Amid the artificial intelligence boom, companies continue to borrow heavily to build data centers and expand computing infrastructure. But the bond market is becoming increasingly selective: investors no longer view all AI-related projects as equally attractive and are demanding higher yields from borrowers with heavy debt loads and risky business models.

This is particularly important for an industry that increasingly depends on debt financing. The more expensive loans and bond issuance become, the harder it is for companies to fund the large-scale infrastructure construction needed to advance AI, Axios reports.

Investors grew more cautious after the CoreWeave sell-off

In early October, bond yields at CoreWeave, one of the most heavily indebted players in the AI infrastructure market, rose sharply. Because bond yields move in the opposite direction to prices, this meant a substantial decline in the value of the company's bonds. At one point, the yield on its unsecured bonds reached roughly 16%.

The corporate debt market later calmed somewhat. Optimism surrounding Anthropic and Meta also helped revive investor interest in AI-related companies.

However, the effects of the CoreWeave sell-off persisted. Investors began scrutinizing not only the financial position of individual borrowers but also the complex web of connections among participants in the AI boom: loans, equity stakes, lease obligations and long-term contracts.

As a result, risk premiums are rising not only for the most heavily indebted companies themselves but also for other participants in the infrastructure chain whose revenues depend on those companies' ability to pay.

Risks spread to data center owners

Companies that build data centers and then lease them to CoreWeave offer a telling example. After CoreWeave's bonds fell, investors also began demanding higher yields on securities issued to finance projects by operators such as Edged Compute, Core Scientific and Applied Digital. Core Scientific and Applied Digital previously mined bitcoin and are now developing infrastructure for computing workloads.

The reason is that the financial stability of such projects depends not only on the value of the facility itself but also on the tenant's ability to meet its contractual obligations. If a major customer runs into financial difficulties, the consequences can affect both the data center owner and the investors who financed its construction.

A similar pattern can be seen in projects linked to Oracle. Yields on bonds issued to finance data centers leased by the company have moved in line with fluctuations in the value of its own debt securities.

In particular, roughly $14 billion in bonds maturing in 2045, issued to build a large data center campus in Saline Township, Michigan, experienced similar fluctuations. A 17-year lease signed by Oracle plays an important role in backing these securities. Rating agencies view it as one of the key factors underpinning the project's financial stability.

Investors assess each borrower individually

Citigroup analysts note that the October sell-off exposed risks tied to specific tenants. CoreWeave, cloud providers linked to it and data center owners were hit particularly hard.

According to the analysts, infrastructure project bonds are not insulated from the risk of a tenant default. If the likelihood that a customer will be unable to meet its obligations increases, the value of the debt securities can fall quickly.

This repricing affects more than just high-yield bonds, which are considered riskier from the outset. In the investment-grade market—that is, bonds issued by borrowers with relatively high credit ratings—investors are also increasingly differentiating between issuers.

Previously, investors might have viewed AI-related bonds as part of a single promising investment theme. Now they are examining individual companies' financial positions, financing terms and the reliability of their contracts much more closely.

“The market has shifted from trading AI as a broad theme to assessing each issuer individually,” Barclays analysts note.

AI debt financing is becoming more selective

The shift in sentiment does not mean investors have stopped believing in the development of artificial intelligence. Rather, they are looking more closely at exactly who will benefit from rising demand for computing capacity and whether a company will be able to meet its financial obligations.

This is an important transition for the market. Building data centers requires enormous investment, while project returns depend on long-term demand, capacity utilization and tenant reliability. If participants in this chain are heavily dependent on one another, financial problems at one company can quickly affect the value of others' debt.

Barclays analysts expect differences in borrower assessments to persist as investors refine their views on the winners and losers in the AI race.

Ultimately, access to capital is determined less and less by simply belonging to a promising sector. For companies building artificial intelligence infrastructure, contract quality, cash flow stability and the ability to service debt are becoming increasingly important.