Overview
Key takeaways
- Nebius provides just enough UI to launch basic GPU infra. They largely compete on price.
- Managed k8s control plane is free (but you still need to add billing details)
- Applications are deployed into managed infra — bunch of stuff abstracted away here but it feels relatively clean & simple
- The application framing is a little loose, e.g. you can deploy applications that cover vLLM (which then asks for a model), or models like DeepSeek-R1 (which bundles vLLM). I assume that most users don’t have dozens of apps so this probably works just fine.
Pricing
From https://nebius.com/prices
Product Info Architecture
Major product concepts
- Compute
- Managed offerings: Kubernetes, Ray, Soperator, MLflow, PostgreSQL
- Applications
- Observability
- Object Storage
- Container Registry
- Network
Here’s the left nav — from an April video they recently added Managed Ray, Managed Soperator, and removed Managed Spark.
Soperator enables you to run slurm over k8s: https://github.com/nebius/soperator.
Compute
Managed Service for Kubernetes
Dashboard
Create a cluster
Cluster dashboard
Deploy App (vLLM)
Create Node groups for Kubernetes
With GPU
Applications
Nebius breaks out applications into 2 broad sections: Standalone Apps and Apps for Managed Service for Kubernetes
Browse applications
Deploy vLLM
Deploy DeepSeek-R1
View Installed Applications
Observability
