Recent H200 rental rates hover near $3.10–$4.62 per GPU-hour across neocloud and hyperscaler providers, reflecting tight HBM3e and CoWoS packaging supply that favors memory-intensive inference workloads over raw training. Trader consensus around the $4.50–$5.50 range at end-October stems from durable demand for the 141 GB H200’s KV-cache advantages amid long-context model growth, tempered by Blackwell (B200) capacity ramp and incremental neocloud deployments that widen availability. Export controls and China shipment caps have had limited near-term global impact, while hyperscaler premiums and spot discounts create the observed spread. Key swing factors include any October cluster activations or frontier-model releases that could shift utilization before month-end resolution.