OpenAI 在 6 周内达到 1 亿美元的广告速度

What Happened

On May 5, OpenAI opened its self-serve Ads Manager to every business in the United States, dropped the prior $50,000 minimum spend requirement, and added cost-per-click bidding to a platform that had previously been cost-per-impression only. The expansion came roughly twelve weeks after ChatGPT’s ads pilot first turned on in February, and forty days after Reuters reported the pilot had already crossed a $100 million annualized revenue run rate inside its first six weeks of operation.

The company has now told marketers it expects $2.5 billion in ad revenue in 2026 and $100 billion by 2030. For context, Google’s entire 2001 revenue, the first full calendar year of AdWords, totaled $86 million. Google needed a full year of AdWords plus a multi-country footprint to put up annual revenue still shy of the run rate ChatGPT’s pilot reached in six weeks, inside one country, behind a $50,000 door.

That door is now open.

Key Details

• The pilot ramp: ChatGPT’s ads pilot launched in February 2026 on the Free and Go tiers. On March 26, OpenAI told Reuters the US pilot had exceeded $100 million in annualized revenue. Sam Altman had publicly opposed ads on ChatGPT in prior years.
• What changed May 5: The $50,000 minimum spend is gone. CPC bidding is live alongside the existing CPM model. OpenAI added a Conversions API and pixel-based measurement to let advertisers track purchases, leads, and sign-ups.
• Penetration: At pilot disclosure, roughly 85% of ChatGPT users were eligible to see ads, but fewer than 20% saw one on a given day, and the program already had over 600 advertisers including Target, Albertsons, and Williams-Sonoma.
• Partners: Agency holding companies Dentsu, Omnicom, Publicis, and WPP are integrated, plus tech vendors Adobe, Criteo, Kargo, Pacvue, and StackAdapt.
• Geographic rollout: OpenAI named Australia, New Zealand, and Canada as the next pilot markets, then the UK, Mexico, Japan, Brazil, and South Korea.

Why It Matters

An ad business of that scale cannot be built without breaking the business model of someone else. That “someone else” was already on the floor before May 5.

Chartbeat’s most recent count put AI platforms at roughly 1% of all publisher traffic. The same period saw Stereogum lose 70% of its 2025 ad revenue, Business Insider’s organic search collapse 55% from April 2022 to April 2025 followed by a 21% staff cut, and travel publisher The Planet D fold entirely after a cascading collapse that started with Google’s May 2024 AI Overviews launch (which halved its traffic), accelerated into layoffs, then dropped another 90% before the closure. The New York Times’ search-traffic share fell from 44% in 2022 to 37% in 2025.

The story for publishers used to be that AI was vacuuming their readers. Now the AI is also vacuuming the ad dollars those readers’ eyeballs used to fund. Traffic was the early casualty; ad revenue is the next one. There is not much else left to lose.

For Consumers

ChatGPT remains free on the Free and Go tiers. The bill arrives as sponsored answers slotted into responses. OpenAI says it has seen “no impact on consumer trust metrics” and “low dismissal rates” during the pilot. Anthropic, by contrast, has kept Claude ad-free. That positioning just became a luxury feature.

For the Industry

The platform that absorbed publisher content and rerouted publisher readers now sells access to those same readers, with none of the dollars flowing back to the people who wrote what trained the model. Google’s overall ad business runs at roughly $300 billion a year, so a $2.5 billion ChatGPT business is a rounding error in 2026. But the velocity is the story. Google’s 2002 revenue reached $440 million, helped by a multi-year AOL distribution deal signed that May (which carried a $100 million revenue guarantee from Google to AOL). ChatGPT cleared a $100 million annualized pace in six weeks with no distribution partners and a $50,000 paywall.

For Investors

At a $2.5 billion 2026 run rate and a $100 billion 2030 target, OpenAI’s ad line projects a Pinterest-to-Meta trajectory in five years, not fifteen. The cost side has not improved (training compute keeps repricing), but the revenue side just acquired a flywheel. For Google, this is the first credible search-ad competitor in twenty years.

The Steelman

Three reasons this might land smaller than the run-rate suggests.

The depth problem. Claire Holubowskyj of Enders Analysis told Raconteur that ChatGPT “has enormous reach but low depth compared to established platforms.” Nicole Greene at Gartner pointed out that the platform “lacks the comprehensive demographic data, behavioural tracking, and audience segmentation offered by Google Ads and Meta.” A prompt is pure intent, which is gold for advertising. But intent without identity is harder to retarget than Meta’s social graph.

The FOMO problem. Digiday’s reporting on the pilot found that several brand marketers admitted to joining out of fear of missing the next platform shift rather than expectation of immediate return. Joseph Levi, CEO of Noise Media Group, called the current creative format “essentially display ads, or some type of ads alongside the answers,” and said it is not something brands “need to overpay for.” A $100 million run rate built on FOMO is not the same as a $100 million run rate built on measured return.

The annualization problem. “Annualized” is a run-rate projection from the most recent period of activity, not realized revenue. The actual cash collected during a six-week pilot is substantially less than the headline figure, and the trajectory only holds if subsequent weeks repeat or exceed the recent pace. That projection compounds beautifully or collapses violently depending on the next leg. The CNBC disclosure noted fewer than 20% of eligible users saw an ad on a given day, which is either enormous room to grow or evidence the algorithm is already throttling.

The boring hypothesis runs like this: OpenAI has built a real but modest ad business at roughly Pinterest scale, the velocity story flatters early traction inside a tightly controlled pilot, and Google’s distribution moat across Search, YouTube, and Android absorbs much of the spillover. None of that requires the headline number to be wrong. It requires the rest of the year to behave nothing like the opening weeks.

What’s Next

Timeline:

• June-July 2026: Australia, New Zealand, and Canada come online. First non-US data point on whether the funnel travels.
• Q3 2026: UK, Mexico, Japan, Brazil, and South Korea expansion. The $2.5 billion 2026 target requires the geography to scale faster than the eligibility throttle.
• 2027: A full calendar year of CPC bidding plus Conversions API attribution. The honest test will be whether advertisers stay after the FOMO fades and the data forces a head-to-head comparison against Google Search and Meta Reels.

The Take

The chart that matters is not OpenAI’s revenue line. It is the publisher referral line, already through 1% of all web traffic, and the publisher ad-revenue line, which until last week had no AI-native competitor bidding for the same brand budgets. See the prior site analysis of the Agentic SEO Wipeout for the upstream traffic collapse. The May 5 launch closes the loop. AI search rerouted attention. AI ads now monetize that attention. The training data ecosystem (the web) gets neither half of the deal.

This is not a complaint about disruption. Disruption is a feature of the technology, not a defect. The complaint is about the accounting. The web is the public commons that trained these models. The dollars the models earn from sitting on top of that commons should partially fund its upkeep. Nothing in OpenAI’s launch suggests they will.

The Bottom Line

OpenAI’s ad business hit a $100 million annualized pace in six weeks behind a $50,000 door in one country. On May 5, the door came off. The $2.5 billion 2026 target is now a function of how fast the company can ship CPC bidding to the rest of the planet, and how long Anthropic can keep saying no. By contrast, Google needed a full year of AdWords self-serve plus the May 2002 AOL distribution deal to bring annual revenue to $440 million. Watch the geographic rollout. The $100 billion path lives there.