The Latest Challenge to Data Centers? Restive Investors.

Building data centers for artificial intelligence is getting more expensive — and not just because of the cost of land or electricity. Interest rates, which are the extra money borrowers must pay back on loans, have been climbing fast. That is making it harder for companies to afford the billions of dollars they need to build powerful computer systems. Now, investors and experts are asking whether these rising costs could slow down the AI boom.

Data centers are giant buildings packed with computers that store and process information. Companies like Microsoft, Meta, and smaller firms called neoclouds rely on them to run AI programs. To pay for these enormous buildings, companies often borrow money by selling bonds — a kind of IOU where a company promises to pay back the money plus extra payments called interest. As interest rates rise, those extra payments grow larger and larger.

The yield on the 10-year U.S. Treasury note recently hit 5.22 percent, a level not seen in many years. When that rate goes up, companies that want to borrow money must offer even higher rates to attract lenders. Morgan Stanley estimates that about $3 trillion will be spent building AI infrastructure through 2028, with about half of that coming from borrowed funds. As rates rise, experts say these projects must earn more money back just to make borrowing worth it.

Real numbers show how fast costs are climbing. Meta and its partner Blue Owl once borrowed money at a 6.58 percent rate to build a huge Louisiana data center called Hyperion. Just months later, another Meta-backed data center in El Paso had to borrow at 7.53 percent. On billions of dollars, that difference means hundreds of millions in extra interest payments.

Smaller companies called neoclouds are feeling the pain the most. Because lenders see them as riskier than giants like Meta or Microsoft, they must pay even higher interest rates. CoreWeave, a neocloud that recently became a public company, warned that every one-percentage-point rise in rates could cost it about $30 million more per quarter. Some of CoreWeave's bonds now carry rates close to 10 percent.

Rising rates are also causing trouble for banks that lend to data center projects. Banks often sell this debt to other lenders, but finding buyers has become harder as rates climb. In some cases, banks have been forced to sell loans at a discount, meaning they get back less than they lent. Credit investors say the market for these loans has grown more unstable.

There is also a tricky cycle making things worse. As big tech companies sell more and more bonds to raise money, they create more competition in the bond market. That extra competition helps push up U.S. Treasury yields, which raises borrowing costs for everyone. In a strange way, the companies trying to build AI are making it more expensive for themselves and others to borrow.

In other business news, Nike reported a bad quarter and its stock dropped more than 9 percent. CEO Elliott Hill said the company would cut jobs and restructure its global operations. Sales in China fell by about 26 percent last quarter, and Nike warned that struggle would continue. The stock is down over 45 percent this year, possibly making it Nike's worst year on record.

Two AI coding start-ups, Factory and Cognition, had a very public argument on social media this week. Factory's CEO accused a shared adviser of leaking secret company information to Cognition, which denied the claims. The fight showed how fierce competition for talent and secrets has become in the AI industry. Top researchers sometimes jump between rival companies for pay packages worth hundreds of millions of dollars.

A database start-up called Supabase announced it raised $150 million in new funding and is now valued at $10.65 billion. The company makes software that stores data for apps, and its business is booming as AI tools build more and more apps on their own. Supabase adds nearly five million new databases each month, and about 70 percent are created by AI — not people. However, a cybersecurity firm found that over 16,000 Supabase databases had exposed sensitive data, raising important security concerns.