USD.AI Underwriting Standards: How GPU Loans Are Scored, Secured and Monitored
Every deal that uses GPU financing through USD.AI is held to the same underwriting standards, from the hardware that qualifies to the coverage test each loan has to pass every quarter. They set what can be financed, how a deal is scored, how the loan is sized and secured, how it is monitored for its full term and what happens when coverage falls short. Compute IQ, USD.AI's agentic underwriting system, applies the same standards to every deal.
This article is written for GPU cloud operators preparing a financing and for the institutions reviewing one. The order of steps from application to funding is covered in our guide to the GPU financing process, and the SPV, escrow and enforcement mechanics in our explainer on how a GPU-backed loan is structured.
What Qualifies for Financing
Financing through USD.AI is for operators that own enterprise GPUs, sell compute from them under an offtake contract or on demand, and run them in a Tier III or Tier IV data center. That covers neoclouds, GPU-as-a-Service providers, HPC operators and data center operators that own the GPUs they sell compute from. Whoever owns the hardware is the borrower, and the loan is repaid from the revenue the hardware earns. On-demand revenue doesn't disqualify a deal; it is priced to reflect its shorter commitment.
The hardware has to be on the approved collateral list, which as of October 2026 covers NVIDIA B200, B300, GB300, GB200 and Vera Rubin (VR200) systems and AMD Instinct MI455X and MI430X. Every deployment also gets a technical evaluation of its data center, and the colocation provider signs a lien waiver before escrow releases, which leaves it with no claim on the servers that could rank ahead of the lenders.
The borrower and its owners have to sit in jurisdictions where the lenders' security interest can be perfected and enforced. Entities on the US Treasury's OFAC sanctions lists aren't eligible.
How Each Deal Is Scored
Compute IQ scores every deal on a 100-point scale, and a deal that scores below 60 isn't financeable. The score covers three pillars: the operator behind the cluster, the customer buying the compute and the offtake contract between them. The operator is assessed on its track record, deployments to date, financial strength, cash runway and equity backing; the customer on its cash runway, earnings, liquidity, leverage and operating history; and the contract on its tenor against the loan term, its termination protection and whether payment is unconditional (take-or-pay).
The score sets the facility size, one of three pricing tiers and the loan's structure, and the rate can be fixed or floating. When a deal falls short, Compute IQ shows which gaps to close. A passing score isn't a financing commitment: the remaining conditions are verified and final approval is given before a loan closes.
Loan Terms and Structure
Every loan covers up to 80% of the deployment cost and is repaid in equal monthly principal payments over three to five years. The operator contributes at least 20% of the cost as equity, a three-month debt-service reserve is funded at closing, and financing starts at about $5 million, with larger facilities available. Longer terms are matched to the tenor of the offtake contract, so contracted revenue covers repayment for the whole term.
The financed assets sit in a bankruptcy-remote special-purpose vehicle (SPV), a company formed only to hold them, which owns the GPUs, the offtake contracts and the revenue accounts. The lender group holds a first-priority perfected security interest over the SPV's assets and a pledge of 100% of its equity. Loan capital is held in escrow until the servers are installed, burned in and independently verified online, and customer payments run through a controlled revenue account for the full term.
How the Loan Deleverages Against the Collateral
In a three-year loan at 80% loan-to-value (LTV), the loan balance as a share of the hardware's value, the balance falls faster than the modeled value of the servers, so the operator's equity cushion grows through the term. How long data center GPUs stay useful is still debated. USD.AI's underwriting models a conservative four-year useful life. CoreWeave's 2025 annual report depreciates its technology equipment over six years, and Alphabet depreciates servers and network equipment generally over six years, while Amazon shortened the useful life of a subset of its servers from six years to five from January 2025. At GTC Berlin in October 2026, NVIDIA CEO Jensen Huang said a GPU lifecycle once assumed to be four years "really is probably eight to maybe more."
The chart models the hardware depreciating straight-line over each of the three lives against the same loan. On USD.AI's four-year life, the hardware is worth 75% of its cost after year one and the loan about 53%, an LTV of about 71%, and after year two LTV is about 53%. The loan is repaid at the end of year three with the hardware still worth a quarter of its cost. On a six-year life, LTV falls to about 64% and then 40%, and the hardware is worth half its cost at payoff. On an eight-year life, LTV falls to about 61% and then 36%, with the hardware at about 63% of its cost at payoff. The figures are illustrative, and how the curve runs for a given deal depends on its tenor and how that maps to modeled depreciation.
Monitoring for the Full Term
Each loan is tested every quarter against a minimum debt service coverage ratio (DSCR) of 1.15x, measured on the cash collected in the quarter. DSCR compares the cash a deployment brings in with the debt service due over the same period, so 1.15x means collections run at least 15% above what the loan requires.
The debt-service reserve covers payments if collections are delayed, and all revenue runs through the controlled account. A monitoring node at the data center reports uptime and server health throughout, which gives the lenders an independent view of whether the collateral is running alongside the payment record. A dedicated servicer manages each loan for its lenders for the full term.
When coverage falls short
A breach of the coverage test sets off a defined sequence, and each step gives the lenders more control over the deployment's cash.
- First breach: cash sweep. From the first quarter coverage falls below 1.15x, a cash trap and sweep applies, and cash the deployment collects beyond scheduled debt service goes toward the loan.
- Two consecutive breaches: cure period. The operator has a cure period to make a mandatory partial prepayment that brings the loan back into compliance.
- No cure: default. If the breach isn't cured, the loan is in default. The lenders can, at their option, accelerate the loan and enforce against the collateral: the servers, the offtake contract and the revenue accounts held in the SPV.
Step-in rights under the colocation agreement let the lenders keep the servers powered and racked in the same data center while they decide whether to keep the deployment running or sell the hardware. GPUs earning revenue under a contract are usually worth more than the same servers pulled out and sold.
Insurance and the Residual-value Warranty
Every loan carries property and casualty insurance on the hardware once escrow is released. Where the offtake customer isn't investment grade, or the revenue is on demand, the deal also requires a residual-value warranty, which covers part of a shortfall if the servers sell below an agreed value, subject to its terms.
Representative Diligence Checklist
Every financing is reviewed against the same set of documents before it closes. The list below is representative, and a specific deal can call for more.
| Document | What it shows |
|---|---|
| Corporate org chart | Who owns and controls the borrower |
| Financial statements | The operator's financial position |
| Evidence of the equity contribution | At least 20% of the deployment cost committed |
| GPU purchase order | Systems on the approved collateral list, from an OEM or authorized supplier |
| Colocation agreement | A Tier III or Tier IV site, the lien waiver and the lenders' step-in rights |
| Offtake contract, or on-demand revenue history | The revenue that repays the loan |
| Lien and judgment searches | No prior claims on the borrower or the assets |
| AML/KYC on every ultimate beneficial owner | The people who ultimately own or control the company |
| OFAC screening | No sanctioned parties |
Every loan also requires a technical evaluation of the data center, a monitoring node installed at the site and property and casualty insurance on the hardware.
FAQ
What DSCR does a GPU loan through USD.AI need?
Each loan needs a debt service coverage ratio of at least 1.15x, tested every quarter on the cash collected in that quarter. A breach starts a cash sweep, and two consecutive breaches open a cure period.
What GPUs qualify for financing through USD.AI?
As of October 2026, the approved collateral list covers NVIDIA B200, B300, GB200, GB300 and Vera Rubin (VR200) systems and AMD Instinct MI455X and MI430X. The servers have to run in a Tier III or Tier IV data center.
What happens if a GPU loan defaults?
If a coverage breach isn't cured, the loan is in default, and the lenders can accelerate it at their option and enforce against the collateral held in the SPV. Step-in rights let them keep the servers running in the same data center while they decide whether to operate or sell the deployment.
Does every GPU loan carry a residual-value warranty?
The residual-value warranty is required where the offtake customer isn't investment grade or the revenue is on demand. Every loan carries property and casualty insurance from the day escrow is released.
Send us your deployment plan and offtake contract, and we'll show you how it measures against these standards. Talk to USD.AI
