Three assets with three different lives
An AI data center is not one asset but several, stacked on top of each other. The land and building last for decades. The power and cooling systems last for many years but are built for one site. The GPUs and servers inside are productive for a few years before newer generations overtake them. Each is paid for by a different stream of cash, and each suits a different kind of lender.
Financing all of them as one project often makes the capital more expensive than it needs to be. A lender asked to fund the whole stack prices it for its riskiest, shortest-lived layer, and matches repayment to none of them well. Financing each layer against its own repayment source, over its own life, is usually cheaper and gives each investor the risk it actually wants.
The layers of an AI data center
| Land and building | Useful for decades. Repaid by long-term leases to data center tenants. Suits real estate and infrastructure lenders with long-dated capital. |
| Power and cooling | Useful for fifteen to twenty years or more, though specialized to the site. Usually financed with the building, or separately where power is supplied under its own contract. |
| GPUs and servers | Useful for three to five years in their primary role. Repaid by compute contracts with customers. Suits lenders comfortable with shorter, contract-backed credit. |
A general guide. Useful lives vary with the design, the technology and the market.
Financing the building
Once a data center has signed long-term leases with creditworthy tenants, the building behaves much like other net-leased real estate: the rent is contracted, the tenant pays most operating costs, and the lease runs for ten years or more. That cash flow can support long-dated senior debt or notes sold to institutional investors, sized on the rent and the tenant’s credit rather than on the building’s appraised value alone.
The terms that matter mirror those in any lease-backed financing: the length of the lease against the debt, the tenant’s credit and any guarantee, how rent escalates, who bears operating and capital costs, and when either side can terminate. Construction risk is a separate question. Lease-backed notes are usually put in place once the building is delivered, replacing construction financing that carried the delivery risk.
Financing power and cooling
Power and cooling systems are long-lived but specific to the site, so their value depends on the building they serve. They are most often financed together with the building. Where power is supplied under a separate long-term contract, for example from on-site generation, that contract can support its own financing on the same principles as the building: a named buyer, a fixed term and predictable payments.
Financing the GPUs
The GPUs belong to a different credit altogether. They are repaid by compute contracts that typically run three to five years, and their value falls quickly after that. Lenders size this layer on the compute contract and expect repayment within its term, as explained in the guide to GPU financing for contracted AI infrastructure. Where the operator wants to finance part of the hardware’s value at the end of the term, that can be done separately, as the guide to financing GPU residual value describes.
Where the layers meet
The layers are financed separately, but they depend on each other, and the documents have to reflect that. The GPU operator’s colocation or lease payments are an operating cost that ranks ahead of its debt service, so the GPU lenders need the lease to run for at least the term of their debt. The building’s lenders, in turn, rely on the operator as a tenant.
The practical link is an agreement between the landlord and the GPU lenders, often called a landlord waiver or access agreement. It confirms that the landlord has no claim on the hardware, gives the lenders access to the site to remove or operate the equipment, and provides notice and time to cure if the tenant falls behind on rent. Without it, a GPU lender cannot be sure it can reach its own collateral.
Where Park Street Global fits
Park Street Global structures financing for each layer against its own contracted cash flow. For data center owners with signed leases, that means long-dated notes backed by the rent. For operators deploying GPUs inside those buildings, it means financing backed by their compute contracts. Where we work on both sides of the same site, we design the financings together so that the lease, the access agreement and the two sets of lenders fit from the start.
We do not lend our own money, so the size of a financing follows the transaction rather than our balance sheet.
Common questions
Can a data center and its GPUs be financed together?
They can, but it is usually more expensive, because one lender has to price the whole stack for its shortest-lived, riskiest layer. Financing each layer against its own cash flow and life generally lowers the overall cost.
What is a landlord waiver?
An agreement between the building owner and the lenders to a tenant’s equipment. It confirms the landlord has no claim on the equipment, gives the lenders access to the site, and provides notice and time to cure if the tenant falls behind on rent.
Can a data center under construction be financed this way?
Lease-backed notes are usually put in place once the building is delivered and the leases have started, replacing construction financing that carried the delivery risk. Signed leases are needed before construction financing can be refinanced on this basis.
Which lease terms matter most to lenders?
The remaining lease term against the debt, the tenant’s credit and any guarantee, rent escalation, who pays operating and capital costs, and the circumstances in which either side can terminate.
Does the GPU lender care about the building’s financing?
Yes. If the building’s lenders enforced against the landlord, the GPU lenders need assurance that their tenant’s lease and their access to the equipment would survive. The access agreement and the lease terms provide it.