entangled with Microsoft's rights. That is not a footnote to the delay. It is part of the reason for it.
Regulatory timing compounds it. EU AI Act obligations for general-purpose models, US frontier-model reporting requirements, and antitrust attention on the chip-to-cloud-to-model capital cycle all remain unsettled. A listing invites scrutiny of exactly these questions under disclosure liability. Deferring is a way of keeping the answers optional.
What the disclosures do not contain is the part an auditor would circle first: the transaction graph.
NVIDIA invests in OpenAI; OpenAI commits to buying NVIDIA silicon. Oracle books OpenAI cloud orders; Oracle's equity rises on that demand; Oracle builds more capacity for OpenAI. Chip vendors exchange warrants for purchase commitments. Three nominally independent firms, transacting in a closed loop where each leg validates the next. Collateral is a lie; math is the only truth โ and the math here is that the loop is self-reinforcing only because private markets impose no requirement to mark it to market. No 10-Q forces the correlation to be disclosed. The loop runs until it doesn't.
Consider each leg separately. NVIDIA's warrants tie future purchases to equity participation: a supplier financing its own customer's demand. Oracle's infrastructure build is underwritten by contracted OpenAI spend, so Oracle's own valuation is levered to OpenAI's. Microsoft, simultaneously the largest investor and the most aggressive competitor through in-house models and Copilot, captures value that would otherwise accrue to OpenAI. None of these relationships is fraudulent. All of them are correlated, and correlation is what a public prospectus is designed to expose.
The compute commitments are the second finding. Stargate alone has been described in the $100 billion to $500 billion range, with additional multi-year cloud and silicon agreements. If any of these carry minimum volumes or take-or-pay terms, they behave as debt. A commitment that cannot be cancelled is a liability wearing a contract's clothing, and public-market accountants will classify it that way the moment it appears in a prospectus. Private investors โ especially the strategic counterparties who benefit from the spending โ are structurally more willing to call it growth.
Run the multiples. A $1 trillion valuation against roughly $30 billion of projected 2026 revenue implies a price-to-sales near 33x. At a more conservative 10x, it demands $100 billion in annual revenue. Either way, the target requires income to grow another three to ten times, and requires buyers to accept a multiple no listed software company sustains. The reported 2025 operating loss sits in the billions before capital expenditure. I do not trust; I verify the hash โ and the hash says the denominator is far from the numerator.
One structural detail is frequently missed. A $1 trillion target is rarely a personal ambition; it is the shape of a term. Staged investments of the size SoftBank has committed typically carry valuation milestones or ratchets. If the target functions as a trigger, then waiting for $1 trillion is not strategy โ it is a contractual obligation disguised as a preference. That distinction matters enormously to anyone modeling the downside.
The pressure valve is the tender offer. Employee shares have been repurchased in secondary rounds at rising marks โ $500 billion most recently. That mechanism provides liquidity without a listing, but it ratchets the expectation with every reset. Each tender raises the floor the IPO must eventually clear, which is precisely why the target keeps moving upward rather than settling.
Here is the finding that should concern anyone exposed to the AI-adjacent token complex โ render networks, decentralized compute markets, the on-chain proxies that trade on the same narrative. The actual event is not a delay. It is the transfer of valuation-validation risk from public markets onto private investors and supply-chain counterparties. SoftBank's staged capital, sovereign funds, and the vendors booking the orders now hold the last hand. If the mark is ever repriced, it reprices everywhere the same dividend of optimism is traded.
The bulls are not wrong about everything. Compute demand is genuine, and OpenAI's developer ecosystem is the strongest in the field โ API share and tooling lock-in are real moats, not marketing. Unlike Terra, this system generates revenue and ships products; it is not a pure reflexivity engine. It is also defensible to argue that postponing a listing is rational when public markets would apply a commoditization multiple and punish the capital expenditure the roadmap requires. A private market willing to fund at 33x sales is, for the company, a better counterparty than a public one at 8x.
The proof is complete; the doubt is obsolete. The strategy is coherent under exactly one condition: that the technology lead and the financing window both persist for the next two to three years. Neither is a guarantee. It is a wager, and the wager is placed with other people's collateral. The private side may be right. The private side may also simply be locked in.
Watch three signals, not the rhetoric. The next tender or primary round โ the mark it clears tells you whether the private side still believes. The earnings calls of the counterparties: Oracle, Microsoft, CoreWeave. Any deferral or restructuring of committed capacity is the first crack. And the copyright litigation calendar; an adverse judgment converts a governance footnote into a disclosure risk. The loop is stable until the day it is not, and the only honest forecast is a probability, not a promise.