**Trader sentiment for H200 rental prices at end-2026 reflects uncertainty around Blackwell ramp-up versus sustained AI demand.** Current on-demand rates on specialist clouds sit near $3.80–$4.50 per GPU-hour, with hyperscalers charging premiums above $5–$10. The closely matched probabilities across $3.50–$6.00 ranges stem from competing forces: expanding B200/GB200 availability is displacing Hopper workloads and easing H200 scarcity, while large-scale training and inference demand for memory-bandwidth-heavy models continues to support pricing. Key swing factors include Blackwell production yields and CoWoS packaging capacity through late 2026, hyperscaler capacity expansions, spot versus reserved contract dynamics, and any shifts in energy or data-center constraints. Market-implied odds embed the view that further meaningful declines below $3.50 remain unlikely without a sharper demand slowdown.