Can You Get GPU Financing Without an Investment-Grade Offtaker?
Summary
- Yes: GPU financing without an investment-grade offtaker works with a sub-investment-grade customer or documented on-demand revenue. Investment grade makes it cheaper and simpler.
- Weaker customer credit shifts the lender's weight onto the offtake contract, the hardware's residual value and the structure, which moves pricing and loan size.
- With no customer revenue, nothing repays the loan, so lenders look first for an executed offtake contract or documented on-demand revenue.
Why the Offtaker's Credit Usually Repays a GPU Loan
Many of the largest GPU financings rest on investment-grade customers, so operators selling to AI startups, inference platforms or on-demand users ask whether they qualify. Investment grade means a credit rating of BBB- or higher at S&P Global Ratings, or Baa3 or higher at Moody's. The offtaker is the customer that contracts to buy the compute, and its rating is the first thing a lender checks.
That check matters because the customer's payments usually repay the loan. A GPU loan is secured by servers that lose value as newer chips ship. Lenders want a payer whose credit lasts through the repayment period. When the customer is a startup with a short cash runway, the lender has to find that comfort elsewhere in the deal. The same applies when there is no contract at all.
This article walks through four offtake types: investment grade, sub-investment grade, on demand and no offtake yet. For each one, it shows how the deal's score, structure and loan size change. It then covers the credit support that stands in for a weak customer, and where the line sits between financeable and not yet financeable.
What an Investment-grade Offtaker Buys in a GPU Loan
An investment-grade customer lets lenders size and price a GPU loan mainly on its credit, so the largest facilities are built around the contract. The hardware still secures the loan. The customer's rating decides how much the lender is willing to advance against it.
CoreWeave shows the mechanism in its filings. Its DDTL 2.0 is a delayed-draw term loan, a facility the borrower draws in stages as it buys hardware. For the quarter ended Sep 30, 2025, the borrowing limit was a percentage of the "depreciated purchase price of GPU servers and related infrastructure," set "based upon the credit rating of the applicable customer." The filing doesn't publish the schedule, and the principle is plain: better customer credit supports more debt per server.
The same approach reached its clearest form in March 2026, when the company closed an $8.5 billion facility described as the first investment-grade-rated GPU-backed financing. The facility is rated A3 by Moody's and A (low) by DBRS, and is secured by HPC infrastructure and what the company called "an associated customer contract" with a leading AI enterprise.
In August 2026, Moody's assigned a Baa2 rating to the $926 million term loan B that Lambda announced. The loan funds a GPU deployment for an investment-grade offtaker and matures Dec 31, 2030, "on a fully amortizing schedule aligned with the contracted cash flows." Under a take-or-pay contract, the customer pays for reserved capacity whether or not it uses it. Our guide to take-or-pay contracts explained covers how those payment terms are drafted.
Put investment-grade credit together with a tenor matched to the contract, and the loan is repaid while a strong customer is still paying. Residual value, what the servers will fetch if they are sold later, then carries less weight in the lender's review. That is the protection an operator has to rebuild when its customer isn't investment grade.
How the Four Offtake Types Change a GPU Financing
The offtake type shifts the lender's weight between the customer, the offtake contract and the collateral. The Compute IQ scoring model scores every deal on one 100-point scale with three inputs: the operator, the customer and the offtake contract. Customer credit and contract terms carry most of the review, and deals scoring below 60 aren't financeable.
The score sets the facility size, one of three pricing tiers and the structure. The Compute IQ page's illustrative sample deal is 256 NVIDIA B200 GPUs serving an AI lab on a 36-month offtake contract. In that sample, the customer adds 34 points, the contract 36 and the operator 18, so customer and contract supply 70 points between them.
| Offtake type | What the lender relies on | Compute IQ effect | Pricing tier (relative) | Structure and tenor | Residual-value warranty |
|---|---|---|---|---|---|
| Investment grade | The customer's credit and the offtake contract | Higher customer score | Moves toward a better tier | Standard structure; 3–5 year term matched to contract tenor | Not required by the customer rule |
| Sub-investment grade | Contract strength, customer financials and the hardware's residual value | Lower customer score; contract terms carry more weight | Higher tier than investment grade | Standard structure; term matched to contract tenor | Required |
| On demand | Documented revenue and the hardware's residual value | Doesn't disqualify; priced for the shorter commitment | Reflects the shorter commitment | Standard structure | Required |
| No offtake | No revenue to repay the loan | Not financeable until revenue is in place | Not applicable | Not applicable | Not applicable |
Investment-grade offtaker
An investment-grade customer earns a higher customer score, which moves the deal toward a better pricing tier. In financing through USD.AI, terms run 3–5 years with straight-line monthly amortization, matched to the offtake contract's tenor. Principal comes down on a schedule the contract is still paying for.
Every deal follows the same structure, set out in USD.AI's underwriting standards. A special-purpose vehicle holds the GPUs, the offtake contracts and the revenue accounts, and lenders hold a first-priority claim on all three. Capital is held in escrow until the servers are installed, burned in and independently verified online.
A three-month debt-service reserve is funded at closing. Property and casualty insurance covers the hardware once escrow is released. The investment-grade customer rule doesn't require a residual-value warranty, because the customer's credit already carries the repayment risk.
Sub-investment-grade offtaker
A sub-investment-grade customer, one rated below BBB- or Baa3, is financeable. The customer is assessed on its financials, backers, cash runway and leverage. The offtake contract is assessed on its term against the loan term, its take-or-pay payments and how well those payments cover debt service.
Weaker customer credit lowers the customer component of the score, so contract strength carries more of the deal. The deal lands in a higher pricing tier and also requires a residual-value warranty. The warranty covers part of a shortfall if servers sell below an agreed value, subject to its terms, as how residual-value warranties work explains.
Rating agencies point to the same exposure. Fitch Ratings wrote in October 2026 that remote AI training centers with concentrated, lower-quality tenants could face greater exposure. In July 2026, KBRA listed re-leasing and residual value, operator replaceability and counterparty concentration among AI compute credit considerations.
In our view, the warranty is how a sub-investment-grade deal rebuilds part of the protection an investment-grade customer gives. The offtake contract still carries the repayment case.
On-demand revenue
On-demand revenue, compute sold by the hour or on short commitments without a long-term contract, doesn't disqualify a deal in Compute IQ. It is priced to reflect its shorter commitment, and the deal requires a residual-value warranty. The revenue has to be documented, because occupancy alone doesn't establish the capacity to repay a loan.
Even the largest operators now sell some capacity on short terms. In Q3 2026, CoreWeave signed "short-dated customer contracts" of about three to six months. By comparison, CoreWeave's 2025 annual report said its initial contract periods generally ranged from one to six years.
Lenders are developing tools for this kind of revenue. Clifford Chance noted in July 2026 that GPU-as-a-Service models carry less track record with counterparties. The firm said dilution reserves that move with performance could emerge over time, alongside triggers and stop-purchase rights. It also flagged that service credits can tip into termination risk.
Re-leasing risk, the chance that GPUs sit idle between customers, sits at the center of on-demand credit. The same agency's June 2026 research on re-leasing risk for GPUs names four drivers: workload suitability, customer extension, third-party demand and chip fungibility. We found no public financing secured mainly on on-demand revenue, so the Compute IQ rule is the clearest published standard.
No offtake yet
A deployment with no customer revenue isn't financeable until it has an executed offtake contract or documented on-demand revenue. Planned usage, product plans and an expected funding round don't count, because none of them makes a loan payment.
GPU financing through USD.AI starts with lining up customer revenue, a hardware purchase order, colocation and the equity contribution. When a deal falls short, Compute IQ shows which gaps to close. A first deployment can still qualify when it pairs a strong customer with a well-drafted offtake contract.
What Stands in for Investment-grade Credit
Large GPU deals replace weak customer credit with a stronger party's support, and middle-market operators usually don't have that option. For them, the offtake terms and the residual-value warranty carry the weight.
Three recent arrangements show how the largest projects do it. In November 2025, Cipher Mining disclosed that Google agreed to backstop part of Fluidstack's lease obligations to support project-related debt. In November 2026, TeraWulf announced contractual credit support from Google for anticipated debt financing tied to its Anthropic lease.
In August 2026, NVIDIA said it may provide residual-value support for up to 25% of an opportunity, decided project by project. NVIDIA described that support as a complement to independent underwriting.
| Credit support | Who provides it | Available to middle-market operators? |
|---|---|---|
| Lease backstop | A hyperscaler such as Google | Mostly reserved for large projects |
| Residual-value support | A chipmaker such as NVIDIA | Project by project, at the chipmaker's discretion |
| Residual-value warranty | A warranty provider | Required in financing through USD.AI where the customer isn't investment grade or revenue is on demand |
| Offtake contract terms | The operator and its customer | Yes |
| Equity | The operator | Yes |
An operator controls the levers that move the customer and contract components of the score:
- Contract tenor. Match the offtake contract's term to the loan term, so the customer is paying through the whole repayment period.
- Take-or-pay and termination protection. Payments that don't depend on usage, and limits on early termination, keep debt service covered.
- Customer financials. Current financials, named backers, cash runway and leverage give the customer score evidence to work with.
- Equity. Financing through USD.AI requires at least 20% equity in the deployment.
In our view, these levers and the residual-value warranty do the work a hyperscaler backstop does in larger deals. They are also the parts of a deal an operator can improve before applying.
Where the Line Sits for GPU Financing Without an Investment-grade Offtaker
Investment-grade credit is the lowest-cost path to GPU financing, and sub-investment-grade and on-demand deals are financeable too. Those deals carry a residual-value warranty and a pricing tier that reflects the added risk. With no customer revenue at all, the first step is lining up an executed offtake contract or documented on-demand revenue.
Every financing through USD.AI is scored by Compute IQ on the same 100-point scale and closes on the same structure. Financing starts from about $5 million. A passing score isn't a financing commitment, since the remaining conditions are verified and final approval is given before a loan closes. Every publicly announced loan appears in the USD.AI loan history.
If your customer isn't investment grade or you sell on demand, Talk to USD.AI about how your offtake contract would score. Bring the offtake contract or revenue records, the hardware purchase order and your colocation plan.
FAQ
Can I get GPU financing if my customer isn't investment grade?
Yes, a sub-investment-grade customer is financeable, scored on its credit and the offtake contract's terms. The deal lands in a higher pricing tier and requires a residual-value warranty.
Does on-demand revenue qualify for a GPU loan?
Yes, if the revenue is documented. On-demand revenue is priced to reflect its shorter commitment, and the deal requires a residual-value warranty.
Can I finance GPUs before I sign a customer?
No, you need an executed offtake contract or documented on-demand revenue first. Planned usage, product plans and an expected funding round don't count.
How much does financing through USD.AI cover?
Financing through USD.AI covers up to 80% of the deployment cost (loan-to-value). You contribute at least 20% as equity.
Editor Notes
Facts held back by brand rules:
- USD.AI minimum DSCR (1.15x, tested quarterly on cash collected). Published on the underwriting standards page and requested by the reviewer as one line. Held back by the Loan terms rule (no minimum DSCR) unless the brief marks it approved by Samuel or another senior team member under the Approved exceptions rule.
- CoreWeave DDTL spreads by customer credit tier, DDTL 4.0 tranche pricing and Compute IQ pricing-tier rates. Held back by the brief (no rate figures) and the Approved exceptions rule.
- Cipher Mining $333M and TeraWulf ~ $19B figures. Held back per the outline; the detail belongs in the planned bitcoin miner AI conversion financing post.
- Warranty provider and reinsurer names. Held back by the Service providers rule.
- USD.AI published deals (for example, the multi-year IG-contracted GB200 NVL72 facility, Sep 2026). Dropped per reviewer, under the Borrower names rule and the brief.
- Internal links to add once live: DSCR covenant, GPU private credit (for CoreWeave pricing), GPU as a service, GPU financing options and bitcoin miner AI conversion financing.
Site fix to flag: the Compute IQ page FAQ uses "residual-value insurance" wording, which conflicts with the Compliance vocabulary rule. The internal link slug <https://usd.ai/blog/gpu-residual-value-insurance> also carries that wording; the anchor and copy here use "residual-value warranty".
