OpenAI's revenue figure shrank by $18 billion. Most of the gap is about counting.
OpenAI told investors it had $50 billion in annualized revenue at the end of September. A month earlier, $68 billion was the figure everyone repeated. Both can be true, depending on what you count.
Revenue math gets messy fast.
Sixty-eight billion dollars is the number that got repeated in September. Fifty billion is the one OpenAI reportedly gave its investors for the end of the month. Both are “annualized revenue,” which means the company’s most recent monthly revenue multiplied by twelve. Two numbers that close should mostly agree. They don’t, and the reason is a definition, not a sudden drop in sales.
On Thursday, CNBC confirmed that OpenAI had told investors it saw $50 billion in annualized revenue at the end of September. The Financial Times had reported the lower figure first, and the Nasdaq fell more than 1% that day, its worst session since mid-August. By Friday morning the index was rebounding, but the sell-off was real enough that it got its own headlines.
Two numbers, one month apart
The public record, as of Friday, looks like this.
| Source | What it says |
|---|---|
| Reports in September | About $68 billion, which CNBC and others widely repeated |
| FT, as reported by AFP | About $50 billion, a gap of roughly $20 billion from earlier reports |
| CNBC, citing a person familiar | The $68 billion figure included gross revenue from partners |
| SiliconAngle | Calls the difference $18 billion, which matches the $68B to $50B arithmetic |
The $20 billion and $18 billion gaps differ only because the earlier figure was $70 billion in one report and $68 billion in the other. Reporters didn’t disagree about the direction. They disagreed about how big the earlier figure was to begin with. AFP’s summary says the $70 billion figure came from information given to investors, so it was not a number OpenAI published itself.
I can’t read the FT piece, because it’s behind a paywall. So the lower figure comes from AFP’s summary of it, and the explanation for the gap comes from one unnamed person who spoke to CNBC. Treat the “gross partner revenue” line as OpenAI’s side of the argument, not a settled fact.
What “gross” means here
The phrase “gross revenue from partners” points at a real accounting problem. When a chip company or cloud provider also pays OpenAI or buys from it, the money can show up on both sides of a deal. A company can report the whole flow as its revenue, or net out the part that just passes back to a partner. Either way, the figure grows or shrinks depending on the choice.
The reason OpenAI would want the gross figure is comparison. Investors track OpenAI against Anthropic using this one number, and the number only works as a yardstick if both companies count it the same way. The reason to want the net figure is that it’s closer to what customers actually pay for ChatGPT, the API and enterprise seats.
I’d rather see the net number, with the partner flows broken out in a footnote, than a headline total that needs a sentence of explanation to make sense. I don’t know which way OpenAI reports its own books, and nothing I read says.
Why the stock market cared
Analyst Adam Crisafulli at Vital Knowledge made the point I found most useful. He wrote that the gross-versus-net question is not the main issue. His bigger worry is that standalone frontier labs look like “increasingly unattractive businesses,” and that markets are starting to push back on the volume of AI-linked debt and equity being issued. He also expects the AI-linked sell-off to take more than a quick bounce to undo.
That’s a claim about margins and financing, and a revenue correction doesn’t prove it. Revenue tells you how big the business is. It doesn’t tell you what it costs to run. OpenAI’s compute bills are large enough that a $20 billion revision matters mostly because it changes how investors read the rest of the numbers.
The same week, Firmus, an Australian AI data center operator backed by Nvidia, withdrew a planned IPO, citing market conditions. One withdrawal doesn’t make a trend. It does fit the mood.
Where I come down
My honest read: the $68 billion number was never a clean figure, so $50 billion isn’t a collapse. OpenAI is still at roughly $50 billion a year on its own accounting, which is a very large business. What changed is the credibility of the comparison. If you’re following these companies through their numbers, the useful habit is to ask which revenue was counted before you compare anyone’s total.
Watch for two things. First, whether OpenAI or the FT publishes a breakdown of partner revenue. Second, whether Anthropic gets measured on the same basis when the next figure comes out. Our earlier post on Anthropic’s $42 billion loss in its IPO filing makes the same point from the other side: a headline loss or revenue figure means little until you know what’s inside it.
Sources
- CNBC live coverage of the stock market, 8 October 2026: OpenAI’s $50 billion figure, the earlier $68 billion, and the gross-partner explanation from a person familiar.
- AFP via Malay Mail, 9 October 2026: summary of the Financial Times report, including the roughly $20 billion gap and the investor-funding context. The FT original is paywalled, so I could not read it directly.
- Quartz, 9 October 2026: Nasdaq move, the Nvidia and Oracle stock reaction, and Vital Knowledge’s Crisafulli quotes.
- SiliconAngle, 9 October 2026: the $18 billion arithmetic and the market reaction.