Domain 2 of 5 · Chapter 3 of 3

AWS GenAI Infrastructure

Unlock the complete study guide + 1,040 practice questions across 16 full exams.

Bundled into the existing AWS Certified AI Practitioner AIF-C01 premium course — no separate purchase.

Included in this chapter:

  • The AWS GenAI service landscape: build vs. consume
  • Mechanics & cost: pricing, customization, AWS edge
  • Exam-pattern recognition: stem → service, traps

Amazon Bedrock vs Amazon SageMaker AI vs Amazon Q

AspectAmazon BedrockAmazon SageMaker AIAmazon Q
Primary useConsume managed foundation models via one serverless API (plus Knowledge Bases, Agents, Guardrails)Build, train, fine-tune, and host custom or pretrained ML/FM modelsUse a ready-made generative-AI assistant (Business for enterprise data, Developer for coding/AWS)
Control levelLow: managed FMs, prompt/RAG/fine-tune without managing serversHigh: full control of data, algorithm, training, and hostingLowest: configure a finished application, no model building
Who operates the modelAWS operates the model; you call the APIYou deploy and run the model on your SageMaker endpointsAWS operates the assistant end to end; you connect data/tools
Pricing modelPer token (on-demand) or committed Provisioned Throughput; also batch and customization pricingPay for the compute/instances used for training and hostingPer-user subscription pricing for the assistant
Best-fit userDevelopers adding GenAI to apps without ML infrastructureData scientists and ML engineers needing custom modelsBusiness users and developers wanting an out-of-the-box assistant

Decision tree

Want a ready-made GenAI assistant (no build)? Yes Amazon Q Q Business: enterprise data Q Developer: coding / AWS No Build, train, or host your own model? Yes SageMaker AI Full control + custom hosting; JumpStart model hub to start No Quick no-code experiment in the browser? Yes PartyRock No-code Bedrock Playground to learn and experiment No Amazon Bedrock Consume top FMs via one serverless API (RAG, Agents, Guardrails) Bedrock cost: on-demand per-token for bursty traffic; Provisioned Throughput (committed hourly) for steady high volume.
Cheat sheet

Cheat sheet

  • Amazon Bedrock serves many providers' FMs through one API
  • Bedrock Agents break a request into steps and call your systems
  • Bedrock Guardrails filter content and PII on inputs and outputs
  • SageMaker AI builds, trains, and hosts your own models
  • SageMaker JumpStart is a hub of pretrained models to deploy or tune
  • Amazon Q Business is a ready-made assistant over enterprise data
  • Amazon Q Developer is a ready-made coding and AWS assistant
  • PartyRock is a no-code Bedrock playground for learning
  • Pick the highest-level managed service that meets the requirement
  • On-demand pricing charges per input and output token, no commitment
  • Provisioned Throughput reserves capacity for steady high volume
  • Batch inference trades latency for a 50%-lower per-token price
  • A custom Bedrock model adds training cost plus Provisioned hosting
  • A smaller model can be the cheaper, correct answer
  • Managed GenAI services lower the barrier and speed time to market
  • Bedrock does not use your data to train its base FMs
  • Managed GenAI inherits the AWS shared responsibility model
  • Amazon Nova is AWS's own foundation-model family in Bedrock
  • Bedrock Data Automation turns unstructured media into structured output
  • Bedrock model evaluation compares FMs: automatic vs human
  • Proprietary JumpStart models require accepting a license/EULA

Unlock with Premium — includes all practice exams and the complete study guide.

Also tested in

References

  1. Amazon Bedrock FAQs FAQ
  2. Knowledge Bases for Amazon Bedrock
  3. Agents for Amazon Bedrock
  4. Guardrails for Amazon Bedrock
  5. SageMaker JumpStart Foundation Models
  6. What is Amazon Q Business
  7. What is Amazon Q Developer
  8. Amazon Bedrock
  9. Amazon Bedrock Pricing
  10. Provisioned Throughput for Amazon Bedrock
  11. Batch inference for Amazon Bedrock
  12. Custom models in Amazon Bedrock
  13. Data protection in Amazon Bedrock
  14. AWS Shared Responsibility Model Well-Architected