Strong demand for Blackwell-architecture GPUs from large-scale AI training and inference workloads continues to support elevated B200 rental rates, with current on-demand medians around $7–8.60 per GPU-hour across providers like CoreWeave, Lambda, and RunPod. Spot and marketplace listings dip lower to $4–6, yet trader consensus favors sustained pricing above $9 by late November due to high utilization, sticky enterprise commitments, and gradual supply ramp-up. Emerging B300 availability adds competitive pressure that could compress rates, though limited near-term capacity and persistent frontier-model demand limit downside risk in the near term.