Take a look at https://rancher.com/docs/rancher/v2.x/en/cluster-provisioning/
In this use case, Rancher sends a request to a hosted provider using the provider’s API. The provider then provisions and hosts the cluster for you. When the cluster finishes building, you can manage it from the Rancher UI along with clusters you’ve provisioned that are hosted on-premise or in an infrastructure provider, all from the same UI.
If you don’t want to use a hosted Kubernetes provider, you can have Rancher launch a Kubernetes cluster using any nodes you want. When Rancher deploys Kubernetes onto these nodes, it uses Rancher Kubernetes Engine (RKE), which is Rancher’s own lightweight Kubernetes installer.
Using the hosted option, the provider creates the nodes and cluster, and presents you (Rancher) with the cluster already created. Some of them (Google does), may include the management (etcd, controlplane) for free, and you only provision the workers. The provider will take care of the version and upgrades and the VM maintenance. I don’t believe you get full access to the nodes themselves, but you get the Kubernetes dashboard and kubectl.
The alternative is that Rancher creates plain old instances and then uses RKE to create the management and workers. You will pay for all of the VMs (management and worker). Since they are normal VMs, you have full control over the VMs, and you are responsible for all maintenance and upgrades of the VMs.