AWS Cost Optimization Checklist 2026

42 items across 7 areas. Each one is specific enough to action immediately — no "consider reviewing your infrastructure" filler. Start with S3 lifecycle rules and Savings Plans — both are high impact, low effort, and take under an hour. The matrix below shows where each section lands.

Work through sections sequentially or jump to the ones most relevant to your architecture. Each item is written to be verifiable — you should be able to open the AWS console and confirm it in under 10 minutes. For a broader multi-cloud review, see the cloud-agnostic checklist. For a structured audit with deliverables, use the FinOps audit guide.

1. EC2 Rightsizing

• Enable AWS Compute Optimizer and review its rightsizing recommendations — focus on instances flagged as "over-provisioned" with CPU <20% over 14 days.
• Identify instances where memory utilization is also low (use CloudWatch agent or third-party APM). CPU alone misses memory-bound workloads.
• Move variable/bursty workloads to T4g or T3a burstable instances. Validate CPU credit balance stays positive under peak load.
• Review instance family generation: M4 → M7g (Graviton3) typically cuts cost 30–40% for the same workload. Re-benchmark before committing.
• For batch/fault-tolerant jobs, shift to Spot Instances with a diversified instance pool. Use EC2 Auto Scaling with mixed instance types to reduce interruption risk.
• Audit stopped instances: you still pay for EBS volumes and Elastic IPs attached to stopped instances. Terminate or snapshot-and-delete if unused >30 days.

2. S3 Storage Classes & Lifecycle

• Enable S3 Storage Lens to identify buckets with low access frequency — these are candidates for tiering.
• Set lifecycle rules to transition objects to S3 Infrequent Access (IA) after 30 days and Glacier Instant Retrieval after 90 days for objects not accessed frequently.
• For compliance archives accessed <1/year, use S3 Glacier Deep Archive ($0.00099/GB/month vs $0.023/GB for Standard).
• Enable S3 Intelligent-Tiering on buckets with unpredictable access patterns (monitoring fee: $0.0025 per 1,000 objects — only worthwhile for objects >128KB).
• Delete incomplete multipart uploads: set a lifecycle rule to abort them after 7 days (AbortIncompleteMultipartUpload). These accumulate silently.
• Audit versioning-enabled buckets: old versions can 10x your storage bill. Add expiration rules for non-current versions.
• Review S3 Requester Pays for buckets serving external consumers — they pay the GET and data transfer costs instead of you.

3. RDS & Database Costs

• Check RDS Reserved Instance coverage via the Cost Explorer RI utilization report. Uncover any on-demand DB instances running steadily for >3 months.
• Audit Multi-AZ deployments: necessary for production, but dev/staging databases rarely need Multi-AZ. Disabling it halves the instance cost.
• Review RDS storage type: GP2 → GP3 migration is free and saves ~20% on storage costs with the same or better IOPS baseline.
• Identify idle RDS instances (0 connections for >7 days via CloudWatch DatabaseConnections metric). Snapshot and delete, or use RDS Aurora Serverless v2 for intermittent workloads.
• For Aurora clusters, check Aurora I/O-Optimized pricing: if your I/O costs exceed 25% of total cluster cost, switching to I/O-Optimized pricing mode saves money.
• Evaluate read replica count: are all replicas actively serving read traffic? Remove unused replicas — each one is a full instance charge.

4. Savings Plans & Reserved Instances

• Open AWS Cost Explorer → Savings Plans → Recommendations. Filter to your look-back period (30 days recommended) and review the suggested hourly commitment.
• Start with Compute Savings Plans unless your instance family is firmly locked in — the flexibility is worth the slightly lower discount.
• Match commitment term to actual runway: 1-year if uncertain, 3-year only for stable baseline workloads with >18 months of track record.
• Check RI expiry dates: expiring RIs flip to on-demand pricing silently. Set a CloudWatch alarm or calendar reminder 60 days before expiry.
• Review the RI Marketplace for unused standard RIs — you can sell unexpired RIs you no longer need.

5. Data Transfer & Egress

• Pull a data transfer cost breakdown from Cost Explorer grouped by "Usage Type" — filter for types starting with DataTransfer-Out.
• Move static assets behind CloudFront: CloudFront-to-internet egress is $0.0085/GB vs $0.09/GB direct from EC2/S3. 10x cheaper for high-traffic assets.
• Ensure traffic between services in the same AZ uses private IPs — cross-AZ traffic costs $0.01/GB each way and adds up fast for microservices.
• Replace NAT Gateway for S3 and DynamoDB access with VPC Gateway Endpoints — these are free and eliminate NAT Gateway data processing fees ($0.045/GB).
• For large on-premise data transfers, evaluate AWS Direct Connect or AWS DataSync vs. internet egress pricing at scale.
• Check if CloudFront logs and access logs are being stored in S3 and generating unnecessary intra-region transfer costs — consider disabling if not actively used.

6. Cost Explorer, Tagging & Billing Hygiene

• Enable AWS Cost Anomaly Detection with an email/SNS alert for spend anomalies >$50 or >20% above expected — catches runaway resources before month-end.
• Set a billing alert in CloudWatch for 80% and 100% of your monthly budget threshold.
• Enforce a mandatory tagging policy via AWS Config or Service Control Policies (SCPs): at minimum, tag every resource with env, team, and project.
• Enable Cost Allocation Tags in the Billing console and activate your resource tags — only activated tags appear in Cost Explorer breakdowns.
• Audit Elastic IP addresses: unused EIPs cost $0.005/hour (~$3.65/month each). Release any not attached to a running instance.
• Delete unused EBS snapshots: use AWS Backup or a Lambda to find snapshots older than your retention policy and not associated with an AMI.
• Review NAT Gateway utilization: high data processing fees with low traffic often mean misconfigured routing. Check if traffic can be rerouted through VPC endpoints.
• Enable AWS Trusted Advisor (Business/Enterprise Support) for automated cost optimization checks — it surfaces idle load balancers, underutilized EC2, and low-utilization RDS.

7. Lambda & Serverless

• Review Lambda memory allocation with AWS Lambda Power Tuning (open-source tool): many functions are over-provisioned on memory, increasing duration costs.
• Check Lambda provisioned concurrency usage — only use it if cold starts are a measurable user issue. Turn it off in non-production environments.
• For high-frequency, short-duration Lambda invocations, evaluate ARM/Graviton2 runtime (arm64 architecture): same performance, 20% cheaper per invocation.
• Set function timeouts appropriately — runaway Lambdas executing to max timeout are a common source of unexpected spend.