Collateral Valuation
Book depreciation does not move with the market
60–70%
H100 rental rate decline, early 2024 → late 2025
Current facilities mark GPU collateral at straight-line book value over 3–5 years. H100 spot rental rates fell by more than 60% in 18 months while book depreciation showed the collateral largely intact. A lender relying on depreciation schedules had no covenant signal until the market had already moved past them.
Reference Rate
Residual value priced on judgment, not a market anchor
No standard
No independent GPU compute reference rate in current deal documentation
Without an independent market reference rate, lenders price residual value uncertainty on internal judgment — estimates that no counterparty can independently verify and that every deal prices differently. CCIR provides the market anchor current documentation lacks.
Workload Migration
The collateral’s revenue profile can change character with no lender signal
Training workloads are contracted, sticky, and predictable. Inference workloads are spot-adjacent, elastic, and reprice faster when the market moves. A standard revenue covenant cannot distinguish between the two — equivalent utilization rates look identical regardless of what the hardware is actually doing. Without workload visibility, the lender has no way to gauge whether reported revenue is durable, what rolloff risk looks like when contracts expire, or whether reported revenue is defensible against what the spot market is currently bearing for comparable hardware.
Generation Convergence
A 36-month facility spans at least one full GPU generation transition
When B200 spot prices converge with H100, the training premium justifying H100 collateral valuations is effectively arbitraged away. This is not a tail scenario — it is a foreseeable transition that a 36-month lender is almost certain to face. No current deal structure has a mechanism that responds to convergence as it is happening. By the time a standard covenant fires, the collateral has already repriced.
Covenant Structure
§ 7.01
Maintenance Test
Borrower’s contracted GPU revenue must cover the on-demand CRI-H100 rate by a negotiated multiple set at underwriting. Rate Cap protects a performing borrower from technical breach when spot prices rise above their contracted rate.
CRI-H100 cited by name
§ 7.02
Collateral Coverage / LTV
Collateral Value = GPUs × CRI-H100 × 8,760 hrs × 72% utilization ÷ 28% cap rate. Maximum LTV 80%. Marked to the live on-demand noninterruptible rate daily via CRI-H100 — the appropriate anchor for redeployment scenarios, where the lender requires structured placement, not interruptible spot listings.
Live mark vs. book value
§ 7.03
Workload Composition
Training Workload Percentage ≥ 50% on two consecutive Calculation Dates. On breach, Collateral Value switches to Blended Utilization Rate weighting training (72%) and inference (65%) tiers.
Novel — not in current market
§ 7.04
Generation Spread
If CRI-B200 converges within 15% of CRI-H100 for two consecutive periods: accelerated amortization at 10%/period and Cap Rate steps up from 28% to 35%, reflecting repricing to inference-tier economics.
Novel — not in current market
§7.03 gives lenders visibility to the borrower · §7.04 gives lenders visibility to the market
Scenario Analysis — How Covenants Respond
| Scenario |
CRI-H100
on-demand |
Maintenance |
Collateral LTV |
Workload |
| Baseline |
$3.20 |
PASS — contracted |
PASS 42% |
PASS 80% |
| Spot decline |
$2.00 |
PASS — locked |
WATCH 67% |
WATCH 65% |
| Spot rise |
$4.50 |
CAP ENGAGED |
PASS 29% |
PASS |
| Training collapse |
$1.40 |
PASS — locked |
BREACH 95% |
BREACH 35% |
| Blackwell cascade |
$1.20 |
PASS — contracts |
BREACH 111% |
BREACH |
No single scenario causes all three covenants to breach simultaneously — the framework is designed to give graduated signals, not a binary default trigger. The Rate Cap matters most in a rising market: without it, a borrower earning a contracted rate below the on-demand market rate would show below-threshold coverage despite performing exactly on contract. Full scenario analysis, collateral value formula derivations, and all drafted covenant language in the downloadable framework above.
Citing CRI rates in a GPU-backed facility?
CCIR provides the independent rate anchor that GPU-backed credit agreements cite by name — the rate definition, the methodology documentation, and the Calculation Date packages that compliance monitoring requires. Counsel sets the thresholds; CCIR provides the market data they are set against.
Rate definition support
Explaining what each CRI index measures, how the methodology governs its computation, and which index applies to a given collateral type. Counsel and lenders determine threshold levels.
Index application guidance
Reviewing GPU mix and collateral type to identify which CRI indices apply and how each should be cited in credit agreement definitions.
Index citation guidance
Providing precise CRI rate definition language, methodology version references, and fallback definition language for use in credit agreement drafting. Counsel is responsible for all operative agreement language.
Ongoing rate monitoring
Calculation Date rate packages showing CRI-H100 and Generation Spread in a format designed for compliance certificate review.
CCIR holds no financial interest in the GPU compute market it measures. Written responses to methodology inquiries do not affect index methodology or published rates. This page does not constitute legal or financial advice.