OpenAI's $50B vs $70B Revenue Discrepancy Explained

OpenAI and Anthropic aren't lying to anyone - they're just counting money differently. When a customer buys OpenAI's models through Microsoft Azure, OpenAI records only its roughly 20-cent cut of every dollar spent; Microsoft keeps the rest ^[1]. Anthropic, by contrast, books the full dollar as revenue even when the sale runs through a cloud partner like AWS or Google Cloud, treating the partner's cut as a separate expense rather than a revenue reduction ^[1]. Both methods are legitimate, GAAP-compliant accounting choices tied to how each company assesses control of the customer relationship and delivery responsibility - but they produce wildly different headline numbers from similar underlying businesses ^[2].

The $70 billion figure that rattled markets never actually came from OpenAI. It traces back to a September 29, 2026 Axios report, citing an anonymous source, that applied Anthropic's gross-revenue method to OpenAI's known growth rate in order to make the two companies comparable ^[1]. Investors ran with that extrapolated number for over a week until the Financial Times reported on October 8 that OpenAI's actual annualized revenue - on its own net-revenue basis - was approaching $50 billion, about $20 billion below what had circulated ^[3]. OpenAI itself declined to comment on the specific gap ^[4].

Even though the gap was methodological, markets treated it like a demand signal. Nvidia shares fell about 3%, erasing roughly $169 billion in market value in a single session ^[5]. Oracle dropped more than 5%, a reaction amplified by its $664 billion in remaining performance obligations with no disclosed breakdown of OpenAI-specific exposure ^[5]. CoreWeave was hit hardest, falling nearly 8%, reflecting how its valuation rests on a $104 billion backlog funded largely through debt ^[5]. Analyst commentary converged on describing the reaction as disproportionate, noting the accounting difference doesn't change the underlying demand trajectory ^[6]. That tension showed up across social platforms too: X and YouTube commentary largely framed the selloff as an overreaction to a bookkeeping quirk, while Reddit's financial community split between a dominant explainer camp walking through the gross-versus-net mechanics (often via a gross-pay-versus-take-home-pay style analogy) and a more skeptical minority who argued that treating 'annualized run rate' as a hard financial metric, rather than an unaudited projection, was itself part of the problem.

The accounting gap obscures a more interesting fact: OpenAI's underlying growth hasn't slowed. The company's own disclosures to investors show 77% total revenue growth and 107% enterprise revenue growth in the third quarter ^[7]. OpenAI has since told investors it expects to reach or exceed $70 billion in annualized revenue by the end of 2026 anyway, driven largely by enterprise expansion ^[8]- meaning the number investors panicked about missing may simply have arrived early in some reports, measured with the wrong yardstick.