**Current A100 rental rates cluster near $0.89–$1.90 per GPU-hour on-demand across providers like Jarvislabs, Vast.ai, RunPod, and hyperscalers, with spot floors dipping to $0.44–$0.68.** Trader consensus splitting evenly between the $0.75–$1.00 and $1.00–$1.25 bins reflects steady but moderating demand for the Ampere-generation cards in cost-sensitive inference and fine-tuning workloads, offset by rising availability of newer Hopper (H100/H200) and Blackwell (B200) GPUs that deliver superior throughput on large language models. Recent indices show A100 medians stable to slightly down over the past 30–90 days as secondary-market supply grows and enterprise buyers shift premium training to higher-performance silicon. Key swing factors through late October include Blackwell production ramp rates, any spot-price compression from hyperscaler reallocations, and sustained AI inference demand that could anchor older cards near current levels without triggering a sharp drop.