OpenAI told investors its annualized revenue at the end of September was about $50 billion, not the roughly $68 to $70 billion that had been widely repeated. Reuters and CNBC both report it, and CNBC says the Financial Times was first. The detail surfaced on October 8, and AI infrastructure stocks fell the same day.
The short version: this looks like a counting difference, not a sales collapse. The story is still worth reading carefully, because a $20 billion gap in a single headline number is the kind of thing that spreads faster than its explanation.
TL;DR: claimed vs verified
| Question | Answer |
|---|---|
| What was claimed? | OpenAI annualized revenue was nearing $70 billion in September (Axios reported "nearing $70 billion"; CNBC cites a $68 billion figure) |
| What is now reported? | OpenAI told investors it was roughly $50 billion at the end of September (Reuters, CNBC, FT) |
| Who confirms? | Reuters cites a person familiar; CNBC says it confirmed the figure with a person familiar. Neither names OpenAI on the record |
| Why the gap? | Per CNBC's source, the $68 billion figure included gross revenue from partners, to compare more directly with Anthropic |
| Is revenue falling? | No. The reported growth figures are 77 percent total run-rate growth in Q3 and 107 percent for enterprise |
| What is not confirmed? | An OpenAI on-the-record statement, the exact partner revenue amount, and the original investor deck |
What was claimed
Late in September, a run-rate figure near $70 billion circulated for OpenAI. Axios reported the company was nearing that mark, with growth above 70 percent since the start of the third quarter. CNBC puts the version that circulated at $68 billion. It spread quickly because it fit a story people already believed: that OpenAI had roughly doubled its pace within a year and was closing on its main rival.
That rival matters here. In August, according to CNBC, Anthropic told investors its annualized revenue run rate hit $65 billion at the end of July. A $68 to $70 billion OpenAI number would have put the two companies neck and neck, and explainx.ai covered the debate over that supposed crossover in Anthropic's $517 billion compute commitments and the OpenAI revenue question.
What is verified
Three outlets now report the lower number. Reuters, citing a person familiar with the matter, says OpenAI told investors its September annualized revenue was "almost $50 billion," a drop from what it had signaled earlier at a separate event. CNBC says it confirmed that OpenAI told investors it "hit roughly $50 billion in annualized revenue at the end of September." The Financial Times, per CNBC, was first to report the number.
All three rely on anonymous sources. That is normal for a private company's investor materials, but it means the figure is reported, not disclosed. OpenAI has no public quarterly filing that would settle it. CNBC notes OpenAI confidentially filed its prospectus with regulators in June, and executives have signaled a 2027 debut, so audited numbers are further out.
Here is what CNBC's source adds about the investor presentation: OpenAI touted 77 percent total run-rate growth during its third quarter and 107 percent run-rate growth for its enterprise business. Those are growth rates on a $50 billion base, not on the larger number.
Why the two numbers differ
According to CNBC, a person familiar said the $68 billion figure included gross revenue from OpenAI's partners, which helps investors make a more direct comparison with Anthropic. Reuters reports only that the September figure was lower than what OpenAI had signaled to investors at an earlier, separate event.
This is the part worth understanding if you read financial AI news regularly. Revenue can be counted two ways:
- Net or direct revenue: what the lab books from its own customers and from what it keeps.
- Gross revenue through partners: the full value of sales made through resellers and cloud marketplaces before the partner's cut.
Per CNBC's source, adding partner gross revenue is what produced the $68 billion figure, which was meant to compare more directly with Anthropic. Leave partner gross revenue out and you get about $50 billion. Neither is wrong in isolation. Comparing one lab's net number to another's gross number is the error.
I could not find the underlying investor deck or an OpenAI statement, so the split between the two bases is the reported explanation, not something I have seen documented. Treat it as the best available account.
What "annualized run rate" does and does not mean
Run rate takes a recent period, usually a month, and multiplies it up to a year. It is a projection, not booked revenue. A fast-growing company's run rate will always run ahead of the revenue it has actually recognized, and a single strong month can inflate it.
That is why the same company can credibly be described as "$50 billion" and "$70 billion" with no one lying: the definition changes the answer. When a number is repeated, ask three things: which month, which basis (gross or net, with or without partners), and who is the source.
Why markets reacted
CNBC reported Nvidia down 3 percent, Oracle nearly 6 percent and CoreWeave nearly 8 percent on Thursday. It also listed AMD and Broadcom down 4 percent, Intel down 5 percent and Super Micro Computer down nearly 5 percent. The common thread is that these companies sell compute to OpenAI, or to the labs that compete with it.
A plausible reading, which is my interpretation and not something CNBC states, is that a lower revenue base makes the gap between OpenAI's compute commitments and its income look wider. We looked at that structure in Nvidia's role as a central bank of AI vendor financing and in the circular revenue and GPU depreciation debate. This is a market reaction to a number, not to a new fact about demand. If the 77 percent growth figure holds, the business is still scaling quickly.
The IPO backdrop
CNBC places the story inside a larger frame. OpenAI is under pressure to justify an $852 billion valuation as it prepares for what is widely expected to be a large IPO. It closed a $122 billion funding round in March, and CNBC previously reported early-stage talks about a new round of around $30 billion, with no term sheet finalized.
Anthropic is preparing its own offering. Reuters previously reported a leaked prospectus showing a $4.6 billion 2025 revenue base and a $42 billion net loss. We covered that in Anthropic's losses and what they mean for usage caps and the earlier Anthropic IPO investor meetings post.
There is a safety thread too. CNBC notes OpenAI recently pulled its plans to launch GPT-6.1 Astra, saying the model did not meet its safety standards, and Sam Altman said in September that "right now would be an ill-advised moment to go public." Our write-up of that decision is OpenAI cancelled GPT-6.1 Astra's October release after safety tests.
What this means for what you build or pay
Finance news only matters here if it changes your costs or your risk. Here is the honest read.
Your bill does not change today. No price or plan change was announced. Do not rewrite a budget off this headline.
Vendor risk is a spectrum, not a switch. A company growing 77 percent a quarter is not in trouble. But two labs competing on a headline revenue number creates an incentive to present the most flattering basis, and that is how the $20 billion gap appeared. When you pick a model vendor, look at product reliability, pricing history and contract terms, not run-rate announcements.
Watch usage limits. Labs that need to justify valuations tend to tighten subsidized tiers. We documented that for plan limits in the Anthropic loss and usage-cap analysis. If you rely on a flat-rate plan for agent work, plan for caps and keep a second provider wired in.
Keep a model-agnostic layer. If a lab's economics shift, you want to swap providers by changing a config, not rewriting your app. This is cheap insurance regardless of whose revenue is whose.
Be skeptical of leaderboards in dollars. The next time a post says one lab "passed" another on revenue, check the basis first. The Anthropic-versus-OpenAI crossover framing depends entirely on whether partner gross revenue is counted on both sides.
How to check a revenue claim yourself
- Find the original source. Is it an on-the-record executive statement, a filing, or an anonymous investor report?
- Identify the metric: annualized run rate, trailing twelve months, booked revenue, or gross bookings.
- Identify the basis: gross or net, with or without cloud-marketplace sales.
- Look for a second outlet with its own sourcing. Here, Reuters, CNBC and the FT each reported the lower number.
- Check whether a later correction exists.
What we still do not know
- Whether OpenAI will comment on the record or publish a reconciliation.
- How large the partner revenue amount is. The roughly $20 billion difference is arithmetic from the two figures, not a reported partner total.
- Whether the $50 billion figure will itself be restated on a different basis in a prospectus.
- Whether the market reaction persists or fades once the counting explanation sinks in.
We will update this post if OpenAI or an investor document clarifies the basis.
Related reading
- Anthropic's $517 billion compute commitments and the OpenAI revenue crossover question
- Anthropic lost $8B operating in 2025: you feel that as usage caps
- Nvidia as the central bank of AI vendor financing
- Circular AI revenue and GPU depreciation
- OpenAI cancelled GPT-6.1 Astra's October release
- Anthropic IPO investor meetings
Figures reflect reporting available on October 9, 2026 from Reuters, CNBC and the Financial Times (via CNBC). Run-rate numbers are projections and sourced to anonymous people familiar with the matter; this post will be updated if OpenAI publishes its own figures.