What is GMI Cloud
GMI Cloud provides NVIDIA GPU infrastructure for training and inference through Cluster Engine and a serverless Inference Engine.
GPU pricing
As of August 5, 2026, published starting prices are US$2 per GPU hour for H100, US$2.60 for H200, US$4 for B200 and US$8 for GB200. GB300 capacity is available for preorder without a public numeric rate.
On-demand, reserved and serverless options
Cluster Engine supports on-demand and reserved dedicated NVIDIA infrastructure, with savings available for longer commitments. Buyers can begin on demand and move predictable workloads to committed capacity. Inference Engine is a separately metered serverless service that charges by request, scales to zero and avoids paying for idle GPU capacity; current per-request model rates are shown in the service at deployment.
Features include Kubernetes-based cluster management, high-speed storage and networking, private model deployment, autoscaling and enterprise support. Final prices vary by GPU configuration, region, reservation and selected services.
Official sources: GMI Cloud pricing, GMI Cloud 2026 pricing guide.