A lender underwrites cash flows, not enthusiasm. When an operator walks into a credit conversation with a plan to buy several million dollars of Blackwell systems, the problem is rarely the demand thesis. The problem is that the asset does not exist yet, its price is not published anywhere, and the only document on the table is a purchase intention that can change shape at any moment. That is not collateral. It is a hope with a spreadsheet attached.
A standardized Rillor forward changes the document. Instead of an open-ended intention to buy hardware at some future spot price, the operator holds a bilateral contract with a fixed price, a contracted delivery month, a deposit sitting in independent escrow, and a seller performance bond standing behind delivery. The future purchase becomes a defined contractual claim. That is precisely the object a lender can examine, value, and lend against well before a single GPU ships.
The financing gap, stated plainly
The capital is there. AI-related companies tapped debt markets for at least 200 billion dollars in 2025, and GPU-backed lending has emerged as a genuine asset class in which high-performance chips serve as loan collateral, usually isolated inside a bankruptcy-remote special purpose vehicle. The structures exist. What stays hard is the underwriting itself, and it stays hard for two structural reasons.
First, the hardware does not exist at the moment the credit is needed. An operator borrows to fund a buildout that is months out. The lender is being asked to advance against an asset that has not been manufactured, let alone delivered. Traditional equipment financing answers this with caution: deals often take 60 to 90 days to approve, carry minimums of 50 million dollars or more, and come back with conservative loan-to-value ratios of only 40 to 50 percent. That conservatism is not stubbornness. It is the honest price of underwriting something you cannot yet see.
Second, the price is opaque. NVIDIA data-center GPUs have no published retail price. They are sold through OEMs and integrators, with the final number left to bundles and contracts, so the sticker on an H100 or H200 system is approximated through resellers and leaks rather than read off a tape. And where a price signal does exist, it moves violently. The H100 rental market fell from rates routinely above 7 to 10 dollars per GPU-hour at its mid-2022 launch to roughly 2 to 4 dollars per hour by late 2025, one of the fastest corrections in modern infrastructure history. A lender trying to set loan-to-value against an opaque, fast-moving spot reference is guessing, and prudent lenders price guesses punitively.
The result is a gap. The operator has real demand and a real plan. The lender has real capital and real appetite. What is missing between them is a document that fixes price and date with enough certainty to underwrite.
What a forward contract supplies
A Rillor forward is a bilateral over-the-counter contract for deferred physical delivery of a complete OEM GPU system at a fixed price in a contracted delivery month. Four of its mechanics map almost directly onto the questions a credit committee asks.
A fixed price answers "what does this cost." The number is set at execution and does not float with spot. The lender is no longer underwriting against a moving target. It is underwriting against a contractual figure that will not change between signing and delivery.
A contracted delivery month answers "when does the cash flow land." The operator can show exactly when the financed hardware arrives, which means the lender can align draw schedules, capitalized interest, and the start of revenue against a date on a contract rather than a vendor's verbal estimate.
A 10 percent deposit in independent escrow answers "is the buyer real." At execution, the buyer places 10 percent of contract value with an independent escrow agent, with the balance due at delivery. That deposit is committed capital sitting outside both counterparties' operating accounts. For a lender, it is evidence of seriousness and a structural buffer that sits clear of either party's insolvency estate.
A seller performance bond answers "what if it never ships." The single largest fear in financing an asset that does not exist yet is non-delivery. The seller posts a performance bond, so a failure to deliver is not just a breach the buyer has to chase through litigation. It is a covered event with capital standing behind it. That converts non-delivery from an open-ended credit hole into a bounded, defined risk.
There is recent proof that lenders move decisively when these elements line up. On March 31, 2026, CoreWeave closed an 8.5 billion dollar delayed draw term loan that was the first investment-grade rated financing secured by HPC infrastructure and an associated customer contract, rated A3 by Moody's and A (low) by DBRS. The lesson is not the headline number. It is the structure: collateral plus a contracted commitment earned an investment-grade rating. A Rillor forward supplies the contracted commitment on the supply side of that same equation, fixing price and date before the hardware exists.
Why standardization is the financial feature
The mechanics above would matter even on a one-off agreement. Standardization is what turns them into something a lender can scale.
Every Rillor contract uses the same six fields: SKU, quantity, delivery month, deposit percentage, settlement currency, and channel-of-record. There is no bespoke legal language per buyer, no clause archaeology, no need to re-underwrite the structure of each agreement from scratch. A lender reviewing one Rillor forward has effectively learned to read all of them. A book of ten contracts is ten comparable claims, not ten separate negotiations.
That comparability is what makes a book of forwards assessable as collateral-adjacent value. The contracts are not the GPUs, and they are not a security. They are standardized delivery claims whose value can be marked, aggregated, and tested against covenants because they share a common shape. This is the same reason commodity markets standardized their instruments long before anyone could finance them efficiently: a heterogeneous pile of private deals resists portfolio underwriting, while a stack of identical contracts invites it. A book of bespoke supply agreements has to be read one contract at a time; a book of standardized forwards can be read once and then counted.
The hardware underneath the SKU is itself standardized, which is what lets the contract reference a known thing rather than a description. A RIL-GX-B200-2T forward points at a complete HGX B200 system whose specification is published by the OEMs that build it. The NVIDIA HGX B200 180GB GPU carries 180 GB of HBM3e, 7.7 TB/s of memory bandwidth, a 1000W total graphics power, 900 GB/s of NVLink, and up to 18 petaFLOPS of FP4 Tensor Core performance with structural sparsity, per Lenovo's official product guide. Whether the system ships as a Supermicro SYS-A22GA-NBRT, a Gigabyte G894-AD1-AAX5, a Dell PowerEdge XE9780, an HPE ProLiant Compute XD685, or a Lenovo SR680a V3, the financed asset is a known, documented configuration rather than a vague "some GPUs." You can browse the full set of contractable systems on the SKU catalog and see live forward references on the marketplace.
Pre-delivery transferability is the recovery path
A lender's first question after "what is it worth" is "what happens if the borrower defaults." With most equipment commitments tied to a future delivery, the honest answer is grim: a supply agreement personal to the borrower, hard to assign, and worth little to anyone else.
A Rillor forward is transferable before delivery. With Rillor and OEM approval, the contract can be assigned to another KYC'd buyer from Rillor's verified book. For a lender, that is a recovery mechanism that does not depend on repossessing hardware that has not been built. If the borrower fails, the underlying claim, a fixed-price, dated delivery of a known system, can move to a creditworthy counterparty rather than evaporating.
This matters most precisely when the forward is in the money. If the contract price sits below the prevailing forward for that SKU, the claim has positive transfer value, and a lender holding a security interest in it has something to realize. Standardization is again what makes this practical: an assignable claim is only useful if there is a book of qualified buyers who can step in, and the common contract shape is what lets a replacement buyer be slotted in quickly. The transfer still routes through Rillor and the OEM, so the end-customer-of-record is updated cleanly rather than left stale.
The index as an independent mark
Loan-to-value and covenants require a number that both sides trust and that neither side sets. For GPU systems, that number has historically not existed, which is exactly why lenders defaulted to 40 to 50 percent advance rates against a price they had to guess.
The Rillor Compute Index supplies the missing mark. It is a 30-day rolling-blend forward price per SKU, computed from active Rillor contracts, owned and controlled by Rillor, and licensed as a settlement feed and API. A lender can write covenants against it: an initial advance rate set off the index value of the financed SKU, a maintenance test that compares outstanding principal to the marked value of the contract book, and a margin or cure mechanism if the index moves against the position.
The index is independent of the borrower in the way a credit committee needs. It is not a self-reported valuation and not a stale appraisal. It is a price derived from real, active forward contracts across the market, which is structurally harder to game than a single quote. That independence is the whole point of building the reference the way Rillor built it, and third-party venues license the same feed to run their own cash-settled products against it. For lenders and funds that want to consume the feed under license, the terms live on the markets page.
One caution worth stating to your risk team: the index marks the contract, not the eventual residual value of the deployed hardware. Depreciation risk, the so-called GPU debt cliff that lenders rightly fear, lives in the asset over its operating life, not in the forward claim. The forward fixes acquisition price and date. It does not promise a resale value years out. Treat the index as a mark for the claim you are financing, not a guarantee about the fleet that claim becomes.
Channel compliance is preserved, not bypassed
A financing structure that quietly broke NVIDIA channel rules would be worse than no structure at all, because it would taint the very delivery the lender is counting on. Rillor forwards are built so that financing leaves channel compliance intact.
Every contract captures the end-customer-of-record for the OEM's NVIDIA business unit. The financed delivery flows through the authorized channel, with the actual end customer named on the contract, so the transaction sits inside the rules of the NVIDIA Partner Network rather than around them. A lender taking a security interest in the contract or in the SPV that holds it is not displacing the end customer of record. The compliance fact stays attached to the contract regardless of who is financing it.
This is a deliberate design choice. The point of routing a financing through a Rillor forward is to make the delivery more certain and more compliant, not to engineer a clever path around the channel that an OEM would later have to unwind.
What this does not do
Two boundaries should be stated plainly, because getting them wrong is expensive.
Rillor is not the lender. Rillor operates the forward market, holds the role of neutral platform between buyer, seller, and the independent escrow agent, and also acts as a seller and underwriter that seeds liquidity. It does not extend credit against contracts and does not arrange the operator's debt. The financing relationship is between the operator and its own lenders. What Rillor supplies is the instrument, a standardized, marked, transferable forward, that makes that financing conversation tractable.
The contract is a delivery obligation, not a security. A Rillor forward is a bilateral commercial contract for deferred physical delivery in which both parties intend that physical transfer of the actual hardware will occur. The CFTC describes forward contracts exactly this way: commercial merchandising transactions in physical commodities where delivery actually occurs but is deferred for commercial purposes. They are distinct from futures, which carry a risk-shifting nature where a party need not expect to make or take delivery and frequently does not, and the Commodity Exchange Act excludes cash-commodity sales for deferred delivery from futures regulation. Rillor sits squarely on the forward side of that line, with physical delivery as the intent on every contract and no cash settlement of its own.
None of this is legal or tax advice for your specific structure. How a forward, the security interest you take in it, and the SPV that may hold it are characterized for your jurisdiction and your covenants is a question for your own counsel. Bring them the standard contract early. Because it is standardized, that review is short, and it does not need to be repeated for the next contract in the book.
The shape of a financeable claim
Put the pieces together and the change is concrete. The operator who used to arrive with an intention now arrives with an instrument: a fixed price, a contracted delivery month, 10 percent committed to an independent escrow agent, a seller performance bond behind delivery, an independent index to mark the claim, and a transfer right that gives the lender a real recovery path, all on a contract whose six fields look identical to every other contract in the book.
| Underwriting question | Before, with a purchase intention | With a Rillor forward |
|---|---|---|
| What does it cost | Unpublished spot, approximated from leaks | Fixed contract price, set at execution |
| When does it deliver | Vendor estimate, subject to change | Contracted delivery month |
| Is the buyer committed | Unsecured intent | 10 percent in independent escrow |
| What if it never ships | Litigation and a credit hole | Seller performance bond covers it |
| How do I mark it | A guess | Rillor Compute Index, drawn from active contracts |
| What is my recovery path | A non-assignable supply deal | Pre-delivery transfer to another KYC'd buyer |
| Is the value comparable | Bespoke, one-off | Standardized six-field contract |
That is the work a forward does for a balance sheet. It does not invent demand and it does not remove depreciation risk from the deployed fleet. What it does is turn the future purchase into a present, defined, marked, transferable claim, which is the difference between a buildout a lender has to wave through on faith and one it can underwrite on terms. If you want a price reference on a specific system to bring into that conversation, request a B300 quote and run it through the standard contract flow.
Trade the forward curve on Rillor.
Rillor is invite only. Verified buyers and sellers transact standardized forward contracts on OEM GPU systems, with physical delivery and independent escrow on every contract.
Become a Partner →- CFTC Letter 97-01: Forward Contracts vs. Futures Contracts
- CoreWeave Closes Landmark $8.5 Billion Financing Facility, Achieving First Investment-Grade Rated GPU-backed Financing
- Financing the AI Buildout (Stijn Van Nieuwerburgh, Columbia Business School, March 2026)
- AI Data Center GPU Debt Financing: CRE Impact 2026 (The AI Consulting Network)
- NVIDIA AI GPU Prices Cost Guide (IntuitionLabs)
- GPU Cloud Price Collapse: H100 Market (Introl, Dec 2025)
- Lenovo ThinkSystem NVIDIA HGX B200 180GB 1000W GPU Product Guide
- NVIDIA Partner Network (NPN)