Investors in the A.I.-fueled stock market have largely shrugged off warnings about a tech bubble, an optimism that has pushed up share prices to repeated new highs this year.
But the debt market is telling a different story, some investors say. New artificial intelligence companies looking to raise funds to supercharge their nascent businesses are being made to pay lofty interest rates on the money they borrow, indicative of investors’ skepticism when new, unproven A.I. businesses take on large debts.
In one debt deal for Applied Digital, a data center builder, the company had to pay as much as 3.75 percentage points above similarly rated companies, equivalent to roughly 70 percent more in interest.
There are other indicators of debt investors’ wariness: Some of the bonds have tumbled in price after being issued, in a sign of increased caution among investors. And the cost of credit default swaps, which protect bond investors from losses, has surged in recent months on some A.I. companies’ debt.
Construction delays at these sprawling data facilities could push out the time it takes before they can start generating revenue from their leases to A.I. companies. Investors also worry that, in the end, there could be less demand for A.I. computing power, creating a glut of unneeded data centers and leading to defaults on the debt used to finance the buildings.
“We just have to be much more pessimistic and not buy into the hype,” said Will Smith, a portfolio manager at Alliance Bernstein.
Debt investors are stereotypically seen as pessimists, while stock investors are viewed as optimists. That is largely because of the difference between the two types of investments. Equity investors have unlimited upside for a company and its stock price to grow and keep rewarding their investment. Debt investors are just looking to get their money back with interest.
“The equity market stands to benefit from a lot of upside if this works well, so they are willing to take a lot of risk because of the upside,” Mr. Smith said. “For us, it’s different. If everything goes really well, we don’t benefit from that upside, so we are much more focused on the downside.”
Companies looking to finance the next leg of A.I. infrastructure have borrowed more than $100 billion this year, according to data from Refinitiv. Much of that has come from highly rated, well-known technology companies that are also sitting on a lot of cash. For example, Amazon raised $15 billion at market rates in November as it ramped up its capital investment in Amazon Web Services.
But there have also been billions raised by some much smaller, newer companies that have had to pay up for the privilege.
In mid-October, Wulf Compute, a subsidiary of Terawulf, raised $3.2 billion through the corporate bond market. Terawulf, which counts Gwyneth Paltrow as an early investor, became a public company in 2021 as a Bitcoin miner and has since expanded into building data centers.
Wulf Compute is on the hook to pay investors 7.75 percent until the deal comes due in 2030. That’s much higher than the average yield for similarly rated issuers, which is around 5.5 percent, based on an index run by ICE Data Services.
In a similar deal, Cipher Compute, a subsidiary of Cipher Mining, sold $1.7 billion of debt in early November, with a similar rating to Wulf Compute and at a similar coupon of just over 7 percent, still markedly higher than comparable bonds in other sectors.
“If you can get through the construction risk, these companies look stable with regular cash flows,” said Naveen Sarma, who leads U.S. media and telecoms ratings at S&P Global. “But you have to navigate through the construction risk.”
The smaller companies bringing debt deals have been either data center builders or renters, like CoreWeave, which rents the data centers and put in the high-powered computer systems to then run large A.I. models for “hyperscalers” like OpenAI and Meta.
Leaning on these contracts with bigger, more stable companies has allowed the newer entrants to the market to obtain middling credit ratings, despite their own lack of a proven, profitable business.
Like Wulf, Applied Digital, which builds data centers, received a double-B credit rating when it sold $2.35 billion of debt in November. Double-B sits roughly in the middle of a ratings range that runs from triple-A down to triple-C.
The debt sold with an even higher coupon of 9.25 percent, suggesting investors viewed the deal as riskier than Wulf Compute’s.
Applied Digital is heavily dependent on CoreWeave as its main tenant. CoreWeave then rents its computing power to the hyperscalers to run their A.I. models and has big contracts with Meta, OpenAI and Microsoft, to name a few. This dependency on just a few large customers has worried investors and ratings agencies.
CoreWeave raised $1.75 billion in July, receiving a lower assessment from the rating agencies of single-B, signaling to investors that there was more risk to the deal. Unlike the data center builders that pledge the building as collateral to investors, CoreWeave borrowed money unsecured, which is inherently more risky. The deal came with a coupon of 9 percent.
The company’s shares were listed on the stock market in March, rising over 350 percent to its peak in June, before steadily falling. The stock remains 70 percent higher this year, while the bonds have lost value since they were issued and are trading around 90 cents on the dollar.
That drop in value has increased the overall yield of the bonds. The yield is the amount an investor would expect to receive over the life of the bonds, assuming all the debt is repaid at its original face value. The yield on CoreWeave’s debt is now above 12 percent, indicative of what the company would have to pay as a coupon if the deal were brought to the market today. The average single-B company trades with a yield of less than 7 percent.
“If things do crack, where are those things going to show up in the economy?” said Mr. Sarma at S&P. “It won’t wipe out Meta, but it could hurt a CoreWeave.”
The yield on Applied Digital’s bonds has also risen, to over 10 percent. In contrast, the implied borrowing costs for Cipher Compute and Wulf Compute have fallen, meaning the value of their debt has risen in recent weeks.
Some investors said this showed the shift in market sentiment over time, with previous deals from the likes of CoreWeave being repriced lower by the market more recently, while newer deals have been priced in line with current expectations.
As concerns in debt markets have risen, equity markets have also responded, and there has been a sharp pullback in recent months, especially among more speculative A.I. companies.
“The debt markets, like the equity markets, have been moving with sentiment,” said K.C. Baer, a co-founder of the fund manager CKC Capital. “To some extent I think they have been driving sentiment.”
The equity sell-off came as more companies turned to debt markets to fund accelerated A.I. plans. When Oracle, which has $100 billion of outstanding debt, borrowed cash in September, it set off a sharp reassessment of the amount of money being bet on an uncertain future.
The cost of buying Oracle credit default swaps, which protect bond investors from losses, has surged higher since then.
“The market is just trying to weigh different possibilities in a market that is very uncertain,” Mr. Baer said.
Joe Rennison writes about financial markets, a beat that ranges from chronicling the vagaries of the stock market to explaining the often-inscrutable trading decisions of Wall Street insiders.
The post As A.I. Companies Borrow Billions, Debt Investors Grow Wary appeared first on New York Times.
