Strong ongoing AI inference demand, alongside sustained training workloads, continues to support H100 rental rates near current neo-cloud levels of roughly $2.50–$3.00 per hour despite Blackwell B200 ramp-up and expanded provider capacity. Recent price rebounds of about 40% from late-2025 lows reflect hyperscaler on-demand premiums exceeding $7 and spot-market spreads from $1.65 to $3.50, driven by HBM constraints and sold-out reservations through year-end. Traders weigh the widening gap between specialized clouds and legacy providers against gradual Blackwell displacement, which may moderate H100 utilization without triggering sharp declines by December 2026. Upcoming capacity additions and any further inference scaling represent the main swing factors in this closely contested range.