What does Kubernetes on Scaleway really cost?

Anyone budgeting a Kapsule cluster for the first time looks at the price per node and is relieved. Then the first monthly bill arrives, and it is twice as high. Not because Scaleway is expensive, but because a cluster consists of more than nodes. This post walks through the five line items every honest calculation needs, ending with the one that is on no price list.

Item 1: The control plane

With Kapsule, the shared control plane is free. You pay nothing for the API server, etcd or the scheduler as long as the mutualized variant is enough for you. Dedicated control planes cost extra – according to Scaleway’s price list from around €80 per month (as of September 2026) – and come with two replicas, a 99.5% SLA and room for more nodes. They pay off when the cluster is business-critical or grows beyond 150 nodes. For most SaaS vendors with a few dozen namespaces, the shared variant is enough for years.

Item 2: Nodes

The obvious item. What matters is not the hourly price of a node type but how many nodes you actually need. Three rules from day-to-day operations:

• Requests determine the number of nodes, not limits. Kubernetes schedules pods by their requests. If you reserve 1 vCPU for every service even though it needs 100 millicores, you pay for ten times as many nodes as necessary.
• Two small nodes are more robust than one large one. If a node fails or is drained during an upgrade, the application keeps running on the other.
• Pool autoscaling saves money at night. A pool with a minimum of two and a maximum of five nodes covers the day and shrinks when nobody is working.

In the Clusterward cockpit, requests and limits are shown separately for each service so that exactly this calculation stays visible: reserved versus maximum; see Deployments.

Item 3: Load balancers

Every ingress controller needs a Scaleway Load Balancer, and it has a fixed monthly price. This is the item that surprises beginners. Two consequences:

1. One controller per cluster, not one per application. All hosts in a cluster go through the same load balancer, and the controller routes by hostname.
2. A second controller, such as Envoy Gateway alongside nginx, means a second load balancer. During a controller switch both run, and the second load balancer costs money until you remove the old controller.

Item 4: Databases, storage, registry

Managed Databases on Scaleway are priced by instance size, regardless of how many databases sit on them. For SaaS, that means one instance, many databases, one per customer. That is cheaper than one instance per customer and still separates the data cleanly through roles. Block Storage for volumes and Object Storage for uploads are billed per gigabyte, the Container Registry by stored images. All three are small compared to nodes and load balancers, but they grow when nobody cleans up old images and snapshots.

Item 5: Operating time

The most expensive item. If you run a cluster yourself, you pay for it in hours: maintaining Terraform, updating Helm charts, keeping an eye on certificates, adapting the onboarding script for the next customer, training the colleague who is supposed to take all of this over. These hours appear on no Scaleway bill, but they are the reason Kubernetes is considered expensive among mid-sized companies. And these hours often depend on a single person. What that means for technical leadership is covered under For CTOs & technical leads; how to change it, in Bus factor 1: when the cluster belongs to a single colleague.

The nodes cost three figures a month. The person who runs them costs five.

This is exactly where a control plane comes in. It does not replace Scaleway – it replaces the hours in between. Clusterward bills per cluster for this; everything else is unlimited; see Pricing.

How are the costs typically distributed?

The shares are empirical values from small and mid-sized installations, not a price list. Calculate with the Scaleway calculator for your node types and regions, and be honest about the fifth item. Leave it out, and you end up comparing Kubernetes with a platform like Vercel or Heroku and wondering why the platform wins.

How do you calculate without a price list?

Take a SaaS with twelve customers, each in its own namespace with one pod, plus an admin interface and an API. That makes fourteen workloads. With honest requests of 200 millicores and 256 MB per pod, together they need just under three vCPUs and four gigabytes, plus headroom for the ingress controller, cert-manager and metrics. Two nodes with four vCPUs and eight gigabytes each carry that with room to spare for an upgrade that drains one node. Add one load balancer, a small database instance with twelve databases and a few gigabytes of Object Storage.

The same installation with requests of one vCPU per pod, because "that was the default in the chart", needs four to five nodes for the same load. The bill doubles without anyone getting more. It is the most common cost mistake on Kubernetes, and it takes ten minutes to fix.

Which three mistakes cost money?

• One load balancer per application. Every ingress controller brings one along. If you install a controller per team or per customer, you pay for a load balancer per team or customer. One controller per cluster is almost always enough.
• Old images and snapshots. The registry grows with every build, snapshots with every schedule. Without a retention rule, within a year you are paying for hundreds of images that nobody will ever start again.
• One cluster per environment. Staging and production on separate clusters sounds clean, but it doubles the load balancers and the minimum number of nodes – and with a dedicated control plane, its cost too. Separate namespaces on one cluster are enough until a customer contractually requires something else.

Conclusion

Kubernetes on Scaleway is inexpensive when the requests are right, one load balancer is enough and nobody sinks hours into operations every week. The first two points are craft. The third you solve with a team that enjoys running Kubernetes, or with a control plane that does it for you. What getting started looks like is covered under Getting started.

Want to run the numbers for your setup? Tell us how many services, databases and environments you run. We will work out with you what that costs per month on Scaleway. Get your setup costed →

Sources and further reading

• Scaleway: Kubernetes Kapsule pricing
• Kubernetes docs: Resource management for Pods and containers
• Scaleway docs: Load Balancer service for Kubernetes
• Scaleway: Managed Databases pricing