AWS Lambda cost calculator

Why Lambda surprises teams in audits

Lambda looks free. The pricing per invocation is $0.20 per million. The pricing per GB-second is six zeros and a three-digit number. Then a single chatty function at 1 GB memory and 200 ms duration, called 200 times per second, runs $5,200 a month and nobody on the team knew.

Three knobs do most of the work: memory tuning, ARM64, and duration. Provisioned Concurrency is a tax most teams pay without needing it. We wrote up the full pattern in the three knobs that matter.

Frequently asked

• How is Lambda priced exactly?

  Two components. First, $0.20 per 1 million invocations on x86. Second, GB-seconds: memory in GB multiplied by execution time in seconds, billed at $0.0000166667 per GB-second on x86. ARM64 (Graviton) is 20 percent cheaper on the GB-second rate. The free tier of 1M invocations and 400,000 GB-seconds per month still applies, indefinitely.

• Why does memory tuning matter so much?

  Lambda allocates CPU proportional to memory. A function set at 128 MB gets a fraction of a vCPU, a function at 1769 MB gets a full vCPU, anything above that gets multi-core. Doubling memory often halves duration on CPU-bound work, leaving total cost unchanged or lower while latency drops by half. The AWS Lambda Power Tuning state machine finds the sweet spot in 30 seconds per function.

• Should I use ARM64 (Graviton)?

  Almost always yes for Python, Node.js, Go, and Rust functions. ARM64 is 20 percent cheaper per GB-second. Performance is equal or better on most workloads. Switch the architecture flag in your Terraform or SAM template, redeploy, you are done. The only blockers are functions that bundle x86-only native binaries, rare in 2026 but worth checking.

• When is Provisioned Concurrency worth it?

  Two use cases. One, latency-critical user-facing endpoints where cold starts hurt. Two, very high steady-state traffic where the GB-second savings on warm invocations exceed the provisioned cost. For everything else, on-demand is cheaper. Calculate the break-even: provisioned concurrency cost per hour divided by the GB-second saving per invocation gives you the invocation rate where it pays off.