Lenders are getting more selective about AI infrastructure. Knowing what they evaluate is the difference between a fast yes and a slow no.
Financing AI infrastructure is not hard because money is scarce. It is hard because lenders have learned what tends to go wrong.
As demand for AI infrastructure financing has grown, lenders have grown more selective. They have seen clusters that could not be energized, hardware that lost much of its value before it was paid off, and utilization forecasts that never materialized. Every one of those lessons is now a question in the underwriting process. The better a company anticipates those questions, the faster and cheaper its capital.
Underneath the paperwork, every lender is asking one thing: how will this loan be repaid, and what happens if it is not? Four factors answer it. Collateral determines what can be recovered if the borrower defaults. Contracts and revenue determine whether repayment is likely in the first place. The utilization forecast tests whether the plan is realistic. And power and deployment readiness determine whether the asset will ever generate the revenue the plan depends on. A weakness in any one can sink an otherwise strong application. Here is how lenders read each.
Collateral: Lending Against a Falling Number
In asset-backed financing, the GPUs secure the loan, so their resale value matters a great deal, and lenders know that value declines over time. Cloud rental rates for the H100 have already dropped 64 to 75 percent from their peak, according to cloud pricing trackers, a signal of how quickly the market reprices compute as supply expands and new chip generations arrive. Resale values fall more gradually than rental rates, but they still fall, and lenders underwrite to that decline.
They account for it by lending against a discounted value of the hardware and by setting a loan-to-value ratio, the share of the asset’s value they will finance, that leaves room for the drop. A borrower who understands that discount, and proposes a realistic ratio instead of arguing with it, structures a request lenders find easy to approve.
Contracts: The Strongest Evidence There Is
Contracted revenue is the most persuasive thing a lender can see. A company with signed customer contracts for the compute it is financing presents far less risk than one building capacity on speculation, because the contracts show the hardware will be used and the repayment money is real rather than projected.
The more of a project’s capacity is committed under contract before financing, the more favorable the terms tend to be.
Utilization: Where Optimism Gets Priced
Lenders scrutinize utilization forecasts because idle hardware repays nothing. A credible forecast is grounded in actual or contracted demand, not in assumptions about growth that has not happened yet.
Overbuilding is the classic failure: a company finances more capacity than it can fill, and the idle portion becomes a cost rather than a source of repayment. Conservative, specific forecasts tied to real demand carry far more weight than ambitious ones, and they price better too.
Power: The Question Borrowers Do Not Expect
A financed cluster that cannot be turned on is a stranded asset, and lenders know it. Power availability has become one of the longest lead-time items in AI infrastructure, with grid connection waits in constrained markets running several years and reaching much longer in the most contested regions.
Before financing a project, lenders increasingly want evidence that power is secured and the deployment can be completed on the stated timeline. Showing confirmed power and a realistic build plan removes one of the largest risks on their side of the table.
How to Strengthen a Financing Application
Most of what makes an application strong can be prepared before the first conversation:
- Assemble clean financials and a clear picture of revenue
- Bring signed or committed customer contracts for the capacity being financed
- Present a conservative utilization forecast tied to real demand
- Show that power and data center capacity are secured, not assumed
- Understand the collateral discount and propose a realistic loan-to-value ratio
What It Takes to Get Financed
Lenders are not looking for reasons to say no. They are looking for evidence that the loan will be repaid and that the asset behind it will hold value and generate revenue.
Companies that prepare for the questions lenders actually ask, about collateral, contracts, utilization and power, close funding faster and on better terms than those that treat financing as an afterthought.
