The number is a negotiation
"Annualized revenue" sounds precise. It is anything but. Take one month's revenue and multiply by twelve, or take the last quarter and multiply by four, or pick a partial quarter and extrapolate, and you get three different numbers for the same company on the same day. Every choice is defensible in isolation. The choice that produces the bigger number is the one that gets shared.
That is the FT's read on what happened: OpenAI's first $70 billion figure was annualized one way, and the $50 billion figure was annualized another way, with the stated goal of comparability to Anthropic. Comparability is doing heavy lifting in that sentence. It is a reasonable goal for an investor deck and a convenient one when your biggest rival is weeks from an IPO and you are raising $30 billion at a $1.4 trillion valuation.
Timing tells the story
Anthropic is expected to go public around November. OpenAI plans its own IPO next year. Both companies' valuations are pegged, in part, to revenue multiples, and multiples are unforgiving when the denominator moves 29%.
This is not the first time OpenAI's private numbers have sprung a leak. The FT reported last week that an internal presentation projects cumulative negative free cash flow of $278 billion from 2026 through 2030. Read the two stories together and the picture is consistent: a company telling one story about scale and another about the cost of that scale, and the documents circulating among investors are where the two stories meet.
The fine print on the fine print
A few things the FT report does not claim. It does not say OpenAI's business shrank. A $50 billion annualized figure, if it holds, is still an extraordinary revenue line for a company founded in 2015. It does not say the $70 billion figure was wrong rather than differently measured; OpenAI declined to comment, so the company's own accounting of the change is absent. And it does not say either figure would survive an audit, because private companies share what they choose with the investors they choose.
What it does say, with a market for a witness: when a $20 billion revision hits the tape on methodology alone, the number was never really about the quarter. It was about the comparison.
What this means for you
If you run a business that buys into the AI supply chain, or a fund that prices one, treat private-market revenue figures the way the FT just treated OpenAI's: ask which months got annualized, what got included, and who the comparison is built to flatter. The number in the deck is the beginning of the diligence, not the end of it.
Sources
- [1] Financial Times reporting via Dow Jones Newswire (Morningstar): Oracle, AMD, Others Slide After Report of Revision to OpenAI Revenue EstimateRead source