AWS & GenAI
AWS Bedrock provides managed access to multiple model families, while SageMaker supports hosting custom or fine-tuned models.
Overview
AWS offers two distinct paths for AI workloads: Bedrock, a managed API to multiple model families without managing infrastructure, and SageMaker, for training or hosting a custom or fine-tuned model on dedicated infrastructure.
Where It Fits
Application
Bedrock
Managed multi-model API.
SageMaker
Custom/fine-tuned hosting.
Key Points
- Bedrock
- A managed API surface across several model families (including Anthropic and Amazon’s own Nova) without provisioning GPU infrastructure.
- SageMaker
- Supports training, fine-tuning, and hosting a custom model with more control than Bedrock offers.
- IAM integration
- Access to Bedrock or SageMaker resources is controlled through the same IAM permissions as the rest of an AWS account.
Interview Question
When would you choose Bedrock over SageMaker for an AI feature on AWS?
Bedrock fits when a hosted model from an existing provider is sufficient — no training or custom infrastructure needed. SageMaker fits when a workload requires fine-tuning a model on proprietary data or hosting a custom model with specific infrastructure control, at the cost of more operational responsibility.
Explain It in 30 Seconds
AWS splits AI workloads into Bedrock, a managed API across multiple model families, and SageMaker, for training or hosting custom or fine-tuned models — the choice depends on whether a hosted model is sufficient or custom infrastructure is required.
Real-World Stack
Technologies commonly used to implement this in production.