The scenario is a familiar one. A company built its production environment on AWS: EC2, RDS, S3, a few Lambda functions, an SQS queue. The bill fluctuates and nobody can explain why, a customer asks where its data is hosted, and the leadership team asks a simple question: could we leave if we wanted to?

The honest answer is rarely a plain yes or no. The real question is not "AWS or a European cloud?". It is: which workloads should stay on AWS, which ones can move, and under what conditions?

Below you will find the concepts not to confuse, the criteria for choosing a provider, the order in which to move applications, a six-step plan, a total cost model and a decision checklist. These are the questions addressed by CERVOX Services' scoped cloud engagements, but the method applies whoever runs the project.

Why companies consider leaving AWS

An AWS exit is rarely driven by a single reason. Six motivations come up most often, and none of them on its own requires a full migration:

  • Data governance: where the data is, who can access it, under which jurisdiction.
  • Dependency on proprietary services: the more the architecture relies on Lambda or DynamoDB, the costlier a switch becomes.
  • Unpredictable costs: transfers, ancillary services and forgotten resources sometimes outweigh compute.
  • Accumulated complexity: simplifying does not always require migrating.
  • Reversibility: being able to switch provider, or restart elsewhere after a major incident.
  • External requirements: customer clauses, public tenders, sector-specific constraints.

To our knowledge, no regulation of general application requires private companies in France or the Benelux to leave AWS. Requirements do exist for specific cases (certain public-sector data, regulated sectors, contractual commitments): check them for your own situation, with legal counsel if needed.

Six concepts not to confuse

ConceptDefinitionWhat it does not guarantee
Data residencyData is stored and processed in a given country or area, for example the European Union.Neither the jurisdiction the provider is subject to, nor control over access.
Digital sovereigntyAn organisation's ability to keep control of its data, operations and technology choices, including in the face of foreign laws.It is not a label: it is measured criterion by criterion.
SecurityTechnical and organisational measures protecting confidentiality, integrity and availability.A secure provider can still be misconfigured by its customer.
GDPR complianceMeeting the controller's obligations: legal basis, processors, transfers, data subject rights.No hosting provider is "GDPR compliant" on your behalf.
ReversibilityA demonstrated ability to retrieve your data and run your service elsewhere, within a known time and at a known cost.A contract clause that has never been tested proves nothing.
SecNumCloud qualificationA qualification issued by ANSSI, France's national cybersecurity agency, to a specific cloud offering, with security requirements and protection against non-EU law.It covers one specific offering, not a provider's whole catalogue.

According to ANSSI's FAQ (in French), SecNumCloud-qualified offerings are listed in its official register: that is what to check, not a sales brochure. France's data protection authority, the CNIL, reminds organisations that transfers of personal data outside the EU remain possible when properly framed. For the United States, the July 2023 adequacy decision (the Data Privacy Framework) was upheld by the EU General Court on 3 September 2025; an appeal remained possible, so check the status of the proceedings.

To measure sovereignty, the European Commission published a Cloud Sovereignty Framework in October 2025: 48 criteria grouped into eight objectives (strategic, legal and jurisdictional, data and AI, operational, supply chain, technological, security and compliance, environmental). Designed for the EU institutions' own procurement, it is not binding on companies but provides a useful grid for questioning a provider.

Can a European cloud really replace AWS?

Yes, for a large share of conventional infrastructure: virtual machines, containers, PostgreSQL or MySQL databases, object storage. The main European providers offer these building blocks, often as managed services. No, or not without redesign, for architectures built on proprietary AWS services with no direct equivalent. The useful comparison is therefore service by service, not brand against brand.

AWS has its own answer too

On 15 January 2026, AWS announced the general availability of the AWS European Sovereign Cloud, an infrastructure physically and logically separate from its other Regions, with a first Region in the German state of Brandenburg, operated by EU residents. The entity that runs it remains a subsidiary owned by Amazon, as InfoQ points out. Depending on your requirements, this option may or may not be enough: it is one scenario to compare among others.

Criteria for assessing a provider

CriterionQuestions to ask
Location and controlWhere are the data centres, backups and support teams? Which legal entity signs the contract, and who controls it?
Certifications and qualificationsWhich ones (ISO 27001, HDS, SecNumCloud…) and on which offerings exactly?
Service catalogueDo the services you need exist as managed services: Kubernetes, databases, object storage, message queues, DNS, secrets management?
ResilienceHow many availability zones per region? Which backup, replication and disaster recovery mechanisms?
Support and contractSupport levels, availability commitments (SLAs), penalties, exit procedure and timelines.
CostsCompute, storage, egress, IP addresses, load balancers, support, licences: what is billed, and how?
PortabilityStandard or compatible APIs, export formats, a maintained Terraform or OpenTofu provider?

This table does not rank providers: the right choice depends on the workload. A few examples of what their documentation lets you verify at the source:

This information changes: check it when you decide.

Which applications should move first?

Start with what is portable and low-criticality, finish with what depends most on AWS. The order follows from two variables: how dependent the application is on proprietary services, and how critical it is to the business.

SituationPortabilityUsual approach
Containerised applicationsHigh, subject to storage, networking and secretsNatural candidates for the pilot
Conventional virtual machines (EC2)GoodRehost (lift-and-shift) or light replatforming
Managed databases and services (RDS, ElastiCache, OpenSearch)Medium: an equivalent often exists, but versions, extensions and settings differReplatform, with thorough testing
Proprietary services (Lambda, DynamoDB, SQS, Step Functions, Bedrock)LowRefactor, replace, or keep on AWS

The vocabulary comes from AWS's own migration framework, which distinguishes seven strategies, the "7 Rs": retire, retain, rehost, relocate, repurchase, replatform, refactor. A lift-and-shift migration (rehosting) moves an application as-is, without changing its architecture: it is the fastest option, but it also carries over the existing flaws.

A container is not a portable application

A container makes the code and its libraries portable. It does not carry what surrounds it: persistent volumes, the load balancer, the IAM roles that grant access to S3, secrets held in AWS Secrets Manager, AWS SDK calls in the code, CloudWatch alarms. A "dockerised" application can therefore remain tightly bound to AWS.

A proprietary service is a service whose API and behaviour exist with only one provider. The dependency is measured by the cost of replacing it: rewriting code, moving data, testing again.

Dependencies to inventory before migrating

  • APIs and SDKs: calls to AWS services in code and scripts.
  • Identity: IAM roles and policies, service accounts, federation.
  • Storage: S3 buckets, policies, versioning, events, lifecycle.
  • Networking: VPCs, VPNs, DNS, firewalls, IPs allow-listed by partners.
  • Secrets and keys: Secrets Manager, KMS, and who holds the keys.
  • Databases: engine, version, extensions, size, write volume.
  • Monitoring and backups: alerts, logs, retention.

The six-step migration method

A controlled migration follows six steps: assessment, classification, target architecture, pilot, phased migration, stabilisation. Each step produces a deliverable that informs the decision on the next one, including the decision to stop if the facts do not justify going further.

StepWork doneDeliverableRisk avoided
1. Assessment of the existing setupInventory of resources, dependencies, network flows, costs, data, criticality and security requirementsInventory, dependency map, cost baselineDiscovering a critical dependency mid-cutover
2. ClassificationFor each workload: migrate, modernise, keep on AWS or retireDecision matrix per applicationMigrating what should have been kept or retired
3. Target architectureProvider and services, networking, security, storage, backups, monitoring, recovery planArchitecture document and running-cost estimateReproducing existing flaws with another provider
4. Pilot migrationOne representative but manageable application; functional, performance, security and restore testsPilot report, validated procedures, measured real costsScaling up an untested method
5. Phased migrationCutover windows, data synchronisation, rollback criteria, monitoring of key indicatorsMigrated applications, cutover logService outage or data loss
6. Stabilisation and reversibilityDocumentation, knowledge transfer, restore test, return of access rights, decommissioning of AWS resourcesOperations handbook, evidence of successful restoresDepending on a contractor, or paying for forgotten AWS resources

No standard timeline fits every migration. The duration of each step depends on scope and can only be estimated seriously after the assessment.

Step 1: measure before you size

Do not copy the capacity provisioned on AWS. Collect several weeks of metrics covering activity peaks (CPU, memory, storage used, latency, network traffic): this baseline is used to size the target, then to check that it delivers. That is the purpose of a read-only AWS account audit: establishing what is in place before proposing anything.

Steps 4 and 5: the database sets the timeline

A cold migration stops the application, copies the data and restarts with the new provider: simple, but it requires a downtime window. Continuous replication copies the data, then ongoing changes until cutover: downtime is short, but replication lag and consistency must be monitored.

With PostgreSQL, logical replication makes the second scenario easier, within documented limits: the schema and DDL commands are not replicated, and sequences must be brought up to date before any switchover. Validation goes beyond counting rows: for invoicing, you also check amounts, payment statuses and the absence of duplicates.

Two indicators are set per application before the project starts. RTO (Recovery Time Objective) is the maximum acceptable downtime. RPO (Recovery Point Objective) is the maximum amount of data the business accepts to lose, expressed as a duration.

Rollback is not a DNS change

As soon as the target accepts writes, going back to AWS requires reverse replication or reconciliation. The rollback plan states who decides, on what criteria, and how long AWS stays available. Do not shut anything down before the target's restore capability and performance have been validated.

How much does an AWS to European cloud migration cost?

There is no universal price. The cost depends on the number of applications, the data volume, the dependencies to replace and the required availability level. What can be standardised is the method: a total cost of ownership calculated over 12 to 36 months, comparing several scenarios.

Total cost of ownership (TCO) adds up everything a solution costs over a given period: setup, running costs, human operations and transition. Seven items to cost:

  1. Assessment and scoping
  2. Design and preparation: architecture, landing zone, infrastructure as code.
  3. Migration and application changes
  4. Data transfer
  5. Running the new infrastructure: compute, storage, networking, support.
  6. Dual running during the transition.
  7. Post-migration operations: monitoring, updates, on-call, skills.

The list price of a virtual machine is therefore not enough to compare two clouds: a vCPU does not perform the same everywhere, disks, egress, IPs and load balancers come on top, and any managed service missing at the target will have to be run by your own team.

Egress fees are no longer necessarily a barrier

Since March 2024, AWS has offered, on request to its support team and after review, free data transfer out to the internet when moving off AWS. The conditions are set out in AWS's FAQ: the scheme covers moving all of your data, within a set period after approval, and AWS may change it at any time. For a partial migration, check eligibility directly with AWS.

The legal framework is moving in the same direction. The EU Data Act has applied to cloud services since 12 September 2025. It caps the notice period for switching provider at two months, provides for a transition period of no more than 30 days (extendable in justified cases), limits switching charges to directly incurred costs until 11 January 2027, then prohibits them from 12 January 2027. Early termination penalties may still apply and amendments have been proposed at EU level: check the rules that apply to your contract.

A worked example, entirely hypothetical

Fictitious figures, which do not reflect any real pricing and are only there to show how the calculation works. Assumptions: current AWS infrastructure at €6,000 per month; estimated European target infrastructure at €4,200 per month; €800 per month of additional operations for services you run yourself; assessment at €8,000 and migration at €30,000; three months of dual running; an AWS optimisation costing €10,000 that brings the bill down to €4,800 per month. 36-month horizon, no discounting, no price changes.

ItemA. Stay as isB. Optimise AWSC. Migrate
One-off costs€0€10,000€38,000
During the project (3 months)Included belowIncluded below€18,000 (AWS) + €12,600 (target running in parallel)
Recurring running costs36 × €6,000 = €216,00036 × €4,800 = €172,80033 × €5,000 = €165,000
Total over 36 months€216,000€182,800€233,600

In this example, migrating costs more than the status quo over 36 months: the net saving of €1,000 per month only pays back the €50,600 of one-off and dual-running costs after about 51 months. Optimising AWS is the cheapest scenario. Yet changing a single assumption is enough to reverse the outcome: with a target at €3,000 per month, the net saving rises to €2,200 and the break-even point falls to about 23 months.

An AWS bill analysis should therefore come before any exit decision. And the financial calculation is not the only criterion: a customer or regulatory requirement can justify a costlier migration. That is the logic behind a "Move to AWS, stay, or leave" assessment, which compares three costed 12-month options with their assumptions.

A hybrid architecture, which keeps some workloads on AWS and moves others to a European provider, is often the rational outcome of this calculation. Its own costs (networking between environments, duplicated tooling) also belong in the TCO.

The mistakes that derail migrations

MistakeConsequencePrevention
Migrating without a dependency inventoryProduction outages, forgotten integrationsFull assessment and mapping before any decision
Comparing prices without networking or operationsExpected savings that never materialiseA 12 to 36-month TCO, including people costs
Underestimating the database migrationPostponed cutover, downtime window exceededA full rehearsal of the migration on a copy
Confusing containerisation with portabilityAWS dependencies discovered after the moveReview of code, IAM, storage and secrets
Neglecting backups and rollbackNo way out if things go wrongA written rollback plan and an independent backup
Shutting down AWS too earlyLoss of the fallback optionRestore and performance validated before decommissioning
Forgetting documentation and knowledge transferDependence on a contractor or a single personAn operations handbook delivered and reviewed by the team

Decision checklist before starting a migration

  • Objectives: the reason (cost, compliance, customer requirement, reversibility) is written down, and quantified where possible.
  • Obligations: actual regulatory and contractual requirements have been checked.
  • Data: sensitive data is identified, with its current and target location.
  • Dependencies: every AWS service in use has an equivalent, a replacement plan or a decision to keep it.
  • Costs: a TCO compares at least three scenarios: stay, optimise, migrate.
  • Availability: RTO and RPO are defined for each application.
  • Provider: location, qualifications, SLAs, support and exit terms are confirmed in writing.
  • Strategy: the pilot is chosen, the batch order defined, the rollback criteria written.
  • Governance: one owner per step, named deliverables and a formal sign-off before AWS resources are decommissioned.

Frequently asked questions

Can a company leave AWS for a European cloud?

Yes. The common building blocks (virtual machines, containers, relational databases, object storage) have equivalents at European providers. The difficulty depends on the proprietary services in use, the data volume and the availability requirements. A prior assessment shows what can move, what must be redesigned and what is better kept.

How long does a migration take?

There is no reliable duration without an assessment. It depends on the number of applications, the dependencies to replace, the data volumes and the acceptable downtime windows. A pilot measures the real pace before the rest is planned.

Does a European hosting provider guarantee GDPR compliance?

No. Compliance depends on processing activities, contracts, access, security measures, processors and transfers. The choice of provider is one component of it, not a guarantee. Conversely, using AWS is not in itself contrary to the GDPR.

Do you need a SecNumCloud-qualified provider?

Only if your obligations or customers require it, or if your risk analysis justifies it. The qualification covers a specific offering: check ANSSI's official list to make sure the offering you are buying is the one that is qualified.

Conclusion: start with an assessment, not a hosting provider

A successful migration starts with an inventory, not with choosing a provider: what is running, what depends on it, what it costs, what the company must guarantee. A phased approach then lets you decide on facts (dependencies, risks, costs, requirements, feasibility). The outcome may be a full migration, a hybrid architecture or an AWS optimisation: three legitimate decisions, provided they are costed.

CERVOX Services works on these topics through scoped engagements, on your own environment and with no subscription. The AWS migration to a European cloud engagement includes an audit of your existing AWS infrastructure, an analysis of dependencies, costs and portability, a migration plan to the target hosting provider, then migration of the resources included in the validated scope, with documentation and handover to your team. You receive a written estimate within 24 business hours, after a first 20-minute conversation.

If the question applies to your infrastructure, describe your situation in a few lines: we will tell you whether we can help, and how.