GCP → European cloud
Migrating from GCP to a European cloud
Leaving GCP for a sovereign European cloud is usually a multi-provider move: compute and Kubernetes go to Scaleway or OVHcloud, but the two services that most often anchor a GCP estate (BigQuery and Spanner) have no EU-sovereign managed twin at all. That inversion is the whole planning problem: the parts you leaned on most are the parts hardest to leave. This page names which GCP services move cleanly and which force a re-platform.
Which GCP services have a true EU equivalent?
Graded conservatively against the EU-sovereign option that exists in production today, across providers.
| GCP | EU-sovereign | Verdict |
|---|---|---|
| Google Cloud Storage | Scaleway / OVHcloud Object Storage S3-compatible object storage; GCS access repoints by endpoint and SDK. | Equivalent |
| Persistent Disk | Scaleway / OVHcloud Block Storage Network block volumes with snapshots. Direct equivalent. | Equivalent |
| Compute Engine | Scaleway / OVHcloud / Hetzner instances General-purpose and GPU instances are a solved problem across EU providers. | Equivalent |
| Google Kubernetes Engine (GKE) | Scaleway Kapsule / OVHcloud MKS CNCF-conformant managed control planes; portable manifests move with little friction. | Equivalent |
| Cloud SQL | Scaleway / OVHcloud Managed PostgreSQL Managed PostgreSQL is well served; there is no AlloyDB-equivalent for its Postgres-compatible acceleration. | Partial |
| Cloud Run | Scaleway Serverless Containers Scale-to-zero container hosting exists for request-driven workloads; the surrounding ecosystem is thinner. | Partial |
| Cloud Pub/Sub | Scaleway Messaging & Queuing / NATS Managed queue/pub-sub exists, but Pub/Sub’s exact delivery semantics are re-tested, not assumed. | Partial |
| BigQuery | No managed twin (self-host ClickHouse / Aiven) No EU-sovereign serverless warehouse. ClickHouse is fast but a different operating model, not a drop-in. | No managed twin |
| Cloud Spanner | No managed twin No managed globally-distributed, externally-consistent SQL twin. Often the single hardest service to leave. | No managed twin |
| Firestore / Bigtable | No managed twin (self-host ScyllaDB) No managed serverless NoSQL twin; you run ScyllaDB/Cassandra. | No managed twin |
| Vertex AI | No managed twin (OVHcloud AI / Scaleway GPU) GPU, training and inference exist, but no one platform matches Vertex AI end to end. You assemble it. | No managed twin |
Where do the GCP APIs actually diverge?
The commodity layer repoints. These are the GCP-specific services with no sovereign twin. Plan the estate around them.
- BigQuery has no EU-sovereign managed twin: its serverless, separated-storage-and-compute warehouse retargets to ClickHouse (self-hosted or via Aiven), a fast but different operating model.
- Cloud Spanner’s globally-distributed, externally-consistent SQL has no managed European equivalent at all. This is often the single hardest service to leave.
- Firestore/Bigtable and Vertex AI have no drop-in: NoSQL moves to ScyllaDB, and end-to-end ML is assembled from OVHcloud AI or Scaleway GPU + Managed Inference rather than one platform.
- Pub/Sub retargets to Scaleway Messaging or NATS; delivery guarantees are re-tested rather than assumed.
Which of your GCP services move cleanly?
21 capabilities
Object storage
Hyperscaler Amazon S3 Azure Blob Storage Google Cloud StorageEU-sovereign path Scaleway Object StorageOVHcloud Object StorageVerdict MatureS3-compatible API. A drop-in target for most SDKs, backups and static assets.
Block storage
Hyperscaler Amazon EBS Azure Managed Disks Persistent Disk / HyperdiskEU-sovereign path Scaleway Block StorageOVHcloud Block StorageVerdict MatureNetwork block volumes with snapshots. Standard building block, well covered.
Virtual machines
Hyperscaler Amazon EC2 Azure Virtual Machines Google Compute EngineEU-sovereign path Scaleway InstancesOVHcloud Public CloudHetzner CloudVerdict MatureGeneral-purpose and dedicated instances are a solved problem in the EU.
Managed Kubernetes
Hyperscaler Amazon EKS Azure AKS Google Kubernetes Engine (GKE)EU-sovereign path Scaleway KapsuleOVHcloud Managed KubernetesVerdict MatureCNCF-conformant managed control planes. Portable workloads move with little friction.
Container registry
Hyperscaler Amazon ECR Azure Container Registry Google Artifact RegistryEU-sovereign path Scaleway Container RegistryOVHcloud Managed Private RegistryHarbor (self-hosted)Verdict MatureOCI registries are commodity. Self-hosted Harbor is a fully sovereign fallback.
Managed PostgreSQL
Hyperscaler Amazon RDS / Aurora Azure Database for PostgreSQL Cloud SQL / AlloyDB for PostgreSQLEU-sovereign path Scaleway Managed DatabaseOVHcloud Managed DatabasesAiven for PostgreSQLVerdict ViableStandard PostgreSQL is well served. Aurora-specific features (e.g. global database, serverless v2 autoscaling) have no exact twin, so plan around them.
Managed cache (Redis/Valkey)
Hyperscaler Amazon ElastiCache Azure Cache for Redis Memorystore for RedisEU-sovereign path Scaleway Managed Database for RedisAiven for Valkey/RedisVerdict ViableIn-memory caching is well covered; Valkey is the emerging open successor to Redis.
Message queue / pub-sub
Hyperscaler Amazon SQS / SNS Azure Service Bus Google Cloud Pub/SubEU-sovereign path Scaleway Messaging & QueuingSelf-hosted NATS / RabbitMQVerdict ViableScaleway offers SQS/SNS-compatible endpoints; self-hosted NATS covers the rest.
Event streaming (Kafka)
Hyperscaler Amazon MSK / Kinesis Azure Event Hubs Managed Service for Apache KafkaEU-sovereign path Aiven for Apache KafkaSelf-hosted Kafka / RedpandaVerdict ViableAiven runs managed Kafka in EU regions. Kinesis's exact API is not replicated, so target Kafka.
Authoritative DNS
Hyperscaler Amazon Route 53 Azure DNS Google Cloud DNSEU-sovereign path Scaleway Domains & DNSOVHcloud DNSBunny DNSVerdict ViableManaged authoritative DNS is well covered. Route 53's routing policies need a manual rebuild.
Transactional email
Hyperscaler Amazon SES Azure Communication Services No first-party service (SendGrid via Marketplace)EU-sovereign path Scaleway Transactional EmailBrevo (FR)Verdict ViableScaleway TEM and Brevo (French) cover sending, reputation and analytics from the EU.
Secrets & key management
Hyperscaler AWS Secrets Manager / KMS Azure Key Vault Secret Manager / Cloud KMSEU-sovereign path Scaleway Secret ManagerHashiCorp Vault (self-hosted)Verdict ViableManaged secret storage exists; self-hosted Vault gives full control including HSM-backed keys.
Application platform (PaaS)
Hyperscaler AWS App Runner / Elastic Beanstalk Azure App Service Google App EngineEU-sovereign path Clever Cloud (FR)Scalingo (FR)Verdict ViableClever Cloud is a production-proven French PaaS (Nantes): git-push or container deploys, autoscaling and managed add-ons (PostgreSQL, Redis, and more). A sovereign home for web apps, though narrower than App Engine’s wider service ecosystem.
Serverless containers
Hyperscaler AWS Fargate Azure Container Instances Google Cloud RunEU-sovereign path Scaleway Serverless ContainersVerdict ViableScale-to-zero container hosting exists and is production-ready for request-driven workloads.
Customer identity (CIAM)
Hyperscaler Amazon Cognito Azure AD B2C Google Cloud Identity PlatformEU-sovereign path Keycloak (self-hosted)ZitadelOryVerdict ViableNo managed Cognito clone, but self-hosted Keycloak and EU-based Zitadel are mature and widely run.
Serverless functions (FaaS)
Hyperscaler AWS Lambda Azure Functions Cloud Run functions (Cloud Functions)EU-sovereign path Scaleway Serverless FunctionsOVHcloud FunctionsVerdict PartialThe runtimes exist, but Lambda's breadth of native event sources and mature cold-start tuning is not matched. Fine for HTTP and cron; harder for deep event-driven fan-out.
Global CDN & edge compute
Hyperscaler Amazon CloudFront / Lambda@Edge Azure Front Door Cloud CDN / Media CDNEU-sovereign path Bunny.netGcoreOVHcloud CDNVerdict PartialEU-headquartered CDNs deliver content well globally. Edge-compute primitives (equiv. to Lambda@Edge/CloudFront Functions) are thinner and less mature.
Metrics, logs & traces
Hyperscaler Amazon CloudWatch / X-Ray Azure Monitor Cloud Monitoring / Logging / TraceEU-sovereign path Self-hosted Grafana / Prometheus / Loki / TempoVerdict PartialThe open-source LGTM stack matches the managed incumbents feature for feature, but you run it yourself. A fully managed, EU-sovereign one-stop observability suite is thin on the ground.
Managed ML platform
Hyperscaler Amazon SageMaker Azure Machine Learning Vertex AIEU-sovereign path OVHcloud AI Endpoints / AI TrainingScaleway GPU Instances + Managed InferenceVerdict GapGPU compute, training and inference endpoints exist, but nothing matches SageMaker's end-to-end breadth (pipelines, feature store, model registry, tuning) as one managed platform. You assemble it.
Cloud data warehouse
Hyperscaler Amazon Redshift Azure Synapse Google BigQueryEU-sovereign path ClickHouse (self-hosted)Aiven for ClickHouseVerdict GapThere is no EU-sovereign, fully managed columnar warehouse with the convenience of Redshift or BigQuery. ClickHouse (self-hosted or via Aiven) is excellent and fast, but it is a different operating model, not a drop-in.
Serverless wide-column NoSQL
Hyperscaler Amazon DynamoDB Azure Cosmos DB Cloud Bigtable / FirestoreEU-sovereign path ScyllaDB (self-hosted)Self-hosted CassandraFerretDB / MongoDB-compatibleVerdict GapDynamoDB's serverless billing and single-digit-ms scale have no EU-sovereign managed equivalent. ScyllaDB/Cassandra deliver the performance, but you run the cluster and lose the pay-per-request model.
Verdicts are conservative and reflect managed, EU-jurisdiction offerings as of 2026. Provider feature sets move quickly, so we re-check on every engagement.
How many engineer-days does this take?
50–120 engineer-days
Higher than a like-for-like AWS→EU move, because the GCP anchors (BigQuery, Spanner) are re-platforms, and the target is usually more than one provider.
- A mid-sized GCP estate: GKE, Cloud Storage, Cloud SQL, a BigQuery warehouse and Pub/Sub messaging.
- The low end assumes a GKE-, GCS- and Cloud SQL-shaped estate with light analytics.
- The high end assumes a BigQuery warehouse to rebuild on ClickHouse and/or a Spanner dependency to re-architect onto partitioned PostgreSQL.
- Excludes application rewrites, egress from GCP and the integration overhead of a multi-provider target.
Where would this actually run?
A sovereign target is usually spread across EU providers’ regions. Pick the mix that matches your data-residency and latency needs.
- FranceScaleway Paris; OVHcloud Gravelines and Strasbourg.
- GermanyOVHcloud Frankfurt; STACKIT (Heilbronn); Hetzner Falkenstein/Nuremberg.
- NetherlandsScaleway Amsterdam.
- PolandScaleway and OVHcloud Warsaw.
- FinlandHetzner Helsinki.
What are the honest limitations?
- BigQuery and Spanner have no sovereign twin The two services most likely to anchor a GCP estate are exactly the hardest to leave: BigQuery becomes self-hosted ClickHouse, and Spanner has no managed European equivalent at all.
- A multi-provider target You typically trade GCP’s single control plane for several (compute on one provider, warehouse self-hosted, ML on another) and own the integration, identity federation and billing spread.
What else does a CTO ask before committing?
- What is the hardest part of leaving GCP specifically?
- BigQuery and Spanner. Everything on the commodity layer (GKE, Cloud Storage, Compute Engine, Cloud SQL) has a solid EU equivalent. But BigQuery has no sovereign serverless-warehouse twin, and Spanner’s globally-consistent SQL has no managed European equivalent at all. Scope those two first.
- Do we have to pick a single European provider?
- Rarely. The pragmatic target is multi-provider: compute and Kubernetes on Scaleway or OVHcloud, a self-hosted ClickHouse for analytics, and a GPU/inference stack for ML. That buys the best fit per workload at the cost of integrating several control planes.
- How do we replace Vertex AI?
- You assemble it. OVHcloud AI (Training, Endpoints) and Scaleway GPU Instances plus Managed Inference cover the compute and serving, but the end-to-end MLOps surface (pipelines, feature store, model registry) is stitched from open-source components rather than one managed platform.