Key Points
It’s set to be a critical year for privately-held AI companies — especially OpenAI — as investors turn their attention to returns, analysts say. It will be “make or break” for companies whose sole business is selling their AI models, Deutsche Bank wrote in a note on Jan. 20. “OpenAI is particularly extended and may be most at risk as it seems not yet to have found a workable business model to cover its reported cash burn of $9bn last year and likely $17bn this year,” Adrian Cox and Stefan Abrudan, analysts at the investment bank, said. They say that of an estimated 800 million weekly users, “only a fraction” are paying. At the same time, the AI bellwether has committed to data center projects worth an eye-watering $1.4 trillion. OpenAI’s revenue was more than $20 billion last year, up from $6 billion in 2024, according to a blog post by its financial chief Sarah Friar. It is widely expected that the company will go public late this year, or early 2027. The company has inked deals with Nvidia and Microsoft , among others, and raised billions of dollars in the process, giving it a possible valuation of $500 billion . It secured $22.5 billion from SoftBank at the end of last year, on top of $40 billion already committed by the investment company. While it partners with many hyperscalers, OpenAI’s moat is “relatively shallow” compared with larger competitors whose AI playbooks are subsidized by sound business fundamentals elsewhere, Cox and Abrudan wrote, adding “Its path to success appears to be looking narrower and narrower.” “The pressure will only increase as it gets nearer to an IPO, mooted for early 2027 and forecast to potentially top $1trn,” they said. In a blow to OpenAI, on Jan. 12, Apple opted to power its AI products with Google’s technology. On Jan. 16, OpenAI announced it would soon test advertising in ChatGPT — a move founder Sam Altman said in 2024 was “a last resort” as a business model. It represents a new phase for foundation model developers, according to Dimitri Zabelin, a senior investment research analyst covering AI and cybersecurity at PitchBook, as “investor scrutiny shifts from scale to returns, or at minimum to credible improvement in unit economics.” “The key question is whether enterprise monetization, pricing power, and inference cost declines can outpace rising compute intensity,” he said, but added that “OpenAI’s access to strategic compute and capital partners remains unusually deep” due to its multi-year capacity agreements that signal support for its scaling roadmap. Competitor Anthropic, which was founded by a group of former OpenAI employees, is also rumored to be targeting a public listing — potentially as soon as this year. The companies benefit from regulatory tailwinds, Zabelin said, “especially as they continue to embed themselves in government operations more domestically and overseas through sovereign AI initiatives.” Market watchers expect the U.S. Federal Reserve to take a more dovish stance on rates, though concerns of interference have wobbled the market, which could accelerate generative AI funding further despite fears of a bubble, according to S & P Global. Deutsche Bank analysts are unconvinced, however. “It will prove almost impossible for smaller independent companies to afford the accelerating compute costs for models,” they said. “It cannot be ruled out that Perplexity and others end up in the arms of the hyperscalers by the end of the year. Anthropic may be the exception, with a slower cash burn than OpenAI, a product that is particularly popular with coders and — paying — enterprises, and a more dynamic pricing model.”
