Trader sentiment on H200 rental prices at end-2026 reflects uncertainty in the balance between sustained AI workload demand and expanding supply of newer Blackwell and upcoming Rubin GPUs. Current on-demand rates cluster around $3.80–$4.50 per hour on specialist providers, with hyperscalers charging premiums above $7, amid volatility from HBM3e and CoWoS packaging constraints that have kept availability tight through mid-2026. Recent production shifts away from Hopper, combined with China export dynamics and early signs of price softening on marketplaces, create competing pressures: new HBM capacity expected late in the year could ease shortages and support declines toward the $3.50–$4.50 range, while persistent enterprise inference and training needs may sustain higher levels. Key swing factors include the pace of Blackwell adoption displacing H200 workloads and any last-minute supply-chain improvements.