The short answer: for EU workloads, pick a provider where the control plane AND the worker nodes sit in EU datacenters under EU jurisdiction, where you can predict the monthly bill from the node count, and where an idle cluster does not quietly burn money. On RareCloud, a managed 3-node starter cluster costs €44.97 per month (3 x g-2vcpu-8gb workers), the control plane is included, and idle worker pools can be scaled down.
Why does "EU Kubernetes" mean more than a datacenter pin?
Three separate things hide behind the label:
- Data residency: your etcd state, volumes and backups physically live in the EU.
- Jurisdiction: the operating company answers to EU law, which is what GDPR processors care about.
- Billing reality: EUR invoices with correct VAT treatment, so finance does not bounce them back.
US hyperscalers' EU regions give you the first. European providers give you all three. RareCloud's cloud region runs OpenStack on our own hardware in Bucharest, operated by an EU company, invoiced in EUR or USD with EU VAT.
What should a managed control plane cost?
Hyperscalers charge for the control plane itself (EKS and GKE bill it per hour). Most European providers include it and charge only for workers.
RareCloud includes the managed control plane. You pay the sum of your worker nodes, period. Our clusters run on Gardener, the open-source Kubernetes-as-a-service engine originally built at SAP, which gives every cluster a managed, self-healing control plane without a per-cluster fee.
How do I estimate the real monthly bill?
Count nodes, multiply, add storage:
- Worker node g-2vcpu-8gb: €14.99 per month each (hourly billed, capped); the smallest worker, g-1vcpu-4gb, is €7.99
- 3-node starter cluster: €44.97 per month
- Block storage volumes: €0.05 per GB per month, metered hourly for as long as the volume exists
Hourly billing with a monthly cap means scaling down mid-month actually saves money, and scaling up for a launch weekend costs cents, not a month of fees.
What should happen to an idle cluster?
Staging clusters spend most of their life idle. Worker nodes are billed by the hour and a pool keeps at least 1 node, so scaling a pool down to that one node overnight or over the weekend cuts the bill straight away, while the included control plane stays up. Scaling back up is one change in the console, the API or the CLI. On a hyperscaler, the equivalent is an engineer remembering to tear things down on Friday.
When is a plain VPS the better answer?
Honest answer: often. If you run one application with a database and a reverse proxy, a €3.99 cloud server with Docker Compose will be simpler, cheaper and easier to debug than any cluster. Kubernetes earns its complexity when you have multiple services, rolling deploys, horizontal scaling or a team that already speaks it. We sell both, so we have no incentive to push you to the expensive one.
Frequently asked questions
Is the control plane really free? You pay for worker nodes only; the managed control plane is part of the product. If you want a highly-available control plane for production, that is an optional add-on at €30 a month, billed by the hour (about €0.041 an hour).
Can my AI tooling manage the cluster? Cluster lifecycle is exposed through the same REST API and console as everything else, with scoped tokens. The MCP server can create a cluster, add or resize node pools and mint kubeconfigs, asking you to confirm anything that costs money or rolls nodes, and the CLI fetches kubeconfigs and scales workers. A new cluster gets the newest Kubernetes version on offer unless you pick another one from the list at GET /v1/catalog/kubernetes-versions. Kubeconfig access works with any standard tooling.
How resilient is the platform? Clusters run on Gardener's self-healing control plane, failed worker nodes are replaced automatically, and the optional €30/month add-on runs the control plane in a highly available setup. Our Terms of Service carry a 99.9% uptime commitment; we don't dress it up as more than that.