Every AWS project eventually runs into the same question: where should this actually run? The answer is not the same for every workload, and picking the wrong one either costs you more than it should or slows your team down with infrastructure work nobody wanted to do.
Here is how the three main compute options actually differ.
EC2 — Full Control, Full Responsibility
EC2 gives you a virtual machine. You choose the operating system, install what you need, and manage patching, scaling, and security yourself.
- Best for applications with steady, predictable traffic
- Best when you need specific OS-level configuration or licensed software
- Requires you to manage scaling groups, patching, and monitoring
- Billed by the hour or second, running whether or not it is doing work
If your application looks the same today as it did last month in terms of load, EC2 is usually the most cost-predictable choice.
Lambda — Pay Only When Code Runs
Lambda runs your code in response to an event — an API call, a file upload, a scheduled trigger — and disappears when it is done. There is no server to patch or manage.
- Best for event-driven workloads, APIs with unpredictable traffic, and background jobs
- Scales automatically from zero to thousands of concurrent executions
- Billed per millisecond of execution, so idle time costs nothing
- Has execution time limits and cold-start latency that matter for some use cases
For workloads that run occasionally or spike unpredictably, Lambda often costs a fraction of an equivalent EC2 setup.
Fargate — Containers Without Managing Servers
Fargate runs your Docker containers without you provisioning or managing the underlying EC2 instances. You define the container, and AWS handles the infrastructure.
- Best for containerized applications that need more control than Lambda allows but less operational overhead than EC2
- Good middle ground for microservices architectures
- Billed based on the vCPU and memory your containers actually reserve
- Removes patching and instance management entirely
A Practical Way to Decide
- Predictable, always-on workload with specific OS needs → EC2
- Event-driven, short-duration, unpredictable traffic → Lambda
- Containerized microservices without wanting to manage servers → Fargate
Many businesses end up running all three side by side — EC2 for a core application, Lambda for background processing, and Fargate for containerized services. The mistake is picking one and forcing every workload into it. Match the compute model to what the workload actually needs, and the AWS bill usually takes care of itself.
