Yes, an application packaged as containers or VM images can run on a cheaper cloud without a rewrite, as long as you replace or keep calling the provider-specific services it depends on. The savings hold only if you also account for cross-cloud egress, the managed services the cheaper cloud does not offer, and the operations work you take back. Whether you are evaluating Hetzner, DigitalOcean, Vultr, Akamai Cloud (Linode), OVHcloud, or Oracle Cloud Infrastructure, the decision comes down to total cost of ownership over your planning horizon. The hourly price of a virtual machine is one input.
Disclosure: Control Plane publishes this guide.
TL;DR
An application built on containers, standard VM images, or Kubernetes manifests can move to a cheaper cloud provider without a code rewrite. Provider-specific managed services are the sticking point.
- What moves unchanged: Container images, VM images, and anything that talks the S3 API for object storage. Kubernetes manifests and Helm charts move with small changes to annotations and storage classes.
- What needs a change: Managed NoSQL databases (DynamoDB, Cosmos DB, Firestore), serverless functions, provider-specific queues, IAM and workload identity, and logging or monitoring agents tied to a specific platform.
- How cheaper clouds differ beyond compute: Managed Kubernetes availability, managed database options, region footprint, bandwidth terms, and compliance attestations vary widely.
- Where savings leak: Cross-cloud egress when the database stays behind, managed services you now operate yourself, support and SLA gaps, compliance re-attestation, and dual-running costs during migration.
- A safe migration order: Audit dependencies, containerize, move stateless services first, measure egress and latency, move or replicate data, cut over DNS, keep the old cloud as fallback, then decommission.
- Where Control Plane fits: Control Plane’s Global Virtual Cloud (GVC), a named set of locations, lets you add a cheaper location to the GVC and run the same container image there from the same workload definition, while Control Plane’s Universal Cloud Identity keeps the workload connected to AWS, GCP, or Azure services it still depends on.
Can I run the same application on a cheaper cloud without rewriting it?
The direct answer is yes for any workload packaged as a container image or a portable VM image. A container runtime does not care whether it runs on an AWS EC2 instance or a Hetzner Cloud server. Kubernetes manifests, Helm charts, and environment variables travel with you, with small changes to annotations, storage classes, and endpoint URLs. If your application talks to a PostgreSQL or MySQL database through a standard connection string, that database can be recreated on any provider that offers the same engine.
The answer becomes no, or at least “not without work,” when the application is deeply coupled to provider-specific services. A backend built around AWS Lambda, DynamoDB, and SQS is one example. The same applies to an application that uses Azure Cosmos DB’s change feed, Google Cloud Firestore’s real-time listeners, or any proprietary API that has no equivalent on the target cloud. In those cases you face a replatforming or refactoring effort.
Most production applications use a mix of portable components and a handful of provider-specific services. The migration work is identifying those dependencies, deciding which ones to replace and which ones to keep calling across clouds, and understanding the cost of each choice. The dependency audit is the first step, and how to avoid cloud vendor lock-in covers what to look for.
What moves to a cheaper cloud without a rewrite?
Each common application component maps to a migration effort. “Effort” reflects the work beyond the initial decision.
| Component | Moves unchanged? | What changes | Effort |
|---|---|---|---|
| Container images | Yes | Registry URL | Configuration |
| VM images (lift and shift) | Yes | Image format conversion if needed | Configuration |
| Kubernetes manifests and Helm charts | Mostly | Cloud-specific annotations, storage classes | Configuration |
| Environment configuration and secrets | Mostly | Secret store references, endpoint URLs | Configuration |
| Managed PostgreSQL or MySQL | Data moves, engine stays the same | Connection strings, backup schedules | Data migration |
| Provider-specific NoSQL (DynamoDB, Cosmos DB, Firestore) | No | Must adopt an alternative data store or keep calling the original | Code change |
| Object storage through the S3 API | Yes, if target supports S3-compatible API | Endpoint URL, credentials | Configuration |
| Provider-specific queues (SQS, Azure Service Bus, Pub/Sub) | No | Must adopt an alternative queue or keep calling the original | Code change |
| Serverless functions (Lambda, Azure Functions, Cloud Run functions) | No | Must containerize or rewrite for a portable runtime | Code change |
| IAM and workload identity | No | Must configure new identity provider or use a cross-cloud identity layer | Configuration or code change |
| Load balancers, DNS, and TLS | Partially | Recreate load balancer config, update DNS records, reissue or transfer certificates | Configuration |
| Logging and monitoring agents | Partially | Replace or reconfigure agents for a new backend (Datadog, Grafana, etc.) | Configuration |
Anything that runs inside a container or a VM moves. Anything that is a managed service outside the container does not, unless the target cloud offers a compatible equivalent. The balance between managed services and portability is the central trade-off in any migration plan.
How do cheaper clouds compare beyond the price of compute?
Production workloads depend on managed services, geographic reach, bandwidth economics, and compliance posture. Six providers often cited as cheaper alternatives to AWS, Google Cloud, and Azure compare on those points.
| Provider | Managed Kubernetes | Managed databases | Region footprint | Bandwidth terms | Compliance attestations (vendor-stated) |
|---|---|---|---|---|---|
| Hetzner Cloud | Not offered as a managed service (community tools like kube-hetzner exist) | Not offered | 6 locations (Falkenstein, Nuremberg, Helsinki, Ashburn, Hillsboro, Singapore) | At least 20 TB included traffic per server in EU locations, 1 TB in US locations, 0.5 TB in Singapore; additional traffic billed per TB | ISO/IEC 27001 (German and Finnish data center parks), BSI C5 |
| OVHcloud | Managed Kubernetes (CNCF-conformant) | Managed databases for PostgreSQL, MySQL, MongoDB, Valkey, Kafka, and others | 46 data centers on 4 continents | Public Cloud instance outbound traffic included in all locations except Asia-Pacific (Singapore, Sydney, Mumbai) | ISO 27001, SOC 1 Type II, SOC 2 Type II, HDS (health data hosting), SecNumCloud (qualified offerings only) |
| DigitalOcean | DigitalOcean Kubernetes (DOKS) | Managed PostgreSQL, MySQL, Valkey, MongoDB, Kafka, OpenSearch | 13 regions (20 datacenters) | Pooled transfer allowance based on droplet size; overage billed per GiB | SOC 2 Type II, SOC 3 |
| Akamai Cloud (Linode) | Linode Kubernetes Engine (LKE) | Managed databases for PostgreSQL, MySQL | 30+ global data centers | Pooled network transfer on most plans (G8 Dedicated and GPU plans have no bundled transfer); overage billed per GB | SOC 2 Type II, PCI DSS (see Akamai Trust Center for scope) |
| Vultr | Vultr Kubernetes Engine (VKE) | Managed databases for PostgreSQL, MySQL, Valkey, Kafka | 30+ locations across 6 continents | Bandwidth allowance per instance; overage billed per GB | SOC 2 Type II |
| Oracle Cloud Infrastructure | OCI Kubernetes Engine (OKE) | Managed Oracle DB, MySQL HeatWave, PostgreSQL, NoSQL | 41 commercial regions in 26 countries, plus government and sovereign regions | First 10 TB per month of outbound data free | SOC 1 Type II, SOC 2 Type II, SOC 3, ISO 27001, PCI DSS, HIPAA, FedRAMP High (US Government regions) |
What to check before moving production
Before committing, verify four things. First, confirm the target provider offers managed versions of the databases and middleware your application needs, or accept that you will operate them yourself. Second, check that the provider has regions close enough to your users and your remaining cloud services to keep latency acceptable. Third, review the support tiers and SLA terms; a cheaper provider with a four-hour response time may cost more during an incident than the compute savings justify. Fourth, match the provider’s published compliance attestations against what your customers and regulators require. A comparison of equivalent cloud services across hyperscalers can help you map what you are leaving behind.
Where do the savings leak?
Moving compute to a cheaper provider does not guarantee lower total cost. Expected savings erode in seven common places.
| Leak | Why it happens | How to contain it |
|---|---|---|
| Cross-cloud egress when the database stays behind | AWS, Google Cloud, and Azure bill data transfer out to the internet per GB. If your database remains on a hyperscaler and your compute moves, every query response crosses the internet. | Move the database to the same provider as compute, or use a read replica on the cheaper cloud to reduce round trips. |
| Chatty calls back to provider-specific services | Microservices that call SQS, SNS, S3, or similar APIs hundreds of times per second generate constant egress and API-call charges on the original cloud. | Replace provider-specific services with portable alternatives (self-hosted queues, S3-compatible storage on the new provider) or batch calls to reduce volume. |
| Managed services you now run yourself (databases, queues, backups) | The cheaper provider may not offer a managed equivalent. You absorb the operations burden: patching, backups, failover, monitoring. | Budget for the engineering time. Use community-supported Kubernetes operators for PostgreSQL, Redis, and similar workloads if the provider lacks managed options. |
| Support and SLA differences | Cheaper providers may offer slower incident response or narrower SLA coverage. | Include the support tier that meets your incident-response target in the cost comparison. Architect for redundancy across locations so a single-provider incident does not take down the application. |
| Compliance re-attestation and audit scope | Moving workloads to a new provider changes the scope of your SOC 2, PCI DSS, or HIPAA audit. Auditors need evidence for the new environment. | Start the compliance mapping early. Choose providers whose attestations overlap with your requirements. |
| Data migration and dual running during the move | You pay for both the old and new environments during migration. Large data transfers incur egress on the source cloud. | Phase the migration. Move stateless services first to start saving on compute while the database migration is planned. |
| Capacity and regional availability limits | Cheaper providers may not have the instance types, GPU availability, or region presence you need. Workarounds (running in a distant region, using smaller instances) add latency or reduce performance. | Validate capacity in the target region before committing. Run load tests on the target provider’s infrastructure. |
Understanding where cloud costs hide is as important as comparing sticker prices. A thorough cost optimization review on your current cloud often recovers enough savings to change the migration calculus.
What is the safest order to move?
A phased migration reduces risk and gives you data to validate the savings before you are fully committed.
- Audit dependencies. Map every managed service, API, database, queue, identity provider, and external integration your application uses. Classify each as portable, replaceable, or locked to the current provider.
- Containerize what is not already containerized. Applications running directly on VMs can be packaged into container images. This makes them portable across any provider with a container runtime or Kubernetes cluster.
- Move stateless services first, with the database left in place. Stateless API servers, web frontends, and background workers are the lowest-risk components to migrate. Keep them pointed at the original database.
- Measure egress and latency. With stateless services running on the cheaper cloud and the database still on the original, you now have real data on cross-cloud latency and egress costs. This tells you whether the database must move too.
- Move or replicate data. Set up the database on the target provider and replicate data. Validate consistency before cutting over.
- Cut over DNS. Point your application’s DNS records to the new environment. Lower DNS TTLs at least one full TTL period before the cutover.
- Keep the old cloud as a fallback for a defined period. Do not decommission the original environment immediately. Run both in parallel long enough to confirm stability, performance, and cost expectations.
- Decommission. Once confident, tear down the old environment to stop paying for it.
How does Control Plane fit?
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, 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 Control Plane’s Global Virtual Cloud (GVC), a named set of locations, runs in every location of that GVC from one workload definition. To run compute on a cheaper provider, you add a location backed by that provider to the GVC, and every workload in that GVC then runs there too. Control Plane’s Managed Kubernetes (MK8s) has a Hetzner provider that sets up Kubernetes clusters on Hetzner Cloud from your Hetzner account, and a Generic provider that turns any Linux server into a worker node. You keep the Hetzner account and pay Hetzner directly for those servers. Enabling the CPLN Platform add-on, an MK8s add-on, turns a cluster built this way into a Control Plane location that you can add to a GVC. You can run the same app on AWS, GCP, and Azure or extend it to Hetzner, or to any Linux servers joined through the Generic provider, with the same workload definition.
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. A workload running on a Hetzner location can keep calling an Amazon S3 bucket, a DynamoDB table, or a Google Cloud Storage bucket while you plan the data migration. Egress charges from the original provider still apply.
Every workload gets its own canonical endpoint that DNS geo-routes clients to the nearest location where that workload passes its readiness checks. Workloads in different clouds reach each other with mTLS and without VPC peering or VPN tunnels you manage. Control Plane handles Day 2 operations for you, including autoscaling, built-in observability, and security.
Control Plane customers typically see 30 to 50 percent lower compute costs compared with running directly on AWS, GCP, or Azure. The 99.999% availability SLA applies to workloads running at least two replicas in at least two locations, and those locations can be on different providers.
Control Plane does not remove egress charges from the provider your data leaves, does not make provider-specific managed services portable, and on Bring Your Own Kubernetes (BYOK) and MK8s locations you keep responsibility for the underlying servers or cluster. Control Plane is PCI DSS Level 1, SOC 2 Type II, HIPAA, and GDPR compliant.
FAQ
Short answers to common questions about moving to a cheaper cloud.
Can I run the same application on a cheaper cloud without rewriting it?
Yes, if the application is packaged as containers or VM images and does not depend heavily on provider-specific managed services. Container images, Kubernetes manifests, Helm charts, and databases running standard engines like PostgreSQL or MySQL move with configuration changes. Components tied to proprietary services like DynamoDB, Lambda, or Cosmos DB require code changes or must continue to be called on the original cloud. The savings are real only after you account for egress, the managed services gap, and the operations work you absorb.
Is Hetzner cheaper than AWS for production workloads?
Hetzner’s published compute pricing is lower than comparable AWS instance types for equivalent CPU and memory. However, Hetzner does not offer managed databases or managed Kubernetes, has six locations, and provides a narrower set of compliance attestations. For a stateless, containerized workload that does not need managed services or global presence, the savings are largest. For a workload that depends on managed PostgreSQL, multi-region failover, or broad compliance coverage, operations and compliance costs can offset the compute savings. Check Hetzner’s pricing page and AWS EC2 pricing for current rates.
What breaks when you move off AWS?
Common breakages are IAM policies and role assumptions that do not exist on the target cloud, references to AWS-specific services (SQS, SNS, DynamoDB, Secrets Manager, CloudWatch), hardcoded ARNs or region-specific endpoints in configuration, and networking assumptions like VPC peering or security group rules. Applications that use the AWS SDK for service calls will fail unless those calls are redirected or replaced.
Can I keep my AWS database and move compute to another cloud?
Yes. Your compute workloads on the cheaper cloud connect to the AWS database over the public internet or through a VPN or interconnect. This works but introduces cross-cloud latency and egress charges on every query response leaving AWS. Measure the egress volume and latency during a pilot before committing. For high-throughput databases, the egress cost alone can erase the compute savings.
How much does egress cost when moving off a hyperscaler?
Egress pricing depends on the provider, the volume, and the destination. AWS, Google Cloud, and Azure each publish tiered pricing for data transfer out to the internet. The cost per GB decreases at higher volumes but remains meaningful at scale. Check the current rates on the AWS data transfer pricing page, the Google Cloud network pricing page, and the Azure bandwidth pricing page. The total bill depends on how much data leaves the provider each month, which is driven by database query volume, API response sizes, and backup transfers.
Do cheaper clouds have managed Kubernetes?
OVHcloud, DigitalOcean, Akamai Cloud (Linode), Vultr, and Oracle Cloud Infrastructure all offer managed Kubernetes services. Hetzner does not offer a managed Kubernetes product, though community tools exist to automate cluster provisioning on Hetzner servers. The maturity, update cadence, and add-on ecosystem vary across providers, so test your specific workload before migrating production.
Is moving to a cheaper cloud worth it for a small team?
It depends on what you are moving away from and what you are moving toward. A small team running containerized workloads on expensive hyperscaler instances with no use of proprietary managed services can see immediate savings. A small team that relies on managed databases, managed queues, and integrated monitoring will spend engineering time replacing or operating those services on the cheaper cloud. The break-even calculation is migration cost divided by monthly savings. If the payback period is longer than your planning horizon, or the operations burden exceeds your team’s capacity, optimizing spend on your current cloud may deliver more value.
How does Control Plane help run the same app on a cheaper cloud?
Control Plane lets you add a cheaper cloud location to a Global Virtual Cloud alongside your existing cloud locations. Control Plane’s Managed Kubernetes can build a cluster on your Hetzner account or on any Linux server, and that cluster can become a location in the GVC. Universal Cloud Identity lets the workload on the cheaper location keep calling AWS, GCP, or Azure services it still depends on, without embedding long-lived keys. You can start with a separate GVC that holds one workload and one cheaper location, measure the results, and expand from there.
To run the same application on cheaper locations from one workload definition, talk to the Control Plane team or sign up free.
Originally published on the Control Plane blog.