GMI Cloud

What is GMI Cloud?

GMI Cloud is an AI infrastructure platform providing GPU clusters, serverless inference, and agent runtime for production workloads. It serves teams deploying large language models and multimodal applications that require scalable compute. Access begins through a web console or by contacting sales.

What are the features of GMI Cloud?

• Serverless inference

  Runs LLM and multimodal model APIs with automatic scaling to zero, request batching, latency-aware scheduling, and multi-tenant isolation.
• Dedicated GPU clusters

  Provides bare metal GPUs with root access, custom software stacks, RDMA-ready networking, and multi-node orchestration via the Cluster engine.
• NVIDIA-based architecture

  Built on NVIDIA Reference Platform Cloud Architecture using validated designs for H100, H200, and Blackwell platforms.
• Flexible scaling path

  Allows workloads to move from API-based serverless inference to full GPU clusters without re-architecting the application stack.

What are the use cases of GMI Cloud?

• Real-time generative video

  Higgsfield runs real-time video workloads on the platform, reporting 65% lower p95 inference latency and 45% lower compute cost.
• Foundational model training

  Mirelo AI uses the infrastructure to scale model development, reporting 40% lower training costs and 20% faster training time.
• Public-sector AI deployment

  WiAdvance supports government and education use cases in Taiwan through flexible allocation across committed and on-demand capacity.

GMI Cloud Pricing Information:

• NVIDIA H100

  $2.60 per GPU-hour; suited for inference and training jobs requiring high memory bandwidth and larger model footprints. Available now.
• NVIDIA H200

  $3.20 per GPU-hour; optimized for training and inference at scale. Available now.
• NVIDIA Blackwell

  Pre-order status; intended for large-scale deployments requiring maximum performance headroom. Coming soon.