According to Silicon Data, the secondary market value of Nvidia B200 GPUs has reached 158% of their original launch price approximately one year after shipment. This represents a 58% premium over the manufacturer's suggested retail price, defying typical depreciation patterns for consumer electronics.
The report highlights that traditional straight-line depreciation models, which spread costs over three to five years, are no longer applicable to the AI accelerator market. Older models like the A100 and H100 also maintain residual values significantly higher than what standard accounting methods would predict.
Performance metrics from SemiAnalysis indicate that the B200 remains highly efficient for AI tasks. When running DeepSeek-R1, the hardware achieves an inference cost of $0.20 per million tokens at an interaction speed of 76 tokens per user per second.
Rental market data confirms the rising demand for computing power. Hourly rental rates for the B200 increased from slightly under $5 in January to a range of $5.50 to $5.80 by August. In non-hyperscale cloud markets, the hourly rate is currently approximately $5.61.
Despite public calls from industry leaders such as Sam Altman of OpenAI and Dario Amodei of Anthropic to potentially slow down the pace of AI model development, the market data shows that demand for GPU capacity continues to grow steadily.
