OpenAI says its advertising business is running at a billion dollars a year. It took under two hundred days to get there, having passed a hundred million within six weeks of the American pilot. On 31 August the self-serve platform opened to advertisers in India, Europe, the Middle East and North Africa, putting it in more than forty countries and in front of tens of thousands of buyers.
The figure worth holding is not the billion. It is that a billion dollars is about two and a half percent of the company's revenue.
A rounding error that grew this fast
Two and a half percent is not a business. It is a pilot that got away from its owners. The company is tracking toward more than forty billion dollars annualised, so advertising remains a line item — and that line item went from nothing to a billion in the time it takes to run a product cycle, on a surface that did not previously carry advertising at all.
For publishers this is the uncomfortable arithmetic. The revenue is small relative to OpenAI and enormous relative to the businesses it draws from. A billion dollars of advertising demand did not appear from nowhere. It moved.
The click was the wrong thing to mourn
The publishing industry spent two years describing its problem as the disappearing click — the answer given on the results surface, the visit that never happens, the audience that reads without arriving. The framing was accurate and incomplete. It treated the answer as an end state: something that consumed demand without producing revenue for anyone.
It was never going to stay that way. A surface with tens of millions of daily sessions and commercial intent is an advertising inventory whether or not its owner intended to sell it, and OpenAI had been notably reluctant. The company resisted advertising for years and reversed. The reversal took under two hundred days to reach a billion.
So the answer is not a dead end for the money. It is a new place where the money lands, and it is not where the reporting is done.
What the self-serve step actually signals
The 31 August expansion matters more than the run rate. A managed advertising business, sold by a team to large accounts, is a test. A self-serve platform open across Europe, India and the Middle East is infrastructure — it means small and mid-sized advertisers can buy without a conversation, which is the mechanism by which advertising markets get large and stay sticky.
That is the same progression search advertising made. It went from a sales-led product to a self-serve auction, and the auction is what made it structural.
Advertisers, for their part, are unlikely to treat this as a single number to be optimised against. They have spent the last few years assembling several measures rather than trusting one, and a new inventory with no independent measurement will be bought carefully and reported optimistically for a while yet.
For publishers, the strategic question moved
If the answer surface now monetises, the argument that publishers should be compensated for feeding it stops being abstract. There is a revenue line to point at. It is early, it is small as a share, and it is growing at a rate that makes the licensing conversation more concrete than it was six months ago.
The defensive work does not change: direct relationships, readers who identify themselves, and formats that do not survive being summarised. What changes is the tone of the negotiation. It is easier to argue about a split when the other side has published a number.
The run-rate figures, the launch timeline, the country count, the self-serve expansion of 31 August 2026 and the revenue-share and full-year targets are as reported by CNBC, Reuters and Digiday on 31 August and 1 September 2026, and as stated by OpenAI. The analysis is our own.
Topics mediaadvertisingaipublishingsearch
