GPU Residual Value Insurance
Definition
GPU residual value insurance, or RVI, is a policy under which an insurer guarantees that financed graphics processors will be worth at least a stated amount, the insured residual value, on a stated future date, and pays the policyholder, usually the lender or lessor, the difference if a sale at that date brings less. It is an adaptation of the residual value insurance long used in aircraft, rail and vehicle leasing to the GPU-backed loans and leases that have financed AI compute since 2023. The insured value is set below the expected resale value on a depreciation curve, commonly 30 to 50 percent of original cost at year three for current-generation accelerators, the premium runs a few percent of the insured amount, and the policy pays only on a genuine sale after the equipment has been returned in the agreed condition. For the lender the policy converts an uncertain terminal value into a rated counterparty's promise, which supports a higher advance rate or a longer amortization; for the borrower it lowers the cost or raises the size of the loan; for the investor in either, it shifts obsolescence risk to an insurer rather than removing it.
Why it matters
Every GPU-backed loan is a bet on what the chips will sell for when the loan ends, and nobody has a long history to bet with: the oldest data center accelerators of the current cycle are four years old and their resale market is thin. Residual value insurance is how the market has begun to make that bet financeable at scale, by paying a specialist to take the tail. It also changes what an investor in a GPU loan, a GPU securitization or the lender itself is exposed to: less chip obsolescence, more insurer credit, and a set of policy conditions that decide whether the guarantee pays when it is needed.
Common misconceptions
- •RVI is not a guarantee of the loan. It guarantees a value for the equipment on a date; the lender's exposure above that value, and before that date, is uninsured.
- •The insured value is not the expected value. It is set below the forecast so that the insurer pays only in a downside; a policy that guaranteed the forecast would cost far more than lenders would pay.
- •The policy does not pay on a mark. It pays on a realized sale, after the equipment has been returned, inspected and marketed under the policy's procedure; a lender that sells badly can forfeit the claim.
- •Insured GPUs are not safer to operate. The policy is indifferent to utilization, contracts and revenue; those remain the borrower's risk and the lender's credit analysis.
Technical details
Worked example: a 36-month GPU loan with and without RVI
Fleet: 2,000 accelerators at $25,000 each, $50 million of cost. The lender's depreciation curve puts resale value at 60 percent after year one, 45 percent after year two and 30 percent after year three ($15 million).
Without RVI: the lender advances 65 percent ($32.5 million) and requires amortization to a balloon of $12 million at month 36, 80 percent of the forecast residual, leaving a 20 percent cushion. Borrower payments: about $20.5 million of principal over three years plus interest.
With RVI: an insurer guarantees 38 percent of cost ($19 million) at month 36 for a $760,000 premium. The lender advances 75 percent ($37.5 million) and sizes the balloon to $17 million, 90 percent of the insured value. Borrower payments: about $20.5 million of principal, on a larger loan.
Outcome A: GPUs sell for 40 percent ($20 million) at month 36. Both loans repay in full; the insurer pays nothing.
Outcome B: a new architecture drops resale to 22 percent ($11 million). Without RVI the lender recovers $11 million against a $12 million balloon and loses $1 million. With RVI the insurer pays $8 million (to $19 million), the lender recovers $19 million against a $17 million balloon, and the insurer bears the loss.
How the policy is written
Insured: the lender, lessor or securitization trust as loss payee; the borrower is usually the premium payer.
Insured value schedule: a value per unit at each measurement date, stepping down over the term; often 30 to 50 percent of cost at three years for current-generation accelerators, lower for older ones.
Measurement: net sale proceeds from an arm's-length sale conducted under the policy's marketing procedure within a window after the measurement date, sometimes by an approved remarketer.
Conditions: equipment returned complete, in working order, with documentation; serial numbers matching the schedule; no prior liens; claims notified within a period.
Premium: single upfront payment, a few percent of the insured value; some policies add a return-condition deposit.
Term: matched to the loan or lease, two to five years; early termination on borrower default may or may not be covered.
What drives the insured value
Model and generation: current flagship accelerators hold value better than the prior generation but face the same cliff when the next one ships in volume.
Secondary market depth: the number of buyers for used accelerators, which has grown with inference workloads that do not need the newest chips.
Cloud rental rates: the hourly price of the GPU's compute, which sets what a buyer will pay to own it; rates for the prior generation fell by roughly half within eighteen months of the successor's release.
Condition and configuration: complete servers with networking sell for more than bare boards; liquid-cooled variants have a narrower buyer base.
Concentration: an insurer with a large book of one model on similar dates faces a correlated loss, which limits capacity and raises price.
Where RVI appears in the capital structure
Bilateral GPU loans: as a condition for higher advance rates or longer tenors from banks and private credit lenders.
Equipment leases: lessors insure the residual they must recover to earn their return, the original use of RVI in aircraft and rail.
GPU securitizations: the residual exposure of the collateral pool is insured or partly insured, and the offering documents describe the insurer, the insured schedule and the exclusions; rating agencies treat the insurer's rating as a cap.
Vendor and cloud financing programs: chip and server vendors have used residual guarantees to support customer financing, which is a related but different instrument (the vendor, not an insurer, stands behind the value).
Reading a policy from the investor's side
Find the insured value schedule and compare it with an independent depreciation view; the gap is the lender's uninsured exposure.
Find the exclusions and the sale procedure; most disputed claims turn on return condition and the marketing process.
Find the insurer and its rating, and ask how much GPU residual exposure it carries; the policy is only as good as the balance sheet behind it.
Find whether the policy survives borrower default before the measurement date, which is when a lender most needs it.
In a securitization, find what share of the pool is insured and by whom; uninsured residual exposure is the notes'.
Frequently asked questions
- What is GPU residual value insurance?
- An insurance policy that guarantees the resale value of financed GPUs at a future date. If a lender repossesses or a lessor takes back the accelerators when the loan or lease ends and sells them for less than the value the policy guarantees, the insurer pays the shortfall. It exists because the biggest uncertainty in lending against GPUs is what they will be worth in two or three years.
- How is the insured residual value set?
- Below the lender's own forecast, from a depreciation curve for the specific model. An insurer might guarantee 35 percent of original cost at 36 months for a current-generation data center GPU whose expected resale value at that date is 45 to 50 percent, so that the policy is a backstop for a bad outcome rather than a forecast. The value is stated per unit or per serial number, and it steps down each quarter or year over the policy's life.
- What does GPU RVI cost?
- Reported terms are private, but residual value cover in other equipment classes runs 1 to 5 percent of the insured amount as a one-time premium, higher for younger asset classes with short data histories, and GPU cover has priced toward the upper end. On a $50 million GPU fleet with $17.5 million of insured residual value, a 4 percent premium is $700,000, paid by the borrower at closing and usually financed into the loan.
- What does the policy exclude?
- Almost everything except market value at the measurement date. Physical damage, theft, missing units, equipment not returned in the specified condition, values realized through a rushed or non-arm's-length sale, borrower default before the measurement date (in some policies), and losses caused by the policyholder's failure to follow the sale procedure are excluded. The insurer also caps its payment at the insured value and is not liable for the lender's loan balance above it.
- Why does RVI let a lender advance more?
- Because it puts a floor under the terminal value the lender must rely on. A lender that underwrites a 3-year loan assuming GPUs will be worth 25 percent of cost at maturity can advance only what that value plus interim cash flows will repay. If an insurer guarantees 35 percent, the lender can size the balloon to the guarantee and lend more up front, or amortize more slowly, and a securitization of the loans can allocate the residual exposure to the insurer rather than to the notes.
- Does residual value insurance remove GPU obsolescence risk?
- No. It transfers the risk for the insured amount to the insurer and leaves the rest with the lender and borrower. If a new architecture cuts resale values to 15 percent of cost, the insurer pays the gap to 35 percent, the lender still loses on any balance above that, and the insurer's ability to pay depends on its capital and on how much GPU exposure it has written. A market where every GPU loan is insured by a few carriers has concentrated the risk, not eliminated it.