Amazon SageMaker AI's serverless model customization feature has been announced #AWSreInvent
This page has been translated by machine translation. View original
Hello, this is Morita.
The serverless model customization feature for Amazon SageMaker AI was announced at the re:Invent 2025 Keynote.
What is the serverless model customization feature
Traditionally, when customizing models in SageMaker AI, it was necessary to select the computing resources (such as instance type) in advance.
However, with this feature, you can start training your custom model simply by selecting the base model and customization method.
Computing resources are provisioned in a managed way by AWS based on model size and data size, eliminating the need to select resources.
For deployment destinations, you can use Bedrock.
Bedrock has been providing model import capabilities for some time, enabling serverless inference with custom models.
Available regions
- US East (Northern Virginia)
- US West (Oregon)
- Asia Pacific (Tokyo)
- Europe (Ireland)
Pricing
You pay based on the number of tokens processed during training and inference, with a pay-as-you-go model.
Please check the pricing page below for details.
Conclusion
In model development with Amazon SageMaker AI, resource selection is often a barrier, so using this feature allows for easier model development.
This is a particularly useful feature for cases where you want to proceed with rapid model development, such as in PoCs.