**H100 rental prices reflect a maturing AI compute market where newer Blackwell GPUs (B200/B300) are displacing H100 from frontier training and high-end inference workloads, while sustained demand for reliable Hopper capacity keeps rates anchored.** Marketplaces and neocloud providers currently list on-demand H100 rates from roughly $1.99–$3.73 per GPU-hour, with spot/interruptible options lower, creating a wide but clustered distribution that aligns with the closely matched Polymarket buckets around $2.50–$3.00. Hyperscalers maintain higher committed rates, but competitive pressure from alternative platforms and expiring long-term contracts is compressing premiums. Key variables through late October include the pace of Blackwell supply additions—which could free additional H100 inventory—and whether inference or multi-agent workloads absorb any freed capacity without pushing utilization (and pricing) higher. Traders appear to weigh these supply shifts against persistent AI demand rather than expecting sharp moves in either direction.