Stocks in some major companies involved in building artificial intelligence data centers have been taking a beating. Oracle is down almost 19% in a month.
The “why” is concern on Wall Street over how much debt these companies are taking on to construct the infrastructure that will power AI.
A lot of these debt deals are complex, involving multiple companies and banks, and private equity. And there’s concern that there’s not enough daylight over potential risks.
Companies are expected to spend nearly $7 trillion to build global AI infrastructure, according to consulting firm McKinsey.
It’s an amount even deep-pocketed Big Tech firms can’t shoulder all by themselves.
“The cash requirements are extremely high,” said Steven Kaplan, a professor of entrepreneurship and finance at the University of Chicago. “And they feel that it’s reasonable to shift some of the risk on to others.”
The way AI firms have shifted risk is through partnerships with other companies, who take out loans from public markets, like banks, and private markets, which can include pensions as investors.
Then, those companies use that money to build AI infrastructure.
“They buy whatever goes in the data centers, which includes the chips, and then they rent that out,” Kaplan said.
Firms like OpenAI will be the tenants, and the rental income will theoretically be enough to pay back the debt and make a profit.
But Andrew Rocco of Zacks Investment Research said some on Wall Street are starting to question these arrangements.
“There’s certainly risks, because a lot of those companies are taking on a lot of debt to get these built out,” he said.
Rocco said they’ve done so with the hopes that AI demand — and demand for data centers — will continue to grow.
“And that’s really the question, of course, is all this debt going to be worth it? Is there a light at the end of the tunnel?” he said.
OpenAI’s Sam Altman was essentially asked this in a podcast published about a month and a half ago, and he bristled at the question: “If you want to sell your shares I’ll find you a buyer. I just. Enough,” he said.
Altman went on to say that he expects the AI business to grow enough to match the spending.
But Kaplan at UChicago said the reason these questions persist is because the pay-off is way in the future. He said there’s no question big tech companies can pay their rents and generate income for data centers for now. But what happens when those rental contracts expire — in say, five years — will those companies be left holding debt?
“So then that’s called the residual risk, is, after the term of the lease, will there be value there?” Kaplan said.
In other words, will AI in the future match hopes in the present?