The trillion-dollar question for OpenAI: will its bubble burst?

ChatGPT maker OpenAIhas, in the space of a few days, shut down its AI-generated video-making app Sora, ended a $1bn partnership with Disney to license its characters, scaled back plans for “instant checkout” shopping and indefinitely shelved an “erotica” feature that would have allowed verified users to have pornographic conversations.

All this raises a question: is the OpenAI bubble about to burst? The company is looking to list on the US stock market, but as that moment draws closer, there are more people asking how it makes its money and how it can possibly justify its valuation.

The latest funding round for the San Francisco-based tech giant valued it at $840bn (£633bn). Amazon, SoftBank and Nvidia collectively put in $110bn. Microsoft has already poured in more than $13bn, mostly in credits to use its cloud services.

OpenAI has been exploring plans to float at a value close to $1tn. That would be about 80 times last year’s revenues; to put that in perspective, oil company Saudi Aramco’s $1.7tn stock market listing was the most highly valued of all time, priced at roughly five times its revenues. So what exactly is OpenAI selling?

“I think it’s going to be a tough year for all these companies that want to go public at $800bn valuations, when the company might only be worth $400bn,” said Ross Gerber, a tech investor and co-founder of US-based Watt Capital Partners. “In the old days, companies would go public to raise money in a growth stage of the business, but in the current markets, companies go public so private equity firms can get out.”

The race to the initial public offering has been turbocharged by several factors. The first is economics. As the Iran war drags on, AI companies face a squeeze on energy and mounting worries about loans from private companies – two pillars of a vast datacentre expansion that consultant McKinsey estimates could cost $5.2tn by 2030. Public floats, it is hoped, will help replenish pools of capital spent on what Fidji Simo, OpenAI’s chief executive of applications, has termed “side quests”.

OpenAI watered down an original plan to build and own its own datacentres, called the Stargate project. Instead, it relies on partnerships with companies such as CoreWeave, Crusoe and Nvidia. The AI startup was previously planning to spend $1.4tn renting datacentres over eight years – a figure that has now reduced to $600bn over four years.

The second factor is the competition. Anthropic’s coding tool Claude Code has reached more than $2.5bn in annualised revenue, according to a company announcement in February. Neither business is profitable – OpenAI lost a reported $8bn in 2025, and Anthropic burned through about $3bn – but Anthropic is expected to break even by 2028, while OpenAI’s projections suggest it will not turn a profit until 2030.

ChatGPT is still a household name – and a verb – making OpenAI a leader among consumers. The company recently hit $100m in annualised revenue from advertising, which was introduced to fewer than 20% of its free users earlier this year,but it has struggled to turn its 900 million weekly users into the kind of high-value recurring revenue that investors – and public markets – reward.

“AI is not going to cost $20 a month. It’s going to be more expensive over time – just like Netflix,” said Gerber.

The third factor is reputation. The trials of Meta and YouTube may have an echo. OpenAI is facing at least eight lawsuits alleging that ChatGPT contributed to user deaths, including the case of Adam Raine, a 16-year-old from southern California, whose family alleges the bot discussed methods of suicide with him more than 200 times before he took his own life.

Those cases also argue that one of OpenAI’s AI models, GPT-4o, was harmful, designed to foster emotional dependency.

A source close to OpenAI says the lawsuits are based on different technologies – social media and generative AI – and it is misleading to draw direct comparisons. But Meetali Jain, executive director of the Tech Justice Law Project, which has filed seven of the ChatGPT lawsuits, thinks the ruling will carry over.

“The legal theories are the same,” she said. “We have asserted that these are tech products that were put on to the market before they were known to be safe and acknowledged that they were not safe. And so these are defective products about which the companies failed to warn consumers.”

OpenAI has denied responsibility for the deaths and mental health harms, arguing in court filings that Raine bypassed the product’s safety features in violation of its terms of use. The company has also said it is working with mental health professionals to strengthen protections, and has updated its models to better respond to users in distress.

The pornography feature, variously known as “adult mode” and “erotica”, invited further scrutiny, including from OpenAI’s own wellbeing council, which voted unanimously against its rollout. It is these knotty issues that a focus on enterprise avoids entirely, said the analyst Jan-Erik Asplund. “In B2B [business-to-business] enterprise, there’s just none of that at all.”

Photograph by Anna Moneymaker/Getty Images