Sam Altman has a lot of things keeping him awake at night

What’s keeping Sam Altman up at night?

Talking to the former Fox News host Tucker Carlson recently, Altman, the chief executive officer and co-founder of Open AI, was only half-joking when he said he hadn’t “had a good night’s sleep since ChatGPT launched” in November 2022. On his mind were the “very small decisions” his company makes that can affect millions of people. He was speaking in the context of a flurry of reports on the sometimes tragic consequences of mentally unwell people turning to ChatGPT for help.

But it’s surely not the only issue swirling around in Altman’s thoughts when he lays down his weary head. To put it mildly, there is a lot coming over the hill for OpenAI for the rest of this year and beyond. None of these struggles looks particularly threatening in isolation. Combine them, though, and you start to wonder whether OpenAI’s challenges, the things it needs to solve to remain competitive, are becoming overwhelming.

First, let’s talk structure. Time is running out to turn into a for-profit corporation. Don’t let the recent memorandum between OpenAI and Microsoft Corp. give the impression that OpenAI’s transition into a public benefit corporation, which technically must be completed by the end of the year, is finalized. It is not. The conscious semi-uncoupling, where both OpenAI and Microsoft are seeking to diversify who they work with on AI and rewrite a revenue sharing arrangement, is just one part of the delicate negotiation in altering OpenAI’s structure. It also has to satisfy regulators in California and Delaware that its new structure doesn’t violate charitable trust laws - complications around which have reportedly had some executives pondering whether the company should move out of California (though OpenAI has denied this). Failure to square this circle puts $19 billion of contingent funding at risk, plus untold future fundraising opportunities.

The company’s costs are skyrocketing. OpenAI’s $300 billion deal with Oracle Corp. for cloud computing over roughly five years from 2027 was enough to propel its chairman, Larry Ellison, back to the top of the list of richest people. But his company will see that money only if OpenAI’s extremely lofty revenue goals are met and its costs drop sharply. By 2030, reports The Information, OpenAI is projecting $200 billion in annual revenue, which is highly ambitious compared with what could be around $13 billion in total revenue this year. How will it make up that difference? More subscribers won’t be enough; it’ll take different revenue models, such as advertising, to fill in the gaps. That’s a great, great uncertainty. While it figures things out, OpenAI estimates it will burn through $115 billion in cash through 2029. Unlike the other companies in the race, like Google and Meta Platforms Inc., there is no highly profitable legacy business underneath it all to keep the lights on. A public listing will surely be needed to buy time.

A significant part of that cost-cutting will be in using its own in-house chips to supplement the ones it buys in from Nvidia Corp. But there’s a problem: It doesn’t yet have its own chips. Its rivals, however, do. Amazon.com Inc., for instance, sells access to its Trainium and Inferentia chips to clients of Amazon Web Services, including a key OpenAI competitor, Anthropic. Google is looking at selling its chips to other cloud providers for AI purposes (in addition to what it sells on its own cloud). OpenAI, in contrast, is still in the nascent stages of this - it recently announced a $10 billion deal with Broadcom Inc. to produce custom chips.

Even if it can do that successfully, there’s a second issue: OpenAI doesn’t own its own cloud-computing platform, either. OpenAI’s lack of a cloud-computing business has become a noted weak spot compared with the already mature (and profitable) cloud platforms belonging to Amazon, Google and Microsoft. Seeking to avoid forever being a tenant, OpenAI came up with the $500 billion Stargate project, announced with great aplomb by Altman and President Donald Trump in January. It has already been beset by delays and such trifling problems as where to build it, how to pay for it and how to power it.

Still, anything feels possible when ChatGPT is as popular as it is today. But what if that wasn’t the case? I’ve written in the past around the muted reaction to the company’s latest model, GPT-5. Extra troubling for Altman is that, on coding, an early breakout use of AI, ChatGPT seems second best to Anthropic’s Claude.

For the mainstream consumer, Altman and former Apple Inc. design guru Jony Ive boldly declared their intention in May to build a device to usher in a post-iPhone world. But it’s unlikely OpenAI can master hardware before Apple can master AI, especially now that the iPhone maker is working with Google to speed up progress on a smarter Siri digital assistant. Both companies have a strong incentive to shut the door to OpenAI’s encroachment on their core businesses.

Google will have been emboldened by the recent court ruling that paved the way for it to continue paying for preferential treatment for its services on iOS, not to mention how it has already tightly integrated its Gemini chatbot into its Android platform and Pixel smartphone. And, just yesterday, Google announced it would finally start integrating Gemini directly into its Chrome browser, by far the market leader worldwide. OpenAI does not have its own browser. It will be extremely difficult for ChatGPT to stay competitive for the consumer if its presence can’t extend much beyond a website and a mobile app. And if the future of AI is through a discreet wearable rather than the screen of a smartphone, how soon can OpenAI come up with something as sophisticated (and broadly adopted) as Apple’s AirPods?

OpenAI employees have until the end of this month to take part in a secondary share sale that values the company at a phenomenal $500 billion - an increase of $200 billion since August. That much of the new money is coming from Softbank Group Corp., with its record of backing stinkers, may be another canary in this particular coal mine. OpenAI’s numbers are detached from reality - especially when you consider much of the top talent that made OpenAI what it is today has left to either start or join competitors.

It increasingly looks as if Altman has learned from his onetime-collaborator-turned-rival Elon Musk, who showed that combining promising technology with untethered rhetoric could send valuations into the stratosphere. The task then becomes about handling the weight of expectations and delivering on those promises. There’s never been a gamble like OpenAI. Someone better get the melatonin.

This column reflects the personal views of the author and does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners. Dave Lee is Bloomberg Opinion’s US technology columnist. He was previously a correspondent for the Financial Times and BBC News.