Amazon SageMaker AI now lets developers give a training or processing job an ordered list of acceptable instance types. SageMaker checks the list in priority order and starts the job on the first option with available capacity.
A preference list can contain up to five instance types. The feature is aimed at workloads that can run on several GPU or CPU configurations but previously had to be submitted against one exact machine type. During periods of tight capacity, teams often waited, resubmitted manually or built monitoring scripts to try alternatives.
The new workflow keeps that fallback logic inside a single job request. A team can put its ideal configuration first and list acceptable substitutes behind it, reducing capacity-watching work and potentially starting experiments or scheduled retraining sooner. It applies to both SageMaker Training Jobs and Processing Jobs.
The feature does not create new capacity, guarantee immediate access or decide whether two instance families are economically equivalent. Users still need to validate that each listed type has enough memory and appropriate performance for the workload, and costs may differ depending on which choice launches. Its practical benefit is automated selection among configurations the customer has already approved.