**Sustained AI training and inference demand has driven H100 rental prices higher through 2026 despite the chip's 2022 launch.** Median on-demand rates currently sit near $3.5–4.1 per GPU-hour across providers, with spot/marketplace floors around $1.0–2.0 and hyperscaler list prices reaching $6.9–12.3. Recent provider increases, including Nebius raising rates to $4.50 effective October 2026, reflect tight high-bandwidth memory supply, power limits, and long-term contracts locking up much of the newer Blackwell capacity. NVIDIA has signaled supply constraints persisting into 2028, while rising B200/B300 rates keep older Hopper GPUs competitive for many workloads. Key near-term catalysts include the CFTC's November 9 decision on H100 futures trading and ongoing Blackwell volume ramps that could ease or redirect pressure on H100 pricing.