**Expanding supply from specialized AI clouds and GPU marketplaces is the main force shaping H200 rental odds, counterbalanced by steady demand for its 141 GB HBM3e memory in large-model inference and training.** Current on-demand medians sit near $4.40–$4.47 per GPU-hour across trackers, with verified lows at $2.60 (and spot/community options below $2.00 on platforms like Vast.ai and Prime Intellect) while hyperscalers remain far higher. Multiple new providers have added capacity in recent weeks, widening availability and capping upside, yet AI workload growth continues to support the $4.50–$5.50 range. With roughly eight weeks until late November, traders are weighing further supply inflows against potential seasonal demand spikes; the closely split probabilities reflect uncertainty over whether competitive pressure or utilization will dominate the final outcome.