AWS Adds Managed Ray Capabilities to SageMaker HyperPod

What AWS released

On August 24, 2026, AWS announced new Ray capabilities for Amazon SageMaker HyperPod. The release integrates Ray with HyperPod's purpose-built infrastructure for large-scale foundation-model training and serving on Amazon EKS.

Managed distributed AI workloads

Ray is widely used to scale Python workloads across GPU clusters for training, inference and data processing. AWS now lets teams create and monitor Ray clusters on HyperPod while continuing to use open-source KubeRay resources and standard Ray APIs. Developers can attach SageMaker Studio JupyterLab and Code Editor environments to running clusters rather than managing separate notebook infrastructure.

Reliability and observability

SageMaker HyperPod already provides node health monitoring and automatic recovery for large ML clusters. The Ray integration combines those infrastructure controls with Ray's distributed execution model and adds out-of-the-box monitoring for Ray workloads. AWS positions the integration for resilient distributed training and accelerated inference without forcing developers to abandon the open Ray ecosystem.

Why it matters

Large-model training and serving increasingly depend on orchestration layers that can recover from failing nodes, allocate GPUs efficiently and keep developer workflows close to the cluster. Native Ray support gives AWS customers a managed path for workloads that already use Ray Train, Ray Serve or KubeRay while benefiting from HyperPod's cluster resiliency.

Availability

AWS describes these capabilities as available now for SageMaker HyperPod on Amazon EKS. Teams should still verify regional availability, supported instance types and current SageMaker pricing before production deployment.