AI borrowers face costly reality

THE artificial intelligence (AI) boom is spreading into the riskier parts of the US debt market, but companies trying to borrow are finding that investors are asking a simple question: when will the money start coming in?

As AI-related businesses take on more debt to fund their growth, lenders are becoming more careful about backing companies whose future earnings are still largely based on forecasts rather than proven cash flow.

According to a Reuters report, low-rated companies have raised US$88bil through AI-related debt this year, with most of the borrowing coming from US issuers.

That compares with just US$20bil in AI-related leveraged finance – mainly junk bonds and loans – during the first 11 months of 2025, based on data from Neuberger Berman cited by analysts.

The sharp increase shows how quickly AI has moved beyond the biggest and most established technology companies.

More borrowers are now turning to the riskier end of the credit market, where investors expect to be paid more for taking on greater uncertainty, the newswire reported.

The problem is that lenders do not get to enjoy much of the upside if an AI business takes off. They get their interest and principal back, but they can be left with losses if things go wrong.

“High-yield people like to know how much cash flow is coming, when that cash flow is coming, and what is the probability that the cash flow doesn’t come,” Larry Holzenthaler, senior portfolio manager, fixed income, at Catalyst Funds, tells Reuters.

He says investors are asking for higher returns because “these investors don’t really participate in the upside because if everything goes well, you get paid back at par, and if it doesn’t, the creditors end up taking the losses”.

This makes the cost of borrowing an increasingly important issue for AI companies that are not yet generating steady cash.

Borrowing gets expensive

Companies that sit further down the credit ladder are likely to feel the pressure most.

According to Reuters, even companies rated BB+, just below investment grade, are paying yields of roughly 9% to 10%. For lower-rated borrowers, the cost could climb as high as 14% to 15%, it notes.

SoftBank Group, which has a BB+ rating, paid yields of 8.625% on 3.5-year notes, 9.25% on 5.5-year debt and 9.75% on 7.5-year bonds when it raised funds last month, Reuters says, citing the company’s filing.

It reports analysts as saying these borrowing costs are more commonly associated with companies with significantly lower credit ratings.

For AI companies, this creates a tricky situation. They need large amounts of money upfront to build infrastructure and expand, but taking on too much expensive debt could make it harder for them to generate enough cash later to service that debt.

Investors are therefore showing a preference for companies where the numbers are easier to understand.

Leveraged finance buyers are clustering around double-B-rated companies, or borrowers just below investment grade. They are also looking for businesses with established customers, long-term contracts, tangible assets and more predictable income.

Data centres (DC) fit much of that description and have become a major source of new borrowing linked to the AI boom.

Erin Brown, head of leveraged finance at BNP Paribas, tells Reuters DCs have been doing much of the heavy lifting in a high-yield market that would otherwise be shrinking.

High-yield volumes are “basically flat” year on year, Brown says, adding that “if it weren’t for the new money supply coming from DCs, we’d have volumes that were materially down year on year”.

AI infrastructure supply in the high-yield market has reached US$40bil so far this year, compared with US$12bil for the whole of 2025, according to BNP data.

Risky business

Another problem for AI borrowers is that not all investors have the freedom to take on large amounts of risky debt.

Investment-grade investors can absorb substantial borrowing from highly rated hyperscalers. But investors in high-yield bonds and leveraged loans have portfolio rules that limit how much exposure they can take to riskier companies, Reuters report analysts as saying.

This is particularly important for collateralised loan obligation (CLO) managers, which are among the biggest buyers of leveraged loans.

CLOs buy large pools of loans and then raise money from investors by issuing their own securities. But AI-related loans can become harder to hold if a company is highly leveraged, continues to burn cash or suffers a ratings downgrade.

“Anecdotally, CLO managers are becoming more cautious on certain names and really making sure they’re checking everything before investing,” says Elizabeth Templeton, senior product manager, fixed income and multi-asset indexes, at Morningstar.

Lotfi Karoui, multi-asset credit strategist at PIMCO, says debt investors face a particularly uneven trade-off.

“For debt investors, the proposition is fundamentally asymmetric,” he writes.

“Returns are largely contractual, driven by coupon, principal, and, at most, some spread compression,” he notes, while the risks range from high debt levels to project delays and rapid changes in technology.

In simple terms, lenders have a fairly limited upside if an investment works, but could take a much bigger hit if the business runs into trouble.

Questions on numbers

Investors’ caution is not just theoretical. The performance of some recent AI-related debt issues shows that buyers are willing to question the numbers after the bonds hit the market.

In August, special-purpose project company Zenith Arc raised US$2.25bil through five-year senior secured notes to finance an Oklahoma DC leased to Jane Street, Reuters notes.

The bonds were sold at 99.50, slightly below their face value, with an 8.875% coupon. They then fell almost three points to a bid price of 96.75 shortly after being issued, analysts say.

By late August, Pender Fund Management wrote to investors that the Zenith Arc bonds had lost more than seven points from their issue price.

Zenith Arc did not have publicly available contact details. A representative for Coatue, the investment firm behind Next Frontier, a joint-venture partner in the Zenith Arc project, said the firm had no comment, Reuters reports.

The episode points to the challenge ahead for AI companies looking to borrow more heavily: having an ambitious growth story may not be enough.

Investors want to see where the money will come from to repay the debt.

And that means the next stage of the AI borrowing boom could increasingly come down to a rather old-fashioned question – whether the business can actually make enough money to pay its bills.

“The joke has become, ‘What’s revenue going to be like next year? I don’t know, but it’s going to be big,’” Holzenthaler tells Reuters.

“If you’re a credit investor, that’s a really bad answer from a borrower.”