Sam Altman spent 18 months drawing a line between OpenAI and the advertising-industrial complex. At a Harvard Forum in May 2024, he called ads-plus-AI "sort of uniquely unsettling" and said he thought of them "as a last resort for a business model." Eight months later, he is running that last resort.
OpenAI has signed up more than 600 advertisers for its ChatGPT ad program, hit $100 million in annualized revenue in under two months, and will begin showing ads to all U.S. users of its free and Go tiers in the coming weeks, according to Reuters. The company will open self-serve advertiser access in April. It just hired Dave Dugan, a former Meta executive, to run the sales operation. The reversal is not subtle.
The most revealing number in the current ad push is not the $100 million ARR. It is the CPM — the cost per thousand impressions. OpenAI initially targeted $60, the kind of rate that puts a product in premium digital territory alongside high-end search. The market is delivering roughly $15. That is not a rounding error. It is a signal that the AI query box does not yet command the intent signal that makes search advertising work.
Search ads succeed because someone types "best running shoes" and the engine knows exactly what they want. ChatGPT, even with browsing access, is a different interaction: people come with questions, not shopping lists. Advertisers are learning that the targeting clarity they get on Google does not automatically transfer to a chat interface. OpenAI is building that inventory and learning what it is worth. The $15 CPM is the opening bid, not the ceiling.
What drove the pivot is equally clear from the numbers. OpenAI had roughly 900 million ChatGPT users as of January. The vast majority, 850 million, pay nothing or nearly nothing. Only a fraction of users cover the subscription line. But paying users account for roughly 66 percent of inference compute, because they use the service more. The free tier is not free to run. OpenAI's inference costs reached $8.4 billion in 2025 and are projected to rise to $14.1 billion in 2026, according to research from Sacra. The company posted a 33 percent gross margin. OpenAI has roughly $1.4 trillion in spending commitments on data centers and related infrastructure. That math does not close on subscriptions alone.
The irony Altman landed in is not subtle. In February, Altman called Anthropic's Super Bowl ads "deceptive" and said OpenAI would "obviously never run ads in the way Anthropic depicts them." He was not wrong about the Super Bowl spots. They showed a character using Claude to write a wedding toast, a product demo dressed as aspiration. But the structural argument for why AI companies should not be in the ad business evaporated the moment the infrastructure bills arrived. What Altman objected to was the style, not the substance.
About 85 percent of U.S. free and Go users are eligible to see ads, but fewer than 20 percent are shown ads on a given day. OpenAI is moving slowly on purpose. The product experience risk of becoming a free ad-supported service is real, and the company knows it. Self-serve access in April is the load-bearing piece: it opens the floodgate from hundreds of advertisers to potentially thousands, which is what turns a pilot into a business.
OpenAI is also raising an additional $10 billion in funding, per CFO Sarah Friar, and told investors it expects to generate more than $17 billion from ChatGPT consumers in 2026. The $17 billion target requires every lever: subscriptions, API, and ads pulling in parallel. Whether the query interface becomes a durable ad vehicle, or whether the CPM stays at $15 while infrastructure costs keep compounding, is what the next year will answer.
