OpenAI’s revenue story changed quickly. In late September, a roughly $68 billion-to-$70 billion annualized-revenue figure circulated as a sign that the company was closing in on rival Anthropic. But the number was designed to make the two businesses look more directly comparable, not to represent OpenAI’s usual top-line reporting.¹

By early October, OpenAI had told investors its annualized revenue was instead approaching $50 billion. The earlier total had effectively included gross sales through partners — a method closer to Anthropic’s treatment of cloud-partner revenue — while OpenAI normally records only its share of certain transactions.²

The distinction is technical, but the stakes are not. Under the partner-sales approach described by accounting professor Francine McKenna, a company can book the full $100 paid by a customer and list the cloud provider’s share as an expense. OpenAI’s approach reflects its view of who controls the customer relationship and delivers the service. Both methods can comply with GAAP; they simply produce sharply different headline figures.¹

Investors, however, saw less nuance and more reason for caution. OpenAI still told backers it achieved 77% total run-rate growth in the third quarter and 107% growth in its enterprise business. Yet the revenue clarification landed as the company faces pressure to justify an $852 billion valuation and the enormous capital being committed to AI infrastructure.³

The market reaction was immediate: Nvidia fell 3%, Oracle nearly 6% and CoreWeave nearly 8%, with other AI-linked stocks also retreating.³ The reset does not mean OpenAI’s business suddenly shrank. It does show how fragile the AI trade can be when a benchmark figure is reclassified — and why public-market investors will want audited-style clarity, not startup-era annualized-revenue constructs, before accepting the next blockbuster valuation.¹