Persistent AI demand for large language model training and inference, combined with tight H100 supply due to HBM memory and silicon constraints, has driven recent rental rate increases of 15-20% at providers like Lambda and Nebius, lifting medians to roughly $3.40–$3.90 per GPU-hour across specialist clouds as of mid-November 2026. Hyperscalers continue charging 50-100% premiums, while spot and reserved options trade lower. Traders see the outcome split because committed capacity and newer Blackwell alternatives could ease pressure before year-end, yet NVIDIA has reiterated shortages extending into 2028 and recent forward indices show firming rather than relief. The closely matched bins around $2.50–$4.00 reflect uncertainty over whether holiday demand or incremental capacity shifts will tip the final average.
基于Polymarket数据的AI实验性摘要。这不是交易建议,也不影响该市场的结算方式。 · 更新于1H6H1D7D30DALL
View resolved

警惕外部链接哦。
警惕外部链接哦。
常见问题