Skill 01: EKS Cluster Provisioning

Purpose

Provision an Amazon EKS cluster configured for GPU-accelerated Physical AI RL training workloads. The cluster must support multi-node GPU training with EFA networking, FSx for Lustre shared storage, and the necessary OIDC/IAM foundations for add-ons.

We use Terraform as the primary IaC tool. Terraform provides state management, plan/apply workflow, module composition, and the ability to bundle Helm releases for open-source plugins (GPU Operator, EFA plugin, KubeRay, etc.) alongside cluster provisioning in a single apply.

Note: eksctl can be used for quick prototyping. See the eksctl quick-start appendix at the end of this skill.

Prerequisites