The Nvidia H100 GPU, launched in Q4 2022, has seen its market value rise by approximately 30% compared to the previous year. This trend follows a period of declining rental prices throughout 2025, driven by sustained demand for AI training and inference workloads.
As of mid-October 2026, median rental rates for the H100 range between $3.3 and $3.6 per GPU-hour. On October 16, 2026, the median on-demand rate was recorded at $3.49 per GPU-hour. Market indices show varying growth, with some reports indicating increases of 13% to 79% from 2025 lows.
Pricing structures differ significantly across providers. Hyperscale cloud services charge premiums 1.7x to 2.7x higher than specialized cloud providers. Meanwhile, on the secondary market, H100 units in new or used condition are trading between $25,000 and $40,000, depending on their specific configuration.
The H100 chip is based on the Hopper architecture, features 80GB of HBM3 memory, and offers up to 989 TFLOPS of FP16 tensor performance with a 700W TDP. Despite the availability of newer hardware, supply chain constraints have kept the H100 relevant in the current market.
