Understanding Your AWS Bill

Every grada command that touches infrastructure tells you what it costs before you pay it: apply shows a pre-flight estimate, add prints the cost impact of each addon, and your README.md keeps a refreshed monthly baseline. This guide explains what those numbers include, what they leave out, and where the cost levers are.

All reference rates below are for us-east-2 (the stack default) and assume a 730-hour month. Other regions typically land within ~5–15% of these figures.

The Fixed Monthly Baseline

These resources bill by the hour (or month) whether your app serves one request or one million. A minimal stack (256 CPU / 512 MB Fargate micro, no database) runs ~$31.68/mo:

Resource	Math	Monthly
Fargate task (0.25 vCPU + 0.5 GB)	(0.25 × $0.04048 + 0.5 × $0.004445) × 730 hrs	~$9.01
Application Load Balancer (base + ~1 LCU)	($0.0225 + $0.008) × 730 hrs	~$22.27
Secrets Manager (1 app secret)	1 × $0.40	$0.40

Optional fixed add-ons to the baseline:

Addition	Monthly
Background worker service (second identical Fargate task)	doubles Fargate to ~$18.02 — but scales to $0 when its SQS queue is empty
RDS Postgres (db.t4g.micro + 20 GB gp3 + managed secret)	~$13.98 + $0.40 secret
Valkey caching (add db:redis, cache.t4g.micro)	~$9.49

So a typical full stack (web + database + Valkey) lands around ~$55.55/mo, and the CLI’s estimate always reflects your actual terraform/ directory — container size, worker, database, secrets, and fixed-cost addons included.

On the Lambda target (--target lambda) there is no Fargate or ALB baseline at all: compute and the API Gateway HTTP API bill purely on use, so a no-database project idles at ~$0.40/mo (one Secrets Manager secret) and an Aurora-backed one at ~$0.80/mo. The pre-flight estimate shows this as Fixed Baseline: ~$0.80/mo (RDS: $0.00, Secrets: $0.80 + API GW & Lambda usage) — the trailing caveat is the reminder that requests, not hours, are the real bill.

On the static target (--target static) the fixed baseline is $0.00/mo: no Fargate, no ALB, no secrets vault. The pre-flight preview shows Fixed Baseline: $0.00/mo (Usage-based only via S3/CloudFront) — S3 storage and requests plus CloudFront requests and egress are the entire bill, all usage-metered.

What the Estimate Leaves Out (Usage Billing)

Anything that scales with traffic is billed on use and intentionally excluded from the fixed number:

• Addons: storage:s3 (storage, requests, CloudFront egress), db:dynamodb (requests, storage, backups), queue:sqs (requests past the 1M free tier), ai:bedrock (per-token inference), email:ses ($0.10 per 1,000 emails sent), cron (Fargate seconds per scheduled run). Each add run prints its own billing drivers.
• Data transfer: outbound traffic and CloudFront egress beyond free tiers.
• Logs & images: CloudWatch Logs ingestion (14-day retention is configured) and ECR image storage (~$0.10/GB-mo) — usually cents, plus gc cleans up orphans.
• Traffic spikes: ALB capacity units above the ~1 LCU baseline, and RDS backup storage past the free allowance.
• Serverless compute (Lambda target): API Gateway HTTP API requests ($1.00 per million) plus Lambda request charges and GB-second compute time. Sporadic traffic costs pennies; sustained high-concurrency traffic can overtake the ~$31/mo Fargate baseline — see the Fargate vs Lambda tradeoffs.

Rule of thumb: the fixed baseline is your floor; side projects with modest traffic typically land within a few dollars above it.

Which target costs less for my workload?

• Side project / internal tool (dozens to hundreds of requests a day): Lambda, by a mile — pennies a month against ~$31+ of idle ECS baseline.
• Steady product API (sustained traffic around the clock): Fargate — the flat baseline undercuts per-request billing once concurrency stops dropping to zero.
• Spiky or unpredictable traffic (launches, webhooks, batch-driven): Lambda absorbs bursts with no capacity planning; just mind the database-connection note in Fargate vs Lambda tradeoffs.
• Static site (docs, marketing pages, SPA exports): the static target — $0.00/mo baseline with pennies of S3/CloudFront usage.

Cost Savers Built Into the Stack

• No NAT gateway. Tasks run in public subnets behind the ALB security group instead of behind a ~$33/mo NAT — the single biggest saving versus a conventional VPC layout.
• Micro defaults, scale up deliberately. Fargate micro, single-AZ db.t4g.micro, and single-node Valkey keep the floor low; grow container size or add read replicas only when metrics say so.
• Workers scale to zero. The SQS-driven worker parks at 0 tasks (and $0 compute) when the queue drains — you pay for background capacity only while jobs exist.
• Sleep idle environments. sleep [env] scales ECS services to 0 and stops RDS, printing the exact hourly/monthly compute savings (e.g. $20.69/mo for a micro web service plus Postgres: $9.01 Fargate + $11.68 RDS compute paused); wake [env] restores everything. Storage ($2.30/mo), the ALB ($22.27/mo), Valkey ($9.49/mo, no pause API), and secrets keep billing while asleep, and AWS auto-restarts stopped databases after 7 days — the CLI shows the exact restart timestamp.
• PR previews self-destruct. Each open pull request runs a full copy of the stack (~$31+/mo each while open, mostly the extra ALB), so previews are destroyed automatically when the PR closes. Close stale PRs and run gc to catch leftovers.
• Serverless-first addons. DynamoDB on-demand, SQS, SES, Bedrock, and scheduled cron jobs cost nothing at rest — prefer them over always-on resources when the workload fits.

Keeping Estimates Accurate

• The README.md estimate refreshes automatically on every add and apply — if you hand-edit terraform/, re-run apply (or --dry-run) to re-sync it.
• Estimates follow your real config: bigger --size at init, extra secrets, and new addons all flow into the number on the next run.
• For hard budget enforcement, pair the CLI estimates with an AWS Budgets billing alarm in the console — the CLI tells you the expected spend, AWS tells you the actuals.