Hourly rates for H100 GPUs on the Token Factory platform will rise from $4.05 to $4.70 effective Oct. 1. This price increase for Nebius's dedicated inference endpoints affects multiple Nvidia chips, with B300 rates climbing 20% to $9.70 per hour and B200 rates moving to $8.70. The move adds approximately $1,152 in monthly revenue per fully utilized B300, or $13.8 million annually across 1,000 GPUs, and follows a previous 17% to 21% hike in standard on-demand cloud compute costs.

Nebius is also introducing dynamic spot pricing for preemptible virtual machines on Oct. 8, replacing fixed discounts with rates based on real-time demand, available capacity, and region. Users can set a maximum price limit or follow market fluctuations, though the company may stop VMs if the spot price exceeds a customer's cap. Existing reserved contracts and negotiated pricing remain unchanged, meaning the immediate revenue benefit will primarily come from new and pay-as-you-go customers.