(Bloomberg) — Artificial intelligence debt will almost certainly reach bubble levels eventually, given the history of periods of heavy investments in areas like railroads and the internet, said DoubleLine portfolio manager Robert Cohen.
High odds of coming froth mean money managers should focus on companies with strong balance sheets, and investments with strong protections, Cohen said, speaking on a panel at the Bloomberg Global Credit Forum in New York on Wednesday. Credit markets are overly exuberant when fund managers buy debt assuming companies will grow into being able to meet their obligations, he said.
“What’s the probability that we will be in an AI bubble? I’ll put maybe 100% on that,” Cohen said.
The market isn’t there yet, and demand for AI debt is still strong, investors and bankers on the panel stressed. Many AI debt deals benefit from the backing of big, profitable tech companies known as hyperscalers, including Alphabet Inc., the parent of Google, or Meta Platforms Inc., the parent of Facebook, the panelists said. That support, along with bonds that might pay down principal over time, can help reduce risk, said Matt Brill, head of North America investment grade credit at Invesco.
You want to focus on “getting your money off the table as fast as possible,” he said.
More than $370 billion of US sales of high-grade notes, junk bonds, loans, and other types of AI debt have hit the market since the start of last year, according to data compiled by Bloomberg News. Bloomberg Intelligence estimates that capital expenditure for AI will approach $5 trillion over the next five years, with much coming from debt markets.
In equities, valuations may be getting more stretched now, DoubleLine’s Cohen said. He defines an equity bubble as prices that reflect unrealistic future growth expectations.
Every corner of credit markets is being tapped to take down the elevated amounts of supply, according to Morgan Stanley’s Anish Shah. Shah estimates that more than 10% to 15% of all debt issuance across credit markets could be tied to AI this year — a shocking amount given AI debt barely existed just two years ago.
Morgan Stanley has said it’s expecting a record $2.25 trillion of US high-grade corporate bond sales this year, up from about $1.81 trillion last year. Factors driving that increase include big bond sales by hyperscalers, which the bank expects to sell about $250 billion to $300 billion this year.
Hyperscalers are selling debt globally. Alphabet in recent months has borrowed in US dollars, Canadian dollars, euros, pounds sterling, Swiss francs, and Japanese yen, raising around $60 billion in total.
Issuance in the junk bond market could help boost the size of that market over time, after a chunk of money allocated to high-yield bonds has migrated to the leveraged loan and private credit markets, according DoubleLine’s Cohen. And spreads on the debt will probably remain around where they are now, according to investors and bankers speaking at the event.
“The only thing that shakes us meaningfully loose from this very tight range is a meaningful downturn in the economic data and a rally in Treasury yields,” said Maureen O’Connor, global head of high grade syndicate at Wells Fargo. That’s despite the fact, she said, that “there’s a huge pipeline of this stuff coming,” referring to AI project finance debt.
With so much debt coming across markets, some companies have been looking to borrow now rather than later, Morgan Stanley’s Shah said.
Cohen said his firm has shied away from buying much AI-linked debt, saying spreads are too tight and that there’s bubble risk. Even so, he added, for the right price his firm would be willing to buy more.
“Our firm has a lot of capacity,” he said. “At the right spread I can easily be bribed to go into some of these higher quality credits.”
More stories like this are available on bloomberg.com
