Stay informed with free updates

Data centre developers are seeking credit ratings — even while facilities are still under construction — as the tech industry tries to open up new sources of capital to fund hundreds of billions of dollars in AI investments.

S&P, Moody’s, Fitch and Kroll Bond Rating Agency have in recent months expanded their coverage to data centre projects that are yet to be completed and in many cases given out private ratings for loans so that banks can offload them to a broader base of institutional investors.

Having the stamp of approval from rating agencies is particularly important for these colossal data centre build-outs, which have overwhelmed the project finance market and need to attract new classes of investors, such as insurance companies, that can only buy into high-grade deals rated by the big three agencies. 

“It’s astronomical growth,” said Roelof Steenekamp, who leads Fitch’s complex credit group that specialises in data centre ratings. “The vast majority of what my team rated has been new hyperscaler-backed facilities being constructed at the moment.”

Fitch has worked on more than 35 ratings for data centre projects in the past nine months, and most of them were private deals with an average size of about $3bn, Steenekamp said. The agency also released a new set of rating criteria for digital infrastructure in August. 

Moody’s and Fitch are among the agencies that privately handed out investment-grade ratings to tens of billions of dollars of data centre construction loans backed by Oracle, according to people familiar with the matter. 

Both firms declined to comment on specific transactions.

Smaller specialist providers, such as KBRA, entered the market even earlier, having rated their first data centre project 15 months ago, said Bill Cox, chief rating officer at the firm. KBRA currently provides ratings for close to $100bn of data centre debt. 

“We expect that to go up, possibly by another $25bn to $50bn in the first half of this year,” Cox said. 

Most data centre projects have sought investment-grade ratings, with bankers often altering deal structures and providing more loan protections to boost their creditworthiness.

Steenekamp said some clients refused to proceed when they were told such a rating was out of reach. About two-thirds of the ratings the firms handed out so far were investment grade.

These ratings are mostly backed by long-term leases signed with Big Tech companies because they have stronger credit profiles than little-known data centre developers. Project ratings are capped by the credit rating of the tenant.

Financial guarantees provided by highly rated tenants could improve credit metrics, said Dhaval Shah, a director for infrastructure ratings at S&P, which gave an A+ rating for $27bn of debt for Meta’s Hyperion data centre in Louisiana.

For instance, Meta promised in the contract to begin to pay rent even if construction is delayed and cover additional costs if the project goes over budget, Shah said. 

“The credit story is very simple,” Shah said. “[Lenders] are only taking risks on Meta.”

Data centres built specifically for AI training are more risky than those used for traditional cloud computing because they are usually located in remote areas and typically cannot be repurposed, making it hard to find another tenant after the initial lease ends, said Steenekamp.

“The biggest risk here is that if one of these hyperscalers fails, there will be a bunch of contracts that cannot be fulfilled,” Steenekamp said.