AWS publishes reference design for sharing SageMaker HyperPod GPU clusters across teams

AWS has published a reference architecture for letting multiple teams share one Amazon SageMaker HyperPod EKS cluster while keeping their workloads separate.

According to the AWS Machine Learning Blog, each team is isolated in its own Kubernetes namespace, with AWS IAM Identity Center handling authentication and a separate SageMaker AI domain per team.

The design relies on HyperPod Task Governance to allocate GPU resources fairly among teams, and on namespace-level cost allocation to give each team visibility into its own spend. The post is a first-party architecture guide, so it describes the intended setup rather than independently measured results.