**Current A100 rental rates cluster near $0.9–$1.8 per GPU-hour across marketplaces and cloud providers, with transaction-based indices like the OCPI-A100 settling at $0.97 as of September 20, driving the tight split between the $0.75–$1.00 and $1.00–$1.25 bins.** Marketplaces such as Vast.ai and RunPod supply lower on-demand and spot capacity (often $0.27–$1.00), while hyperscalers and established neo-clouds command premiums, creating variance that keeps outcomes closely matched. Downward pressure stems from newer accelerators (H100, H200, B200) absorbing frontier training workloads, releasing A100 capacity for inference and fine-tuning use cases that sustain a durable floor. Recent data show modest declines—roughly 5% week-over-week and 9% over 30 days—reflecting steady supply growth without major demand shocks. With resolution just days away, traders weigh whether spot surges or sustained listings will push the effective rate into the upper or lower leading range before month-end.