The Great Depreciation Myth
For years, the AI industry has debated the 'useful life' of high-end hardware. Detractors argued that a three-year shelf life was the most realistic estimate for expensive NVIDIA GPUs, while proponents suggested these assets would remain relevant much longer. As of mid-2026, the market is siding with the proponents. Instead of falling, rental rates for flagship NVIDIA A100 and H100 GPUs are trending upward, defying conventional depreciation models.
Why Rental Rates Are Climbing
The sustained strength of older hardware isn't just a byproduct of high demand; it is driven by a complex interplay of substitution and infrastructure scarcity. As the cost of newer, cutting-edge hardware like the B200 continues to hover near the $6/hour mark, many companies are finding that older architectures remain the more economically viable choice for specific workloads.
- Substitution Demand: As H100 supply becomes less elastic, buyers are sliding down the performance curve, keeping A100 prices propped up.
- Performance per Dollar: While H100s offer superior power for specific tasks, the A100 remains the 'sweet spot' for fine-tuning and inference.
- Sovereign AI Spending: Global investment nearing $100 billion in 2026 is creating a supply crunch that prevents prices from cooling down.
The Warning Signs: Looking Beyond the Index
While current rental rates paint a picture of stability, analysts warn against complacency. Data suggests that once H100 units hit a certain age, depreciation may accelerate sharply, particularly in the resale market. Operators who treat these GPUs as assets with permanent value may find themselves facing significant write-downs once newer architectures, such as the upcoming Rubin generation, hit the market in late 2026 and 2027.
Wait too long to offload H100s could mean recovering just a fraction of their former worth.
— Silicon Data Analysis