Sustained demand for AI training and inference workloads has anchored H100 rental rates near recent highs, with on-demand medians holding between $3.39 and $3.86 per GPU-hour across dozens of providers as of late October 2026. Specialist clouds list single-GPU on-demand rates from roughly $1.80 to $4.50 while hyperscalers command premiums up to $11+, reflecting differences in managed services, availability, and interconnect options like NVLink. Year-over-year gains of 13–28% and recent weekly upticks of 9% underscore supply constraints for the Hopper-generation GPU even as newer Blackwell options emerge. With only days left in the month and no major capacity releases reported, trader consensus heavily favors outcomes at or above $3.00.