The best tools for migrating workloads between clouds depend on the layer you are moving. Use a VM replication tool like AWS Application Migration Service or Azure Migrate for virtual machines, Velero or a GitOps redeploy for Kubernetes workloads, a change-data-capture service like Debezium or a managed database migration service for databases, and a transfer service or rclone for object storage. No single tool covers every layer well, so pick one or two specialized tools per layer, run the workload in both clouds simultaneously, and shift traffic gradually so no single cutover moment carries all the risk. This guide includes a tool-by-layer matrix, a decision table for each kind of workload, and a cutover checklist.

Disclosure: Control Plane publishes this guide.

TL;DR

The best cloud-to-cloud migration tools are chosen one layer at a time.

  • VMs: AWS Application Migration Service, Azure Migrate, Google Cloud Migrate to Virtual Machines, or RackWare RMM for continuous replication and orchestrated cutover.
  • Kubernetes workloads: Velero for backup-and-restore, Konveyor (Crane) for resource transformation, or a GitOps redeploy with Argo CD or Flux.
  • Databases: AWS Database Migration Service, Azure Database Migration Service, Google Cloud Database Migration Service, or Debezium for open-source change data capture.
  • Object storage: rclone (free, supports every major cloud), AWS DataSync, Google Cloud Storage Transfer Service, or Azure Storage Mover.
  • Infrastructure as code: Terraform or OpenTofu (with import blocks for existing resources), or Pulumi to re-target providers.
  • The hardest layers are databases with strict uptime requirements, managed services with no portable equivalent, and serverless functions that must be rewritten or repackaged as containers.
  • Egress fees apply every time you copy data out of a cloud. Model them before you start.
  • The parallel-run approach replaces a single big-bang cutover with a staged one. You run in both clouds, shift traffic gradually, and then remove the old location.
  • Control Plane lets you add a new cloud location to a workload’s Global Virtual Cloud (GVC), a named set of locations, run it in both clouds at once, and remove the old location when you are ready. It does not move data layers, so the matrix tools still apply.

What are the best tools for migrating workloads between clouds?

A cloud-to-cloud migration moves compute, data, identity, networking, and configuration, each with its own tooling, and this matrix maps the most widely used tools to the layer they move, the clouds they support, how they handle cutover, and whether they are free, included with the target cloud, or paid.

Tool Layer it moves Source clouds Target clouds How it handles cutover Cost
AWS Application Migration Service (now AWS Transform MGN) VMs Any (agent-based), including Azure and GCP AWS Continuous replication Free period per server, then paid
Azure Migrate (server migration) VMs AWS, GCP, VMware, Hyper-V, physical Azure Continuous replication Free for 180 days per server
Google Cloud Migrate to Virtual Machines VMs AWS, Azure, VMware Google Cloud Continuous replication Included with Google Cloud
RackWare RMM VMs AWS, Azure, GCP, Oracle Cloud Infrastructure (OCI), physical AWS, Azure, GCP, OCI Continuous replication Paid
Velero Kubernetes workloads Any Kubernetes cluster Any Kubernetes cluster Backup-and-restore (File System Backup for cross-cloud volumes) Free / open source
Veeam Kasten (formerly Kasten K10) Kubernetes workloads Any Kubernetes cluster Any Kubernetes cluster Backup-and-restore (export to object storage for cross-cloud) Free edition (Veeam Kasten Free); paid Enterprise
Konveyor (Crane) Kubernetes workloads Any Kubernetes cluster Any Kubernetes cluster Snapshot-and-restore with resource transforms Free / open source
GitOps redeploy (Argo CD or Flux) Kubernetes workloads Any Kubernetes cluster Any Kubernetes cluster Redeploy from Git Free / open source
AWS Database Migration Service Databases Azure, GCP, on-premises, self-managed AWS managed databases Continuous replication (CDC) Paid (replication instance hours)
Azure Database Migration Service (SQL Server) and the Azure Database for PostgreSQL migration service Databases SQL Server on premises or on other clouds; PostgreSQL from Amazon RDS, Aurora, Google Cloud SQL, on premises Azure SQL, Azure Database for PostgreSQL Continuous replication (online mode, supported pairs) Included with Azure
Google Cloud Database Migration Service Databases AWS RDS and Aurora, Azure, and self-managed MySQL, PostgreSQL, SQL Server, Oracle Cloud SQL, AlloyDB for PostgreSQL Continuous replication Free for same-engine migrations; paid per GB for cross-engine
Debezium Databases Any database with a supported connector (MySQL, PostgreSQL, MongoDB, SQL Server, Oracle, others) Kafka, or other sinks via Debezium Server Continuous replication (CDC) Free / open source
rclone Object storage AWS S3, Azure Blob, GCS, S3-compatible, 70+ backends AWS S3, Azure Blob, GCS, S3-compatible, 70+ backends Sync or copy Free / open source
AWS DataSync Object storage Azure Blob, GCS, S3-compatible, NFS, SMB AWS S3, EFS, FSx Scheduled sync Paid (per GB)
Google Cloud Storage Transfer Service Object storage AWS S3, Azure Blob, S3-compatible, HTTP/HTTPS sources Google Cloud Storage Scheduled sync Included with Google Cloud
Azure Storage Mover Object storage NFS, SMB, AWS S3, Google Cloud Storage (preview) Azure Blob, Azure Files Scheduled sync Included with Azure
Terraform / OpenTofu Infrastructure as code Any provider with a Terraform provider Any provider with a Terraform provider Redeploy Free / open source
Pulumi Infrastructure as code AWS, GCP, Azure, and others AWS, GCP, Azure, and others Redeploy Free tier and paid tiers

A few patterns stand out. Each hyperscaler offers a VM migration service, free or free for an initial period, that pulls servers in from competing clouds; none of them moves VMs out to a competitor. Open-source tools like Velero, Debezium, and rclone are cloud-agnostic and free but require you to operate them yourself. RackWare RMM is the paid option in the table for replicating VMs between any pair of AWS, Azure, Google Cloud, and OCI.

Which tool should you use for each kind of workload?

The right tool depends on what you are moving and how much downtime you can tolerate. Each row gives the simplest path and when to use something heavier.

Moving Simplest path When to use something heavier
A handful of VMs The target cloud’s built-in migration service (MGN, Azure Migrate, or Google Cloud Migrate to VMs) When VMs span multiple target clouds, use RackWare RMM for a single pane of glass
A fleet of VMs Same built-in services with wave-based orchestration When you need cross-cloud replication or automated testing across waves, add RackWare or a migration factory partner
Kubernetes workloads Velero backup-and-restore or a GitOps redeploy from your existing Git repo When you need resource transforms (renaming storage classes, updating annotations), use Konveyor Crane. For policy-driven backup, use Veeam Kasten
Stateless containers outside Kubernetes Repackage as container images and deploy to the target platform When the target is Kubernetes, add Helm charts or Kustomize overlays to your CI/CD pipeline
A production database with minimal downtime tolerance A managed CDC service (AWS DMS, Azure DMS, or Google Cloud DMS) that replicates continuously until cutover When you need open-source CDC or must fan out to multiple targets, use Debezium with Kafka
Large object stores rclone for cloud-agnostic sync When the target is a single hyperscaler and you want managed scheduling, use that cloud’s transfer service (DataSync, Storage Transfer Service, or Storage Mover)
Serverless functions No portable tool exists. Plan a rewrite in the target cloud’s function runtime or repackage as a container When the function count is large, invest in a container-based abstraction so future moves are easier
Managed queues and caches No portable tool exists. Re-create the service in the target cloud and drain or replay messages When message ordering or delivery guarantees matter, run both services in parallel and dual-write during the transition

Databases and managed services are the hard part of any cloud-to-cloud migration. VMs and containers are relatively mechanical to move because the compute layer is portable. Databases require continuous replication with careful validation, and managed services like queues, caches, and serverless functions often have no direct equivalent on the target cloud. Our guide to comparing equivalent cloud services across AWS, Azure, and Google Cloud maps the closest matches.

What does a cloud-to-cloud migration cost beyond the tools?

Most tools in the matrix are open source or cost little next to egress and parallel-run spend, so licensing is rarely the largest expense. Three other costs dominate.

Egress fees. Every major cloud charges for data transferred out to the internet or to another cloud. Egress pricing varies by provider, by volume tier, and by whether you use a direct interconnect or the public internet. Model your total data volume before you start, because a multi-terabyte database replication stream or a petabyte object store copy can generate a significant bill on the source cloud.

Parallel-run costs. During the transition you will run compute, storage, and networking in both the source and target clouds simultaneously. This period can last days or weeks depending on your validation and rollback requirements. Budget for both environments’ compute plus cross-cloud replication and request egress for the whole window.

Re-platforming managed services. If you use managed queues, caches, search clusters, or serverless functions, you will spend engineering time re-creating those services on the target cloud. This is often the largest hidden cost because it requires application code changes as well as infrastructure work. Our cloud migration pillar guide covering the 7 Rs breaks down how to classify each workload and estimate the effort.

Which vendors have experience migrating between cloud platforms?

Each hyperscaler operates a formal migration program with tooling, funding, and partner networks.

AWS offers the AWS Migration Acceleration Program (MAP), which provides credits, methodology, and access to AWS Migration Competency Partners. Its tooling centers on the AWS Application Migration Service, AWS Database Migration Service, and AWS Transform for assessment and tracking.

Microsoft Azure pairs Azure Migrate (assessment and server migration) and its database migration services with funding programs and a partner ecosystem recognized through Microsoft partner designations such as Azure Expert MSP.

Google Cloud provides the Google Cloud Migration Center for discovery and planning, along with Migrate to Virtual Machines and Database Migration Service. Its Rapid Migration and Modernization Program (RaMP) pairs customers with Google Cloud partners.

Among large systems integrators, Accenture , Deloitte , Wipro , Infosys , and HCLTech each run cloud migration and modernization practices and partner with AWS, Microsoft, and Google Cloud. These firms typically bring migration factory methodologies that combine the native tools with their own automation and governance layers.

For guidance on reducing long-term dependence on any single vendor, read how to avoid cloud vendor lock-in.

How do you migrate between clouds without a big-bang cutover?

A parallel run with gradual traffic shifting avoids a big-bang cutover, where all traffic flips at once.

Step What to verify Common miss
Inventory dependencies including managed services Every service, API, queue, cache, and external integration the workload touches Forgetting managed services that have no equivalent on the target cloud
Map equivalent services Target cloud service for each source cloud service, including SKU and region Assuming feature parity when the target service has different limits or behavior
Copy data and keep it in sync Replication lag, error counts, and data integrity checksums Starting the cutover before replication has fully caught up
Re-create identity and secrets IAM roles, service accounts, secrets manager entries, and certificates Hardcoded credentials or long-lived keys that bypass the new identity layer
Re-create networking and DNS VPCs, subnets, security groups, routes, peering, and DNS records Overlapping CIDR ranges between source and target clouds
Run both in parallel and compare Functional tests, performance baselines, and data consistency checks Skipping load testing in the target environment
Decide the database write strategy Which cloud holds the primary during the ramp, cross-cloud write latency, or dual-write and conflict handling Ramping app traffic while the database stays single-writer in the source cloud
Lower DNS TTLs and update allowlists TTLs lowered a day or two ahead; partners and payment providers allowlist the target’s egress IPs Third parties that still allowlist only the source cloud’s IPs
Validate data after the final sync Row counts, checksums, and sampled comparison queries Declaring success on replication status alone
Cut over traffic gradually Weighted DNS, load balancer rules, or application-level routing percentages Sending 100% of traffic at once instead of ramping from 1% to 10% to 50% to 100%
Keep the source warm until rollback is no longer needed Reverse replication (target to source) running after the database cutover, source environment healthy, rollback runbook tested Decommissioning the source before the rollback window closes
Decommission and close egress paths Source resources deleted, DNS records cleaned up, egress firewall rules removed Leaving orphaned resources that continue to accrue cost

Monitor error rates and latency at each traffic step, and remove the source only after the target is stable. This approach to running an app across multiple regions with routing and failover is the same pattern used for multi-region resilience.

How Control Plane fits

Control Plane is the unbreakable platform for running production workloads. It runs serverless, standard, stateful, and cron workloads natively on AWS, GCP, Azure, and Oracle Cloud Infrastructure (OCI), VM workloads on locations joined through Control Plane’s Managed Kubernetes (MK8s), and workloads on Kubernetes clusters you already run or your own machines, including on premises, active-active across regions and clouds.

A workload deployed to a GVC runs in every location of that GVC from one workload definition. Locations can be Control Plane-managed regions on AWS, GCP, Azure, and OCI, Kubernetes clusters you already run joined through Control Plane’s Bring Your Own Kubernetes (BYOK), or your own servers joined through MK8s, including on premises.

Moving a container workload to another cloud means adding that cloud’s location to the GVC, letting the workload run in both clouds at once, then removing the old location. Every workload has its own canonical endpoint that DNS geo-routes clients to the nearest location where that workload passes its readiness checks. Clients keep reaching the workload while both locations run, so the compute move does not depend on a single DNS cutover. Databases and other data still need their own cutover.

Control Plane’s Universal Cloud Identity lets workloads reach AWS, GCP, and Azure services through one identity, with no long-lived cloud keys in your code. Control Plane handles Day 2 operations for you, including autoscaling, built-in observability, and security. Control Plane is PCI DSS Level 1, SOC 2 Type II, HIPAA, and GDPR compliant.

An honest scope note: Control Plane changes where a container workload runs. VM workloads move between locations joined through MK8s. It does not replicate databases, copy object storage, or convert serverless functions. The tools in the matrix still apply to those layers. Our guide on modernizing VM applications for Kubernetes and multi-cloud covers moving VM applications to a multi-cloud Kubernetes model.

FAQ

Common questions about cloud-to-cloud migration tools, costs, and cutover.

What are the best tools for migrating workloads between clouds?

The best tools depend on the layer you are moving. For VMs, use AWS Application Migration Service, Azure Migrate, Google Cloud Migrate to Virtual Machines, or RackWare RMM. For Kubernetes workloads, use Velero, Konveyor Crane, or a GitOps redeploy with Argo CD or Flux. For databases, use a managed migration service from the target cloud or Debezium for open-source CDC. For object storage, use rclone or the target cloud’s transfer service. For infrastructure as code, use Terraform, OpenTofu, or Pulumi to re-target providers.

Can I migrate VMs from AWS to Azure or Google Cloud?

Yes. Azure Migrate supports AWS and GCP VMs as sources with continuous replication. Google Cloud Migrate to Virtual Machines supports AWS and Azure VMs as sources. RackWare RMM supports any combination of AWS, Azure, GCP, and OCI as both source and target. Each uses an agent or agentless connector on the source VM to replicate data continuously to the target cloud, letting you test before cutover.

How do I move Kubernetes workloads to another cloud?

The simplest path is a GitOps redeploy: point your Argo CD or Flux configuration at a new cluster on the target cloud and let it reconcile. If you need to carry persistent volume data, use Velero with File System Backup (Kopia) or CSI snapshot data movement to copy volume data to object storage, then restore on the target cluster. Native cloud disk snapshots cannot be restored on another provider. Konveyor Crane adds resource transformation so you can rename storage classes and update cloud-specific annotations during the move.

How do I migrate a database between clouds with minimal downtime?

Use a service that supports continuous replication via change data capture. AWS Database Migration Service, Google Cloud Database Migration Service, and Azure’s migration services (SQL Server through Azure Database Migration Service, PostgreSQL through the Azure Database for PostgreSQL migration service) replicate data continuously from a source database on another cloud. Debezium provides the same capability as an open-source project that streams changes through Kafka or Debezium Server. In all cases, the source database stays live during replication. You cut over by stopping writes to the source, letting the final changes replicate, and pointing the application to the target database.

How much does it cost to move data out of a cloud?

Every major cloud charges egress fees for data transferred out. Pricing varies by provider, by volume, and by transfer method (public internet vs. direct interconnect). There is no fixed dollar amount because it depends on how much data you move and how long you run continuous replication. Model your total data volume and expected replication duration before you start, and check each provider’s current pricing page for the applicable tiers.

Which vendors have experience migrating from one cloud platform to another?

AWS, Azure, and Google Cloud each operate formal migration programs with dedicated tooling, funding credits, and certified partner networks. Among systems integrators, Accenture, Deloitte, Wipro, Infosys, and HCLTech run cloud migration practices and partner with multiple hyperscalers. These firms combine the native migration tools with their own automation and governance frameworks.

Can serverless functions be moved between clouds?

No portable migration tool exists for serverless functions. AWS Lambda, Azure Functions, and Google Cloud Functions each use proprietary runtimes, event models, and deployment mechanisms. The two practical options are rewriting the function in the target cloud’s runtime or repackaging it as a container image that can run on any cloud’s container platform. If you have a large number of functions, investing in a container-based abstraction now will make future moves easier.

How does Control Plane help with moving workloads between clouds?

Control Plane lets you add a new cloud location to a workload’s Global Virtual Cloud so the workload runs in both clouds from one definition. The workload’s canonical endpoint DNS geo-routes clients to the nearest location where the workload passes its readiness checks, so the new location starts receiving traffic once it is ready. When the new location is stable, you remove the old one. Control Plane does not replicate databases, copy object storage, or convert serverless functions, so you still use the specialized tools in the matrix for those layers.

What is the biggest risk in a cloud-to-cloud migration?

Hidden dependencies on managed services that have no direct equivalent on the target cloud. VMs and containers move mechanically, but a workload that depends on a proprietary queue, cache, search cluster, or serverless function requires re-platforming. Comprehensive dependency mapping during the inventory step is the most effective mitigation.

Do I need to shut down my application during a cloud migration?

Usually not for long. Continuous replication and a parallel run shrink downtime to a brief write freeze at the database cutover. Replicate your database and object storage continuously, run the application in both clouds, and shift traffic gradually. Keep the source environment operational until your rollback window closes. The trade-off is paying for both environments during the transition.

Run the same workload in two clouds at once and retire the old location when you are ready. Talk to the Control Plane team or sign up free.

Originally published on the Control Plane blog.