Stop, start or reboot an instance

Power an instance off, on again, shut it down cleanly or reboot it.

Last updated 18 November 2026View as Markdown

Replace i-3f9a2c07 with your instance's id. stop cuts the power; shutdown asks the guest OS to halt cleanly; reboot restarts the machine.

Prerequisites

  • In the console, also ec2:DescribeInstances to open the instance's page (without it the page is not found).
  • Permission for ec2:StartInstances (start), ec2:StopInstances (stop and shutdown) or ec2:RebootInstances (reboot).
  • start needs the instance stopped (or suspended); stop, shutdown and reboot need it running. Otherwise the request is refused with 400 "Cannot start instance in state: …" ("Cannot stop instance in state: …" for stop and shutdown, "Cannot reboot instance in state: …" for reboot). In the console the action is unavailable, with the reason, until the instance is in that state.
  • An instance that is a node of a Kubernetes cluster is refused with 409 "This instance is node … of the Kubernetes cluster …. Scale or delete its node group instead."
  • An instance on your own hosts whose machine does not exist yet is refused with 409 "Instance is not yet backed by a machine".

Change the power state

  1. Open the instance. The header shows Stop while it runs and Start while it is stopped.
  2. Choose Stop or Start, or open the ⋯ menu and choose Reboot.
  3. For Stop and Reboot, choose Stop or Reboot in the dialog that asks to confirm. Start asks nothing.

The console has no separate shutdown; its Stop cuts the power.

Result

The answer comes at once with the transitional state (pending, stopping or rebooting); the final state follows when the provider is done (running or stopped). If the action fails, the instance returns to its previous state and the page shows the error. A stopped instance keeps its disk, addresses and volumes. Instance hours are billed only while the instance is running, so a stopped instance costs nothing per hour; its volumes are still billed, and its snapshots and elastic IPs are still metered.

A machine of an auto scaling group that you stop here is replaced by the group.

Did this page help you?