AWS AI & Machine Learning Services

Managed AI APIs, foundation models and the custom ML platform on AWS, and how to choose between them.

  • AI Services Overview How Bedrock, task-specific APIs, and SageMaker fit together.
  • Amazon Bedrock Managed foundation models (Claude, Llama, Mistral, Titan) with Knowledge Bases, Agents, Guardrails.
  • Amazon SageMaker End-to-end ML platform for training, tuning, hosting, and monitoring custom models.
  • Amazon Comprehend NLP — entities, sentiment, PII redaction, medical concept extraction.
  • Amazon Textract Document AI — OCR, forms, tables, Analyze Expense / ID / Lending.
  • Amazon Rekognition Computer vision — labels, faces, moderation, custom image classifiers.

Overview

AWS groups its machine learning offerings into three layers. Task-specific AI services expose pre-trained models behind simple APIs: send a document to Textract and get text, forms and tables back; send text to Comprehend and get entities, sentiment or redacted PII. Amazon Bedrock provides managed access to foundation models from several providers through one API, along with Knowledge Bases for retrieval-augmented generation, Agents for tool use and Guardrails for content and PII filtering. Amazon SageMaker is the platform for teams that build, train, tune and host their own models, including open-weight models deployed from JumpStart.

How the Pieces Fit

Three Layers of AWS AI and ML AI servicesComprehend, Textract Amazon Bedrockfoundation models SageMaker AIbuild your own API call, no ML skillsfixed tasks Prompt, RAG, agentsgenerative apps Train, tune, hostfull control Move right only when the layer to the left cannot do the job: each step adds flexibility and operational work.

The layers are complementary rather than competing. A document pipeline might use Textract to extract text from scanned forms, Comprehend to detect and redact personal data, a Bedrock model to summarise the result, and a SageMaker endpoint hosting a custom classifier trained on the company's own labels. Choose the highest-level service that meets accuracy, cost and compliance needs, because it carries the least operational burden.

Comparison

OptionYou provideBest forPages
AI servicesinput dataOCR, NLP and vision tasks with standard outputsoverview, Comprehend, Textract, Rekognition
Amazon Bedrockprompts, documents, toolschat, summarisation, RAG, agentsoverview, Knowledge Bases, Agents, Guardrails
Amazon SageMakerdata, code, model choicecustom training, fine-tuning, hostingoverview, JumpStart
QuickSight Qa datasetnatural-language questions over BI dataQuickSight

Cost models differ too: AI services and Bedrock charge per request, page, character or token, while SageMaker charges for the instances you run, so steady high-volume inference can be cheaper on SageMaker and spiky or low-volume use cheaper on the per-call services.