Deploy Based on Application Requirements

Different workloads need different deployment approaches.

Lambda is useful for on-demand execution.

Request
  ↓
Lambda
  ↓
FM/API
  ↓
Response

Useful when:

• execution is event-driven
• workload is intermittent
• server management should be minimized

Provisioned throughput can be considered when predictable capacity/performance is important.

Conceptually:

Application
    ↓
Amazon Bedrock
    ↓
Provisioned Capacity
    ↓
FM

Think:

predictable workload → predictable model capacity requirement.

SageMaker AI endpoints can be used when deploying models with greater control over the serving environment.

Application
     ↓
SageMaker Endpoint
     ↓
Model

Useful when deployment requires more control over model serving.