Traders see sustained AI demand for large language model training and inference as the main driver pushing H200 rental rates toward the $5.50+ band by month-end. Current on-demand medians sit near $4.20–$4.72 per GPU-hour across providers, with spot listings as low as $2 and peaks above $7, reflecting tight Hopper supply and high utilization in cloud marketplaces. Recent price tracking shows modest 90-day stability or slight gains amid ongoing capacity constraints, while spot volatility and reserved-term discounts highlight how utilization spikes can quickly lift effective rates. With only two weeks left, any surge in enterprise workloads or reduced spot availability could reinforce the market-implied odds favoring the higher tier.