Current on-demand H100 rental rates show wide dispersion, with spot and community options as low as $0.94–$1.99 per GPU-hour and provider medians near $3.3–$3.4, as of early October 2026. Trader sentiment for end-of-November pricing hinges primarily on the pace of Blackwell B200 and GB200 deployments, which offer higher performance per rack and are displacing equivalent H100 workloads in training and inference clusters. Fixed H100 supply—after NVIDIA redirected wafers—and ongoing HBM memory constraints limit downward pressure, while neocloud capacity additions and hyperscaler optimizations could increase available spot inventory. Persistent demand from large language model serving and fine-tuning supports the higher outcome buckets, but any accelerated Blackwell ramp or seasonal slowdown could push averages below $2.25. The fragmented odds capture these competing supply and demand signals over the next eight weeks.