Recent provider price hikes and sustained AI training demand have anchored H100 rental rates near current medians of $3.49–$3.90 per GPU-hour across specialist clouds, with hyperscalers commanding premiums above $6. Strong large-language model workloads and limited immediate displacement by higher-memory H200 or Blackwell B200 chips support trader consensus around the $3.00–$4.00 bands. Marketplace spot offers remain softer at $1.70–$2.70, but on-demand and reserved commitments reflect tighter availability. New neocloud capacity expansions and any accelerated Blackwell volume in December represent the main swing factors that could compress or stabilize rates before year-end resolution.
基於Polymarket數據的AI實驗性摘要。這不是交易建議,也不影響該市場的結算方式。 · 更新於1H6H1D7D30DALL
View resolved

警惕外部連結哦。
警惕外部連結哦。
Frequently Asked Questions