OpenAI revenue is nearly 50 billion dollars, not 70 billion

For a few days in late September, the AI industry had a new headline number. OpenAI, according to several reports, was running at close to 70 billion dollars in annualized revenue. That would have put the ChatGPT maker neck and neck with Anthropic, which has been growing faster than almost anyone expected. Then, on October 8, the Financial Times reported a very different figure. OpenAI had told its investors that its annualized revenue was approaching 50 billion dollars at the end of September. That is roughly 20 billion dollars less than the number that had been making the rounds.

The story was picked up quickly by TechCrunch, Investing.com and Bloomberg Law, and it briefly weighed on shares of companies whose fortunes are tied to OpenAI's spending, including Oracle, AMD, Nvidia and Microsoft. But the most interesting part of the story is not that a number went down. It is why there were two numbers in the first place.

Where the 70 billion dollar figure came from

According to the FT, OpenAI never put out the 70 billion dollar number itself. It was built by OpenAI's own investors, using information the company had shared with them, in an attempt to compare OpenAI directly with Anthropic. TechCrunch summarized it this way: the higher figure came from "attempts by OpenAI's own investors to produce a direct comparison with Anthropic's annualised revenues".

That sounds harmless. Investors want to know who is winning, and the obvious way to find out is to put both companies on the same scale. The problem is that the two companies do not report revenue the same way, and the investors' adjustment apparently went in the wrong direction. Reports from Crypto Briefing and SiliconANGLE describe the adjustment as an attempt to apply Anthropic's accounting style to OpenAI, which pushed the OpenAI figure up to around 68 to 70 billion dollars. OpenAI's own figure, the one it actually gave investors, is close to 50 billion dollars.

There is a second detail worth noticing. OpenAI reportedly also told investors that its revenue grew by more than 70 percent over the period being discussed. So the company is still growing very fast. It is just smaller than the most optimistic headline suggested.

Gross versus net: the boring detail that moves billions

The whole gap comes down to one accounting choice. When a business customer buys Claude through Amazon Web Services or Google Cloud, Anthropic records the full amount the customer pays as its own revenue, and then pays the cloud partner a share as a cost. That is called booking revenue gross. Reuters reported that partner sales made up about half of Anthropic's revenue in 2025, and that Anthropic pays the partners roughly 16 cents of every dollar earned through them.

OpenAI, by contrast, counts only its own share for certain partner deals, the part that actually reaches OpenAI after the partner takes its cut. That is booking revenue net. Neither approach is wrong in itself. Accounting rules allow both, depending on who is considered the main seller to the customer. But the result is that the same underlying business can produce very different headline numbers.

A simple example makes it clearer. Imagine a company buys 100 dollars of AI model access through a cloud marketplace. Under the gross method, the AI lab reports 100 dollars of revenue and 16 dollars of costs. Under the net method, the AI lab reports about 84 dollars of revenue. Nothing about the actual cash changes. Only the headline does. Multiply that by tens of billions of dollars in sales and you get a gap large enough to rattle the stock market.

So who is actually bigger?

This is where things get uncomfortable for anyone who wants a simple scoreboard. Anthropic reported an annualized revenue of about 65 billion dollars in July. If you compare that with OpenAI's own figure of about 50 billion dollars at the end of September, Anthropic looks roughly 15 billion dollars ahead. If you compare it with the investor built estimate of 70 billion dollars, OpenAI looks about 5 billion dollars ahead.

Both comparisons are flawed. They compare different months, and they compare a gross number with a net number. A fair comparison would need either Anthropic's revenue restated on a net basis or OpenAI's restated on a gross basis, and neither company has published that. What we can say is that, on the numbers each company gives out about itself, Anthropic currently reports the larger figure. Reuters also reported that Anthropic's quarterly revenue exceeded OpenAI's for the first time in the second quarter, which points in the same direction, again with the caveat about partner sales.

We looked at a related part of this rivalry earlier this week, when OpenRouter data suggested that business spending on OpenAI and Anthropic had become roughly even after a summer of price cuts. You can read that analysis in our article on the AI price war.

Why markets cared so much

At first glance, a revenue run rate is just a projection. It takes recent monthly revenue and multiplies it out to a full year. So why would a change in that number move shares of companies that are not OpenAI?

The answer is that OpenAI's revenue has become a kind of proxy for the whole AI buildout. Oracle has signed huge data center contracts with OpenAI. AMD and Nvidia sell the chips that go into those data centers. Microsoft is OpenAI's biggest partner and investor. When OpenAI's revenue looks bigger, those commitments look safer. When it looks smaller, investors start asking how a company with about 50 billion dollars in annual revenue will pay for hundreds of billions of dollars in infrastructure.

That question is not new. OpenAI raised 122 billion dollars in a single funding round in March. Its leaked 2025 financials earlier this year showed about 13 billion dollars in revenue and much higher spending. The company has signed long term deals for compute that dwarf its current sales. We recently looked at the same pattern at Anthropic, whose IPO prospectus shows at least 518 billion dollars in compute commitments, most of them impossible to cancel. The leading labs are spending ahead of revenue on a scale that has few precedents.

The 70 billion dollar number may come back

There is an ironic twist. According to Bloomberg, as summarized by The Decoder on October 9, OpenAI expects its revenue pace to reach at least 70 billion dollars by the end of 2026. In other words, the number that turned out to be wrong for September could become correct, on OpenAI's own counting, within about three months.

That would require very fast growth. Going from about 50 billion to 70 billion dollars in a quarter means adding roughly 40 percent. OpenAI has a lot of levers: ChatGPT subscriptions, the newer GPT 6 models that rolled out to paying users in early October, enterprise deals, Codex for developers and the API. It also cut prices sharply on some models, which can grow usage but squeeze revenue per customer. Whether the target is realistic will be one of the most watched numbers in the industry over the next few months.

What this means for the IPO

OpenAI's IPO, which some had expected this year, is now reportedly pushed to early 2027. Before a public listing, a company must publish audited financial statements under strict rules. At that point, the question of gross versus net will not be left to investors and journalists. It will be written down, checked by auditors and open to everyone.

That is good news for anyone trying to understand the AI market. Right now, almost every big number in this industry comes from leaks, selective disclosures to investors or reports based on anonymous sources. Anthropic's own prospectus is still confidential, and its figures reach the public only through outlets like Reuters. Once both companies file public documents, it will be much easier to compare them on equal terms.

Until then, a few practical rules help when you see the next big revenue headline.

• Ask who produced the number. A company's own disclosure, an investor's estimate and a journalist's calculation are three different things.
• Ask how partner sales are treated. Gross and net figures for the same business can differ by billions of dollars.
• Check the month. Run rates move quickly in AI. A July figure and a September figure are not a fair head to head comparison.
• Remember that a run rate is not annual revenue. It is a snapshot multiplied by twelve, and it can swing with a single large contract.

Dany's take

I find this story more reassuring than alarming. OpenAI did not lose 20 billion dollars of business. A rough estimate built by outsiders was replaced by the company's own figure, and the company is still growing at more than 70 percent. What it does show is how fragile the public picture of the AI economy is. Trillions of dollars in market value are moving on numbers that nobody outside these companies can check. The sooner OpenAI and Anthropic publish audited accounts, the sooner we can stop arguing about yardsticks and start talking about what these businesses are really worth.

The bottom line

OpenAI's annualized revenue was nearly 50 billion dollars at the end of September, not 70 billion. The difference comes from investors applying Anthropic's gross accounting to OpenAI's net figures. On each company's own numbers, Anthropic currently reports more revenue, though the comparison is imperfect. OpenAI is aiming for at least 70 billion dollars by the end of the year, and audited numbers will only arrive with an IPO, now expected in early 2027.

Sources

The core figure was first reported by the Financial Times on October 8, 2026, and summarized by TechCrunch, Investing.com and Bloomberg Law. The accounting background is based on Reuters reporting and on Crypto Briefing. The original FT article is here (paywall). All numbers are as reported and were not independently verified.