Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
An initial public offering is like a gummy bear rescued from behind a sofa cushion. It may be sweet and delicious, but it also tends to come with a film of miscellaneous detritus. The would-be consumer must decide how much is too much.
CoreWeave, a US data centre operator and the year’s biggest tech IPO to date, has fallen foul of this not-quite-scientific test. On Thursday, it sharply downsized what would have been a $32bn listing to $23bn and cut the amount of money it planned to raise in half to $1.5bn. That seems like a fair, if belated, reflection of some unusual potential pitfalls.
Newly listed companies always carry some grit. They have a limited record, and executives may be unknown quantities. Technology IPOs often bring wrinkles such as rapid growth forecasts, founders with super-voting shares and hard-to-eject boards. More than half of US IPO candidates have admitted to “material weaknesses”, mostly involving their accounting, according to a KPMG report.
CoreWeave has all of these and more. The company is essentially a vehicle through which companies engage in rental access to high-end chips for artificial intelligence. But CoreWeave has turned that simple-sounding idea into a cat’s cradle of stakeholder interests. It depends on two customers, Microsoft and OpenAI, and one chip supplier, Nvidia. These last two are also shareholders, and Nvidia is taking more stock in the IPO.
To comfort investors over these dependencies, CoreWeave has signed up customers for contracts that last several years. As a result, it expects $27bn of revenue between now and 2030. After that, it is hard to predict what demand will look like. Meanwhile, the company carries a heap of debt — interest payments ate up its entire operating profit in 2024.
Those are not necessarily reasons not to go public. CoreWeave is the most direct way for investors to get large-scale equity exposure to the data centre industry. Its reduced valuation equates to less than 10 times ebitda for next year, assuming that profit measure roughly doubles — far below the 14 times of other big-data handlers Meta Platforms and Amazon.
It is just that, to repeat a phrase often heard in financial circles, the company comes with hair on it, and it is not the only one. Cerebras, an AI chipmaker, filed for a listing last year despite getting 87 per cent of its revenue from one Abu Dhabi-based customer. That IPO seems to be on hold for now, saving investors from trying to price in that unusual risk.
Finance has an answer to these questions, in theory. Traditionally, underwriters would price the shares at a discount to fair value of, say, 15 per cent to reflect new stock issues’ inherent dangers.
But when spirits are rowdy, as they have been whenever AI is involved, that adjustment can dwindle to nothing. That seemed the case when CoreWeave initially pitched itself. Its climbdown is a hopeful sign that IPO buyers once again want their gummy bears with less fluff.
john.foley@ft.com
