Sustained AI training and inference workloads, combined with constrained Blackwell supply through long-term contracts, continue driving H100 rental rates upward and position the $3.25+ outcome as the leading market-implied probability. Recent October 2026 data show median on-demand prices at $3.30–$3.80 per GPU-hour across providers, with indices rising 8–10% month-over-month amid high-bandwidth memory shortages and data-center power limits. Traders see limited near-term relief from newer chips, as NVIDIA supply bottlenecks extend into 2028 and elevated memory costs raise replacement economics. Spot and reserved tiers remain differentiated by provider type, but the overall upward trajectory through November aligns with the current 54% consensus on prices holding or climbing further.