GPU-Backed Loans
The definitive guide for private credit underwriters evaluating GPU-secured facilities. Covers collateral valuation, depreciation modeling, and covenant structuring.
How do GPU-backed loans work?
GPU-backed loans are equipment-secured credit facilities where NVIDIA H100, A100, or similar AI accelerators serve as collateral. Lenders typically advance 50-70% LTV against appraised GPU value, with DSCR covenants of 1.25-1.5x. Key risks include rapid depreciation (40-60% in Year 1) and collateral redeployability. The market emerged in 2023-2024 as AI compute demand outpaced traditional financing.
Key Data Points
- Typical LTV: 50-70% of appraised value
- Year 1 Depreciation: 40-60%
- Standard DSCR Covenant: 1.25-1.5x
- Key Risk: Rapid technological obsolescence
What is a GPU-Backed Loan?
A GPU-backed loan is an equipment-secured credit facility where AI accelerators (NVIDIA H100, A100, H200, or AMD MI300X) serve as the primary collateral. These loans emerged in 2023-2024 as AI compute demand outpaced traditional equipment financing capacity.
Unlike traditional server financing, GPU loans present unique underwriting challenges: rapid technological obsolescence, concentrated asset pools, and volatile secondary markets. However, the strong demand for AI compute creates attractive lending opportunities for sophisticated credit funds.
Collateral Valuation Approaches
Lenders typically employ one of three valuation methodologies:
Cost Approach
Original acquisition cost less accumulated depreciation. Conservative but may undervalue GPUs with strong secondary demand.
Market Approach
Comparable transaction analysis using secondary market data. Our GLRI index tracks real-time market pricing.
Income Approach
DCF based on projected lease revenue. Preferred for operational GPU clusters with contracted revenue.
Depreciation Modeling
GPU depreciation is non-linear and generation-dependent. Key factors:
- H100 (Hopper): 40-50% Year 1, stabilizing at 70% cumulative by Year 3
- A100 (Ampere): 50-60% Year 1, steeper decline as Hopper dominates
- B200 (Blackwell): Premium pricing expected to hold longer due to supply constraints
The “Blackwell cliff” risk: H100 residuals may drop sharply when Blackwell GPUs reach volume production in late 2025. Underwriters should stress-test portfolios against this scenario.
Covenant Structuring
Standard GPU-backed loan covenants include:
| Covenant | Typical Threshold | Purpose |
|---|---|---|
| DSCR | 1.25-1.5x | Ensure cash flow coverage of debt service |
| LTV Cap | 70% initial, 80% cure | Maintain equity cushion as GPUs depreciate |
| Utilization Floor | 75-85% | Ensure revenue-generating deployment |
| Concentration Limit | 25% single customer | Diversify counterparty risk |
Step-In Rights & Remedies
In default scenarios, lenders need clear step-in rights to:
- Access and relocate physical GPU hardware
- Assume or terminate hosting/colocation agreements
- Novate customer contracts to alternative operators
- Liquidate via secondary market brokers or OEM buyback programs
Key Risk: Colocation Dependencies
GPUs are often deployed in third-party colocation facilities. Ensure your security agreement includes rights to access the facility, and confirm the colo provider will recognize lender step-in rights under their standard agreement.
The 'GPU Lending Playbook'
- Model term sheet for GPU-backed facilities
- UCC-1 filing best practices for AI hardware
- Secondary market liquidation strategies
Need format proof first? View sample output
Explore More
Underwriting Tools for GPU Loans
GPU Residual/LTV Calculator
Calculate GPU depreciation and residual values
Open Speed-to-Power WatchlistGPU Residual/LTV Calculator
Calculate GPU depreciation using physics-based decay models
Open Speed-to-Power Watchlist