Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
Imagine this set-up for a new stock market listing. A transformational new technology has sparked an infrastructure spending boom. Entrepreneurs from outside the tech industry have spotted the opportunity to borrow heavily to build a new type of infrastructure company, narrowly focused on feeding the new demand. With Wall Street hungry for pure-play ways to invest in the new technology, the conditions for an IPO would seem opportune.
That could be a description of CoreWeave, the wholesaler of AI computing power. Its shares are set to start trading on Wall Street on Friday in a litmus test for the state of the AI capital spending boom.
But it could also describe Global Crossing, a hot telecom start-up from the late 1990s. At a time when the financial markets were transfixed by the potential of the early internet, Global Crossing amassed undersea fibre optic cables capable of handling a surge in traffic — much as CoreWeave has amassed banks of powerful graphics processing units made by Nvidia.
Global Crossing boomed as internet fervour took off, only to crash into bankruptcy four years after it launched. Internet demand was slow to pick up and telecoms companies were left with massive overcapacity. Echoes from the bubble in networks a quarter of a century ago are hard to avoid as the AI infrastructure boom continues apace. But like all comparisons, it is instructive as much for the differences as the similarities.
CoreWeave, unlike Global Crossing, arrives on Wall Street with significant demand: its revenue leapt eight-fold last year, to $1.9bn. It has another $26.5bn in future revenue already under contract, much of it from giant tech companies which have strong balance sheets and seem unlikely to renege. Yet, like Global Crossing, it is a wholesaler, vulnerable to the spending decisions of a handful of customers. Some of these build most of their own data centres, meaning they look to lease equipment from companies like CoreWeave as a release valve for their excess demand. That makes it something of an arbitrage play for GPUs. And, ultimately, its fortunes depend on its customers’ customers quickly finding productive and profitable uses for all that new computing power — otherwise its GPUs will be as unwanted as those undersea cables at the turn of the century.
So far, the company has shown enviable timing, pivoting from its initial business of crypto mining into AI infrastructure just as ChatGPT fever took hold. And it has manoeuvred nimbly between the tech giants: just as it fills a need for big cloud players such as Microsoft, which accounted for much of its revenue last year, it serves a strategic purpose for Nvidia, which has every reason to channel scarce GPUs to new companies like CoreWeave to reduce its dependence on the biggest clouds.
Two big timing issues loom. One is the speed at which Nvidia’s technology is advancing, potentially rendering older generations of its GPUs obsolete. CoreWeave recently stretched the depreciation schedule for its AI servers, extending their expected life to six years. That echoes accounting practices at companies like Microsoft and Google, though it could leave it holding unproductive and undepreciated assets. However, CoreWeave claims that in cash terms, its spending on tech equipment pays for itself within two and a half years. That relatively quick payback reflects enviable pricing power — something that will be difficult to maintain if supply of the hottest new GPUs catches up with demand.
The other timing issue has to do with whether generative AI can live up its hype quickly enough to justify the leap in capacity. Microsoft chief executive Satya Nadella recently told an interviewer that he was very happy to be leasing a lot of his company’s infrastructure in 2027 and 2028 rather than owning it outright. “The only thing that’s going to happen with all the compute build is the prices are going to come down,” he said.
A smart trader might see the sense in taking some of their chips off the table. Gary Winnick, who founded Global Crossing, sold hundreds of millions of dollars’ worth of his stock in the company while it was riding high. The three founders of CoreWeave, who come from the world of energy trading, have also sold nearly $500mn of shares between them, even before their company goes public. Plenty of investors are likely to welcome a new chance to bet on the AI boom — but at the right price.
richard.waters@ft.com
