The $20 billion gap that shook AI stocks reveals a problem with the industry’s numbers
OpenAI says its annualized revenue run rate is about $50 billion, not the $70 billion cited in earlier reports. The dispute highlights how a widely followed metric can shape investor sentiment even when the underlying figures are difficult to compare.
A report that OpenAI’s annualized revenue run rate was $20 billion lower than previously reported rattled markets over the weekend, weighing on stocks across the AI ecosystem. But the figure, which has become a key metric for evaluating private AI companies, is difficult to interpret. Companies calculate it independently, using criteria that are not always disclosed, making it an unreliable basis for assessing and comparing their actual business performance.
In recent months, as OpenAI and Anthropic have prepared for potential IPOs, discussions of their financial performance have increasingly centered on annualized revenue run rate. Reports about the two companies’ growth and success repeatedly focus on the metric. In January, for example, OpenAI reported a 233% surge in its annualized revenue run rate.
In August, Anthropic overtook OpenAI after reporting an annualized revenue run rate of $65 billion. Then, over the weekend, stock markets came under pressure after the Financial Times reported that OpenAI’s revenue run rate was $20 billion lower than previously reported, at approximately $50 billion in September.
Although OpenAI and Anthropic are private companies, their performance has implications across the technology industry. Microsoft, Oracle, Google, Amazon and Nvidia have significant equity stakes in one or both companies, or maintain major commercial relationships and partnerships with them. As two of the leading players in AI, their financial performance can influence investor sentiment across the broader ecosystem.
Without quarterly financial reports providing a consistent picture of revenue, operating profit and net income, investors have few ways to assess and compare the companies. Figures disclosed by the companies themselves, or reported by journalists, have therefore become central to the debate. But differences in how the figures are calculated can create uncertainty about what they actually reveal.
A revenue run rate is calculated by taking revenue from a short period, such as a month or a quarter, and extrapolating it over a full year, assuming the same pace continues. It is a snapshot rather than a measure of revenue actually earned over 12 months, and it depends on assumptions about whether current sales levels will persist, including whether existing customers will continue to use the company’s services.
The metric is not standardized, and companies have considerable discretion in how they calculate it. That can make comparisons misleading, particularly when companies use different accounting treatments for revenue generated through cloud partnerships.
According to the Financial Times and Bloomberg, Anthropic counts the full gross revenue from product sales made through cloud partners such as Amazon because it acts as the principal party in those transactions. OpenAI, by contrast, counts only its share of sales made through cloud partnerships, most notably with Microsoft. The different approaches can make OpenAI appear to be trailing Anthropic, even when the figures may not be directly comparable.
Investors have tried to recalculate the figures to establish a like-for-like comparison, but have concluded that there is not enough information to do so accurately. According to the Financial Times, reports that OpenAI had reached an annualized revenue run rate of $70 billion stemmed from investor efforts to compare its performance with Anthropic’s.
A company’s revenue run rate can also differ substantially from its actual revenue over a calendar year. According to sources, OpenAI is projected to generate $35 billion in revenue in 2026, about half the run rate it expects to reach by the end of the year. Anthropic offers a similar example: it generated $4.6 billion in revenue last year, while its annualized revenue run rate had reached $9 billion by year-end.
