AWS Fargate
AWS's serverless container runtime — run containers on ECS or EKS without managing the underlying servers, trading per-node control for zero cluster ops.
What is Fargate?
AWS Fargate runs containers without you managing the servers under them. You define a task (container image, CPU, memory) and Fargate provisions the compute, runs it, and bills for exactly the resources requested for exactly the runtime — no EC2 instances to size, patch, or scale. It’s the serverless container answer, sitting between Lambda (functions, event-driven) and self-managed EC2/EKS nodes (full control, full ops), and it works as a launch type for both ECS and EKS.
The trade it makes
Fargate removes the entire node layer from your responsibility: no capacity planning, no OS patching, no cluster autoscaler tuning, no bin-packing math — each task gets its own isolated (Firecracker microVM-backed) environment. The cost of that convenience is twofold. Price per unit of compute is higher than raw EC2 — you pay a premium for not running nodes, which is worth it until sustained scale makes the premium exceed the ops savings, the familiar serverless crossover. Less control: no daemonsets in the classic sense, no custom kernel/host tuning, no GPU on Fargate (GPU workloads stay on EC2 nodes), and per-task overhead that makes very large fleets of tiny tasks inefficient.
Where it fits
Fargate shines for teams that want containers without a platform team: variable or spiky workloads, batch jobs, dev/staging environments that shouldn’t run nodes 24/7, and organizations minimizing operational surface. It’s less ideal for steady high-density fleets (EC2 nodes bin-pack cheaper), GPU/AI workloads, and anything needing host-level access. The common mature pattern mixes launch types — Fargate for bursty and low-ops services, EC2 node groups for the dense, steady, or GPU-bound core.
What people get wrong
- Fargate for large steady fleets — the per-task premium exceeds the ops savings past a scale threshold.
- Expecting GPU or host access — those workloads belong on EC2-backed nodes.
- Ignoring task right-sizing: Fargate bills the CPU/memory you request, so over-requesting is pure waste on every task, every run.
Primary source: AWS Fargate documentation
Historical figures and technical concepts for informational purposes only. Not technical, professional, legal, or financial advice. Sources: Official Documentation.