How much AWS costs a startup per month, line by line.

The line items that surprise founders.

• NAT gateways. $32.85 a month each just for existing, plus $0.045 for every GB that passes through. That per-GB charge includes container images pulled from ECR and objects fetched from S3, unless you add VPC endpoints. A NAT gateway per AZ per environment adds up fast.
• Public IPv4 addresses. $3.65 a month each, idle or in use. Every load balancer AZ, NAT gateway and public task has one.
• Cross-AZ traffic. $0.01 per GB in each direction between availability zones. A chatty service in one zone talking to a database or cache in another pays it on every request.
• CloudWatch Logs ingestion. $0.50 per GB is more than ten times the cost of storing the same GB in S3. Debug-level logging in production is the classic cause of a sudden jump.
• Log groups with no retention. The default is to keep logs forever, so storage only ever grows.
• Multi-AZ on non-production. It doubles the database line and staging doesn't need it.
• Forgotten resources. A load balancer from a deleted preview environment, unattached volumes, old snapshots, a test database someone stopped (RDS restarts a stopped instance after seven days).
• EKS control planes. $73 a month per cluster before any workload, more on extended support. See ECS vs EKS for startups for the full comparison.

What credits do and don't cover.

AWS Activate credits pay most of the lines above automatically, so a startup on credits often sees a $0 invoice for a year or more. That's exactly why the bill tends to grow unwatched. The gross usage on the invoice is the number to track, not the amount due.

Credits don't cover past charges, and marketplace software billed through AWS is generally not covered either. They also expire, and they run out early if usage grows faster than planned. The tiers, eligibility rules and exclusions are covered in our AWS Activate credits guide, so we won't repeat them here.

If you are close to the end of your credits, the page on what happens when AWS credits run out walks through the cliff month and what to check first.

What to do when the bill grows.

Work through these in order. The first three cost nothing and take an afternoon.

1. Turn on cost allocation tags and a budget alert. Tag everything with an environment and a service, then set an AWS Budgets alert at the number you expect. You can't fix a bill you only see once a month.
2. Read Cost Explorer by usage type, not by service. "EC2-Other" hides NAT gateway hours and data processing. Grouping by usage type shows them as separate lines.
3. Set log retention on every log group. Thirty days is plenty for most application logs. Lower production log levels to info.
4. Add VPC gateway endpoints for S3 and DynamoDB. They're free and take that traffic off the NAT gateway. Interface endpoints for ECR and others cost money per hour, so compare them with your NAT data charge first.
5. Shrink staging. One NAT gateway, Single-AZ database, smaller tasks, and scale it to zero outside working hours.
6. Right-size from metrics. If tasks sit at 10% CPU, halve them. Try Fargate on ARM (Graviton), which lists at $0.03238 per vCPU-hour against $0.04048 for x86, if your images build for it.
7. Only then commit. Compute Savings Plans and RDS reserved instances cut the price of what's left, but lock in a year of spend. Trim first, commit second.

Re-run the numbers in the runway calculator after each change to see what it's worth in months.