GKE provides a hosted control plane for your cluster. The "master" for a cluster is really a collection of standard services (kube-apiserver, kube-scheduler, kube-controller-manager) along with some environment specific controllers (ingress, autoscaling, etc). GKE hides the implementation of these services behind the cluster endpoint, which is how they are exposed to your nodes (kubelet and kube-proxy) and also you as the cluster user. GKE handles the provisioning of a public IP for the cluster endpoint and the configuration needed to make the the nodes communicate with it. The control plane is not run in your project so you don't see any of the underlying resources that are consumed running it.

If you would like full control over the configuration and deployment of your control plane (on a single "master" machine or spread across multiple "master" machines) you can use a community sponsored tool like kops or kubespray to deploy your cluster.