**Current H100 GPU rental prices, tracked via neo-cloud and marketplace indices, sit near $2.50–$2.72 per hour as of mid-August 2026, with on-demand rates commonly ranging $2.19–$3.50 across providers like Thunder Compute, RunPod, Vast.ai, and Lambda Labs.** Hyperscaler list prices remain higher ($3.50–$7+), but competitive spot and reserved offerings pull the effective market toward the $2.50–$2.75 band. This positioning explains the 47% market-implied probability for that range at end-October, reflecting trader consensus on near-term stability. Sustained AI training and inference demand continues to support pricing, while expanded capacity from neo-cloud operators, shorter lead times (now 6–12 weeks versus prior peaks), and competition have prevented sharp rebounds seen earlier in 2026. Blackwell (B200) ramp-up introduces some substitution risk but has not yet materially displaced H100 workloads. Minor seasonal demand fluctuations or new capacity additions could nudge outcomes into adjacent buckets ($2.25–$2.50 or $2.75–$3.00), consistent with their combined ~40% probability. The market aggregates real-capital bets on these supply-demand dynamics rather than forecasting exact future prices. Resolution will likely reference standardized indices such as Silicon Data’s neo-cloud H100 rental metric.