ChatGPT reaches 1.2 billion people a week

OpenAI says 1.2 billion people now use ChatGPT every week. Sam Altman gave the number on stage at the company's DevDay event in San Francisco on 29 September 2026, and OpenAI repeated it in writing on 5 October, in the first line of a post launching a new visual ad format: "ChatGPT reaches 1.2 billion people each week, making it the world's largest AI-native consumer platform."

The comparison point is July, when ChatGPT passed 1 billion weekly users, as The Verge reported. That is roughly 200 million more people a week in about two and a half months, or about a fifth more. In early September OpenAI's finance chief said the company's products reach more than a billion weekly users and 2.5 million businesses, a figure that covered all of its products rather than ChatGPT alone. The new number is for ChatGPT itself.

Why this matters for the map. The ninth of the nine conditions AI Map tracks is whether consumer and enterprise AI agents become a real, token-hungry category. Brad Gerstner put it this way at the All-In Summit on 15 September: "We're going to have consumer agents in everybody's pocket … This could be another trillion dollar category, but it's definitely going to consume massive tokens." Until now the row has had no evidence that AI Map would count. Its rule is AI Map's own and strict: only company-disclosed usage, paid conversion or transactions, never downloads. A weekly user count stated by the company, on stage and then in writing, is the first figure that clears that rule.

So the row turns green. That is a statement about evidence, not a forecast. It means there is now one recent, sourced, company-disclosed data point pointing the way the condition needs to go, and none pointing the other way.

What the number is, and what it is not. It is a measure of reach: how many people opened ChatGPT in a given week. It is not a count of agent tasks completed, purchases made or hotels booked, which is closer to what the condition describes. It is not a count of paying subscribers, and OpenAI did not say how many of the 1.2 billion pay anything. It did not say how it counts a person, whether logged-out visitors are included, or how often the average user returns. Those gaps matter, because a free user who asks one question a week consumes a very different amount of compute from a professional running an agent all day.

That is where the second half of the 5 October post comes in. OpenAI is building advertising into ChatGPT, with a new visual format and expanded measurement for advertisers. In its own words, advertising "allows us to serve a much larger market than subscriptions alone." Put plainly, OpenAI intends to earn from that reach through advertising as well as subscriptions. No ad revenue figure was given.

This connects to the money rows. On the lab-revenue row, OpenAI's annualized revenue was reported near $70 billion on 29 September, from anonymous sources rather than the company (already on the map). On the infrastructure row, the question is who generates the revenue that pays for a buildout measured in hundreds of billions of dollars of compute commitments. A user base of this size is the strongest argument that demand exists. Whether it converts into enough revenue per user to pay for the data centers is a separate question that this number does not answer.

Other figures from DevDay were reported but are not part of this signal. The Decoder reported that OpenAI shared more than 35 million weekly users of ChatGPT Work and Codex, its business and coding products. AI Map could not find that number in an OpenAI written source, so it is context here, not evidence.

On the map, OpenAI's own color does not change. It stays yellow, because the Federal Trade Commission's investigation into OpenAI and other labs, confirmed on 30 September, and the company's decision to rule out a 2026 IPO are still inside the 90-day window. A large user number does not cancel those.

What would move this row next: a company-disclosed paid-subscriber count or conversion rate; a disclosed figure for agent tasks or transactions from OpenAI, Meta or Anthropic; OpenAI breaking out advertising revenue; or a quarter in which a company reports agent revenue separately. A fall in weekly users, disclosed by the company, would count against the row.