**Sustained demand for H200 GPUs in AI training and inference workloads, paired with stable but constrained on-demand supply across providers, underpins trader consensus favoring rental rates at or above $5.50 by month-end.** Current medians hover near $4.50–$5.00 per GPU-hour on platforms like Vast.ai, RunPod, and Hyperstack, with spot listings dipping lower to $1.33–$3 while hyperscalers command $7–$10+; recent 90-day trends show modest upward pressure of 4–20% amid persistent large-model needs for the H200’s 141 GB HBM3e memory and 4.8 TB/s bandwidth. No major capacity surges or Blackwell displacements have materialized in the past week to ease tightness, keeping the $5.00–$5.50 band as a secondary outcome while lower tiers reflect limited probability of near-term softening.