OpenAI counts $50 billion in revenue, not $70 billion, and chip stocks fall 3.4%

OpenAI has told its investors that its revenue is running more than a quarter below the figure the market had been carrying. No sales vanished: the gap comes from how the sales are counted. That was still enough to send chipmakers from Nvidia to SK hynix sharply lower in a single session, after the Financial Times reported the number and CNBC confirmed it.

That reaction is the story for anyone who does not own a piece of OpenAI, which is almost everyone. The data centres, power contracts and chip orders behind the AI boom are justified by a handful of private-company figures, and those figures feed straight into the stocks that dominate index funds and retirement accounts. When Nvidia, Broadcom, Micron and Oracle fall together on one number, a saver holding a plain index fund feels it.

OpenAI put its annualized revenue at close to $50 billion at the end of September. Reports a few weeks earlier, built on information given to investors, had it near $70 billion, and the Philadelphia Semiconductor Index fell 3.4% on the correction.

Annualized revenue is a startup metric: one month of sales multiplied by twelve. According to a person familiar with the matter cited by CNBC, the higher figure counted gross revenue from OpenAI’s partnerships, an adjustment investors wanted so they could line it up against Anthropic. Anthropic books the full price customers pay through cloud platforms such as Amazon Web Services and Google Cloud, then records the provider’s cut as a cost. OpenAI books only its own share of partner sales. Both methods sit inside accounting rules.

The selling ran down the whole supply chain. Micron lost 4.8%, Broadcom 4.4%, AMD 3.9% and Nvidia 2.9%. Oracle, whose cloud contract with OpenAI has been reported at around $300 billion, fell close to 6%. Outside the US, TSMC dropped 3% and SK hynix 4.4%, which carried the shock to savers in Taiwan and South Korea, where the home markets lean heavily on those two names. The Nasdaq Composite had its worst day since mid-August, although rising oil prices and Treasury yields added to the pressure.

Several analysts argued the market overreacted. Gil Luria of D.A. Davidson said the confusion came from comparing two revenue-recognition methods rather than from any slowdown, and that the pace of the underlying business matters more than the headline. On that pace, OpenAI told investors its run rate grew 77% in the third quarter, with its enterprise business up 107%. The skeptical reading cuts the other way as well. A run rate is a snapshot of momentum, not cash in the bank, and a number shared with investors stood uncorrected in the press for weeks. Public-market buyers tend to discount the next disclosure after that. OpenAI declined to comment to the FT.

The gap matters most to the people about to price the company. OpenAI is negotiating a funding round that would value it at about $1.4 trillion, according to the FT. Against $50 billion of annualized revenue, that is roughly 28 times sales; against $70 billion it looked closer to 20. Suppliers carry the same exposure. Broadcom, which according to the Wall Street Journal has been arranging more than $50 billion in financing for the custom chips it designs with OpenAI, is betting on whichever number proves durable.

OpenAI confidentially filed for a stock-market listing in June and has since postponed it, and chief executive Sam Altman has ruled out an IPO this year. Anthropic is expected to list in November, and the Federal Reserve’s next rate decision follows on October 28.

When Anthropic files, the gross-method revenue line that investors used to inflate OpenAI’s number will face an auditor for the first time, and the AI trade will find out which way of counting it has been pricing all along.

Tags: OpenAI, Anthropic, nvidia, AI infrastructure, semiconductors, Broadcom