ap-south-1 (Mumbai) Fargate rates
| Resource | On-Demand (USD) |
|---|---|
| per vCPU / hour — Linux x86 | 0.04656 |
| per GB / hour — Linux x86 | 0.00511 |
| per vCPU / hour — Linux ARM | 0.03725 |
| per GB / hour — Linux ARM | 0.00409 |
| per vCPU / hour — Windows | 0.09317 |
| per GB / hour — Windows | 0.01011 |
Billing is per second (1-minute minimum) from image pull to task stop. Formula per task: (vCPU × vCPU_rate + GB × GB_rate) × hours.
Valid vCPU / memory combinations
| vCPU | Memory options |
|---|---|
| 0.25 | 0.5, 1, 2 GB |
| 0.5 | 1–4 GB (1 GB steps) |
| 1 | 2–8 GB (1 GB steps) |
| 2 | 4–16 GB (1 GB steps) |
| 4 | 8–30 GB (1 GB steps) |
| 8 | 16–60 GB (4 GB steps) |
| 16 | 32–120 GB (8 GB steps) |
Fargate vs EC2
Fargate removes cluster capacity management and you pay only for requested task resources, but the per-vCPU rate is higher than an equivalent EC2 instance. Fargate usually wins for spiky, bursty or low-density workloads and small teams; a densely packed EC2 Auto Scaling group with Savings Plans is cheaper at steady high utilisation. Graviton and Spot narrow the gap significantly.
Questions
How is Fargate priced?
Per vCPU-hour and per GB-hour, per-second billing with a 1-minute minimum. Mumbai Linux/x86 ≈ $0.04656/vCPU-hr + $0.00511/GB-hr.
Is ARM cheaper?
Yes, about 20% lower for the same size, and often faster per vCPU. Images must be built for arm64.
How much is Spot?
Roughly 65% off on-demand, but tasks can be reclaimed with a 2-minute warning. Linux/x86 only.
What's missing from this estimate?
ALB/NLB, NAT Gateway, egress, ephemeral storage over 20 GB, ECR, CloudWatch, and the EKS control-plane fee.