**Current median on-demand H100 rental prices sit near $3.38 per GPU-hour across dozens of providers, with a wide spread from lows around $1.30 on platforms like Lium to over $10 on hyperscalers.** Recent entry of additional neocloud and marketplace capacity, including community and spot offerings on Vast.ai and RunPod, has kept the floor low and contributed to relatively flat 90-day medians despite year-over-year gains of roughly 9-15%. Strong ongoing demand for large language model training and inference supports higher tiers, yet the proliferation of lower-cost dedicated GPU clouds exerts downward pressure. Trader sentiment across the dispersed bins reflects uncertainty over whether new supply additions or sustained AI workloads will dominate by month-end, with no single outcome exceeding 35% implied probability. Key swing factors include spot market volatility, regional availability shifts, and any last-minute hyperscaler pricing adjustments.