Tech in Diagrams

AI in the Cloud

Cloud AI services expose trained models as APIs, so applications can use them without owning accelerators.

Beginner 5 min Cloud Computing
Prompt in, inference, response out
  • Request
  • Context
  • Response
USER · APPLICATIONYour appComposes the request, ownspermissions and the interface.prompt · user inputGUARDRAIL · POLICYChecked in, checked outAuthentication, quotas and ruleson what may be sent and returned.request ✓ · response ✓OPTIONAL CONTEXTYour data, this request onlysearch / database lookupAttached as input. The model doesnot permanently learn from it.MANAGED AI SERVICE · PROVIDER OPERATEDMODEL ENDPOINTAuthenticated model APIRouting, model version, throttling and quotas.model v3 · servingINFERENCETrained model runsShared accelerators, abstracted.OUTPUTGenerated responsetext / structured dataA suggestion, not a guarantee.OPERATIONAL CONCERNSTRAINING · FINE-TUNINGA different jobNot run on every prompt.WHAT THE CLOUD GIVES YOUYours: data, permissions, UX.
AI in the cloud is a managed model endpoint you call like any other service: your app orchestrates the request, the provider runs the inference, and training stays a separate workflow.

Hover the app, the policy step, context retrieval, the model endpoint, inference, the response, the operational indicators or the training rail to isolate related elements; click any component for a short explanation.

Next conceptMicrosoft Cloud & Modern WorkMicrosoft 365 is SaaS for everyday work; identity ties it together, and Copilot works on top of that same content.