AWS Lambda pricing calculator

Why the compute line grows

Lambda charges for memory multiplied by time. A function with 1 GB running for 300 ms uses 0.3 GB-seconds, and the meter runs while your code waits on anything: a database query, a Stripe call, an LLM that takes eight seconds to answer. In a typical API function most of the billed time is that waiting, not computation. A function that calls an LLM synchronously pays for every second the model is thinking.

Memory is the other half. Raising it also gives the function more CPU, so CPU-bound code often finishes faster and costs about the same or less. Code that mostly waits gets nothing from extra memory and simply pays more per millisecond. AWS Lambda Power Tuning, an open source tool, runs your function at several sizes and shows where the cost bottoms out, which beats guessing.

Since August 1, 2025, AWS also bills the INIT phase, the cold start, for functions packaged as ZIP files on managed runtimes. Before that it was free for them. A Node or Python function that loads a big SDK or connects to a database at startup now pays for that time on every cold start, which is why the calculator has its own line for it.

Arm is about 20% cheaper per GB-second and most Node and Python functions run on it without changes. Functions that ship native binaries need an Arm build first, so test before you switch.

We go through each of these lines, plus the API Gateway and CloudWatch charges that sit next to Lambda on the bill, in AWS Lambda pricing: where the bill actually comes from.

If Lambda is a bigger share of your AWS bill than this page says it should be, send us the bill. The fix is usually in the code: a slow query, a sync call that should be a queue, or a cron firing every second.