AI Guides › Playbooks
By Nigel Guy · 6 min read
Most people reading about "OpenAI's IPO filing" assume the headline numbers came out of the filing. They did not. The filing is confidential, so almost every figure you see is a press report, an analyst estimate, or a company projection, repeated until it sounds official. That feels fine while you read it and fails you the moment you try to decide anything from it.
The rule: before you use any OpenAI number, say which rung of the ladder it sits on, and never let a number from a lower rung borrow the authority of a higher one.
Four rungs, strongest first.
| Rung | What it means | Example from this story |
|---|---|---|
| 1. Filed | In a document the SEC can see and the public can read | Nothing financial yet; the S-1 is not public |
| 2. Announced | The company said it, on the record | The confidential submission; the March 2026 funding round |
| 3. Reported | A journalist or analyst says sources or estimates | Monthly revenue, run-rate figures, loss projections |
| 4. Projected | A forecast of something that has not happened | Profit around 2030, IPO valuation above $1tn |
Put a rung number next to every figure you copy into your notes. If you cannot find the rung, it is rung 4 until proven otherwise.
OpenAI said it had confidentially submitted a draft S-1 registration statement to the US Securities and Exchange Commission. Reporting dates the announcement to 8 June 2026, with the submission itself around 22 May. The company's own wording, as quoted by press, was that it expected the news to leak, so it was announcing it, and that it had not decided on timing and it "may be a while". Goldman Sachs and Morgan Stanley are widely reported as lead banks (rung 3; the banks have not been confirmed in a document we could read).
A confidential draft is a legitimate route: the SEC staff review it privately. For an IPO, the draft and its amendments must be made public at least 15 days before the company starts its roadshow (the investor sales tour). So the real numbers, audited and in the company's own words, arrive late, and only if OpenAI proceeds. As at 4 October 2026 we found no public S-1, no ticker and no IPO date.
These are rung 3 figures, from press and the company's finance chief as reported by others, so treat the direction as firmer than the decimals:
Annualised means a recent month multiplied up. It is not revenue earned in the calendar year, and it is not profit. A full-year 2026 revenue target of roughly $29bn was also reported, which is the sort of figure that shows how far run-rate and booked revenue can differ.
Rung 2, announced: on 31 March 2026 OpenAI closed a $122bn round at an $852bn post-money valuation (about £640bn at time of writing, at roughly 75p to the dollar). Reported investors include Amazon at $50bn, with $35bn of that reported as contingent on an IPO or a technology milestone, plus Nvidia and SoftBank at $30bn each. "Raised" here includes commitments, not necessarily cash in the bank.
Rung 3 and 4, the cost side:
Growth and losses are the same story. A business spending ahead of revenue can look brilliant at run-rate and fragile at the cash line, and you cannot tell which from a headline.
We could not verify that comparison against a dated, sourced calculation, so we do not repeat it. What we can check: Walmart was reported to pass a $1tn market capitalisation in January 2026. OpenAI's IPO valuation talk of $852bn to $1tn is a rung 4 range, and the comparison depends on which day each company is priced. If a post gives you that claim, ask for the date and source for all three market caps.
Until a public S-1 exists, you cannot know audited revenue, the real loss, the share classes and voting control, the nonprofit foundation's stake and rights, customer concentration, or contractual risks with partners. Those sit in sections called Risk Factors, Use of Proceeds and the financial statements. Nothing you read now replaces them.
Imagine a friend sends you a post saying "OpenAI's filing proves $40bn revenue and a $1tn value." You would tag it:
Result: one usable fact, one useful label, one claim to discard. Takes two minutes.
Use this order: the company's own post, an SEC page, a major newswire, then everything else. Several SEO-style summary sites mix rungs freely and sometimes contradict each other on numbers. We used them only to find leads, then looked for a firmer source.