**Strong trader consensus on the $4.00–$4.50 per GPU-hour band for H200 rentals at end of August 2026 stems from observed on-demand rates across neocloud and marketplace providers.** Late-August snapshots showed competitive listings clustered tightly in this range, including Akash Network at $4.45, Nebius at $4.50, RunPod near $4.31–$4.59, and Jarvislabs or Vast.ai floors around $3.82–$3.99. H200’s 141 GB HBM3e memory and 4.8 TB/s bandwidth sustained demand for memory-intensive inference and long-context workloads, while expanding neocloud capacity and platform competition capped upside versus hyperscaler premiums above $7–$10. Realistic challenges include accelerated Blackwell (B200) substitution easing Hopper demand, sudden large-scale capacity additions, or sharp shifts in spot/interruptible pricing that could pull the effective on-demand median below $4.00.