AI data-center borrowers face higher costs as lenders struggle to absorb more debt

Financing the next wave of AI data centers is becoming significantly harder, even before the industry seeks much of the debt it expects to need. In an October 28 report, Bloomberg found that more than 30 bankers, investors and other financing participants agreed raising cash had become tougher than just months earlier.

Citadel Securities is not predicting an imminent credit crunch; technology companies raised tens of billions of dollars in new financing this week. Its warning is narrower: finding a financially strong company to back a project may no longer be enough. Investors must also be willing to absorb more exposure to those same backers.

Why a bank’s existing loans can block its next deal

Banks can free capacity for new lending by selling debt or transferring some of its risk to other investors. But that process is slowing. A senior lending executive at a large European bank told Bloomberg that the firm was approaching its sector limit. Selling debt or transferring risk could take five months, tying up capacity for fresh deals.

Insurers are another potential destination for that risk, but their appetite is finite too. A Willis Towers Watson survey found that some insurers were indicating months ago that they were nearing, or had already reached, their limits on data-center exposure.

Customer contracts create a separate obstacle. Investors worry that cancellation rights could leave them exposed if a major customer walks away. One hedge fund told Bloomberg it rejected a deal after discovering the customer could exit if construction ran more than three months late. Lawyers working on other projects described exit rights triggered by delays of 180 days or more.

Those concerns intensified after Oracle issued a force-majeure notice in September seeking to defer payments if a large New Mexico data center were delayed. The notice brought the risk of delayed customer payments into sharper focus.

Higher yields reach new borrowing

Higher financing costs are particularly difficult for operators whose rental income is fixed under long-term leases. Campbell said issuing bonds becomes significantly more challenging as borrowing costs approach 8%. She put this year’s data-center issuance in asset-backed and commercial mortgage-backed securities—debt supported by assets or property loans—at roughly $22 billion, barely keeping pace with last year.

At least 18 data-center financings across junk-bond and leveraged-loan markets now carry double-digit yields, according to Bloomberg’s data. The pressure also reaches new loans: JPMorgan Chase and Goldman Sachs offered steeper discounts and higher interest payments to attract buyers for Volta Infrastructure Holdings’ $5 billion loan, pushing its yield above 11%.

Big-tech backing does not eliminate concentration

Citadel analyst Tucker Roberts said specialist AI cloud providers’ future borrowing may increasingly depend on technology giants with substantial financing needs of their own. Explicit guarantees, long-term contracts or expectations of a rescue could leave chipmakers and giant cloud operators shouldering more risk. That is why a strong backer and a willing debt buyer are not interchangeable.

Investors are responding with different forms of selectivity:

• Impax Asset Management says it is underweight data-center debt.
• KKR is avoiding remote facilities over concerns about repurposing them after leases expire, a person familiar with the matter said.
• Carlyle has passed on several recent deals while emphasizing diversification and selectivity.