A100 rental prices sit near a balance point as traders weigh sustained demand for cost-effective inference and fine-tuning against substitution pressure from newer Hopper and Blackwell accelerators. Current marketplace medians hover around $1.45–$1.79 per GPU-hour on-demand, with spot floors near $0.40, reflecting ample secondary supply from operators upgrading fleets while hyperscalers maintain premiums above $3. Competitive dynamics hinge on workload migration—frontier training shifts to higher-bandwidth H100/H200 and B200/B300 parts, yet A100 retains appeal for 70B-class models and established pipelines where its 80 GB HBM2e delivers strong unit economics. Key near-term catalysts include Blackwell capacity ramps and any seasonal demand spikes that could stabilize or compress rates through November.