Sagemaker

Amazon SageMaker is AWS's fully managed machine learning platform, launched in 2017, designed to cover every stage of the ML lifecycle without leaving the AWS ecosystem. SageMaker's breadth is unmatched among cloud ML platforms: SageMaker Studio (unified IDE for ML development), SageMaker Training (managed distributed training with 150+ built-in algorithms and custom containers), SageMaker Pipelines (CI/CD for ML workflows), SageMaker Feature Store (centralized feature management), SageMaker Model Registry, SageMaker Endpoints (real-time and batch inference with auto-scaling), SageMaker Autopilot (automated ML), SageMaker Ground Truth (human-in-the-loop data labeling), SageMaker Canvas (no-code ML), and SageMaker Experiments (experiment tracking). SageMaker's managed training eliminates infrastructure management — teams specify instance type, training script, and S3 data location, and SageMaker provisions, trains, and stores the model automatically. SageMaker HyperPod introduced distributed training on purpose-built clusters with auto-recovery for multi-node GPU training. SageMaker JumpStart provides 400+ pre-trained foundation models (Llama 2, Mistral, Stable Diffusion) deployable with one click. SageMaker's tightest integration is with the AWS data stack: S3, Glue, Athena, Redshift, EMR, and EventBridge. Pricing is consumption-based — training and endpoint instances billed by the second. SageMaker's complexity has earned a reputation for a steep learning curve, with many teams using it selectively (endpoints only, or pipelines only) rather than adopting the full suite.