AI's Credit Stress: A Real Warning, or an Overdue Repricing?

The sharpest argument in the market this week has moved past the Federal Reserve. It is whether the companies borrowing to build artificial-intelligence data centers can keep paying for the debt they are taking on. Credit markets have begun pricing real doubt. Reuters reported Thursday that the cost of insuring the debt of several AI-linked firms has climbed, with cloud operator Oracle, Japan’s SoftBank and data-center builder CoreWeave all drawing wider credit default swap spreads. Stocks followed. The question dividing investors is easy to state and hard to answer: is the credit market catching a problem early, or repricing a risk the buildout will simply grow through?

Insuring CoreWeave’s debt now costs distress prices

The clearest stress sits in the balance sheets built fastest. CoreWeave, a heavily indebted data-center operator, has seen the five-year cost of insuring its debt top 1,100 basis points, according to S&P Global Market Intelligence data cited by Reuters. That implies about a 63% chance of default over five years, up from near 30% a year earlier. Its 2031 bonds have traded at yields above 16%, a level that usually signals distress, and analysts expect a loss of $2.13 billion this year on about $13 billion of revenue. It carries about $50 billion in debt.

The worry runs wider than one firm. A Brookings paper by Columbia Business School’s Stijn Van Nieuwerburgh estimates that 2026 capital spending by Oracle, Microsoft, Amazon, Meta and Alphabet will exceed their combined operating cash flow for the first time, meaning the buildout can no longer be funded from internal cash alone. Once the spenders lean on borrowed money, pricier insurance can lift their own borrowing costs, and that can feed on itself.

Demand still runs well ahead of supply

The other reading starts with where the stress is not. None of the companies whose swap costs rose has defaulted or said it might, and an investment-grade credit index was trading near 62 basis points, far below CoreWeave’s. The swaps on the cash-generative giants sit much lower, with Nvidia and Meta near 90 and 105 basis points. CoreWeave’s chief executive told staff its deal backlog “more than covers” its obligations, and Nvidia has agreed to buy any capacity the company cannot sell through 2032. The giants look different.

Bulls read the selloff as the price of money resetting. A Saxo Bank strategist described it as a repricing of the cost of capital with technology fundamentals intact. Wedbush’s Dan Ives called the drop “way overdone” and framed the issue as a financing question, not a demand question, arguing the infrastructure layer stays supply-constrained through at least 2027. For that camp, the cure is cheaper capital or stronger earnings, and the next few weeks could supply either.

What the data shows

The tape through Wednesday’s close looked orderly. The S&P 500 sat at 7,735, up 0.21%, and the Dow rose 1.03%, even as money left the AI-heavy corners. Technology stocks fell 2.98% and communication services dropped 1.30%, while defensive health care gained 2.53% and consumer staples added 2.22%. The best and worst sectors finished more than five percentage points apart. The VIX, Wall Street’s volatility gauge, closed below 17, well under levels that mark real fear. The 10-year Treasury yield sat at 5.21%, high enough to make debt-funded spending expensive and to pay investors for waiting.

What would settle it

Three scheduled tests arrive soon. The largest cloud spenders report quarterly results later this month, and the market will watch whether they raise or trim their capital-spending plans; Van Nieuwerburgh’s cash-flow math and Ives’ bull case both hinge on that line. CoreWeave’s next report, and whether its bond yields ease from distressed levels, will show whether the credit alarm was specific to it. And the price of renting graphics chips, along with any canceled data-center contracts, would be the clearest sign of a demand slowdown underneath the financing scare.

For now the two signals point in opposite directions. The debt behind the fastest builders is getting more expensive to insure, and the spending has outgrown the cash coming in. The demand for computing power still outstrips what the industry can deliver, and the firms with the strongest balance sheets are not the ones the credit market is questioning. Both are true at once. Which one matters more is what the next round of earnings will begin to tell.

This is a general-market summary for information only — not investment advice, and not a recommendation regarding any security.