SageMaker Canvas & SageMaker Unified Studio

Yes, in SageMaker Canvas, you can train models directly using data from external databases without relying on S3. Canvas supports importing data from various sources including Amazon Redshift provisioned clusters, AWS Glue Data Catalog through Amazon Athena, Amazon Aurora, Amazon RDS, Salesforce Data Cloud, Snowflake, and other databases through JDBC connectors. You can import this data directly into Canvas and use it for model training without first storing it in S3.

SageMaker Canvas is actually integrated into the SageMaker Unified Studio experience. In the Unified Studio environment, you can access Canvas which provides the AutoML capabilities (formerly known as Autopilot). Through Canvas in Unified Studio, you can use data from multiple sources for AutoML workflows, not just S3. These sources include Amazon Redshift, AWS Glue Data Catalog, Amazon RDS, and other external databases mentioned above.

Canvas is not a separate service but rather an application within SageMaker Unified Studio. You can launch and use Canvas directly from within the Unified Studio environment. In fact, the Autopilot UI (AutoML capabilities) has been integrated into the SageMaker Canvas application in the updated Studio experience. You can access Canvas through the Unified Studio interface, which provides a seamless experience for data preparation, model building, and deployment.
Sources
Data import - Amazon SageMaker AI
Amazon SageMaker Unified Studio - Autopilot & Model Monitor | AWS re:Post
Connect to data sources - Amazon SageMaker AI
(Optional) Migrate data from Studio Classic to Studio - Amazon SageMaker AI
A single data and AI development environment - Amazon SageMaker Unified Studio - AWS