AWS AI Updates: July 8, 2026

1. SageMaker Adds OpenLineage-Compatible Data Lineage to IAM-Based Domains

AWS. Amazon SageMaker Unified Studio now supports OpenLineage-compatible data lineage tracking in IAM-based domains, a capability previously limited to IAM Identity Center environments. It captures lineage events from Apache Spark jobs on Amazon EMR, AWS Glue, SageMaker Visual ETL, and notebooks, rendering an interactive graph with configurable depth and column-level detail, plus a new DeleteLineageEvent API for managing published events. The feature is rolling out to all regions where SageMaker Unified Studio is available. Source

2. SageMaker Unified Studio Can Now Import Existing MWAA Environments

AWS. Amazon SageMaker Unified Studio now lets teams connect existing Amazon Managed Workflows for Apache Airflow (MWAA) environments to Studio projects without recreating configuration or migrating DAGs. Users add the connection via the Workflows tool by referencing their domain and project, after which they can sync, trigger, and monitor Airflow workflows alongside analytics and ML tools, including visual drag-and-drop authoring for Airflow 3.0 and later. The capability is available in all regions where SageMaker Unified Studio operates. Source

3. ECS Managed Instances Cuts GPU Management Fees Up to 60 Percent

AWS. Effective July 1, Amazon ECS Managed Instances reduced management fees for accelerated compute by 35 percent on G-series instances and 60 percent on P-series and AWS Trainium instances, applied automatically with no customer action. The service auto-provisions and operates optimal EC2 instances for containerized tasks, surfacing GPU utilization, memory, and temperature metrics through CloudWatch Container Insights and automatically detecting and replacing unhealthy GPU hosts. AWS says the change targets ML and AI inference and training workloads run on containers. Source

4. EKS Auto Mode Cuts GPU Management Fees Up to 60 Percent for ML Workloads

AWS. Amazon EKS Auto Mode reduced management fees for accelerated instances by 35 percent on G-series and 60 percent on P-series and AWS Trainium instances, effective July 1 and applied automatically across all regions where Auto Mode runs. The mode automatically provisions and manages GPU nodes, including parallel image pulling and unpacking on local NVMe storage to speed deployment of large model images, and accelerator-aware node repair that replaces unhealthy GPU hosts. AWS positions the update for inference, model fine-tuning, and batch processing, matching the same-day fee cut for ECS Managed Instances. Source