What the AWS free tier covers, and where the bill starts

The AWS free tier is not one offer. It is three separate mechanisms sharing a name, and most unexpected first invoices come from mistaking one for another.

The three kinds of free

Always free. Permanent monthly allowances that do not expire with the account. Lambda requests and compute time, DynamoDB storage, CloudWatch metrics and alarms, and SNS publishes all fall into this group. These reset monthly and continue indefinitely.

Time limited. Allowances that run from the date the account was created and stop on the anniversary. Historically this covered the classic combination of a small EC2 instance, EBS storage, S3 storage and RDS hours. The clock runs whether or not you use them.

Short term trials. A fixed window, often 30 to 90 days, or a fixed quantity of a specific service. These are per service and start when you first use the service rather than when the account was created.

AWS has revised the shape of the new account experience more than once, including moves toward a credit based welcome plan for new sign ups. Because the terms change, treat the AWS free tier page as the authority for current numbers and treat this article as a map of how the mechanisms behave.

Where the bill usually starts

In practice, a small number of things account for most surprise charges on an otherwise free tier account.

• Public IPv4 addresses. AWS charges hourly for every public IPv4 address, including the one attached to a running instance. It is a small number that runs continuously, and it is the single most common line on a first invoice.
• EBS volumes on stopped instances. Stopping an instance stops compute billing. The attached volume keeps existing and keeps costing.
• Snapshots. Easy to create, easy to forget, billed per gigabyte month.
• NAT gateways. Created almost by accident when you follow a VPC wizard. They bill per hour plus per gigabyte processed and are not free tier eligible.
• Data transfer out. There is a monthly free allowance, but egress from a busy service passes it quickly. Our guide to data transfer costs covers this across providers.
• Load balancers. An idle Application Load Balancer still bills hourly.
• Instance size drift. Free tier EC2 hours apply to specific small instance types. Launching one size up puts the whole instance on standard rates.

How to check what you are consuming

The Billing console has a Free Tier page that lists each allowance, your month to date use and a forecast. It is genuinely useful and underused. Two habits help:

• Enable Cost Explorer early. It takes up to 24 hours to populate, so turning it on when you already have a problem is too late.
• Tag everything from the start. Without cost allocation tags, a multi project account becomes a single unexplained number.

What the free tier is good for

It is a genuinely capable learning environment. You can run a small web application, a managed database, a queue, a Lambda based pipeline and the monitoring around all of it without spending anything meaningful. For certification study or for evaluating whether a service fits, it is more than enough.

Where it stops being the right tool is when the work needs capacity rather than access. Free tier eligible instance types are small by design, and the vCPU quota on a new account is low, so a build farm or a batch job runs into a wall quickly. That wall is explained in AWS vCPU quotas explained.

A short checklist before you close the console

• Terminate, do not just stop, anything you were only testing.
• Delete orphaned EBS volumes and old snapshots.
• Release Elastic IP addresses you are not using. An unattached one costs more than an attached one.
• Check for NAT gateways and load balancers left behind by a tutorial.
• Look at the Free Tier page once a week while you are learning.

If you would rather compare what the equivalent starting offers look like elsewhere, the Google Cloud trial and Azure’s free credit work on quite different principles, and Oracle Cloud’s always free tier is the most generous of the group for persistent small workloads.