OpenAI filed a confidential S-1 registration statement with the SEC on approximately June 8, 2026, targeting a listing above $1 trillion in market capitalization. The financial picture available to the public before that S-1 becomes public consists entirely of leaked estimates.

Those estimates, which have circulated through analyst notes and press reports since the filing, show $24 billion in annualized revenue and a projected $14 billion net loss for 2026 [1]. The loss-to-revenue ratio the estimates imply — roughly 58 cents in losses for every dollar earned — is not an unusual figure for a company investing heavily in compute and infrastructure. It is, however, an unusual figure to carry into a trillion-dollar IPO when the books are not open.

The public S-1, when filed, will require audited financials that include revenue composition, customer concentration, compute obligations to suppliers including SpaceX and Oracle, data center energy contracts, and GPU delivery schedules. None of those details are currently available [1]. What is available is what OpenAI has chosen to let circulate: a revenue number, a loss number, and a 2030 cash-flow-positive projection. The S-1 will test all three.

The 2030 projection is the one that carries the most structural weight. OpenAI's business model requires that its compute costs grow more slowly than its revenue as model performance improves and inference efficiency increases. If that efficiency curve does not materialize on schedule — if training costs stay high, if inference remains expensive at scale, if enterprise customers prove more price-sensitive than projected — the 2030 cash-flow timeline extends. The leaked projections do not include the compute-cost assumptions that underlie the 2030 target [2].

Sam Altman has told advisers and interlocutors that a valuation below $1 trillion is a "non-starter" for the IPO. That floor creates a specific tension with the market. The company's most recent private-round valuation was approximately $852 billion [2]. A trillion-dollar listing price requires that public market investors apply a higher multiple than private-round investors did — in a market that has, across mid-2026, shown increasing skepticism about AI revenue projections that have not yet been tested by earnings reports.

The company's capital position adds urgency. OpenAI has raised approximately $122 billion across its funding history [3]. At a burn rate that analysts estimate at roughly $6.8 billion per month — driven primarily by compute, engineering, and data center costs — the capital runway is finite. The IPO is not a reward for reaching profitability; it is a mechanism for converting equity into cash before burn outpaces revenue growth.

The sealed S-1 is itself a communication. Companies filing confidentially are typically within six months of going public; the confidential process allows regulators to raise concerns privately before the public prospectus locks in the disclosure record. OpenAI filed in June; a public S-1 release in September remains possible under that timeline, though recent reports suggest Altman may push the listing to 2027 rather than accept a sub-trillion valuation this year.

What the paper can observe from outside the sealed room is the structure of the ask. OpenAI is requesting that public investors commit to a trillion-dollar valuation based on leaked revenue estimates, a 2030 profitability target that has not been audited, and a compute obligation structure that has not been disclosed. That is not a critique of the company's prospects. It is a description of what the current information environment permits investors to assess. The S-1 is what changes that. Until then, the trillion-dollar number is the ask. The books are the answer.

-- THEO KAPLAN, San Francisco