NEW YORK, Nov 9 (Reuters) - Banks and credit investors are upping their scrutiny of AI-related debt deals while showing reluctance to increase exposure to borrowers, jolted by the recent market selloff and changing sentiment toward the technology's growth trajectory.
Some banks, which just a few weeks ago were lining up to fund AI deals, are now telling market participants they would be reluctant to finance additional deals even when they involve guarantees from the sector's marquee technology companies,
according to several banking and credit sources.
The lending pullback has been exacerbated by a recent drop in the stock market, which erased $2.5 trillion in market value on Thursday from the Nasdaq 100, and sent the index into a correction. The US midterm election results are also adding anxiety about more oversight of data centers.
A potential slowdown in AI data center construction could curb loan growth and dent investment banking fees from the loans, bond sales and stock offerings that have helped finance the sector's rapid expansion.
The change, according to two sources, has come in just the past few days. Some lenders have shown reluctance to finance deals even if they are backstopped by names such as Anthropic and Google , the sources said, without specifying names of the lenders.
Anthropic declined to comment on its recent borrowing. Google declined to comment.
"Over the past year there was unbridled enthusiasm to provide debt to neoclouds, anyone who was building data centers or some sort of AI asset," said Chris Poggi, finance partner at Ropes & Gray, referring to companies that rent out powerful AI computing capacity, often built around advanced chips, to businesses developing AI.
"Debt investors' books have become full and now they are assessing if they have extended themselves as much as they're going to extend themselves this year."
Even so, much more supply is coming to the AI loan market because borrowing needs are so high. Goldman Sachs Research projects global hyperscaler debt issuance will reach $420 billion in 2027, more than 60% above its estimate of $250 billion for fiscal 2026.
Investors and lenders are also stress-testing cash flows for economic downturns and scrutinizing contractual details underpinning AI infrastructure projects, including service obligations, operational deadlines and the conditions under which large technology customers could reduce or terminate commitments, two of the sources said.
Many of these deals are fairly complex. Chipmakers typically backstop the senior debt that will finance the purchase of advanced chips for an AI company and also guarantee payments on the long-term lease of the data centers. Banks and private credit players have until recently been attracted to such loans because of the guarantees that ensure debt payments.
As hyperscalers and neoclouds race to raise additional debt for data centers, chips and computing infrastructure, they are increasingly running into lenders questioning how much additional exposure they want to a sector that is already taking up a large part of their loan book.
At JPMorgan Chase, for instance, loans backed by other income-producing properties, which includes data centers, were at $29 billion outstanding at the end of the third quarter. They are the bank's second-largest commercial real estate exposure at 12%, following apartment buildings. JPMorgan declined to comment.
That was up from 8% of the bank's CRE loans three years ago, when the data center category ranked fourth after apartments, industrial properties and office buildings.
Like direct lenders, investors are becoming far more selective about which deals they will back. The spread on 10-year hyperscaler debt has widened by about 50 basis points year to date, versus roughly 20 basis points for high-quality industrial companies not tied to AI, according to Bank of America credit analyst Yuri Seliger. Bond prices of Meta, issued earlier this year, and SpaceX are trading below par.
In Asia, some banks are making sure deals are backed by cash flow and scrutinizing risks closely.
"The kind of deals we look at, we are very focused on the risk aspect. So we structure the deals around to make sure that the cash flow is there," said Singapore's OCBC Group CEO Tan Teck Long on Friday during an earnings briefing.
WEAKER CREDIT
The pullback has been particularly pronounced among riskier AI borrowers such as neocloud companies and standalone data center operators.
Oracle and CoreWeave are among the neocloud companies that could be impacted by possible construction delays or moratoriums, according to Baird Equity Research. CoreWeave's stock has fallen 35% year to date, and its bond prices have also dropped sharply on the secondary market.
Some planned bond offerings linked to data centers and chips are also being shelved.
Nasdaq-listed neocloud firm Nebius had been exploring a high-yield bond issue in September, but decided not to go ahead with it last month, one of the sources familiar with the matter said. The AI cloud company did not respond to a Reuters request for comment.
Nebius in October entered into a $3.5 billion senior secured loan facility provided by Citigroup, to use for data center development and GPU procurement and general corporate purposes.
"Most banks and credit investors will take a wait and see approach," said Jay Bala, co-founder and CEO of $300-million alternative asset manager AIP, referring to how AI growth will fare following the Democratic sweep of Congress in the midterms.
(Reporting by Saeed Azhar and Nupur Anand; additional reporting by Suzanne McGee, Tatiana Bautzer and Rae Wee in Singapore; Editing by Stephen Gandel, Megan Davies and Nia Williams)
