Sustained AI inference and fine-tuning demand for cost-effective large language models continues to anchor A100 rental rates near current marketplace and neo-cloud medians of roughly $1.00–$1.50 per GPU-hour. Persistent Blackwell supply constraints from high-bandwidth memory shortages keep newer B200/B300 capacity scarce and expensive, spilling workloads onto Ampere hardware that remains widely available and efficient for mid-sized open-weight models. Recent indices show A100 on-demand prices stable or modestly higher over the past 30–90 days, reversing typical generational depreciation as token consumption rises faster than new capacity arrives. With no major A100 supply additions or regulatory shifts expected before year-end, trader consensus clusters on the $1.00–$1.25 band as the most probable outcome for end-of-December levels.