OpenAI has launched Ads Manager for ChatGPT, opening the product to direct advertiser buying as well as agency demand through Dentsu, Omnicom, Publicis, and WPP. According to the source material, the move came less than three months after the ad pilot began on February 9. The revenue target is large by any standard: $2.5 billion in ad revenue by 2026 and $100 billion by 2030.
Reuters reported that the pilot generated more than $100 million in annualized revenue within six weeks. For now, ads are limited to free users and Go plan users, do not alter ChatGPT responses, and do not involve sharing user data with marketers. That still leaves a core business problem. OpenAI is trying to monetize attention from free users, and that audience may not carry the level of commercial intent advertisers want.
Most ad inventory comes from users who are not paying
ChatGPT is said to have 900 million weekly active users, with about 50 million paid subscribers. That puts the free-to-paid conversion rate below 6%. In practical terms, the ad inventory comes from the remaining 94% of users who are not paying. Scale is obvious. Buying intent is less so.
The source argues that advertisers able to commit at least $50,000 often sell enterprise software, SaaS tools, or B2B services. Those buying decisions are more likely to be made by users already paying $20 to $200 per month for stronger models and larger context windows. Those same users will not see the ads. The mismatch is hard to ignore.
Intent inside ChatGPT is real, but often not transactional
OpenAI’s pitch leans on the idea that users arrive in a conversation box with clear intent. That is only partly true. A user searching for hotels, tax software, or noise-canceling headphones may be close to a purchase. A user opening ChatGPT is often writing emails, translating text, debugging code, building a plan, or sorting through thoughts. The need is real. The buying signal is much weaker.
That leaves ChatGPT in an awkward middle ground. It can sit closer to user needs than a social feed, yet it can be harder than search to classify commercial intent. It is also more private, which makes attribution tougher. The source notes that OpenAI’s move from CPM toward CPC is not just a product update. Advertisers will still ask where a click came from, where conversion happened, and how much budget should actually move from Google, Meta, or TikTok.
Heavy infrastructure spending sits behind the ad push
The ad business also needs to be read against OpenAI’s cost structure. HSBC analysts estimated in late 2025 that OpenAI could face a $207 billion funding gap before 2030. Cloud and AI infrastructure spending between the second half of 2025 and 2030 could reach $792 billion, while long-term compute commitments through 2033 could approach $1.4 trillion.
Under that setup, subscriptions, enterprise API revenue, and fundraising all have limits. Ads look like a faster non-dilutive source of income. They do not require free users to pay directly, and they are easier to present to investors. Still, ad growth will depend on whether marketers believe this audience converts, not on weekly active user numbers alone.
Buying exposure does not guarantee favorable AI judgment
The source also points to a shift in brand strategy. Before Ads Manager, a major concern was how to get cited inside AI-generated answers. With paid placement now available, brands can buy targeted exposure. But the next step for users is often simple: they ask the AI whether the product is actually good.
That changes the conversion gate. Advertisers can purchase impressions, but they cannot purchase positive model judgment. If an AI system draws on public information and gives a negative answer, the ad spend may end up accelerating user drop-off rather than driving conversion. Product quality, review volume, and third-party coverage start to matter more than paid reach by itself.
Anthropic is taking the opposite route
The same source contrasts OpenAI with Anthropic, which said in a February 4, 2026 blog post that Claude would never carry ads, sponsored links, or third-party placements. Its business model is framed around enterprise trust. According to the source, more than 80% of Anthropic’s revenue comes from enterprise customers, and annual recurring revenue rose from about $9 billion to $19 billion. Claude Code and Cowork have contributed at least $1 billion in revenue.
That path is not risk-free either. Stanford AI Index data cited in the source says the cost of reaching GPT-3.5-level performance fell by 280x in two years, from $20 per million tokens in November 2022 to $0.07 in October 2024. If model capabilities keep converging and API price pressure grows, the durability of enterprise subscription premiums becomes a harder question.
OpenAI is bringing ads into the center of its consumer strategy, while Anthropic is turning the absence of ads into part of its value proposition. The approaches are very different, but the underlying issue is the same: if free usage cannot carry inference costs over time, someone still has to pay for the model economics.