Amazon SageMaker

SageMaker started as a way to run machine learning on AWS without managing servers. Data scientists use notebooks to prepare data and write code, then train models on managed compute that scales up as needed. Once a model is trained, SageMaker can deploy it behind an endpoint, monitor its performance, and handle versioning, which turns a notebook experiment into a production service. It supports popular frameworks such as PyTorch and TensorFlow.

The product has broadened. SageMaker AI covers model training and deployment, including foundation models, with features for distributed training on large clusters, fine-tuning, and hosting open models. SageMaker Unified Studio is a single environment for data engineering, analytics, and AI work, and SageMaker Lakehouse brings data from S3, Redshift, and other sources under one set of permissions and governance. Pipelines, feature stores, and experiment tracking support teams building repeatable workflows.

SageMaker is billed by usage on compute, storage, and managed features, and a free tier covers small amounts for the first months. Its breadth is also its drawback: many services, settings, and pricing dimensions make it complicated and easy to overspend. Teams already on AWS value the integration with IAM, S3, and security controls. Google Vertex AI and Azure Machine Learning are direct competitors, while Databricks is a common alternative for data-heavy teams. For calling hosted models without managing training, Amazon Bedrock is simpler.