Sustained demand for H200 GPUs from memory-intensive inference and fine-tuning workloads, particularly large language models requiring 141 GB HBM3e capacity and high bandwidth, continues to anchor trader sentiment around the $4–5 range by end-October. Current on-demand rates cluster between $2.60–$4.50 across neocloud and marketplace providers, with hyperscalers commanding premiums above $7–10, reflecting utilization pressure and limited incremental supply as production shifts toward Blackwell. Recent limited H200 shipments to Chinese firms like ByteDance and Tencent offer modest global relief but remain constrained by export rules and domestic competition. Key swing factors include neocloud capacity additions, any acceleration in China approvals, and utilization trends ahead of Nvidia’s earnings.