**H100 rental prices reflect a tight balance between persistent AI inference demand and ramping Blackwell supply.** As of mid-2026, on-demand and contract rates for reliable H100 capacity cluster around $2–$3.50 per GPU-hour on specialized providers, with hyperscalers commanding premiums up to $7+, while spot and community tiers dip lower. Recent data show a partial rebound from 2025 lows, driven by surging token consumption from open-weight models, multi-agent workflows, and media-generation workloads that keep utilization high even as NVIDIA ships millions of Hopper units. Blackwell (B200/GB200) deployments, accelerating through 2026, are the key swing factor: they offer substantially better performance per dollar for many tasks, positioning H100 as a value-tier option and exerting 10–20% downward pressure on rentals as capacity comes online. At the same time, HBM constraints, long lead times for new clusters, and inelastic demand from hyperscalers and AI labs have supported prices against earlier oversupply expectations. Regional differences, commitment length, and spot availability add further variance. With roughly 2.5 months until end-October resolution, trader sentiment remains split because any acceleration in Blackwell availability or softening inference growth could push averages below $2.50, while sustained demand or supply hiccups could hold them near or above $3. The closely matched probabilities capture this uncertainty around the pace of generational transition versus ongoing AI workload expansion.