How to run Kubernetes on premises for beginners

So you have taken my advice…
and decided to migrate your Kubernetes cluster to a European bare-metal provider, but you don’t know where to start. Let me give you a few tips.
Bare-metal or virtual machines
The first decision to make is whether you want to run the Kubernetes cluster on bare-metal or virtual machines. At Digitalis, we have been steadily moving away from virtual machines toward bare-metal, so bear in mind I may be a bit biased.
Bare-Metal Kubernetes
✅ Pros:
- Better performance (no virtualization overhead)
- Direct hardware access (ideal for latency-sensitive workloads)
- More control over resources and configurations
- Higher pod density — Can run more than 110 pods per node (unlike many VM-based setups, which hit limits due to virtualization overhead).
❌ Cons:
- Higher maintenance (manual hardware provisioning, updates, and scaling)
- Less flexibility (harder to migrate workloads or resize clusters)
Virtual Machines (VMs) for Kubernetes
✅ Pros:
- Easier scalability (quick VM provisioning and resizing)
- Simpler maintenance (hypervisors handle hardware abstraction)
❌ Cons:
- Performance overhead (due to virtualization layer)
- Less control over underlying hardware
- Higher cost
Deployment phase
You have made your decision and you’re sticking to it. Now, it’s time to start deploying but you don’t know how. Let’s explore some of the most popular options available.
Rancher
Rancher is a complete software stack for Kubernetes that addresses the need to manage multiple clusters. It can provision to all the hypervisors, and it does a pretty good job for on-premises too.

If you’ve gone the VMs route, their Terraform provider is solid, and you can orchestrate the entire cluster creation with Terraform. In my book, that’s a huge plus.
Another interesting thing is you can choose between RKE and k3s. See below.
What I don’t like about Rancher is their UI. It’s very slow and clunky. Also, upgrading Rancher versions is sometimes difficult. There are quite a few bugs, and the team is slow to fix them.
K3s
K3s is a lightweight, production-ready Kubernetes distribution certified for high availability. It’s ideal for resource-constrained environments, remote locations, and IoT appliances, even in unattended scenarios.
Because of its simplicity, it is straightforward to install and maintain. You can have a cluster up and running in no time from Rancher or from the command line using:
curl -sfL https://get.k3s.io | sh -
Remember, this is not recommended for large production workloads but for IoT or development environments.
Kubespray
Kubespray is a set of Ansible playbooks designed to deploy a production-ready Kubernetes cluster. They are well-maintained and relatively easy to use, especially if you have experience with Ansible.
For Virtual Machines, you’ll still need to use either your own Ansible playbooks or terraform code to create and manage them.
Like any other Ansible code, it’s not hard to create pipelines to run it and automate the process.
Kops
I haven’t used Kops in years, but this was how I started deploying Kubernetes clusters. It’s very powerful and flexible. The downside is it has no support for bare-metal.
Kubeadm
Kubeadm is essentially the official way to install Kubernetes. It’s the method taught in CNCF’s Kubernetes certifications, and for good reason — it works reliably and efficiently. That said, while you can find approaches using Ansible or Terraform to automate cluster creation, these methods aren’t as straightforward or well-integrated.
Bonus tip
If you need a Web UI for managing the clusters, you can still use Rancher or the Kubernetes Dashboard though I recommend having a look at https://kubesphere.io/

Final words
You have many options and many decisions to make. It all boils down to your level of commitment to the project, your skills and how much you would to have it automated.

Since I’m the type who would automate even beer opening if I could, I prefer solutions like Kubespray or Rancher that offer full orchestration capabilities. As for bare-metal versus virtual machines, both are viable, but bare-metal wins every time when it comes to raw power and cost efficiency.
Contact our team for a free consultation to discuss how we can tailor our approach to your specific needs and challenges.
I, for one, welcome our new robot overlords