That’s not true. If you want to set up a complete data platform on European infrastructure today—with a lakehouse, pipelines, governance, dashboards, and notebooks—all the components are already in place.
Mature, open source, and in production at petabyte scale at companies you’re familiar with. So the question isn’t whether it’s possible, but which components you choose and how you connect them.
Step 1: Start with the infrastructure
If your computing and storage run on a U.S. hyperscaler, then your “sovereign” data platform is little more than a label. An EU region doesn’t change the legal reality: U.S. providers are subject to the CLOUD Act, regardless of where the data center is located. Microsoft confirmed this as recently as 2025 during a hearing in the French Senate.
Fortunately, the European offerings have become more robust. For a data platform, you only need two things from your cloud:
S3-compatible object storage, where your tables will be stored
Managed Kubernetes
Several European providers now offer both of these out of the box. In addition, there’s a tier of Dutch providers that specialize by sector, from healthcare to government. They know the compliance requirements inside and out, and the lines of communication are shorter than you’ll ever get with a hyperscaler. Be sure to check out the local options as well!
Practical tip: Define your infrastructure in code from day one, using OpenTofu, for example. An exit plan without portable code often remains just a plan.
Step 2: Realize that the major platforms won’t play along
Most teams figure this out too late. You’ve chosen your European cloud, and now you want to set up your trusted data platform on it. That’s not possible. The major commercial data platforms run exclusively on U.S. hyperscalers, and some don’t even leave their home cloud at all. The “sovereign regions” that the providers offer as an alternative aren’t worthless, but they still run on hyperscaler infrastructure with a control plane managed by a U.S. company.
Let’s clear up a misunderstanding: this isn’t about data formats. Open table formats are now well supported by the major platforms, so that’s no longer a distinguishing factor. It’s about the layer surrounding them— the control plane, your pipelines, your governance, and the jurisdiction under which all of that falls.
And that’s where the opportunity lies. Because everything those platforms do, open source can do as well.
Step 3: Choose the components, layer by layer
At its core, a data platform consists of six or seven components wrapped in a single framework. For each layer, there’s a mature open-source option that runs wherever containers run.
Storage: Your tables are stored as open files in a bucket that belongs to you, with schema evolution, time travel, and ACID transactions. Apache Iceberg has become the de facto standard here; it originated at Netflix and is used on a petabyte scale by companies including Apple. Delta Lake and Apache Hudi are the best-known alternatives.
Catalog: Tracks schemas, snapshots, and permissions. Lakekeeper, Apache Polaris, and Project Nessie are the options—all self-hosted and offering fine-grained access control across multiple query engines. Pay close attention to who controls the project governance, as this is the prime area where lock-in occurs.
Query engine: Trino runs distributed SQL directly on your tables, without copies or extracts. It’s just SQL, so your analysts can work with it from day one. Be honest about your data volume, though: most workloads fit perfectly on a single machine—without a cluster—using, for example, DuckDB.
Orchestration: Scheduling, retries, and lineage. Apache Airflow is the standard; Dagster is the modern challenger. More important than the choice of tool is ensuring your pipeline code is stored in your own Git repository. Code that exists only in your vendor’s UI will be lost in the event of an exit.
Identity: One login for everything—otherwise, no one will use the platform securely. You want SSO, MFA, and role-based access control across all components. Keycloak is the veteran and somewhat heavier to operate; Zitadel is more modern and is itself a European open-source project. That takes care of authentication, but not authorization: if you want to specify exactly which data and actions are permitted for each user, you’ll need to add a policy layer , such as OpenFGA or OPA.
Dashboards: Apache Superset connects to Trino and provides the organization with charts, dashboards, and SQL exploration. Metabase is more user-friendly for non-technical users, while Grafana excels at operational metrics.
Notebooks: JupyterHub is the self-hosted standard. Marimo is the modern upgrade to that: reactive notebooks stored on disk as regular Python files, making them Git-friendly and easy to deploy to production. The polished alternatives are almost all American SaaS offerings.
Seven layers, with multiple mature options per layer. None of these are a compromise or a European alternative that’s almost there. These are the projects the rest of the world uses as well.
Step 4: Put it together
First, lay the foundation: managed Kubernetes, an object storage bucket—all in Terraform code. Then deploy the components, as each tool provides an official Helm chart. Use ArgoCD for GitOps. Connect your identity provider to each component via OIDC, and define your authorization rules as code in OPA or OpenFGA, so that permissions are managed centrally rather than reconfigured for each tool. Next, point your query engine and orchestrator to the catalog, register the bucket as a warehouse, and connect your dashboards and notebooks.
That initial Helm chart takes a good week’s worth of work, and with some AI assistance, you can have a POC up and running within a day. Then comes the real work: identity management, authorization, TLS, network policies, backups, monitoring, failover, a tested upgrade path, and so on, and so on. Every project has its own release cadence, and security patches don’t wait for your sprint planning.
That doesn’t have to hold you back, because it’s platform management—the same work that every serious IT team already does. It’s just something to budget for honestly up front, in addition to the licensing costs of the alternative. In many migration discussions, the focus is exclusively on the price of software and hardly at all on the cost of management.
Practical Choices vs. Contractual Agreements
You can’t buy sovereignty with a checkmark in a contract. Contractual agreements are also important, but it remains primarily an architectural choice.
And you build it in by keeping three things under your own control:
Your data, in your own bucket, in open formats, with keys managed by you
Your pipeline code, in your own repository
Your infrastructure, in code that you can redeploy elsewhere. Cloud-agnostic.
If you do that, (almost) every tool is replaceable. And you don’t have to wait for anything, because the components are already available today.
Buy or build?
You can assemble everything mentioned above yourself, and for teams with platform engineering capabilities, that’s a great approach. At Databaas, we deliver this stack as a single package, running on the European cloud of your choice or on your own Kubernetes cluster. We handle the deployments, upgrades, and security patches. You keep the keys, because your data is in your bucket and your pipeline code is in your repository. No lock-in, so you can leave whenever you want. We’ll even help you leave. You won’t want to, but it’s possible.
Databaas x Data Expo
On September 9 and 10, we’ll be at Data Expo at the Jaarbeurs Utrecht. Stop by if you want to know what a platform like this would look like for your organization—and how quickly it can be up and running. I’d be happy to discuss it with you there.
This blog post is a contribution from Databaas, the sovereign data and AI platform built on open source and open standards. Databaas provides a complete data stack—from lakehouse and pipelines to dashboards, notebooks, and AI—running on European infrastructure and governed by European law. Built and managed from Utrecht by Wolk. Find more inspiration at www.databaas.eu or visit Databaas at Data Expo.
Author: Stijn Meijers, Co-founder, Chief Product Officer @ Wolk & Databaas