Persistent AI demand for large language model inference and training—leveraging the H200’s 141 GB HBM3e memory and high bandwidth—has driven recent rental price gains, with medians rising 8-22% over the past month to roughly $4.5–$5.4 per GPU-hour across neoclouds. Trader uncertainty stems from competing forces: ongoing supply constraints from power shortages and HBM allocation versus accelerating Blackwell B200/B300 deployments that could ease H200 utilization. Recent provider hikes, such as Nebius’s 20% increase effective October 1, support the higher bands, while spot availability and potential export policy shifts introduce downside risk. Key near-term catalysts include hyperscaler pricing adjustments, frontier model training surges, and Blackwell capacity reports through November that could shift the Ornn-style index across the evenly distributed outcome ranges.