**Abundant secondary-market supply and steady migration toward newer Hopper and Blackwell GPUs are anchoring A100 rental rates near $1 per hour through late October.** Live benchmarks as of early September show on-demand listings clustered between roughly $0.89 and $2.00 across providers such as Jarvislabs, RunPod, and Vast.ai, with spot capacity often dipping below $0.70, producing a median near the $1.00–$1.25 band that traders currently price at 50% implied probability. Demand for the Ampere-generation cards remains concentrated in cost-sensitive inference, fine-tuning, and mid-scale workloads that do not require the higher throughput or memory bandwidth of H100/H200 or B200 systems, preventing sharp upward pressure. No major capacity releases, regulatory shifts, or AI training surges have materialized in recent weeks to disrupt this equilibrium, while historical patterns indicate modest seasonal stability rather than volatility ahead of year-end. The next 7–8 weeks therefore favor continuation of the current $0.75–$1.25 range unless a sudden hyperscaler fleet expansion or unexpected large-model training wave emerges.