Debt options proliferate as lenders back data centres

“Building out the data centres needs so much capital that all markets essentially are being tapped in one way or another,” said John Medina, a senior vice president at Moody’s Ratings. Companies will spend nearly USD 7tn on data centre infrastructure globally by 2030, consultants McKinsey & Co forecast in an August 2025 report.

Much of the capital is going into the development stage, fuelling bank lending in the form of construction loans, project finance, commercial real estate loans and more, Medina said. Moody’s estimates the size of this market alone at more than USD 200bn.

Data centre operators have also issued high-yield bonds, and some market sources expect the broadly syndicated loan market will be next. Last autumn, TeraWulf raised USD 3.2bn from the high-yield bond market to finance the expansion of a data centre in New York state, with Google backstopping the lease obligation of the centre’s tenant.

Private credit managers are also diving deep into the sector via infrastructure debt deals and private asset-backed finance. Trinity Capital has backed hyperscalers — operators of giant data centres — by financing graphics processing units (GPUs), networking and other key equipment, according to CEO Kyle Brown.

Last October, Blue Owl formed a joint venture with Meta to develop a USD 27bn data centre and said it was funding part of its commitment via a private bond issued to PIMCO and other investors.

Some private credit managers offer a tighter spread than banks or the public securitisation market. They are able to do so by structuring long-term financing that covers multiple stages of development for a combined rate; they also evaluate risk differently than the public market or heavily regulated banks. A private lender may also place more stock in the triple A tenant that is locked into a long-term lease than a bank or public debt market would.

“Infrastructure-oriented credit can price tighter than cashflow lending because it’s typically supported by real assets, long-term contracts and essential service demand,” said Harlan Cherniak, head of infrastructure debt for Macquarie Asset Management. “As you move up the risk spectrum into development or construction projects, or non-investment grade tenants, required returns increase accordingly.”

As more data centres are completed in 2026, activity in both the ABS and private credit markets is expected to increase. At the same time, the cost of building an enormous, bespoke data centre for a hyperscaler is rising rapidly, motivating lenders to think up novel financing ideas or forge partnerships to provide the huge sums needed.

Use of the master development company arrangement, which involves multiple lenders banding together to write larger cheques, has also increased, according to David Ridenour, a partner at international law firm King & Spalding.

“We’ve done a bunch on the data centre lending side, both private and public,” said David Colla, head of capital solutions at CPP Investments. These include asset-backed debt deals, GPU-related financing, public bond issues and fully private financing.

The rush to develop data centres has prompted worries about oversupply. Many of the new projects assume the use of AI will continue growing exponentially. Comparisons have been drawn with the boom in installing internet infrastructure in the run-up to the telecom bust in the early 2000s. But optimists point to key differences.

“These are being built with long-term leases, long-term [power purchase agreements], long-term contracts and very little being built on spec,” Medina said. “Even if it does not materialise, this is a 15-year lease: the expectation is that tenants will pay. The tech company tenants are also rated higher and have better balance sheet strength than the telecom companies did.”

Lenders, of course, are doing what they can to protect their investments. “While we don’t control the board, management, compensation or capital allocation, we control what we believe is important — credit agreements and covenants,” Cherniak said.

If a project gets into trouble, lenders can foreclose on the asset and take it over. Many data centre campuses are built for AI computing, but their energy- and water-supply agreements make them attractive even to non-AI tenants. Moreover, once built, they have overcome any local opposition or regulatory obstacles, Ridenour said.

“In constrained markets where power is hard to come by, if a tenant falls away the space gets absorbed quickly,” said Maniesh Khatri, senior director of project finance at NORD/LB, a German bank.