Unlock Editor’s Digest for free
Roula Khalaf, editor-in-chief of the FT, selects her favorite stories in this weekly newsletter.
The volatility in U.S. stocks this week, triggered by the revelation that OpenAI’s annualized revenue was about $20 billion less than previously reported, exposed markets’ reliance on a vague metric used by AI start-ups to convey their growth to investors.
The FT reported Thursday that OpenAI’s annualized net revenue was close to $50 billion, a figure the company recently shared with investors as part of a fundraising effort. That’s about $20 billion less than the roughly $70 billion first reported by Axios, and then by other media outlets, including the FT, in late September.
In the September article, the FT reported that OpenAI’s annualized net revenue was around $70 billion after the company confirmed that figure was correct.
That gap and the resulting market fallout have put a spotlight on annualized revenue, a fuzzy number used by fast-growing startups to help investors understand their financial trajectory. The focus on the measure has renewed calls for OpenAI and rival Anthropic to become public companies or formalize their financial reporting.
Indeed, publicly traded U.S. companies must regularly disclose their performance to the public in a standardized format.
OpenAI and Anthropic are competing fiercely to develop better AI models, and their revenue projections have become a closely watched barometer for investors and the public trying to gauge the strength of the AI boom.
Although neither company’s shares are publicly traded, shares of AI-related companies including Oracle and Nvidia fell sharply on Thursday after the FT’s revenue gap report. The AI sell-off pushed the tech-heavy Nasdaq 100 down 1.4 percent on Thursday, with the index partially recovering on Friday.
Despite the attention investors give them, the annualized revenue figures are partial, extrapolating recent performance and expectations about the sustainability of customer contracts. Adding to the confusion is the fact that OpenAI and Anthropic use different accounting methodologies to calculate their annualized revenue.
Anthropic recognizes gross revenue on sales through cloud partners. Indeed, Anthropic is the principal involved in the transaction, according to a person familiar with its accounting practices. Claude’s maker-generated revenue rate was calculated by annualizing the previous 28 days of consumption-based revenue, plus recurring monthly subscription revenue, they added.
OpenAI, meanwhile, only counts its own share of revenue from deals with partners such as Microsoft.
The $70 billion figure announced in September resulted from investors’ attempts to compare the performance of Anthropic and OpenAI on a like-for-like basis, according to a person close to OpenAI. However, OpenAI described the $70 billion as a “net” revenue measure when approached by the FT at the time.
OpenAI has always used the net figure in its internal accounting, and even on a gross basis the $70 billion estimate would be inaccurate, the person close to the company said.
In a statement, OpenAI said: “Last week, we shared a remarkable 70 percent annualized revenue growth in the third quarter. Subsequent reports that we had achieved a run rate of $70 billion were not accurate. We can confirm that our current run rate is approximately $50 billion.”
Wall Street’s sensitivity to revenue numbers underscores the importance of OpenAI and Anthropic to the broader AI boom, which has propelled U.S. stocks to record highs and helped fuel economic growth.
Annualized figures give private market investors an idea of a startup’s dynamics – usually more important to venture capitalists than profitability. But OpenAI and Anthropic have reached unprecedented scale and importance for such young, unlisted companies that they both face calls to go public or provide more detailed accounts of their performance.
Anthropic is expected to do so later in 2026, while OpenAI has delayed its listing until next year.
OpenAI’s business performance has taken off in recent months since the company’s GPT-5.6 model was released in July. The company told potential investors in an upcoming funding round that it had revised its full-year 2026 revenue estimate upward to $35 billion, from $30 billion forecast in the first quarter.
Gn bussni

