Amazon SageMaker AI Adds Instance Preference Lists for Training Jobs

They are a new feature for training and processing jobs that let you specify an ordered list of up to five instance types, so SageMaker starts on the first type with available capacity.

Previously you picked a single instance type and retried manually or ran capacity-watching scripts; now SageMaker handles the fallback automatically.

Yes, especially for teams facing GPU capacity shortages in certain regions or during peak hours, as it cuts waiting time and job failures.

Yes, instance preference lists apply to both training and processing jobs.