Cost-effective resources

Using the appropriate services, resources, and configurations for your workloads is key to cost savings. Consider the following when creating cost-effective resources:

You may employ internal guardrails (built using AWS Organization Service Control Policies) to allow a limited set of services to be provisioned to contain costs. If a workload requires services outside of the allow list, they need to centralized, created, and shared with an individual account, or created by an administrator.

Prefer managed generative AI services such as Amazon Bedrock or Amazon SageMaker AI JumpStart to minimize total cost of ownership by reducing the overhead of managing infrastructure, scaling, and updates. Where self-managed model hosting is required, benchmark deployments on AWS Graviton-based instances for price–performance gains and use Spot Instances for stateless batch jobs (for example, embeddings generation, evaluation, and model fine-tuning). Implement guardrails to prevent persistent provisioning of high-cost model endpoints.