Persistent high demand for H200 GPUs in AI training and inference, paired with constrained Hopper supply as vendors prioritize Blackwell ramps, keeps November rental prices closely contested around current medians near $4.50–$5.50 per GPU-hour. Marketplace platforms like Vast.ai offer lower spot and on-demand rates through flexible hosting, while established providers command premiums for guaranteed capacity and support. HBM memory shortages and steady large-model workloads support resilience in pricing, though accelerating B200/B300 availability could exert downward pressure if deployment timelines hold. Traders weigh these supply dynamics against potential year-end demand spikes when assessing the narrow spread across outcome brackets.