Recent supply expansions from neo-cloud providers and datacenter builds have stabilized H100 rental rates near $2.70 per hour in the competitive segment as of mid-August 2026, creating a tight contest between the $2.50–$2.75 and $3.00+ bins. Sustained demand for large language model training and inference workloads continues to support pricing power, while the accelerating rollout of Blackwell-series GPUs introduces downside risk by shifting some workloads and increasing overall capacity. Hyperscaler rates remain significantly higher, highlighting segmentation in the market, and near-term catalysts such as quarterly capacity reports or major AI infrastructure announcements could tip sentiment either way before October resolution. Traders are pricing in modest volatility around current equilibrium levels amid these competing supply and demand forces.