How they use Nebius

Saturn Cloud’s hosted tiers use Nebius hardware. The joint offering launched in June 2025 with NVIDIA Hopper GPUs, notebooks, jobs and deployments. Saturn Cloud’s hosted pricing page lists Nebius CPU and H100 instances, with underlying hosting charges collected through Saturn Cloud.

Enterprise installations run in customers’ Nebius accounts on Managed Kubernetes. Saturn Cloud can provision and maintain the cluster, or customers can install its Helm chart themselves. Hugo Shi reports installing Saturn Cloud in an unnamed customer’s Nebius account after the customer had primarily used AWS. Development workspaces use Nebius persistent disks. Jobs support multi-node training, deployments serve models and dashboards, and all three resource types can attach Nebius shared filesystems. Its Kubernetes operator reconciles Helm releases every two minutes. Compute, storage and networking remain in the customer’s Nebius account under its IAM and VPC policies. Running workloads continue if Saturn Cloud’s API server becomes unavailable. Its inference guide describes vLLM, NVIDIA NIM, FastAPI and BentoML serving on Nebius GPUs, with demand-based GPU node scaling and automatic replacement of unhealthy nodes.

Saturn Cloud also documents a supported workflow connecting development and training on Nebius AI Cloud with model artifacts in Nebius Object Storage and production inference through Nebius Token Factory. In July 2026, Saturn Cloud announced self-service deployment through the Nebius marketplace, including managed fine-tuning, distributed training, inference endpoints and per-token billing. The offering supports full-weight and LoRA fine-tuning and OpenAI-compatible inference endpoints.