How to Launch an EC2 Instance With Terraform

The one resource that launches an instance

Everything else on a DevOps job (security groups, load balancers, auto scaling) builds on a single primitive: the aws_instance resource. It needs two required fields - an AMI id to boot from and an instance_type for the hardware shape:

resource "aws_instance" "web" {
  ami           = "ami-12345678"
  instance_type = "t2.micro"
}

The aws_instance.web label is how you reference this instance elsewhere in your config. The AMI is the machine image (OS + preinstalled software) the instance boots from; t2.micro is a small, cheap general-purpose type - the right default when you just need compute to exist.

Expose the instance id as an output

A bare instance isn't much use until something else can find it. Terraform outputs surface a resource attribute so other modules, scripts, or a terraform output call can read it. The most useful one to publish is the instance id:

output "instance_id" {
  value = aws_instance.web.id
}

Now aws_instance.web.id is available to wire into a security group attachment, a target group, or a DNS record later. This is how real stacks compose - one resource's output becomes another's input.

Apply and verify

With the provider already initialized, apply the plan. -auto-approve skips the interactive confirmation so it runs unattended:

cd /workspace
terraform apply -auto-approve

Terraform prints the plan, creates the instance, and writes it to terraform.tfstate. Read the output back with:

terraform output -raw instance_id

Then confirm the instance actually came up by querying it directly. A freshly launched instance should report state running:

aws ec2 describe-instances \
  --query 'Reservations[].Instances[].[InstanceId,InstanceType,State.Name]'

You should see your instance id, t2.micro, and running. If describe-instances returns nothing, the apply didn't create the resource - re-run terraform apply and check for a syntax error in main.tf.

Why declare it in code at all

You could click the same instance together in the AWS console in a minute. The reason every team past a handful of people uses Terraform instead: the config is the source of truth. terraform apply is repeatable and reviewable, terraform destroy tears the whole thing down cleanly, and the diff in a pull request shows exactly what infrastructure is changing before it changes. Console clicks leave no record and don't scale.

A production-grade version of this instance adds more: a subnet_id to place it in a VPC, vpc_security_group_ids to control traffic, tags for cost tracking, and often user_data to bootstrap software on first boot. Start with the two required fields, get a green apply, then layer the rest on.