**Persistent demand for large language model training and inference kept typical on-demand H100 rental rates above $3 per GPU-hour through October 2026.** Multiple trackers reported monthly averages around $3.30 and medians near $3.49 across specialist clouds and marketplaces, with several providers implementing modest increases amid steady AI workloads. Hyperscalers continued quoting substantially higher list prices, reinforcing the floor set by dedicated GPU clouds. While spot and marketplace options dipped lower, the volume-weighted typical rate stayed elevated due to capacity constraints and ongoing model development cycles. A sudden surge in Blackwell supply or sharp pullback in training demand could pressure prices downward, though no such shift materialized by month-end.