A panic about OpenAI’s revenue briefly threatened the whole artificial-intelligence trade. It’s a reminder of how much hangs on the ChatGPT developer’s spending and how nervous investors are that the AI boom is built on a fragile foundation.

Chips, networking, and cloud-computing stocks all dropped Thursday after a Financial Times report stated that OpenAI’s annualized recurring revenue was nearly $20 billion lower than previously signaled—a mere $50 billion rather than the previously reported $68 billion.

Cue panic. Is ChatGPT ‘washed’ or ‘cooked,’ as the kids say? Is the AI bubble popping? Have the past four years of market moves all been a massive mistake?

OpenAI didn’t immediately respond to a request for comment early Friday but there may be an innocent explanation.

“It looks as if the initial knee-jerk reaction has been tempered to some degree, possibly as an insider explained that the $68 billion figure included gross revenues from partners,” wrote David Morrison, senior market analyst at Trade Nation in a research note.

The source of the confusion is that some estimates of OpenAI’s revenue have apparently included business generated via its cloud-computing partners, in an effort to make the numbers comparable with those reported by rival Anthropic.

OpenAI has been happy to let the market concentrate on the largest figure available. But that meant when the FT reported annualized net revenue excluding that figure, it spread fear quickly.

Stocks that were hit hard on Thursday such as CoreWeave, Oracle, Coherent, and Broadcom were only recouping part of their losses early Friday. News that Nvidia-backed cloud-computing firm Firmus has just canceled its plans to launch an initial public offering might be focusing minds on the risks of depending on two mega-IPOs in the form of OpenAI and Anthropic to justify the entire AI trade.

It’s a strange situation where two private companies are essentially driving the wider market. Investors are having to extrapolate snapshots of OpenAI and Anthropic’s revenue rates to annual rates, combined with bits and pieces of third-party data such as payment-processor tracking of spending on AI services.

There has rarely been so much at stake in estimating whether rapid growth is sustainable. OpenAI has committed to roughly $1.4 trillion in spending and has major cloud-computing and chip deals with companies including Microsoft, Oracle, CoreWeave, Advanced Micro Devices, and Broadcom. Anthropic’s spending over the next decade is pegged at $518 billion, according to a Reuters report.

“Our main message in client meetings was the following: the biggest systemic risk in the market next year is…“something going wrong” in the AI ecosystem: a safety event, a failed IPO, or disappointing revenues,” wrote Deutsche Bank analyst George Saravelos in a research note Friday. “Concentration risk is immense, and it is this dollar negative (and very bond positive) event risk that is most under-priced in markets at the moment.”

Until Anthropic and OpenAI go public, a huge amount of market value is vulnerable to any leak or fragment of information that suggests AI spending might weaken—which means bubble concerns will never be far away.