AWS AI Updates: September 16, 2026

1. SageMaker Training Jobs Can Now Take a Ranked List of Acceptable Instance Types

AWS. SageMaker AI training and processing jobs accept a prioritized list of instance types instead of a single hardcoded choice, and SageMaker works down the list and launches on the first configuration that has capacity. A job can mix capacity sources in one submission, combining on-demand instances with reserved Flexible Training Plans, so a run configured for two ml.g6.48xlarge instances can fall back to four ml.g5.48xlarge rather than sitting in a queue. The feature is live in every region where SageMaker runs and is exposed through the CLI, APIs, SDKs, and console. Source

2. Amazon Q Console Can Query CloudTrail Logs in Plain English

AWS. Amazon Q in the AWS Management Console now reads CloudTrail trails, CloudWatch log groups, and event data stores, so questions like “who accessed this IAM role” or “which API calls returned errors” resolve without hand-writing a query. AWS says answers are grounded in actual account activity rather than generic documentation, which also covers configuration questions about whether logging coverage has gaps. It is available in all commercial regions where Amazon Q Console is supported. Source