NVIDIA AI Cluster Runtime (AICR) Provider: Installation & Configuration

Installation

The NVIDIA AICR provider is available as a package in all Pulumi-supported languages:

• JavaScript/TypeScript: @pulumi-labs/nvidia-aicr
• Python: pulumi-labs-nvidia-aicr
• Go: github.com/pulumi-labs/pulumi-nvidia-aicr/sdk/go/nvidiaaicr
• .NET: Pulumi.Labs.NvidiaAicr
• Java: com.pulumi.labs:nvidia-aicr

In normal usage, installing one of the language SDKs above is enough. The first pulumi up automatically downloads the matching provider binary from the URL embedded in the SDK, so no separate step is required.

Provider Binary (optional)

If you need to install the provider binary explicitly (for example, in an air-gapped environment or to prewarm a CI cache), use the pulumi plugin command:

pulumi plugin install resource nvidia-aicr <version> --server github://api.github.com/pulumi-labs/pulumi-nvidia-aicr

Replace <version> with your desired release (see the releases page).

Configuration

The NVIDIA AICR provider is a Pulumi component provider and does not require any provider-level configuration. The deployed software stack is configured per-resource via the inputs to ClusterStack (accelerator, service, intent, platform, kubeconfig, and optional componentOverrides). See the API Reference (left navigation) for the full input surface.

ClusterStack requires only a Kubernetes kubeconfig, which can be:

• A literal kubeconfig string passed via the kubeconfig input.
• A path to a kubeconfig file on disk via the kubeconfigPath input.
• An Output from another Pulumi resource (e.g. EksCluster.kubeconfig from pulumi-eks).

For runnable per-cloud examples (EKS, AKS, GKE, OKE, CoreWeave, kind), see the examples/ directory in the provider repository.