**Sustained AI training and inference workloads continue to anchor H100 on-demand rental medians near $3.39–$3.77 per GPU-hour across specialist providers as of late October 2026.** Recent list-price adjustments at platforms such as Nebius and RunPod, combined with limited new H100 supply amid the shift toward Blackwell GPUs, have supported these levels despite softer spot and marketplace rates often falling in the $1.80–$2.50 range. Hyperscalers maintain premiums above $6–$12, widening the spread but contributing less weight to typical resolution indices. With only days remaining until month-end and no major supply shocks or demand shifts reported in the past week, trader consensus favors the $3.00+ band while acknowledging potential softening in lower-tier or interruptible segments.