OpenAI’s annualized revenue run rate is about $50 billion, not $70 billion as media outlets reported in September. The $20 billion difference is not due to a drop in sales, but to how the company accounts for revenue from partner deals. The previously published estimate was adjusted so investors could directly compare OpenAI’s financial metrics with those of its competitor Anthropic.

Why the numbers turned out differently

According to Axios, the $70 billion estimate of OpenAI’s annual revenue was based on information provided to investors. The company was trying to align its metrics with Anthropic’s methodology, which treats revenue from sales through cloud partners differently.

Both companies follow U.S. generally accepted accounting principles, or GAAP. However, the rules for accounting for partner sales allow them to report the same economic transaction differently.

Imagine a customer pays $100 for an AI service through a cloud provider. Anthropic may record the full amount as revenue and reflect the provider’s share as an expense. In certain partner deals, OpenAI includes in revenue only the portion of the amount that it actually receives.

As accounting professor and The Dig newsletter author Francine McKenna explains, the difference depends on the role the company plays in the transaction: who controls the customer relationship and who is responsible for delivering the service.

Thus, the $20 billion difference primarily reflects different ways of presenting revenue in financial statements, rather than necessarily a difference in the volume of services sold.

Why OpenAI increased its revenue estimate

The $70 billion figure was an attempt to recalculate OpenAI’s revenue using an approach close to the one Anthropic uses. Investors wanted comparable figures that would allow them to assess the market position of the two AI competitors.

Such a recalculation may be useful for comparing businesses, but it is important to distinguish it from the actual revenue reflected in financial statements. The annualized revenue run rate — annualized revenue, or ARR in this context — is calculated by extrapolating sales over a certain period to a full year. It shows the assumed scale of the business if the current pace continues, but it does not mean the company has already earned that amount over the past 12 months.

In Silicon Valley, this metric is widely used to assess fast-growing companies. However, in public markets investors usually prefer actual revenue supported by financial documents rather than calculated metrics, which are especially common among early-stage startups.

Anthropic’s profitability also comes with caveats

A separate issue is the claims about Anthropic’s profitability. According to the Financial Times, the company told investors it posted an operating profit in the second quarter. However, these calculations did not include stock-based compensation expenses.

In addition, a significant portion of these figures relates to the period before the company entered into a number of major computing capacity agreements. According to The Information, the total value of such deals amounts to hundreds of billions of dollars, and their impact on the company’s future expenses could be substantial.

That is why claims about AI companies’ profits should be assessed cautiously. Without financial documents similar to those public companies publish in earnings reports, it is difficult to understand which expenses are included in the calculations and how sustainable the reported financial result may be.

The story of OpenAI’s revenue shows how important it is to consider methodology when comparing the largest AI companies. Even if the figures formally comply with accounting standards, different approaches to accounting for partner sales can create the impression that one company’s business is significantly larger than another’s.