Sustained demand for AI training and inference continues to support elevated H100 rental rates, with current medians across specialist clouds and marketplaces holding near $3.3–$3.9 per GPU-hour as of mid-November 2026. Recent provider actions, including Nebius’s October price increase to $4.50 and CoreWeave’s earlier rate hikes, have driven a 6–12% rebound over the past quarter and roughly 30% year-over-year gains from 2025 lows, reflecting tight capacity and limited near-term supply relief. Traders appear to price in continued firmness through year-end, with the $3.25+ outcome leading at 57% implied probability, though potential Blackwell ramp effects or demand moderation could still introduce downside risk before December 31.