A company needs to deploy a custom-trained classification ML model on AWS.

For near real-time inference with low latency and variable traffic, AWS recommends deploying models to managed SageMaker endpoints. By enabling auto scaling, the endpoint automatically adjusts the number of instances based on request volume, ensuring consistent performance while optimizing cost.

Amazon SageMaker Endpoints abstracts infrastructure management, health checks, scaling, and model deployment. This provides lower operational overhead than managing EC2 instances manually.

Batch transform is for offline inference. API Gateway with S3 cannot serve ML models. EC2-based deployments require manual scaling and monitoring.

Therefore, a SageMaker endpoint with auto scaling is the correct solution.