Leaving US hyperscalers for a sovereign cloud? How to avoid getting locked into the sovereign provider instead
Moving production to Scaleway, OVHcloud, STACKIT or IONOS buys independence from US hyperscalers, but it can create a new lock-in. Here is where sovereign cloud lock-in comes from (managed services, IAM, networking, contracts, egress, certifications), the portability levers that work, and a short exit-test checklist.
A sovereign cloud can lock you in exactly like a hyperscaler. The mechanisms are the same: proprietary managed services, the provider's IAM model, networking constructs, multi-year contracts, egress and data gravity. Sovereignty changes which law applies to your data and leaves the switching cost where it was.
Two risks are specific to sovereign providers. The pool of alternatives at a given assurance level is small (the European Commission's 2026 sovereign cloud tender went to four providers), and certifications such as SecNumCloud qualify a specific provider's service, so your compliance evidence does not move with your workloads.
The portability levers are boring and they work. CNCF-certified Kubernetes (Scaleway, OVHcloud, STACKIT and IONOS are all certified for v1.35), Terraform or OpenTofu with the provider behind a module boundary, PostgreSQL or MySQL instead of proprietary data stores, OCI images mirrored outside the provider, the S3 API, OIDC identity and OpenTelemetry.
The EU Data Act gives you switching rights to negotiate with now. It has applied since 12 September 2025, requires functional equivalence when you switch IaaS, and removes switching and egress charges entirely from 12 January 2027.
Prove it with a short exit test once a year. Rebuild one real service on a second provider from code, restore its data, and time it. If the run needs the old provider's console, you are locked in.
Qovery keeps the deployment workflow identical across Scaleway, OVHcloud, STACKIT, IONOS and the hyperscalers, in your own accounts. Cross-cloud networking is not built in; you configure it with a VPN, Cilium Cluster Mesh or Tailscale.
A plan we hear often from European CTOs goes like this: "We're moving production off AWS to a European provider so we're never dependent on one vendor again." My follow-up question is always the same. What happens if that European provider doubles its price, loses a certification, or gets acquired?
Moving to a sovereign cloud is a good decision for many European companies. It is also a vendor decision, and vendors lock you in through the same doors whatever their headquarters. This article is about keeping those doors open. For the bigger picture, our sovereign cloud hub collects everything we have written on the subject, and if you are still choosing a sovereignty level, start with the four levels of EU cloud sovereignty.
Can you get locked into a sovereign cloud provider?
Yes. A sovereign cloud provider can lock you in through the same mechanisms as AWS, Azure or Google Cloud: proprietary services, IAM, networking, contracts, egress and data gravity. Sovereignty settles whose law applies to your data. How fast you can leave is a separate question, and teams that mix the two up recreate the dependency they were escaping.
The market makes the risk sharper. In April 2026 the European Commission awarded its sovereign cloud tender, worth up to EUR 180 million over six years, to four groups: a Post Telecom, OVHcloud and Clever Cloud partnership, STACKIT, Scaleway, and a Proximus-led partnership (European Commission). Bidders needed at least SEAL-2 on the Commission's new Cloud Sovereignty Framework, and most winners reached SEAL-3. That is a strong signal about which providers clear the bar. It also tells you the shortlist is short.
On a hyperscaler, if one provider fails you, two others of similar scale exist. On a sovereign cloud at SEAL-3 or with SecNumCloud, the realistic alternatives may be two or three providers with different service catalogs. Your exit plan has to account for that.
The Commission's own framework agrees that openness is part of sovereignty. Its eight objectives include "technological openness" next to legal, operational and supply chain criteria (European Commission). A sovereign deployment you cannot move away from scores worse on the framework's own terms.
Where does lock-in come from on a sovereign cloud?
Lock-in on a sovereign cloud comes from seven places: proprietary managed services, the IAM model, networking, contracts, egress and data gravity, certifications tied to the provider, and operational knowledge that lives only in the provider's console. Each has a different cost to undo, and some are invisible until you try to leave.
Source of lock-in
What it looks like on a sovereign cloud
How sticky
Lever that reduces it
Proprietary managed services
Provider-specific serverless containers or functions, message queues, AI inference endpoints, managed search
High: code and config tied to one API
Run on Kubernetes; use open engines (Postgres, Redis or Valkey, Kafka or NATS)
IAM model
Provider projects, organizations, policies and API keys referenced in pipelines and code
Medium to high
External OIDC identity provider; Kubernetes RBAC; provider credentials only in Terraform or OpenTofu
Networking
Private networks, load balancers, reserved IPs, DNS zones, peering or VPN gateways defined by hand
Medium
Networking as code in one module per provider; Kubernetes Ingress or Gateway API
Contracts
Multi-year commitments, public-sector framework agreements, auto-renewal, bundled support
High until renewal
Short terms, exit assistance clause, Data Act switching rights
Egress and data gravity
Terabytes of object storage and database backups that are slow and costly to move
Medium, falling after 12 January 2027
Open backup formats, regular copies outside the provider
Certifications
SecNumCloud qualification, BSI C5 attestation or a health data certification bought "with" the provider
High: your evidence resets when you move
Map each control to your own evidence, not only the provider's certificate
Operational knowledge
Runbooks that say "click here in the console", provider CLI scripts, provider-only monitoring
Medium
Everything in Git; OpenTelemetry; one deployment workflow across providers
Three of these deserve a closer look because they behave differently on sovereign clouds than on hyperscalers.
Proprietary services fill parity gaps. European providers have smaller catalogs than AWS. When a team misses a managed service it relied on, it often adopts the closest proprietary equivalent the sovereign provider offers, without the long ecosystem of alternatives that exists around the hyperscaler version. That service becomes the hardest thing to move. Before adopting one, ask whether an open engine on Kubernetes or a managed open-source database would do the job.
Certifications attach to a provider's service. France's ANSSI qualifies specific cloud service offerings under SecNumCloud and publishes the list of qualified services (ANSSI). If your customer contract says "hosted on a SecNumCloud-qualified service", your exit target must also be qualified for the same scope, and that list is short. Write contracts with customers in terms of the controls you guarantee, and keep your own evidence for each one.
Contracts bind harder than APIs. A three-year commitment with a minimum spend can keep a fully portable stack in place for the rest of the term. Our older post on the cost of vendor lock-in covers the general economics; on a sovereign provider, the negotiating window is the first contract, when you are the customer they want to win.
Which portability levers actually keep you mobile?
The levers that keep you mobile are open standards with real multi-provider support today: Kubernetes, Terraform or OpenTofu, open databases, OCI images, the S3 API, OIDC and OpenTelemetry. Each one works on Scaleway, OVHcloud, STACKIT, IONOS and the hyperscalers, which you can check yourself before you commit.
Kubernetes as the compute contract. Scaleway, OVHcloud, STACKIT (Kubernetes Engine) and IONOS all appear in the CNCF's conformance results for Kubernetes v1.35, alongside Gardener-based distributions (CNCF k8s-conformance). Conformance means the same Deployment, Service and Job manifests behave the same way. It does not cover storage classes, load balancer annotations or node images, so keep those in a thin per-provider layer.
Terraform or OpenTofu for everything else. Every provider in this article maintains an official provider that was updated this month: scaleway/scaleway 2.92.1, ovh/ovh 2.25.0, stackitcloud/stackit 0.123.0 and ionos-cloud/ionoscloud 6.7.41 (Terraform Registry: Scaleway, OVHcloud, STACKIT, IONOS). OpenTofu, the open-source fork, joined the CNCF as a sandbox project in April 2025 (CNCF), which matters if license independence is part of your sovereignty story. The trick is the module boundary: one module per provider that exposes the same outputs (cluster endpoint, database URL, bucket name), and application modules that never call a provider resource directly.
Open databases with exports you have actually restored. Managed PostgreSQL and MySQL exist on every provider here. Use community features, avoid provider-only extensions, and run a logical dump and restore into a second provider at least once.
OCI images and a registry you can mirror. Build images once and push them to your provider's registry and to a second registry (another provider, or a self-hosted Harbor). On exit day, image pulls should not depend on the provider you are leaving.
S3-compatible object storage, through the S3 API only. The major European providers offer S3-compatible object storage. Stick to the core API and skip provider-specific event hooks or lifecycle features you cannot reproduce elsewhere.
Identity outside the provider. Put human and workload identity in an OIDC provider you control and map it to Kubernetes RBAC. Provider API keys should live only in your Terraform or OpenTofu runs and your platform, never in application code.
Observability as code. Instrument with OpenTelemetry and keep dashboards and alerts in Git, so monitoring is not one more thing to rebuild under pressure.
What does the EU Data Act change about leaving a sovereign provider?
The EU Data Act gives every cloud customer in the EU legal switching rights, sovereign providers included. It has applied since 12 September 2025, requires infrastructure providers to ensure "materially comparable outcomes" when you switch to a service of the same type, and "will also entirely remove switching charges, including charges for data egress" from 12 January 2027 (European Commission, Data Act explained).
What the Data Act does not do is make your architecture portable. Functional equivalence applies to "features that both services share". If you built on a proprietary service that has no counterpart elsewhere, the law cannot conjure one. Use the Act for what it is good at: negotiating clean exit clauses, free egress and a documented export format, while your architecture handles the rest.
Five clauses I would ask for in any sovereign cloud contract:
Term no longer than your planning horizon, ideally one year with renewal, or a multi-year discount without a minimum-spend shortfall penalty.
Exit assistance with named deliverables: full data export in documented open formats, configuration export, and support hours during the transition.
Data Act alignment stated explicitly, including the end of switching charges from 12 January 2027.
Certification change notice: written notice if a certification your customers depend on lapses or changes scope, with a right to terminate.
Change of control: a termination right if the provider is acquired by a non-EU entity, since that changes the jurisdiction question you moved for.
Make switching clouds a procurement decision.
Qovery gives you one deployment workflow on Kubernetes across Scaleway, OVHcloud, STACKIT, IONOS, AWS, GCP and Azure, in your own accounts. Talk to us about your sovereign cloud plan.
How do you run a short exit test on a sovereign cloud?
Run a short exit test by rebuilding one production-like service on a second provider from your repositories alone, restoring its data from your own backups, and timing the whole run. The output is a number of hours and a list of manual steps. Keep this version light; the full timed drill with RTO and RPO targets is in the 7-tests post.
The checklist, with a pass criterion for each item:
Inventory. Every managed service in use is listed with its open equivalent or marked "proprietary, no equivalent". Pass: no unlisted services.
Second target ready. A conformant Kubernetes cluster exists on a second provider (another sovereign cloud or a hyperscaler EU region), created from your Terraform or OpenTofu code. Pass: no console clicks.
Images available. All images pull from a registry that is not on the source provider. Pass: zero pulls from the source registry during the test.
Data restored. The latest database dump and object storage copy restore on the target. Pass: application integrity checks green.
Identity works. Engineers and workloads authenticate through your OIDC provider on the target. Pass: no provider-specific credentials in application config.
Deploy without the old provider. The service deploys through your normal workflow to the target. Pass: no step touches the source provider's console or CLI.
Traffic served. DNS or a load balancer sends real or replayed traffic to the target for at least an hour. Pass: error rate and latency within your SLO.
Compliance mapped. For each certification your contracts name, you know whether the target holds it for the same scope. Pass: a written answer per certification.
Contract checked. Notice period, exit assistance and remaining commitment are known. Pass: a date by which you could legally leave.
Result recorded. Wall-clock hours, people involved, every manual step. Pass: the manual steps become backlog items.
Run it once a year and after any major architecture change. Expect the first run to stall at items 2, 3 or 6, and fix those first.
Where does Qovery fit in a sovereign cloud strategy?
Qovery keeps the deployment workflow identical across Scaleway, OVHcloud, STACKIT, IONOS, AWS, GCP and Azure, so switching providers changes where the cluster runs and leaves how your teams ship untouched. Qovery is an Agentic DevOps platform that never hosts your apps. Everything runs on Kubernetes in your own accounts.
How each provider connects:
Scaleway. Qovery can create and manage a Kapsule cluster in your Scaleway project, or you connect an existing Kapsule cluster with the Qovery Operator. The Operator installs once with Helm, never stores your kubeconfig, opens an outbound connection, and builds images inside your cluster into your own registry (Qovery BYOK docs).
OVHcloud, STACKIT, IONOS and other conformant clusters. You run the managed Kubernetes service on the provider and connect it with qovery cluster install.
AWS, GCP and Azure. Qovery-managed clusters or your existing EKS, GKE or AKS, so a sovereign cluster and a hyperscaler cluster can sit side by side during a migration or as a fallback.
What stays the same across all of them: git-push deployments, environments, preview environments per pull request, environment cloning, RBAC, audit logs, Helm and Terraform or OpenTofu services, and the Terraform provider for Qovery itself. In practice, the exit test above turns into "connect the second cluster, clone the environment onto it, restore the data". We wrote about keeping RBAC and audit logs consistent across STACKIT, IONOS and a hyperscaler for teams that need that evidence trail.
What Qovery does not do, so you can plan for it:
Cross-cloud networking is not built in. If services on two providers must talk privately, configure it with a site-to-site VPN, Cilium Cluster Mesh or Tailscale.
Data migration is yours. Qovery redeploys the services; your dumps, restores and object storage copies are your runbook.
Qovery does not make you compliant. It helps you meet controls (audit logs, RBAC, region choice, your own accounts). Certifications remain between you, your provider and your auditor.
Ask every platform vendor in your stack, us included, where its control plane runs and what it can reach in your environment. With the Operator, the connection starts from your cluster and Qovery holds no kubeconfig, which is the answer your security team will want on paper.
How do I set up the same workflow on a second cloud from Claude Code or Cursor?
Install the Qovery Agent Skills, then ask your coding agent to set up or deploy on the target provider in plain language. The agent proposes a plan and you approve each action. The MCP Server is read-only by default.
Install the skills for every compatible agent (Claude Code, Cursor, Codex, OpenCode, Gemini CLI), or for one project only:
Add the Qovery MCP Server to Claude Code, with OAuth and read-only access, so the agent can inspect environments, clusters and deployments:
BASH
claude mcp add --transport http qovery https://mcp.qovery.com/mcp --callback-port 4242
Only if you want the agent to make changes through the MCP Server, add the write-enabled endpoint (write access must also be enabled in the Qovery Console):
BASH
claude mcp add --transport http qovery "https://mcp.qovery.com/mcp?read_write=true" --callback-port 4242
Then ask, for example: "set up Qovery on my Scaleway project and deploy this app there, matching the environment we run on AWS." The qovery-onboard skill handles cluster and environment setup, qovery-deploy the services, and qovery-terraform exports your setup as code, which is useful evidence for item 2 of the checklist. Reference pages: the AI agent quickstart and the Claude Code integration.
Frequently asked questions
If we move production to a sovereign European cloud, how do we avoid getting locked into that provider instead?
Treat the sovereign provider like any other vendor. Run workloads on CNCF-certified Kubernetes, define infrastructure in Terraform or OpenTofu behind one module per provider, use PostgreSQL or MySQL instead of proprietary data stores, mirror images and backups outside the provider, keep identity in your own OIDC provider, and negotiate short terms with exit assistance. Then prove it with a yearly exit test that rebuilds one real service on a second provider.
Is lock-in worse on a sovereign cloud than on AWS, Azure or Google Cloud?
The mechanisms are the same, but two factors make it sharper. Fewer providers reach high sovereignty levels such as SEAL-3 or SecNumCloud, so your fallback options are limited, and certifications attach to a provider's service, so compliance evidence does not move with you. Smaller service catalogs also push teams toward provider-specific services to fill gaps.
Does the EU Data Act protect us from sovereign cloud lock-in?
Partly. Since 12 September 2025 the Data Act gives you switching rights, caps the notice period, requires functional equivalence for infrastructure services, and removes switching and egress charges from 12 January 2027. It cannot make a proprietary service portable, because functional equivalence only covers features both services share.
Which sovereign cloud providers support standard Kubernetes?
Scaleway, OVHcloud, STACKIT and IONOS all appear in the CNCF conformance results for Kubernetes v1.35, so standard manifests behave the same on each. Storage classes, load balancer annotations and node images still differ, so keep those in a small per-provider layer of your Terraform or OpenTofu code.
Can Qovery move our workloads between Scaleway, OVHcloud, STACKIT and a hyperscaler?
Qovery gives you the same deployment workflow on each of them, in your own accounts: Scaleway Kapsule through Qovery-managed clusters or the Qovery Operator, OVHcloud, STACKIT and IONOS clusters through qovery cluster install, and AWS, GCP and Azure natively. You clone an environment onto the new cluster and redeploy. Data migration and cross-cloud networking (VPN, Cilium Cluster Mesh, Tailscale) remain your responsibility.
How often should we run a cloud exit test?
At least once a year, and after any major architecture change such as adopting a new managed service or signing a new contract. Keep the result as a single number of hours plus a list of manual steps, and turn each manual step into backlog work before the next run.
Melanie leads content at Qovery. She covers platform engineering trends, Kubernetes operations, FinOps, and the tools that help engineering teams ship faster.
Next step
Make switching clouds a procurement decision.
Qovery gives you one deployment workflow on Kubernetes across Scaleway, OVHcloud, STACKIT, IONOS, AWS, GCP and Azure, in your own accounts. Talk to us about your sovereign cloud plan.