I Tested Managed Kubernetes Across EU Providers - Most Are Cooking With Water

In January 2026, I needed managed Kubernetes for a new project. The requirements were simple: European provider, managed control plane, reasonable pricing, and good enough security to not keep me up at night. I come from a regulated environment, so “good enough security” means something specific to me: private endpoints, API access controls, audit logging, the basics.

I didn’t want a hyperscaler. Not because AWS or GCP are bad at Kubernetes (they’re good), but because I didn’t want to explain CLOUD Act exposure in a compliance review, and I didn’t want to pay hyperscaler prices for a cluster that runs in Frankfurt anyway. Surely there’s a European provider that can do managed K8s well, at a fraction of the cost. Right?

I couldn’t find an honest answer. Every comparison I found was either marketing from the providers themselves, or a blog post from 2023 with outdated pricing. Nobody was actually spinning up clusters, running benchmarks, and reporting what they found. So I built a spreadsheet.

That spreadsheet became EU Cloud Cost.

From Spreadsheet to 31 Providers

The first version was embarrassingly simple. A static page built with Astro, some JSON files with pricing data, and a basic table that let you configure a Kubernetes cluster and see what it would cost across a handful of providers. I built it in a few days.

Then it grew. European cloud is surprisingly crowded. I kept finding providers I’d never heard of. The JSON files got bigger. The comparison table got wider. Today it covers 31 providers, generates ~560 static pages, and builds in under 3 seconds. The pricing data is open source on GitHub - and to my surprise, people actually started sending PRs. Seven stars and counting, which feels both modest and oddly motivating.

But the interesting part isn’t the tool. It’s the data.

The Numbers Nobody Puts Side by Side

Once you normalize pricing across 31 providers for the same cluster configuration, patterns emerge that no individual provider’s marketing page will show you.

The price spread is absurd. A 3-node cluster with 2 vCPU and 4 GB RAM per node costs anywhere from ~€19/month (DIY on Aruba Cloud) to ~€249/month (AWS EKS). That’s a 13x difference for equivalent compute. Even within managed-only offerings, you’re looking at a 4-5x spread.

Control plane pricing varies wildly. 12 out of 21 managed K8s providers offer a free control plane, including OVHcloud, Scaleway, IONOS, and Azure. Others charge between €30 and €75/month, roughly in line with AWS EKS at $73. The spread is wide enough that control plane cost alone can swing your monthly bill by the price of another worker node.

Egress is the hidden tax. Unless it isn’t. 53% of EU providers offer unlimited or effectively unlimited egress: OVHcloud, Scaleway, netcup, Civo. All unlimited. Hetzner gives you 20 TB free. Meanwhile, AWS charges per GB after 100 GB/month. If you’re running anything with meaningful outbound traffic, this single line item can dwarf the compute cost difference.

Storage IOPS vary by 400x. This one caught me off guard. The spread across providers goes from 1,000 IOPS (Infomaniak) to 400,000 IOPS (Azure). Within European providers alone, netcup benchmarks at 99,500 IOPS while Infomaniak caps at 1,000. That’s not a rounding error. It’s the difference between a responsive database and one that stalls under load. No provider’s marketing page puts this number front and center, but it matters more than CPU pricing for most workloads.

Certifications are a mess. 90% of providers have ISO 27001. It’s table stakes. But only 4 have BSI C5 (the German government standard), only 2 have HDS (French healthcare), and exactly 1 has FINMA (Swiss financial regulation). If you’re in a regulated industry, your shortlist gets very short very fast. HIPAA? Hyperscalers only. SOC 2? Eight providers, most of them not European.

GPU availability is broader than expected. 23 out of 31 providers offer GPU instances. Exoscale alone has six GPU families, including RTX 6000 PRO. This isn’t hyperscaler-exclusive territory anymore.

Actually Testing the Clusters

Numbers on a page are one thing. I also started doing hands-on reviews: spinning up the same 3-node cluster at each provider, deploying ArgoCD, running fio benchmarks on the storage, timing provisioning, and opening support tickets to see how fast they respond. Same methodology, every provider. I even threw in a self-managed Talos Linux setup on Hetzner partly to see if DIY on cheap VMs could compete with managed offerings, and partly because I’d been wanting to try Talos anyway.

The results were humbling. Every provider has trade-offs that aren’t in the marketing material. One provisioned a cluster in under 2 minutes but had the API endpoint open to the internet with no way to restrict it. Another had a free control plane but storage so slow that database workloads would choke. A third had solid networking and security but their Terraform provider was still at v0.x after years.

I even went back and softened one of my reviews because my regulated-environment background made me too harsh. The gaps were real, but I was grading a development-friendly provider against enterprise security standards it never claimed to meet.

They’re all cooking with water. Some have better recipes than others, but nobody has figured out the perfect managed Kubernetes offering in Europe yet.

Unexpected Conversations

Building a comparison tool that covers 31 providers means some of those providers notice. I’ve had conversations with people I’d normally never meet, including CEOs of cloud companies. These are real teams trying to build competitive infrastructure in a market dominated by three American giants. You develop a certain respect for what they’re doing, even when you’re writing about their shortcomings.

What’s Next

More providers to test, more benchmarks to run. The hands-on reviews take about a week each (setup, benchmarking, writing, fact-checking), but they’re the most useful part. Knowing that a provider’s block storage has 2.5-second p99.99 write latency is the kind of detail that saves you from a production incident.

Next up is IONOS. They offer a free control plane, show up in a lot of German enterprise conversations, and I genuinely don’t know if they’re decent or just well-marketed. That’s exactly the kind of question this project exists to answer.

I also want to be honest: this is fun. Spinning up clusters, breaking things, writing about what I find. The regulated environment taught me to care about the details. Building EU Cloud Cost lets me put that instinct to use.

The pricing data is open source on GitHub. If your favourite EU provider is missing or the data is wrong, PRs are welcome.

---

Compare providers at eucloudcost.com. Provider reviews: OVHcloud | Exoscale | Infomaniak | Hetzner + Talos