Updated October 25, 2026
· ~3 min read
ECS has three places to run a task, and since September 2025 the middle one changed the decision. Fargate runs each task on capacity you never see. EC2 capacity providers run tasks on instances you launch, patch and replace. ECS Managed Instances sits between them: AWS chooses, launches, patches (every 14 days, in a maintenance window you can set) and replaces Bottlerocket EC2 instances for you, and you keep EC2's instance choice, GPUs and pricing options, for a per-instance management fee.
For most teams that removes the old trade-off — "Fargate for simplicity, EC2 for cost or GPUs" — and replaces it with a question about workload shape.
Fargate charges for each task's requested vCPU and memory by the second, which is ideal for spiky services, scheduled jobs and anything that scales to near zero. It is less ideal for many small, always-on tasks: you pay for each one's reservation separately, and nothing packs them together. That is where Managed Instances earns its fee — ten small services can share a couple of instances, Spot or Savings Plans apply to the EC2 part, and AWS still handles the patching. As a rule of thumb, look at Managed Instances for services that run around the clock at steady load or need more memory per vCPU than Fargate's sizes offer; keep bursty and low-volume services on Fargate.
It is a capacity-provider change, not a rewrite: the task definition and service stay; you add a Managed Instances capacity provider to the cluster and move the service's capacity provider strategy to it, ideally with a mixed strategy first so some tasks stay on Fargate while you watch. AWS documents the steps in the ECS developer guide. Check anything that assumed Fargate's isolation — tasks now share an instance with other tasks — and set the maintenance window so the 14-day instance replacement lands at a quiet hour.
Question 1 of 2
Does it need a GPU, a specific instance family or a custom AMI?
Start on Fargate: per-second billing for exactly what a task asks for, nothing to patch, and Fargate Spot at up to 70% off for interruptible work. Move a service to ECS Managed Instances when it needs a GPU or a particular instance family, when it runs steadily enough that packing tasks onto EC2 beats paying per task, or when you want EC2 Spot and Reserved pricing without operating the fleet. Keep EC2 capacity providers you manage yourself only for what the other two cannot do — custom AMIs, host-level agents you must control, or very specific kernel and networking setups.
Serverless Migration
Move from expensive always-on compute to serverless architecture: reduce costs, eliminate patching, and scale automatically.
AWS Cost Optimization & FinOps Sprint
Reduce your AWS bill by 20–60% while improving performance and reliability.
CI/CD Pipeline Build
Automated, secure deployment pipelines that get code to production faster without manual steps or weekend deployments.
We work with startups and growing companies on AWS architecture decisions. Tell us about your situation and we will recommend the right approach.
As of October 25, 2026. Provided as is, for general information, and accurate to the best of our knowledge on the date shown. Not legal, compliance, tax, financial or other professional advice. Prices, limits and vendor services change; check current vendor documentation before relying on anything here. Terms of use.