Wall Street’s AI Debt Machine Is Starting to Choke on Supply
(Bloomberg) — A new anxiety is gripping Wall Street as it bankrolls the vast network of data centers and chips behind the artificial intelligence boom: Early signs suggest that the credit market is approaching the limits of how much more debt it can absorb.
One after another, yields are soaring into the double digits on bonds and loans funding the AI infrastructure buildout. Bankers are finding it more difficult to offload risk to insurers and other investors. Elon Musk himself warned the other day that hundreds of so-called neoclouds risk going under as firms get cut off from fresh financing.
With those jitters growing louder each week — and spurring, in turn, angst about the potential economic fallout — Citadel Securities is cautioning clients that even the biggest names in tech could get ensnared as they put their balance sheets behind scores of projects critical to their AI ambitions. That risk is increasingly becoming evident in the market for credit default swaps, where banks and other lenders hedge their exposure. The cost of insuring against losses on the debt of companies from Broadcom to SpaceX and Oracle has spiked in recent days, reaching levels more typical of junk-rated borrowers. And the damage is already starting to spread to broader credit markets.
Citadel, to be clear, isn’t predicting an imminent credit crunch, and tech giants have managed to cobble together tens of billions of new financing this week alone. But in some ways that’s beside the point. The concern is that, with trillions more in debt needed over the coming years, there appears to be a growing disconnect between what Silicon Valley’s top executives believe can be raised, what financiers say they can deliver, and what money managers are willing to take down.
“When you have such a large quantum of debt coming, at some point the challenge could be less about finding an investment-grade backstop, and more about finding the actual buyers willing to absorb more of the risk coming from the same handful of investment-grade backstop providers,” said Citadel analyst Tucker Roberts.
In conversations with more than 30 bankers, money managers, private equity executives and others involved in recent and current data center financing talks, all acknowledged that raising cash has become significantly more difficult than it was just months ago.
A key reason is the growing concern among investors over cancellation rights and other contractual off-ramps that could leave them exposed if a major customer walks away. Those worries only intensified after Oracle issued a force-majeure notice last month seeking to defer payments if a massive New Mexico data center were delayed.
One hedge fund said it turned down a recent data center deal when it discovered that the firm under contract had the option to back out if the project were delayed by more than three months. Partners at a top 10 law firm who’ve worked on development financing for data centers have seen deals where compute customers can walk away if there are holdups of 180 days or more.
Another major challenge has been offloading risk to free up capacity for new deals.
The head of technology, media and telecom lending at a large European bank said the firm is approaching the limit of what it can lend to the sector given its already sizable exposure. Selling the debt or transferring the risk can take five months, limiting its ability to recycle capital into new financing, the person said, asking not to be identified discussing internal matters.
Some insurers, which often take on a share of the credit risk on such loans, were indicating months ago that they were nearing or had already reached their limits on data center exposure, according to a survey from Willis Towers Watson.
Data center operators are also being hurt by higher interest rates, particularly in the asset-backed securities market, where rental income is often locked in under long-term leases.
“At the beginning of 2026, the highest-rated data center bonds offered yields below 5%. Today similar bonds are yielding closer to 7%, and that increase matters enormously for companies trying to finance multibillion-dollar facilities,” Morgan Stanley ABS strategist Carolyn Campbell said in a podcast earlier this month. “When borrowing costs approach 8%, issuing these bonds becomes significantly more challenging.”
She noted that data-center debt issuance in the ABS and CMBS markets has totaled roughly $22 billion this year, barely keeping pace with last year despite the massive sums needed to bankroll the buildout.
Citadel’s warning, in many ways, echoes those from other Wall Street heavyweights.
Diameter Capital Partners, after spending time in the Bay Area earlier this year, came away wondering if the tech industry was “too complacent” in believing there “was unlimited debt capital available to finance a $5.5 trillion build out” of artificial intelligence.
“Not all of the business models are sound,” managing partners Scott Goodwin and Jonathan Lewinsohn wrote in a letter to investors.
And Oaktree Co-Chief Executive Officer Armen Panossian said the firm has become “very, very cautious” about the structure and quality of assets backing data-center deals, citing risks from technological change and construction delays to permitting hurdles and access to power.
“We’re seeing in the last several weeks a widening of those spreads that indicates that the credit markets are becoming either full, saturated of the loans that are out there, or concerned about some of these factors that are lurking in the background,” Panossian said in a recent client webcast.
Some of those factors suddenly burst into view in recent weeks, jolting markets. Oracle’s force majeure notice was followed by confusion over OpenAI’s revenue projections and a rout in the stock and bonds of CoreWeave Inc., the once obscure neocloud that’s become the new proxy for AI bubble fears. More recently, a trio of public offerings or planned IPOs from AI infrastructure companies Elice Inc., Firmus Grid Ltd. and Nscale Ltd. have stumbled, adding to signs of weak investor demand.
Tony Trzcinka an investment-grade portfolio manager at Impax Asset Management, said AI has become the main topic of conversation at internal credit strategy meetings, and that the firm is now underweight data-center debt.
“The market reaction confirmed how interconnected these credits are and how heavily they rely on the same underlying collateral,” he said.
At least 18 data-center financings across the junk-bond and leveraged loan markets now carry double-digit yields, according to data compiled by Bloomberg. Even new deals are feeling the strain. JPMorgan Chase & Co. and Goldman Sachs Group Inc. had to offer steeper discounts and higher interest payments to drum up demand for neocloud Volta Infrastructure Holdings Ltd.’s $5 billion syndicated loan, pushing its yield above 11%.
Credit investors are starting to recognize that much of the debt neoclouds still need to raise may ultimately depend on backing from tech giants already burdened with their own massive financing needs, Citadel’s Roberts said. Whether via explicit guarantees, long-term contracts or simply the expectation that they’ll step in if projects run into trouble, chipmakers and hyperscalers could find themselves shouldering an increasing share of the risk, he added.
Nowhere is that unease more pronounced than in the CDS market. The cost of insuring Oracle Corp.’s debt, widely seen as a barometer of risk across the AI boom, reached a fresh record last week. So did default protection costs for SpaceX, Nvidia and Meta.
The pressure is showing up in their borrowing costs, too. The average yield on bonds in the Bloomberg Investment Grade Technology Index has surged to about 6%, about 1.4 percentage points higher than a year ago.
“The banks and companies raising capital have been relatively transparent that they do not care about pricing it wider and marking down paper you’ve bought,” said John Yovanovic, co-head of leveraged finance at MetLife Investment Management. “So you’re being asked to finance a deal that has likelihood of being less than par in the near term, even when it has attractive long returns. Who wants to step in front of that?”
Credit markets, of course, could once again prove willing to bankroll even the most speculative projects should hyperscalers show earnings growth is keeping pace with the massive upfront capital spending.
But for the time being, money managers are becoming more choosy about where, and at what cost, they’re willing to lend.
KKR & Co. is steering clear of data centers in remote locations amid concerns that the facilities could be difficult to repurpose once their leases expire, according to a person familiar with the matter, while Carlyle Group Inc. has passed on a number of recent deals as it emphasizes portfolio diversification and selectivity.
“People are just getting more conscious of their concentration risk in the sector,” said Felicity Juckes at TwentyFour Asset Management. Investors “are going to be increasingly picky.”
–With assistance from Aaron Weinman, Scott Carpenter and Katherine Schwartz.
©2026 Bloomberg L.P.
See more Swissinfo articles in your Google results. Learn more
