Data centers have become one of the largest financing themes in the market, and they sit at an unusual intersection. They are real estate, they are infrastructure, and the revenue depends on technology tenants whose own needs change quickly. In my work advising companies and asset owners on debt and equity, I find the clearest way to think about a data center deal is to ask three questions: who is the tenant, will the power arrive, and what is the building worth when the lease ends.
The capital stack, layer by layer
At the top are the hyperscalers themselves, which fund a large share of their spending from cash flow and from corporate bonds. Bloomberg reported that Alphabet, Amazon, Meta and Oracle borrowed $93 billion in the U.S. investment-grade bond market in 2025.
Below that sits project finance. IJGlobal data cited by Bloomberg put data center project finance debt at about $170 billion in 2025, roughly 18 percent of that market's debt issuance and up 57 percent from the prior year. A typical structure uses a bankruptcy-remote special purpose vehicle that owns the project, a construction loan that converts to a mini perm of roughly three to four years after completion, and a refinancing once the lease is producing steady income.
Next comes securitization. JPMorgan projects data center ABS and CMBS issuance of $30 billion to $40 billion in each of 2026 and 2027, up from about $27 billion in 2025. Insurers and other long-term investors also buy privately placed debt backed by 15 to 20 year leases to highly rated tenants. Private credit funds, which have lent more than $200 billion to AI-related companies, fill the spaces that banks and bond markets leave open, and GPU-backed loans have emerged as their own niche.
What the biggest deal tells us
The Hyperion project in Louisiana, developed by Meta with Blue Owl, is the reference case. As reported in October 2025, it was financed with almost $30 billion: more than $27 billion of debt and about $2.5 billion of equity, with Blue Owl holding 80 percent and Meta 20 percent. The debt was a Rule 144A private placement of bonds issued by a special purpose vehicle, rated A+ by S&P, maturing in 2049 and priced about 225 basis points over Treasuries, with Pimco as anchor investor. The structure puts the credit of the tenant at the center and keeps the debt in a separate vehicle.
Few sponsors have a tenant like that. But the logic of the structure applies at every size: strong tenant, long lease, a ring-fenced asset and a debt tenor that matches the cash flows.
What lenders underwrite
Tenant credit comes first. A lease with a hyperscaler is a different asset from a lease with a start-up cloud, and pricing reflects it. Equipment finance for GPUs shows the difference clearly: because the chips lose value fast, lenders typically advance only about 50 to 70 percent of their value, with pricing reported in the range of 12 to 15 percent.
Lease term against debt tenor comes second. The danger is a loan that matures before the lease does, or a lease that ends before the loan is repaid. Where lease terms are shorter than the debt, lenders look hard at residual value, and in some structures the tenant guarantees it.
Power is third, and it is where many projects fail. Grid connections can take years. Lenders ask for utility will-serve letters, interconnection agreements and, increasingly, on-site generation or power purchase agreements that make costs predictable. A completed building without power produces no rent.
Completion risk is fourth. Lenders release construction funds in tranches against progress, expect experienced contractors and often require completion support from the sponsor. Lenders also look for debt service reserves, commonly six to twelve months of lease payments, along with reserves for taxes and insurance.
Refinancing is the risk people discuss least. Mini perm loans have to be refinanced at maturity, and some market commentators warn that the volume coming due could be large compared with the securitization market that is meant to take it out. Sponsors should plan the exit at the start and not at maturity.
Where the middle market fits
Not every data center is a multi-billion-dollar campus. A large part of the opportunity for sponsors in the middle market lies in powered land, colocation and edge facilities, the conversion of industrial and logistics buildings, and the supporting power and cooling infrastructure. These deals depend more on mezzanine debt, preferred equity and joint ventures to complete the stack. Investors I speak with in Europe, the Middle East and Asia are also looking for ways into U.S. digital infrastructure, and they care about the same questions of tenant, power and residual value.
A checklist for sponsors before they go to market
Have a signed or near-final lease, or a credible pre-leasing pipeline, with the tenant's credit documented. Have a power plan with utility correspondence in writing. Have a construction budget with contingency and a contractor with a record on comparable work. Know your tax position, because New York, California and Massachusetts do not conform to every federal depreciation rule. Decide how the project will be refinanced. And be prepared to show lenders what happens to the asset if the first tenant leaves.
Good financing does not remove risk from a data center. It allocates each risk to the party best able to carry it, and prices it honestly. That is the work, and it is why the advisory side of these transactions matters as much as the capital itself.
This article is general information, not legal, tax, investment or insurance advice, and it is not an offer of securities or investment banking services. The views are my own and are not those of Oberon Securities. Figures are third-party estimates or company statements as reported, and they change.

