OpenAI's $50bn annualised revenue differs from $70bn headline by $20bn

The discrepancy underscores the challenge of comparing private AI developers, whose top-line figures influence valuations in some of the biggest private funding rounds to date. For investors with exposure to listed AI firms, from cloud providers to chipmakers, this episode serves as a reminder that annualised revenue numbers from private labs are not audited and can fluctuate based on accounting decisions rather than true demand. Should either company move toward a public listing, accounting methods will face more rigorous scrutiny, as reported revenue and standard earnings disclosures take over from investor-facing metrics. Profit claims that exclude stock-based compensation and precede major computing commitments are likely to face similar scepticism.

OpenAI's reported $70 billion in revenue is actually $50 billion when calculated using its own method, providing a $20 billion lesson on why headline figures from AI companies rarely align.

Summary:

• Axios reported that OpenAI's annualised revenue stands at about $50 billion, roughly $20 billion under the $70 billion figure cited in September.
• According to sources cited by Axios, the higher figure was inflated to allow a direct comparison with Anthropic for investors.
• Anthropic recognizes the entire value of cloud partner sales as revenue and lists the partner's share as an expense, whereas OpenAI only records its own portion of certain partner sales.
• Both firms comply with US GAAP; the distinction depends on which party controls the customer relationship and delivers the product.
• Anthropic reportedly informed investors about an operating profit in the second quarter, but that figure excluded stock-based compensation and mostly preceded large computing agreements.

Axios reported that OpenAI's annualised revenue is about $50 billion, approximately $20 billion lower than the widely circulated figure from last month, attributing the discrepancy to differences in how AI firms account for cloud partner sales.

According to the report, the $70 billion figure, which Axios and other outlets reported in September, was derived from data provided to investors to enable a more direct comparison with competitor Anthropic. Sources told Axios that the number was adjusted upward because investors sought a comparable figure using Anthropic's methodology.

Both companies adhere to US GAAP but handle revenue from partner sales in distinct ways. An accounting expert cited by Axios explained that under Anthropic's method, if a customer pays $100 for an AI service via a cloud provider, Anthropic can book the entire $100 as revenue and record the provider's portion as an expense. OpenAI, in contrast, only books its own share of certain partner sales. The difference hinges on each company's view of its role, including which party controls the customer relationship and is responsible for delivering the product.

This situation underscores the limitations of annualised revenue, which extrapolates sales from a set period to a full year. While commonly used in Silicon Valley, the report noted that public-market investors generally favor reported revenue over measures designed for early-stage start-ups.

Caution is also warranted regarding profit claims. According to the Financial Times, Anthropic reportedly informed investors of an operating profit in the second quarter, but that figure excluded stock-based compensation. The Information also noted that the profit largely predates a series of computing agreements the company has signed, which are anticipated to cost hundreds of billions of dollars.

Until these figures are released in audited form, as is standard for earnings reports, headline revenue and profit numbers from private AI developers will probably continue to be difficult to compare consistently.