Sustained demand for AI training and inference workloads continues to support A100 rental rates near the $0.75–$1.25 per GPU-hour range at end of November. Persistent token consumption growth has kept older Ampere GPUs relevant for fine-tuning and inference even as newer Hopper and Blackwell chips face supply constraints and higher pricing. Current on-demand medians across providers range from roughly $1.40–$2.25 for 80GB variants, with spot and marketplace listings frequently available below $1.00, while hyperscaler rates remain elevated. Tight HBM memory supply and limited new A100 capacity additions reinforce trader consensus that prices will hold in the leading buckets rather than drop sharply or spike beyond $1.50.