GPU Residual Value Guide
How fast do NVIDIA H100 GPUs depreciate?
AI GPUs follow a 'performance-linked' depreciation curve. Unlike general servers, their value is tied to the release cycle of next-gen silicon (e.g., Blackwell). An H100 typically retains 60-70% value after 12 months, dropping to 25-35% after 36 months. Residual value is highly sensitive to total available HBM3e supply.
Key Data Points
- 12-Month Residual: 65% of MSRP
- 24-Month Residual: 45% of MSRP
- 36-Month Residual: 30% of MSRP
- Primary Decay Driver: Performance density (TFLOPS/Watt)
- Secondary Decay Driver: Memory capacity (GB of HBM)
Understanding the Decay Curve
Physics-Based vs. Market-Based
Lenders typically use a linear 5-year depreciation. However, the secondary market for GPUs is non-linear. Values hold steady until the next architecture becomes widely available, then experience a "step-down" adjustment of 20-30%.
- • Inflection Points: Release of B200, MI400X, etc.
- • Floor Price: Set by inference demand for smaller models.
- • Utilization Impact: 24/7 training slightly accelerates thermal-induced decay.
Frequently Asked Questions
How does GPU memory affect resale value?
Higher VRAM (e.g., H200 with 141GB vs H100 with 80GB) maintains significantly higher residual value because it can run larger parameter models that smaller chips cannot.
What is a typical "Loan-to-Value" (LTV) for GPUs?
Lenders usually advance 60-70% of the hardware purchase price, ensuring that the remaining loan balance is always lower than the projected residual value of the chips.
Do AMD and NVIDIA chips depreciate differently?
Historically, NVIDIA chips have held value better due to the CUDA software moat. AMD MI300X residual values are more sensitive to the specific adoption of ROCm by major hyperscalers.
Explore More
Related Tools
GLRI (GPU Lease Rate Index)
Track H100/A100/B200 lease rate trends - core market data
Open Tool