
“In the end,” wrote Bethany McLean in The Free Press last month, “the fundamental issue is simple: Can the debt be repaid when it’s supposed to be repaid?”
The debt she was referring to was the billions of dollars—nay, hundreds of billions—being spent on AI infrastructure build-out, primarily the rapid construction of data centers. Thanks to the money they are throwing at the boom, the so-called hyperscalers—Amazon, Alphabet, Meta, Microsoft, and Oracle—are no longer cash-flow positive, despite the billions they reap in revenue each quarter. The neoclouds—that is, the relatively new companies that were created to actually develop data centers and then rent out the “compute” to the hyperscalers and others—are tens of billions in debt. A company called CoreWeave, for instance, reported that it had nearly $90 billion in debt and liabilities, and a net loss of $709 million in the third quarter alone. Its stock price is down more than 50 percent in the last six months.
And then, of course, there’s Nvidia, the AI Goliath. The company’s cutting-edge semiconductors are the beating heart of the AI revolution, and its money is the driving force behind the build-out. On Tuesday evening, Nvidia released a staggering earnings report: $116 billion in third-quarter revenue, and net income of $65 billion. But it also disclosed that it has $32 billion in debt and another $118 billion in liabilities.
Under former CEO Jensen Huang, Nvidia had what The Wall Street Journal called an “open checkbook” policy, loaning money to the neoclouds, which would use some of that money to buy Nvidia’s chips. (Critics call this tactic “circular financing.”) And as part of these deals, Nvidia essentially guaranteed that it would cover the neoclouds’ debt if they were unable to pay it back—hence the liability. Much of that debt was resold to Wall Street and held by insurance companies, but Nvidia remained the guarantor of last resort.
Read more
