OpenAI’s annualised revenue falls $20bn short, questions over AI boom

OpenAI’s annualised revenue was approaching $50 billion at the end of September, around $20 billion below figures previously reported to investors, according to financial documents cited by the Financial Times. The discrepancy has renewed questions about how the artificial intelligence industry measures growth as technology companies commit hundreds of billions of dollars to computing infrastructure.

The Financial Times reported on Thursday that OpenAI had recently shared the figure of nearly $50 billion with investors, compared with approximately $70 billion widely reported last month. OpenAI declined to comment. The gap appears to stem largely from differences in how the company and rival Anthropic account for revenue generated through cloud-computing partners.

The distinction matters because annualised revenue is a run-rate estimate: revenue generated over a shorter period is projected across a full year. It is not the same as audited annual revenue, nor does it indicate that a company is profitable. For fast-growing private AI businesses, however, investors use the measure to assess commercial momentum and compare competitors.

According to a person familiar with the matter cited by the Financial Times, investors had attempted to adjust OpenAI’s figures to make them more directly comparable with Anthropic’s. Anthropic includes revenue from sales through cloud partners such as Amazon Web Services and Google Cloud in its reported measure, while OpenAI records only its share of certain partner sales. Efforts to adjust the figures upwards contributed to the higher estimates circulating last month.

The revised figure does not mean OpenAI’s business has stopped growing. The company’s annualised revenue was close to $30 billion in July, according to the investor presentation described by the Financial Times. The September figure of nearly $50 billion would therefore still represent substantial growth over the period, although it falls short of earlier expectations.

The immediate concern for markets is whether the commercial returns from AI will justify the enormous cost of building and operating the infrastructure required to serve it. OpenAI has committed to major spending on computing power and data-centre capacity while competing with Anthropic, Google, Meta and other developers. Its expansion also depends on capital and strategic support from major technology partners.

Investors reacted to the report by selling shares in several AI-linked companies. Reuters reported that the Nasdaq Composite fell 1.25 per cent on Thursday, while the semiconductor index declined 3.4 per cent. Nvidia lost 2.9 per cent and Oracle fell 5.5 per cent, amid wider concerns about technology spending, rising oil prices and borrowing costs.

The market response highlights how expectations about private AI companies can influence publicly traded businesses. Chipmakers, cloud providers and data-centre operators have benefited from expectations of sustained demand for AI computing. If revenue growth proves slower than anticipated, investors may reassess the scale and timing of infrastructure investment. However, a change in accounting presentation should not be confused with a corresponding fall in customer demand.

The comparison with Anthropic also underlines a wider problem in evaluating private AI companies: headline figures may not always measure the same thing. Differences in the treatment of partner sales can produce large gaps even when the underlying commercial activity overlaps. Investors need comparable definitions, clear disclosure and actual financial results to distinguish genuine demand from estimates constructed for comparison.

OpenAI’s reported annualised revenue still points to a business expanding rapidly. But the $20 billion discrepancy illustrates the risks of treating a projected revenue run-rate as a definitive measure of the AI sector’s health. As investment commitments grow, the central question for investors is not only how quickly AI sales are rising, but whether those revenues can support the long-term cost of developing and delivering the technology.