The $200 billion forecast for OpenAI and Anthropic is a pricing story, not just a growth story

The speed at which frontier AI labs have accumulated revenue is, by now, widely cited. What is less often examined is the mechanism. Strip away the superlatives, and the story that emerges is about pricing architecture, not just product demand.

Elad Gil put the baseline simply: both Anthropic and OpenAI each reached $1 billion in revenue in approximately one year. That figure is striking on its own. Marc Andreessen gave it scale, arguing that the two companies are currently adding more revenue per month than Meta, Google, or Microsoft. Whether or not every reader accepts those comparisons at face value, the directional point is hard to dispute: the ramp is faster than almost any prior technology business at equivalent stage.

Andreessen went further, saying he would not be surprised if the combined revenue run-rate of the two companies reached $200 billion by the end of this year. Gavin Baker arrived at a similar number by a different route. His argument is structural: “The shift to usage based pricing is probably why you will see OpenAI and Anthropic exceed well over $200 billion in ARR this year.” That distinction matters. A subscription model caps revenue at the number of seats sold. A usage-based model scales with how deeply the product is embedded in workflows, and that ceiling is meaningfully higher.

The usage-based point deserves more attention than it typically gets in coverage of AI revenue growth. When a company bills by the token, by the call, or by the task completed, its revenue curve is not driven by net-new customer acquisition alone. It is driven by intensity of use among existing customers. That is a different kind of compounding, one that is both faster and more durable when the underlying product becomes load-bearing in a customer’s operations.

The shift to usage based pricing is probably why you will see OpenAI and Anthropic exceed well over $200 billion in ARR this year. Gavin Baker

Brendan Foody extended the bull case to a horizon that will read as aggressive to most: at least one of the two companies reaching a $10 trillion valuation within five years. That is a different category of claim. It depends not just on sustaining the current growth rate but on the terminal size of the market these companies are addressing. Foody’s framing treats the current trajectory as a floor rather than a ceiling, which is a bet on continued model improvement, continued enterprise adoption, and a pricing regime that holds. Each of those is a real uncertainty.

What is not uncertain, or at least not seriously contested by the people tracking the numbers closely, is that the current pace is exceptional. The comparison to Meta, Google, and Microsoft is not flattery. Those are companies that have had decades to build revenue infrastructure, distribution, and enterprise relationships. The AI labs are doing something faster than those companies did it, and the usage-based model is a plausible explanation for why.

The risk to the forecast is also visible in the same structure. Usage-based pricing rewards intensity of use, but it also exposes revenue to any shift in how customers use the product. If a major customer finds a cheaper alternative, switches models, or brings capability in-house, the revenue impact is immediate rather than spread across a contract term. The same elasticity that accelerates the ramp can accelerate a reversal. That is not a reason to dismiss the $200 billion figure. It is a reason to watch the customer concentration and churn data as carefully as the headline run-rate.

The honest read on the evidence is this: the growth is real, the pricing shift is real, and the forecasts are internally consistent given the inputs. The $10 trillion valuation claim sits further out on the risk curve and requires assumptions that the current data cannot confirm. For now, the more grounded story is the one Baker is telling: the structure of how AI is priced has changed, and that change is doing more work in these growth curves than most of the commentary acknowledges.