Elastic Compute (ECS)

Cloud servers that provision in minutes and scale on demand. Multiple instance families cover general-purpose, compute-optimized, memory-optimized, and AI-accelerated workloads, so you spend time on your business — not on server upkeep.

Core capabilities

What cloud servers are actually for

Use a cloud server whenever you need control of the operating system: web apps and backend APIs, self-managed databases, crawlers and batch jobs, or anything that depends on specific kernel parameters or a custom image.

If you are only hosting a static site or a purely functional API, object storage plus a CDN — or a fully managed service — usually costs less. A cloud server bills for the whole machine, and idle time is billed too.

Choosing a spec you will not have to keep changing

Start from the memory-to-CPU ratio. General-purpose covers most web and middleware workloads; compute-optimized suits heavy builds and transcoding; memory-optimized fits caches and large in-memory databases; GPU instances handle training and inference.

Do not size for peak traffic. Pick the tier that covers your normal load and let autoscaling absorb the spikes — on pay-as-you-go billing, permanently paying for one tier higher usually costs more than a handful of scale-out events.

Bandwidth and disks are the costs people forget

On many accounts, public egress and block storage end up costing more than the instance itself. Decide the bandwidth billing model up front: traffic-based pricing wins for egress-heavy workloads, fixed bandwidth wins for steady, predictable traffic.

Pairs well with

Frequently asked questions