Current on-demand H100 rental medians sit near $3.30–$3.42 per GPU-hour across tracked providers, with spot and community tiers as low as $0.94–$1.60, reflecting abundant decentralized capacity from platforms like Vast.ai, Lium, and RunPod. This positioning underpins the leading $3+ outcome, as hyperscaler list rates remain far higher while competitive neocloud floors have stayed flat over the past 90 days despite modest year-over-year gains. Differentiating factors include SXM versus PCIe variants, interruptible availability, and regional stock levels, with new low-price listings from Lium recently compressing the bottom end. End-of-October resolution hinges on any late demand spikes from AI workloads or additional supply inflows that could shift the observed median before month close.