EKS in 60 seconds: AWS runs the brain, you choose the muscle
An EKS cluster is split across two accounts' worth of responsibility. The control plane, meaning the Kubernetes API servers and etcd, runs in an AWS-managed account across several Availability Zones. You never see those machines: no SSH, no etcdctl, no patching. AWS keeps it highly available, patches it and backs it up. Reveal the steps with the right arrow.
Your side is the VPC. The worker nodes run in your subnets, and EKS creates network interfaces (ENIs) in subnets you designate so that the AWS-operated control plane can reach your kubelets for exec, logs and webhooks, and so the nodes can reach the API server. Every request to the API is authenticated against AWS IAM, which is the first thing that feels different from other platforms.
The practical result is a spectrum of ownership. With managed node groups you still choose instance types and trigger node upgrades. With Karpenter you add intelligent node provisioning but run it yourself. With Auto Mode AWS operates the nodes, the node autoscaling and several core add-ons too. With Fargate there are no nodes at all, but also constraints. Module 2 compares them.
Cost model in one line: you pay a per-cluster hourly fee for the control plane (the price depends on whether the Kubernetes version is in standard or extended support) plus the compute, storage and network you use. Check the AWS pricing page for current figures.