AWS Cost Hygiene: S3 Lifecycle Savings at Careem

Case study repo: aws-cost-infra-case-study

Cloud spend often hides in defaults: log buckets that never expire, idle compute nobody owns, Terraform changes that merge without a cost diff. This page summarizes two real workstreams: S3 lifecycle savings at Careem and SageMaker pricing in Infracost.

Careem: S3 log storage

Application and service logs accumulated in S3 Standard storage without tiering or expiration. After lifecycle policies (transition to IA/Glacier, expire beyond retention, noncurrent version cleanup):

Reference Terraform pattern: s3-log-lifecycle.tf

Full checklist (EC2/EKS right-sizing, idle resources, tagging): INFRA-HYGIENE-CHECKLIST.md

Infracost: SageMaker endpoint configuration

GitHub PR: infracost/infracost#3567

Three billing paths mapped for aws_sagemaker_endpoint_configuration:

• Provisioned instances: instance_type and initial_instance_count, plus optional hosting storage
• Serverless inference: GB-second compute and provisioned concurrency when configured
• Shadow variants: Parallel shadow production variants costed separately from live traffic

Longer write-up: Shift-Left FinOps: Adding SageMaker Support to Infracost

Sample audit artifact

Illustrative monthly waste table (format only, not Careem billing): sample-waste-audit.md