OpenAI may be in too deep

This article first appeared as part of the Daily Sensemaker newsletter – one story a day to make sense of the world. To receive it in your inbox, featuring content exclusive to the newsletter, sign up for free here.

OpenAI has cancelled a multibillion-dollar data centre project in the UK that formed part of Labour’s plan to “mainline AI into the veins” of the economy. The company cited high energy costs and copyright rules.

So what? This is only part of the story. OpenAI is streamlining its strategy as it prepares for a stock market listing, which could value the company at $1tn. The plan is to shed what a top executive has called “side quests” in the face of

• •

  competition from rivals;

• •

  relatively low revenues; and

• •

  limited available computing capacity.

We want it all. OpenAI has spent the last year raising enormous sums of money. In March its latest investment round secured $122bn, mostly from Amazon, Nvidia and SoftBank. This took the company’s valuation to $835bn and broke the previous fundraising record for a private tech firm of $40bn, also set by OpenAI in 2025.

The good side. In the next few months, its ChatGPT tool will reach a billion users. It took Facebook eight years to reach this milestone. TikTok took five. ChatGPT was released just over three years ago.

And the bad. A key concern is the gap between how much money OpenAI makes and amount of cash it is spending to build the computing capacity needed to train its models.

By the numbers. In November, the CEO Sam Altman

• •

  outlined plans to invest $1.4tn over the next eight years; while

• •

  his company was enjoying annualised revenue of $20bn.

Belt tightening. Those spending plans have now been cut back to $600bn over four years. OpenAI has scrapped several projects, including Sora, a video tool underpinning a $1bn deal with Disney, and Instant Checkout, a feature that allowed users to buy products using ChatGPT.

Going to waste. These failed bets consumed precious computing power and failed to make enough money. Forbes estimates that Sora was costing ChatGPT $15m a day. By one count, Walmart purchases with the chatbot were three times lower than on the supermarket’s own site.

In the rearview. Anthropic has become the AI firm of choice for high-value business users thanks to its Claude Code and Claude Cowork, which automate programming and admin tasks. It is much smaller than OpenAI but expects to turn a profit sooner, having focused on fewer products.

Copycat. OpenAI is now turning away from its all-things-to-all-men strategy and doubling down on coding and business users. It has also started running ads in ChatGPT, and there is talk of taking royalties from pharmaceutical firms that make new discoveries with the tool.

The underlying motivation is an expected IPO, which seeks to raise even more capital to fund OpenAI’s computing needs. Investors wary of a bubble will want to see a solid business model and path towards profitability.

The man who would be king. They will also want a steady hand at the tiller. But a New Yorker investigation has uncovered memos that allege that Altman frequently lied and misrepresented facts to his board, which tried to oust him. Several insiders called him “sociopathic”.

King of something. The investigation also details how Altman allegedly rolled back safety guardrails in the pursuit of growth and allegedly lied about elements of a safety review, despite his fears that AI could destroy humanity. Altman has disputed many of the claims and called the article “incendiary” after his home was allegedly targeted by a Molotov cocktail.

What’s more… OpenAI was founded as an ethics-driven nonprofit, “unconstrained by a need to generate financial return”. The subsequent restructuring is central to a lawsuit filed by Elon Musk and going to trial this month. It could be another stumbling block for the company.

Photograph by Florian Gaertner/Photothek via Getty Images