Traders see closely matched probabilities across H100 rental price bands near $2.50–$3.25 because current on-demand medians hover around $3.40–$3.50 per GPU-hour with limited near-term catalysts. Steady AI training and inference demand from large language models continues to support utilization, yet expanding capacity from specialist clouds and marketplaces has kept prices relatively flat over the past 90 days. Spot and reserved tiers already trade lower, creating downward pressure, while hyperscalers hold premiums. Key swing factors over the next six weeks include any Blackwell rollout updates that could ease Hopper shortages, new model launches that spike short-term demand, or seasonal shifts in cloud utilization. The market-implied odds capture this balanced uncertainty in supply growth versus persistent AI compute needs.