NVIDIA Run:ai

What it does

NVIDIA Run:ai is infrastructure software for Kubernetes clusters that share GPUs. It takes in workload submissions for development workspaces, training, distributed training and inference, plus the quotas, priorities and access rules an organisation sets. It then decides where and when each workload runs on available GPU capacity, and reports cluster, node, GPU and workload utilisation back to a central control plane.

Platform teams use it to pool GPUs across departments, projects and clusters, with fair-share scheduling, preemption, topology-aware placement and fractional GPU allocation. Developers submit work through a web UI, CLI or REST API. It sits on top of Kubernetes rather than replacing it, and parts can be adopted on their own, such as the KAI Scheduler, the GPU fractioning operator and the Model Streamer SDK for faster model loading.

It suits platform and infrastructure teams running shared GPU estates on Kubernetes or OpenShift, in cloud, hybrid, on-premises or air-gapped setups. The code is open source under Apache 2.0, but you need your own cluster and GPUs. It is not a hosted app for individual users, and a team with dedicated GPUs may find native Kubernetes scheduling enough. Supported options are sold through NVIDIA AI Enterprise, DSX OS, Mission Control or as standalone software.

What it can do

• AI-aware GPU scheduling with fair-share, preemption, bin-packing, priorities and topology-aware placement
• Fractional GPU allocation and dynamic sharing of available capacity
• Quotas and policies across departments and projects
• Multi-cluster management from one control plane
• Workspaces, training, distributed training and inference management
• Role-based access control plus monitoring of clusters, nodes, GPUs and workloads

Works with

• Kubernetes
• OpenShift
• NVIDIA GPU Operator
• KServe
• Knative
• Kubeflow
• Ray

Example workflows

Share one GPU cluster across teams

Inputs: a Kubernetes cluster with GPUs and a list of departments and projects. Steps: install the control plane and cluster components, set quotas and priorities per project, then have teams submit workloads. Output: workloads scheduled under fair-share rules, with utilisation visible centrally.

Run inference on fractional GPUs

Inputs: inference workloads that do not need a full GPU. Steps: enable fractional GPU allocation, submit the workloads via UI, CLI or API, and monitor per-workload resource use. Output: several workloads sharing GPUs within defined resource limits.

Before you choose

• Requires you to operate Kubernetes and GPU infrastructure; the software does not supply compute.
• Several adoption routes exist (open source, NVIDIA AI Enterprise, DSX OS, Mission Control, standalone), so check which includes the support you need.
• Performance figures on the product page (such as 10x GPU availability) are marketing claims without stated test conditions.

Listed under

• Coding

Get started

Apache 2.0 code on GitHub; supported editions via NVIDIA

1. Check the support matrix and system requirements for Kubernetes or OpenShift versions, GPU operators and networking.
2. Install with the runai-installer CLI and a configuration YAML (guided), or deploy the Helm charts manually.
3. Define departments, projects, quotas and access roles, then submit a first workload via the web UI, CLI or REST API.

The source code is available under Apache 2.0. You provide Kubernetes and GPUs; commercially supported editions have separate terms.

Source: github.com/dsx-ai-factory/run-ai, checked 28 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 28 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Accelerate AI & Machine Learning Workflows | NVIDIA Run:ai ↗www.nvidia.com
• GitHub - dsx-ai-factory/run-ai: Run:ai is the AI workload orchestration platform for Kubernetes. This monorepo contains the control plane services, in-cluster Kubernetes operator/controllers, shared libraries and clients, the installer, and the local dev environment. · GitHub ↗github.com
• NVIDIA Run:ai Solution Overview ↗resources.nvidia.com

Company website

nvidia.com: product, pricing and policy pages

Web search

Confirms the official site

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 28 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.