vCPU-hours and memory-hours for serverless containers.
Oversized Fargate tasks quietly overpay on vCPU and memory every second they run. Right-size the task and scale count with demand. Compare a self-managed cluster on the Kubernetes cost calculator and container hosting calculator.
The AWS Fargate Cost Calculator estimates your monthly serverless container bill by combining the four inputs you provide: the number of running tasks, the vCPU per task, the memory (GB) per task, and the hours per month each task runs. Fargate prices compute and memory separately, so the tool multiplies your total vCPU-hours and GB-hours by their respective rates and sums them across all tasks. The main cost drivers are therefore straightforward: more tasks, larger vCPU or memory allocations, and longer running hours each push the total up in a roughly linear way, making it easy to see which lever moves your bill most.
The key trade-off to watch is right-sizing versus uptime. Because you pay for allocated vCPU and memory whether or not the task uses them, over-provisioning a task that sits mostly idle wastes money every hour it runs. Before committing, model a few scenarios: trim vCPU and memory to what a task actually needs, and question whether tasks truly require full-month, always-on hours or could scale down during quiet periods. Small per-task reductions multiply across many tasks and many hours, so modest cuts often produce the largest savings.
Per vCPU-hour (about $0.04048) and per GB-hour of memory (about $0.004445), billed per second with a one-minute minimum. You choose vCPU and memory per task, and pay only while tasks run — there are no idle EC2 hosts underneath.
Fargate removes server management and is cheaper for spiky or low-utilisation workloads because you pay per task-second. For high, steady utilisation, running your own EC2 hosts under ECS/EKS can be cheaper per container. It is a utilisation question.