The Third Platform

Two numbers arrived in The Information this morning. One is already famous: $102 billion. The other is the one that explains it: 36%.

OpenAI's internal documents project $102 billion in advertising revenue by 2030. That's the headline. The 36% is the document's actual disclosure — by 2030, advertising would represent 36% of OpenAI's projected total revenue of roughly $283 billion. The headline says how big the ad business gets. The percentage says what kind of company OpenAI has decided to become.

The Baseline

$102 billion in ad revenue would make ChatGPT one of the two or three largest advertising platforms in the world. For context:

The comparison isn't Google's $240 billion — that took a quarter century. The more instructive comparison is the ramp rate. TikTok, the fastest advertising business in recent history, took roughly four years to reach $14 billion in US ad revenue. OpenAI's projection puts it at $102 billion in four years. That's not extrapolation from TikTok's curve. It's a different curve entirely.

The Velocity

The existing pilot grounds the projection.

OpenAI launched ad testing on January 18, 2026, placing ads below ChatGPT replies for free and Go tier users. The launch came with an early revenue expectation: "low billions" in 2026. Six weeks later, the pilot had already crossed $100 million in annualized revenue with 600+ advertisers. Self-serve advertiser access — the mechanism that transforms a pilot into a market — launches this month.

The $2.4B full-year target implies roughly 24x the pilot rate. That's not a measurement error — it's the difference between a closed pilot and a self-serve market. The pilot rate was set by 600 manually onboarded advertisers. The self-serve launch is what makes the projection competitive with a media plan.

The pricing structure signals that OpenAI isn't competing at the bottom of the market. The $60 CPM it charges — on par with live NFL broadcasts, above Meta's sub-$20 — reflects the bet that ChatGPT intent is more valuable than social intent. An advertiser paying $60 per thousand impressions is paying for a user who is already asking questions, not scrolling past content.

The Mix

The 36% figure is the structural disclosure. By 2030, OpenAI is projecting that advertising will be its second-largest revenue stream — substantial enough to fund scale, but not large enough to define the company the way it defines Google or Meta.

Google generates roughly 80 cents of every dollar from advertising. Meta generates roughly 97 cents. At 36%, OpenAI's 2030 projected revenue mix is more diversified than either — because subscriptions and API/enterprise revenue are growing at the same time. The remaining 64% comes from those two streams.

The ad business doesn't have to carry the company. The ad revenue doesn't fund OpenAI's research or infrastructure — the subscriptions and enterprise contracts do that. The ad revenue funds free access. The $20-per-month subscribers and the six-figure enterprise contracts pay for the model. The advertisers pay for the users who don't.

The Assumption

The $102 billion projection requires something that isn't in the document: that ChatGPT advertising becomes structurally important to media buyers in the same way that Google and Meta are today. Advertisers would have to find that ChatGPT users convert — that the intent signal in a question is more valuable than the interest signal in a scroll. The platform would have to maintain its scale and engagement advantages against Gemini, Meta AI, and Perplexity. And the self-serve launch this month would have to go as well as the pilot.

The $2.4B for 2026 is grounded. The $11B for 2027 is a bet on the self-serve ramp. The $102B for 2030 is a prediction about structural importance. Internal projections at this stage are not commitments. OpenAI's revenue projections have been revised upward at every interval since 2024. They have also been wrong in the other direction — the 2025 loss projections exceeded initial estimates by several billion dollars.

What 36% Means

In December 2024, OpenAI's CFO described the company as "weighing an ads model." In September 2025, Fidji Simo began meeting with potential advertisers. In January 2026, the ads launched. In March 2026, the pilot crossed $100 million annualized. Today, internal documents project a $102 billion business by 2030.

The sixteen-month arc from weighing to projecting follows the same logic as every attention platform. The sequence is consistent: free product → scale → advertising → the advertising funds the free product. OpenAI described the ads as a way to fund "expanded access." That's accurate. It's also the description Google would give of its own model. Today's documents don't change the architecture. They name the destination.

At 36% ad dependency, OpenAI is not building Google. It's building the company Google would build if it could start over with a product users paid for.

The premise of OpenAI was that powerful AI should be broadly accessible. The documents don't contradict the premise. They price it.