CCIR — The Reference Standard for GPU Collateral

Independent reference rates for GPU-backed credit — built for citation in loan covenants, collateral appraisals, and compliance certificates. The structural independence and IOSCO-aligned methodology credit documents require.

CRI-H100 — / GPU · hr

On-demand spot · loading… See all rates ↓

Home / Covenant Structuring / Reference Rates / Changelog / Contact

Novel Covenants

Rate-Anchored Covenants That Move With the Market

Two covenants that don’t exist in current GPU-backed deal documentation: a Workload Composition trigger that detects training-to-inference migration before it becomes a coverage breach, and a Generation Spread covenant that responds to Blackwell convergence as it happens — not after the collateral has repriced. See the full framework →

Reference Rates

H100 Today. Your Collateral Tomorrow.

CRI-H100 and the full CCIR index family give covenants a named, independent market anchor — one that neither party controls and any counterparty can verify. Collateral marked to the live on-demand rate daily, not book depreciation. The structural independence credit documents require.

Bespoke Structuring

Calibrated to the Specific Deal at Underwriting

CRI indices give counsel and lenders the named, independent market anchor covenants require. CCIR provides the rate definition language, methodology documentation, and Calculation Date rate packages that put independently verifiable numbers behind each parameter. Threshold levels are determined by counsel and lenders. Engage CCIR →

Where Current Documentation Falls Short

Three gaps that leave lenders exposed on active GPU-backed facilities.

01

Collateral marked to the past, not the market

Current facilities mark GPU collateral using straight-line book depreciation over 3–5 years. H100 rental rates fell more than 60% in 18 months while book value remained largely intact. A lender relying on depreciation schedules had no covenant signal until the market had already moved past them.

02

No trigger for the transition already underway

H100 collateral was underwritten as training hardware — and training contracts carry materially less revenue volatility than inference. As Blackwell displaces Hopper for large-model training, a borrower’s revenue mix can shift from contracted and sticky to spot-adjacent and elastic. A lender monitoring utilization figures has no line of sight to that migration until it shows up as a coverage problem.

03

GPU collateral moves faster than the documentation was built for

A 36-month facility spans at least one full GPU generation transition — a foreseeable event, not a tail risk. When B200 spot prices converge with H100, the training premium justifying H100 collateral valuations is effectively arbitraged away. No current deal structure has a mechanism to respond to it as it happens.

The only covenant framework built to respond to all three — with full drafted language available for download.

Full drafted covenant language, scenario analysis, and a hypothetical $50M facility framework available for download.

See the Covenant Framework →

The Reference Rates the Covenants Cite

01

Maintenance Test

Structure revenue covenants around an independent market rate, not the borrower’s projections. A Rate Cap protects a performing borrower with locked contracts from technical breach when the spot market rises — while preserving the lender’s protection when it matters.

02

Collateral Coverage / LTV

Collateral value marked to CRI-H100 daily — not book depreciation. The live market rate moves with the redeployment value of the hardware. When the market declines, the collateral marks down in real time and the lender has a covenant signal before the damage is done.

03

Generation Spread

The Generation Spread — CRI-B200 versus CRI-H100 — measures the training premium in real time. When B200 supply expands and its price falls toward H100 levels, that premium erodes. CCIR publishes both rates daily. The §7.04 Generation Spread covenant responds to convergence as it happens — not after the collateral has repriced.

Which index for which deal H100 SXM → CRI-H100 · H200 → CRI-H200 · B200 → CRI-B200 · A100 SXM → CRI-A100 · Mixed-GPU facility → each tranche references its own index

Loading…

Generation Spread

Generation Spread — Live Signal

The Generation Spread measures whether next-generation hardware is converging on H100 pricing in real time. When CRI-B200 falls toward CRI-H100, the training premium that justifies H100 collateral valuations erodes — and the §7.04 covenant responds to that signal before the collateral has repriced. The H100/A100 spread shows the longer-term commoditization trajectory: where H100 is headed as inference demand absorbs prior-generation capacity.

CRI-B200 vs CRI-H100 · §7.04 Trigger

—

Loading…

§7.04 Covenant Anchor · Threshold: 15%

CRI-H100 vs CRI-A100 · Commoditization Trajectory

—

Loading…

Commoditization Signal · 18–24 mo horizon

IOSCO IOSCO Principles for Financial Benchmarks

CC BY 4.0 Open data — freely citable

Daily Published nightly to ccir-index/ccir-data

v1.0.0 Open methodology — reproducible pipeline