OpenAI’s annualized revenue run rate reportedly approached $50 billion at the end of September, roughly $20 billion below an earlier estimate that drew attention across the AI industry.
According to the Financial Times, the ChatGPT maker shared the updated figure with investors. The difference does not necessarily indicate a decline in sales. Instead, reporting suggests it reflects how revenue generated through cloud partners is counted when comparing OpenAI with rival Anthropic.
The discrepancy highlights a challenge for investors and channel partners evaluating the AI market: headline revenue figures may not measure the same thing, even when companies appear to be competing for similar customers.
The illusion of apples-to-apples
The sudden $20 billion gap did not stem from a collapse in sales, but from how start-up financial metrics were dressed up for comparison.
Rival Anthropic includes gross partner sales made through cloud providers such as Amazon Web Services and Google Cloud in its top-line run rate. In contrast, OpenAI traditionally records only its net share of cloud partner transactions.
According to Axios, if a customer pays $100 through a cloud provider, Anthropic records the full amount as top-line revenue and lists the provider’s cut as an expense. OpenAI records only its portion.
To manufacture an apples-to-apples benchmark against Anthropic—which claimed a $65 billion annualized run rate by late July—investors attempted to “gross up” OpenAI’s figures, which is how the $70 billion figure emerged. OpenAI’s July annualized revenue was closer to $30 billion, per the FT.
While both firms remain GAAP-compliant, the conflation of gross and net projections artificially elevated OpenAI’s reported pace.
Tech stocks tumbled Thursday following the report. The Nasdaq 100 closed down 1.4%. Nvidia fell 2.9%, Oracle dropped 5.5%, CoreWeave slipped nearly 8% and Micron declined 4.8%.
The run-rate trap
Start-ups love annualized revenue run rates because multiplying a stellar single month by 12 turns early momentum into headline-grabbing projections. Yet this accounting episode exposes the fragility of using speculative constructs to justify hundred-billion-dollar bets.
When private entities construct vanity benchmarks to match rivals, they mask the operational reality underneath. Real revenue requires disciplined customer retention, recurring usage, and solvent unit economics, not mathematical extrapolation designed to flatter investor pitch decks.
As these frontier labs court trillions in public market capitalization, Wall Street is signaling that synthetic projections will no longer substitute for verified cash flows.
Enterprise budgets and consumer realities
The difference between OpenAI’s reported revenue figures highlights the financial pressures facing AI companies as they invest heavily in infrastructure and compete for enterprise customers.
OpenAI reportedly projects substantial spending through 2030, including on computing infrastructure, while Anthropic also faces significant costs as it expands. These commitments increase pressure on AI providers to turn rapid adoption into sustainable revenue.
For cloud providers, resellers and systems integrators, the discrepancy also underscores the importance of looking beyond headline revenue figures. Gross customer spending does not necessarily reflect how much revenue an AI vendor recognizes or retains, making margins, infrastructure costs and partner agreements important considerations.
Enterprise buyers could eventually face changes to pricing, discounts or contract terms as AI companies seek to improve profitability. However, the reported revenue gap does not establish that OpenAI or Anthropic has announced such changes.
For channel partners, the bigger question is whether the economics behind AI’s rapid growth can support sustainable pricing, reliable services and profitable long-term partnerships.
Other news: Anthropic is expanding its Claude Startups program, offering eligible startups a free year of Claude Team, $1,000 in API credits and up to $45,000 in partner benefits.