GMI Cloud

GMI Cloud provides computing infrastructure for artificial intelligence work. Its site describes one platform that covers compute, inference and agents, built for production use rather than experiments, and it follows an inference-first design with automatic batching, latency-aware scheduling and isolation between tenants.

The offer has three main parts. Serverless inference gives teams ready-made APIs for large language models and multimodal models, scaling down to zero when idle. GPU clusters are dedicated bare metal machines with root access and custom configurations, for customers who want full control of the hardware. Studio is an environment for developing and working with models.

Capacity is published with prices. NVIDIA H100 GPUs are listed at 2.60 dollars per GPU hour and H200 GPUs at 3.20 dollars, both available now, while NVIDIA Blackwell machines can be pre-ordered. The company says its platform is validated on the NVIDIA Reference Platform Cloud Architecture, and it reports gains in throughput, inference speed and cost against its own baseline.

Customers shown on the site include Mirelo AI, Higgsfield, HeyGen, Eigen AI, LegalSign, Utopai and OXMIQ. The pages read do not state where its data centres are or give a head office address.

Key facts

• Products: serverless inference, bare metal GPU clusters, Studio
• GPUs: NVIDIA H100, H200, Blackwell on pre-order
• Pricing: H100 2.60 dollars and H200 3.20 dollars per GPU hour
• Named customers: HeyGen, Higgsfield, Mirelo AI, Eigen AI