Recent median on-demand H100 rental rates sit near $3.39 per GPU-hour across dozens of providers, with marketplaces like Vast.ai and RunPod offering $2.13–$3.49 while hyperscalers command premiums above $6. Trader sentiment around the $2.50–$3.00 band reflects flat 90-day trends amid steady AI inference demand and new low-end capacity from providers such as Lium at $1.30, which has broadened supply without yet shifting the overall median. Differentiating factors include SXM versus PCIe variants, spot versus reserved contracts, and marketplace liquidity versus enterprise reliability. With only weeks until month-end, any surge in Blackwell deployments or major model training announcements could tilt near-term pricing.