On Thursday October 8, 2026, the Financial Times reported that OpenAI had told investors at a recent private briefing that its annualized revenue was approaching $50 billion at the end of September — about $20 billion below the $70 billion figure that had been circulating in the press for a fortnight. The next trading day the Nasdaq 100 fell 1.7%, Nvidia dropped 2.9%, Oracle lost almost 6%, and Micron slid 4% (Digg aggregating FT, Reuters via KSL).

A person close to OpenAI pushed back, calling the $70 billion figure "in the right ballpark" (Reuters via KSL) — a sentence that moved $20 billion by refusing to specify which direction the ballpark lay in. OpenAI itself declined to comment to the FT.

Here is the thing most of the day-after coverage missed, and the reason the story is a Loop piece rather than a wire. The $20 billion didn't disappear. It was never there, in the specific form the Anthropic comparison required.

What's in the gap

Annualized revenue is a run-rate estimate — the strange unit of private-company commerce where you take one recent month, multiply by twelve, and tell investors what that would look like if nothing changed for a year. Nothing is ever changing, in this business, by less than 70% a year, so the number is a projection in the shape of a reporting line.

The difference between OpenAI's $50 billion and the $70 billion figure in circulation is a methodology dispute, not a shortfall. OpenAI does not include revenue from models sold through cloud partners — primarily AWS and Google Cloud Vertex AI — in its headline number. Anthropic does. Anthropic pays its cloud partners roughly 16 cents on each dollar earned through those channels, and sales via AWS and GCP make up about half of Anthropic's revenue (Reuters via KSL). The sales exist for both companies. Only one of them books them as revenue in the number it hands to investors.

When OpenAI's own investors — the same group currently sitting on a round that would value the company at $1.4 trillion pre-money — tried to produce a direct comparison with Anthropic earlier this year, they had to "gross up" OpenAI's figures to add the cloud-partner tier. The grossed-up number got to roughly $40 billion for July. With OpenAI telling its investors the business is growing above 70% since then, the Sept gross-up lands at about $70 billion (Digg aggregating FT). That is where the headline figure came from. It was Anthropic-apples compared to Anthropic-apples, which looks a lot like Anthropic.

The $50 billion is OpenAI-apples. Both numbers are true; they measure different things. The FT's reporting is that the smaller one is the one OpenAI itself is using when it talks to its own investors.

Why the re-measurement happened now

Both companies are inside the IPO window. Anthropic filed its S-1 confidentially on June 1; OpenAI followed a week later. Anthropic is targeting the Nasdaq, with Morgan Stanley, Goldman Sachs and JPMorgan Chase on the deal, and secondary-market demand for its shares is reportedly outstripping OpenAI's by a visible margin.

The Q2 numbers are the subplot the market is actually watching. Anthropic reported $11.5 billion in Q2 revenue; OpenAI reported $6.7 billion — the first quarter in which Anthropic has out-earned OpenAI on a quarterly basis (Reuters via KSL). Anthropic's annualized run rate crossed $65 billion in July, with internal projections for $100 billion by year end. OpenAI started 2026 at $20 billion annualized, up from $6 billion in 2024. The growth is spectacular. It is also now, on the OpenAI-apples basis, running behind.

That is the context for investors doing the gross-up. The comparison matters because at IPO it will be the only number anyone can compare, and the gap between what OpenAI says and what a GAAP-tier auditor is going to let it say is the gap the private market is pricing in now.

What this does to the $1.4 trillion round

OpenAI is in talks to raise at least $30 billion at a pre-money valuation of roughly $1.4 trillion. At the circulated $70 billion annualized figure, the multiple on that raise worked out to about 20 times run rate. At the $50 billion figure OpenAI gave its own investors, the multiple is about 28 times (Digg aggregating FT).

Neither number is unreasonable for a company at this growth rate. 28x of a run-rate figure that will double inside two years is 14x of next year's number, which is in the range the public market has been willing to pay for the top of the AI stack. The problem is not the multiple. The problem is that the multiple has been re-struck in public, mid-round, by a leak that OpenAI didn't want and can't formally deny without disclosing more than it wants to disclose.

The private market will absorb this. The IPO market will remember.

The ballpark that moved

The one direct quote from an OpenAI-adjacent source is the "right ballpark" line. It is a classic non-denial — a sentence engineered to feel like an answer without carrying one. If the actual answer were "the $70 billion figure was the gross-up and the $50 billion figure is what we book," you say that. You don't call the gross-up "right ballpark."

The silence of OpenAI's PR office in response to the FT is the second tell. A company that disputed the FT's reading of what it told investors would issue a correction inside the news cycle. A company that didn't dispute it, but didn't want to confirm it either, lets the "right ballpark" line do its work and moves on. That is what happened on Oct 8 and 9.

Anthropic, for its part, has said nothing. It doesn't need to. The number is on its own filings and the comparison is doing the talking.

What to watch

1. OpenAI's next investor update. If the next round of leaked investor communications shows OpenAI quietly adopting the gross-up methodology — including cloud-partner revenue in the headline figure — the correction has happened, and the IPO prospectus will carry two numbers for two reasons. If it keeps the OpenAI-apples number, the private round is being underwritten on figures the public market will have to re-translate at listing.
2. The S-1s. The question that matters is whether OpenAI's confidential filing uses the OpenAI-apples basis or the grossed-up basis. The answer is in the document; the document will become public somewhere between now and spring 2027.
3. Anthropic's year-end print. The $100 billion projection is now the number the IPO market is going to measure OpenAI against on an apples-to-apples basis. If Anthropic lands at $90-100 billion annualized on the Dec 31 figure, OpenAI's grossed-up ~$70 billion stops being the ceiling and becomes the floor of the comparison. If Anthropic undershoots, the gap closes and the "right ballpark" line was correct.
4. Nasdaq AI capex names through October. The Oct 9 sell-off — Nvidia -2.9%, Oracle -6%, Micron -4% — was the market pricing in a scenario where the AI-infrastructure buildout is being bought by companies with less revenue than the sell-side had assumed. If the sell-off reverses inside a week, the market has decided the methodology dispute is noise. If it holds, something else is happening.

The $20 billion is not missing. It is in the way the two companies agreed, separately, to hold the ruler. The reason the story is a Loop story and not a wire note is that at IPO everyone has to use the same ruler, and OpenAI's investors have just been shown what their company looks like through Anthropic's.