Nov 17, live webinar: ECS to EKS without freezing the roadmap for a quarter

Leaving AWS for Google Cloud: What Actually Reduces Operational Overhead, and What Stays the Same

Google Cloud is a realistic alternative to AWS for a web app when you use what is different about it: GKE Autopilot, Cloud Run, a global VPC and automatic sustained use discounts. Most day-2 work stays the same on both clouds, and a platform layer removes it, not the vendor switch.

Mélanie Dallé
Senior Marketing Manager
NOV 18, 2026 · 17 MIN
Leaving AWS for Google Cloud: What Actually Reduces Operational Overhead, and What Stays the Same

Key points:

  • Google Cloud is a realistic alternative for a web app, but only if you adopt its managed defaults. GKE Autopilot has Google manage "your nodes, scaling, security, and other preconfigured settings" (Google Cloud), Cloud Run scales to zero and bills per request, and GKE "automatically upgrades all clusters over time" (Google Cloud). Lift your EC2 and self-managed habits over unchanged and you get a different vendor with the same work.
  • Four differences are real. Autopilot bills Pod resource requests instead of nodes. Cloud Run gives 2 million free requests a month and timeouts up to 60 minutes. A Google VPC is a global resource with regional subnets, where an AWS subnet "must reside entirely within one Availability Zone". Sustained use discounts apply automatically, up to 30% on eligible machine series.
  • Most of the overhead stays. IAM design, network security, observability, database operations, CI/CD, secrets, cost governance and on-call look almost identical on both clouds. Cloud SQL is to Google Cloud what RDS is to AWS: a managed database you still size, back up, patch on a schedule and pay egress for.
  • AWS has closed part of the gap. EKS Auto Mode now manages nodes, load balancing and storage for a fee on top of EC2 (AWS). AWS App Runner, the closest thing to Cloud Run, "is no longer open to new customers" (AWS).
  • Staying on AWS is the right call if your app is built on Lambda, DynamoDB, SQS or Kinesis, if you carry Savings Plans or a spend commitment, or if your team's skill is AWS. Move if you want serverless containers, a simpler network model and Google's data and AI services next to your app.
  • A platform layer removes the day-2 work a vendor switch leaves in place. Qovery deploys the same application definition to your own Google Cloud or AWS account, operates the cluster including upgrades, and lets you run both clouds in parallel during the move.

Qovery · Infrastructure control plane
Kubernetes, operated through one governed API
Learn more

Every few weeks I talk with a team that wants to leave AWS. The reasons vary: a bill that keeps creeping up, a CTO who prefers BigQuery, startup credits on Google Cloud, or a feeling that AWS has become too many services to keep track of. The question underneath is always the same. Will Google Cloud make our life easier, or will we spend six months moving and end up with the same pager, the same Terraform and the same spreadsheet of costs?

My answer is that Google Cloud removes some specific work, and only if you use what is different about it. The bulk of operational overhead comes with running production software on any cloud. This article covers where GCP is genuinely different, where the work is identical, when staying on AWS is the better call, and what removes the overhead instead.

If you already decided to move and want the tools and the cut-over sequence, read our guide to migrating from AWS to GCP without rewriting your app. This page answers the question before that one.

Is Google Cloud a realistic alternative to AWS for a web app?

Yes. Google Cloud runs the same building blocks a web app needs (containers, managed Kubernetes, managed PostgreSQL and MySQL, object storage, load balancing, a CDN) in regions across the Americas, Europe, Asia Pacific, the Middle East and Africa. It becomes a better choice than AWS, rather than an equal one, when your app fits Cloud Run or GKE Autopilot and you want Google's data services close by.

It is not a realistic alternative if you expect the switch alone to cut your operations work. A team that runs EC2 instances with Ansible on AWS and moves them to Compute Engine VMs with Ansible on Google Cloud has changed its invoice and kept its workload. The savings come from the managed defaults: Autopilot instead of node pools, Cloud Run instead of VMs, Cloud SQL instead of a self-run database. AWS has its own versions of most of these, and many teams on AWS do not use them either.

So the useful question is narrower. Which of your operational chores does Google Cloud take over by default, and which ones does it leave with you?

Where is Google Cloud operationally different from AWS?

Google Cloud is different in four places that matter for a web app: GKE Autopilot, Cloud Run, the network model and automatic discounts. Each one removes a specific chore. None of them removes on-call, security design or cost governance.

Does GKE Autopilot remove Kubernetes operations?

GKE Autopilot removes node operations. In Google's words, Autopilot is "a mode of operation in GKE in which Google manages your infrastructure configuration, including your nodes, scaling, security, and other preconfigured settings", and Google calls Autopilot clusters "the recommended way to use GKE" (Google Cloud).

Three details change your week:

  • Billing follows Pods, not nodes. General-purpose Autopilot workloads "use a Pod-based billing model, in which you're charged based on Pod resource requests" (GKE pricing). You stop paying for half-empty nodes. You start paying for over-sized requests, so right-sizing still matters.
  • The SLA covers your Pods. GKE commits to 99.95% for the control plane of Autopilot and regional Standard clusters, and 99.9% for "Autopilot Pods in multiple zones" (GKE pricing). AWS's EKS SLA covers the control plane, 99.95% on the standard tier (AWS), and your worker nodes fall under the compute SLA.
  • Upgrades happen to you. "GKE automatically upgrades all clusters over time, including those not enrolled in a release channel", and you steer the timing with maintenance windows and exclusions (Google Cloud). On EKS, a version gets 14 months of standard support, then 12 months of extended support at $0.60 per cluster per hour instead of $0.10 (AWS), and only at the end of extended support does AWS upgrade the cluster automatically (AWS docs).

Both providers charge the same $0.10 per cluster per hour management fee. GKE adds a free tier of $74.40 in monthly credits per billing account, "equivalent to one free Autopilot or zonal Standard cluster per month" (GKE pricing).

The fair caveat: AWS answered Autopilot with EKS Auto Mode, which automates compute autoscaling, Pod networking, load balancing, DNS and block storage, billed per instance on top of the EC2 price (AWS docs, AWS pricing). If you are on EKS with managed node groups, turning on Auto Mode closes much of the gap without leaving AWS. We compared both modes in detail in GKE vs EKS without a platform team.

Upgrades still need you on both clouds. Google upgrades the cluster, but Kubernetes API removals still break old Helm charts and manifests. Someone checks deprecations before each minor version, whatever the logo on the console.

Is Cloud Run a real option for a production web app?

Cloud Run is the strongest reason to pick Google Cloud for a stateless web app. You hand it a container image, it scales instances with traffic, down to zero, and bills only while a request is in flight on the default request-based billing (Cloud Run pricing). The free tier covers 180,000 vCPU-seconds, 360,000 GiB-seconds and 2 million requests a month per billing account. Request timeouts default to 5 minutes and extend "up to 60 minutes" (Google Cloud).

AWS's closest equivalent has gone in the other direction. The App Runner documentation now opens with "AWS App Runner is no longer open to new customers", and AWS points new users to Amazon ECS Express Mode (AWS). ECS with Fargate is a solid service, but it asks you to think about task definitions, services, target groups and VPC subnets, which Cloud Run hides.

Cloud Run has ceilings: long-running background workers, stateful services, sidecar-heavy setups and anything needing node-level control fit GKE better. My colleague Alessandro mapped those limits in Cloud Run vs App Runner vs Azure Container Apps. For a web front end plus an API, I think Cloud Run removes more operations work than anything AWS currently sells to new customers.

How is Google's network model different?

A Google VPC is global. "VPC networks, including their associated routes and firewall rules, are global resources. They are not associated with any particular region or zone. Subnets are regional resources" (Google Cloud). On AWS, a VPC lives in one Region and "each subnet must reside entirely within one Availability Zone and cannot span zones" (AWS).

In practice, a two-region app on AWS means two VPCs, peering or a Transit Gateway, and per-AZ subnets in each. On Google Cloud it means one VPC with a subnet per region, and the firewall rules apply to both. Google's global external Application Load Balancer adds "global anycast external IP addresses over Premium Tier" and can reach backends in several regions behind one IP (Google Cloud). On AWS you combine regional load balancers with Route 53 or Global Accelerator.

If your app runs in a single region, which is most web apps, this difference shrinks to fewer subnets to plan. It matters once you go multi-region.

Do sustained use discounts make Google Cloud cheaper to run?

Sustained use discounts make Google Cloud cheaper to run without paperwork, on some machine types. Compute Engine applies them automatically when a resource runs for a large part of the month, "up to a 30% net discount off of the resource cost for virtual machine (VM) instances that run the entire month" (Google Cloud).

Read the limits before you count on them:

  • Only N1, N2, N2D, C2, M1 and M2 machine types (plus f1-micro, g1-small and GPUs attached to N1) are listed as eligible. N1, C2, M1 and M2 get up to 30%, N2 and N2D up to 20%. Series not on that list get no sustained use discount.
  • "Only Self-serve (or Online) Cloud Billing accounts are eligible", so invoiced enterprise billing accounts do not receive them.
  • They do not stack with committed use discounts, and only VMs created by GKE and Compute Engine qualify.

AWS has no automatic equivalent. Its Savings Plans lower prices "in exchange for a specific usage commitment (measured in $/hour) for a one- or three-year period" (AWS). Google also sells one- and three-year committed use discounts. So the real difference is for a team that does not want to commit: on eligible machine types, Google discounts steady usage anyway. Romaric ran the per-service numbers in Is Google Cloud cheaper than AWS?.

DimensionAWSGoogle CloudDoes GCP remove work?
Kubernetes nodesEKS managed node groups, or EKS Auto Mode for a per-instance feeGKE Autopilot manages nodes, billed per Pod requestYes, unless you already run EKS Auto Mode
Cluster upgradesYou upgrade within 14 months, or pay $0.60/h extended supportGKE upgrades all clusters automatically, inside your maintenance windowsPartly; API deprecations stay yours
Serverless containersApp Runner closed to new customers; ECS Express Mode or FargateCloud Run, scale to zero, 2M free requests/monthYes, for stateless web apps
NetworkRegional VPC, subnets per AZGlobal VPC, subnets per region, global anycast load balancerYes for multi-region, little for single-region
DiscountsSavings Plans and Reserved Instances, 1 or 3 yearsAutomatic SUDs on eligible series, plus CUDsYes, if you avoid commitments and use eligible series
Managed PostgreSQL / MySQLRDS, AuroraCloud SQL, AlloyDBNo, same operational model
IAMIAM users, roles, policies, SCPsIAM principals, roles, org policiesNo, different syntax, same design work
ObservabilityCloudWatch, X-RayCloud Logging, Cloud Monitoring, Cloud TraceNo
Run your app on GCP, AWS, or both, without a DevOps hire.
Qovery deploys and operates your applications in your own Google Cloud, AWS, Azure or Scaleway account, or on your existing Kubernetes cluster, with git-push deployments, preview environments and managed cluster upgrades.

Which operational work stays the same after you switch to Google Cloud?

Almost everything that is not compute provisioning stays the same: identity and access design, network security, databases, observability, CI/CD, secrets, cost control, compliance evidence and on-call. These belong to running production software, and Google Cloud asks you to do them with different nouns.

Here is the list I walk teams through:

  • Identity and access. You redesign roles, service accounts and organization policies. Workload Identity on GKE replaces IRSA or Pod Identity on EKS. The work of deciding who can do what is identical.
  • Databases. Cloud SQL and AlloyDB replace RDS and Aurora. You still choose instance sizes, high availability, backup retention, maintenance windows and read replicas. Moving the data is its own project with its own tool.
  • Observability. Dashboards, alerts, log retention, tracing and SLOs get rebuilt in Cloud Monitoring and Cloud Logging, or stay in Datadog or Grafana, in which case nothing changes.
  • CI/CD. Your GitHub Actions or GitLab pipelines change their deploy step and credentials. Build, test and approval stay the same.
  • Infrastructure as code. Terraform modules for AWS do not apply to Google Cloud. You rewrite them with the Google provider, and keep the same review process.
  • Secrets. Secret Manager replaces AWS Secrets Manager or Parameter Store. Rotation policies and who reads what remain your job.
  • Cost governance. Budgets, labels, idle resources, over-sized requests. Pod-based billing on Autopilot makes over-requested CPU the new waste.
  • Compliance. Your SOC 2 or ISO 27001 controls need new evidence from new logs. Auditors do not care which cloud you picked, only that the controls are evidenced.
  • On-call. Incidents, runbooks and postmortems keep their shape. The 3am page is still yours.

If those nine items take most of your team's operations time today, moving to Google Cloud will not free that time. It will add a migration project on top.

What does the move itself cost?

The move costs engineering time first, and the size depends on how much of your app uses AWS-only services. Containers on ECS or EKS move with little code change. Lambda handlers, DynamoDB access patterns, SQS and SNS integrations and Kinesis consumers are rewrites.

Data transfer used to be the scary line item. Both providers now waive it for customers who leave. AWS offers "free data transfer out to the internet when they move all of their data off of AWS", as credits approved through AWS Support, and says you do not have to close your account (AWS News Blog). Google offers the same when you migrate all workloads and data away from a Google Cloud service (Google Cloud). The symmetry is useful: if the move does not work out, coming back does not carry an egress penalty either.

The other hidden cost is running two clouds at once. A safe cut-over keeps both sides live for weeks while you replicate data, shift traffic and prove the new side under load. That means two sets of dashboards, two bills and two sets of credentials. The AWS to GCP migration guide has the step-by-step sequence and the database tools.

When should you stay on AWS?

Stay on AWS when the move would turn into a rewrite, when you would walk away from money already committed, or when your team's expertise is AWS and nothing on Google Cloud solves a problem you have today. Any one of those usually outweighs Google Cloud's operational advantages for a web app.

Concretely, stay if:

  • Your app is built on AWS-native services. Lambda, Step Functions, DynamoDB, SQS, SNS, EventBridge and Kinesis have no drop-in equivalent. Moving them means redesigning the app.
  • You hold commitments. Savings Plans and Reserved Instances run one or three years, and private pricing agreements carry spend commitments. Moving before they expire means paying twice.
  • You can get the same benefit on AWS. If node management is your pain, EKS Auto Mode addresses it. If your problem is the bill, rightsizing and Savings Plans are cheaper than a migration.
  • Your customers or partners require AWS. Marketplace listings, co-selling programs or customer contracts sometimes name the provider.
  • Your team knows AWS deeply and Google Cloud not at all. Expertise is a real operational asset. Losing it costs incidents.

Move to Google Cloud if your app is mostly stateless containers that fit Cloud Run or Autopilot, you want BigQuery or Vertex AI next to your production data, you plan a multi-region footprint, or you have Google Cloud credits that cover the migration period. Romaric wrote a decision guide for teams picking a first cloud without a DevOps engineer, GCP or AWS for your first production deployment, which applies to this choice too.

If the cloud switch does not remove the overhead, what does?

A platform layer removes the overhead. The work that stays the same on both clouds (deployments, environments, cluster upgrades, access control, preview environments, cost visibility) is work a platform does for you, on whichever cloud you pick. Change the vendor and the platform stays.

At Qovery, this is what we build. Qovery is an Agentic DevOps platform that deploys and operates your applications on Kubernetes inside your own Google Cloud, AWS, Azure or Scaleway account, or on a Kubernetes cluster you already run. Qovery never hosts your app: the control plane is ours, the workloads and data stay in your account. On Google Cloud, Qovery-managed clusters run on GKE Autopilot, so you get Google's node management and Qovery's deployment workflow together.

What that changes for a team moving off AWS:

  • One application definition for both clouds. The same service, its environment variables and its database dependency deploy to an AWS cluster or a GCP cluster. You run both in parallel during the move and clone environments from one to the other.
  • Managed cluster lifecycle. Qovery provisions the cluster in your account and handles upgrades and day-2 operations, on EKS and on GKE.
  • Databases stay managed services in your account. Qovery can provision managed cloud databases such as Amazon RDS on AWS, and its Google Cloud migration paths pair GKE with Cloud SQL and Memorystore in your own project (Qovery), so moving clouds does not mean running databases yourself.
  • Git-push deployments and a preview environment per pull request, with RBAC per environment and audit logs, identical on both clouds.
  • Terraform provider for teams that want the whole setup as code, for AWS and GCP clusters alike.

Qovery does not connect the two clouds' networks for you. If services on AWS and GCP must talk privately during the transition, that is configurable with a site-to-site VPN, Cilium Cluster Mesh or Tailscale. And Qovery does not rewrite Lambda functions or DynamoDB tables; those remain migration work.

With that in place, the choice between AWS and Google Cloud is reversible. You pick the cloud for its prices, credits, data services or regions, and the operational model stays the same.

How do I set this up with Qovery from Claude Code or Cursor?

Install the Qovery Agent Skills and your coding agent can set up a cluster in your Google Cloud or AWS account and deploy your app to it. Four commands, and you only need the first one to start.

BASH|1. Agent Skills, globally (Claude Code, Cursor, Codex, OpenCode, Gemini CLI and more)
curl -fsSL https://skill.qovery.com/install.sh | bash
# 2. Or for the current project only
curl -fsSL https://skill.qovery.com/install.sh | bash -s -- --project
# 3. MCP Server in Claude Code over OAuth, read-only by default
claude mcp add --transport http qovery https://mcp.qovery.com/mcp --callback-port 4242
# 4. Read/write, only when you explicitly want it
claude mcp add --transport http qovery "https://mcp.qovery.com/mcp?read_write=true" --callback-port 4242

Your agent asks you to approve the install command. The MCP Server opens an OAuth login in your browser, starts read-only, and write mode also needs write access enabled in the Qovery Console.

Then ask in plain language: "set up Qovery on my GCP project and deploy this app" or "deploy this app to my AWS account with Qovery". The skills deploy (qovery-onboard, qovery-deploy, qovery-preview, qovery-terraform and others); the MCP Server operates what already exists. Start with the AI agent quickstart or the Claude Code integration.

Frequently asked questions
If I move a web app off AWS, what makes Google Cloud a realistic alternative instead of the same overhead with a different vendor?

Google Cloud reduces overhead when you adopt its managed defaults: GKE Autopilot for node management and automatic upgrades, Cloud Run for stateless services that scale to zero, a global VPC for multi-region setups, and automatic sustained use discounts on eligible machine types. If you recreate your AWS setup with VMs and self-managed tooling, the overhead stays the same. Identity, databases, observability, CI/CD, secrets, cost governance and on-call remain your work on either cloud.

Is GKE Autopilot easier to operate than Amazon EKS?

GKE Autopilot manages nodes, scaling and security settings, bills per Pod resource request, and upgrades clusters automatically within your maintenance windows. EKS with managed node groups leaves node and upgrade work to you, but EKS Auto Mode now automates nodes, load balancing and storage for a fee on top of EC2. Against EKS Auto Mode the gap is smaller than most comparisons suggest.

Are sustained use discounts automatic on Google Cloud?

Yes, on eligible resources. Compute Engine applies sustained use discounts automatically, up to 30% for N1, C2, M1 and M2 machine types and up to 20% for N2 and N2D when they run the whole month. Only self-serve billing accounts receive them, they do not stack with committed use discounts, and machine series outside the eligible list get none.

Does AWS charge egress fees when you migrate to Google Cloud?

AWS waives data transfer out to the internet for customers moving all of their data off AWS, or off a specific service, through credits requested from AWS Support. Google Cloud offers the same waiver for customers leaving Google Cloud. You still pay for running both environments during the cut-over.

When should I stay on AWS instead of moving to Google Cloud?

Stay on AWS if your app depends on Lambda, DynamoDB, SQS, SNS or Kinesis, if you hold Savings Plans, Reserved Instances or a spend commitment, or if your team's expertise is AWS. If your pain is node management, EKS Auto Mode may solve it without a migration. Move when your app fits Cloud Run or GKE Autopilot and you want Google's data and AI services close to it.

Can I run the same app on AWS and Google Cloud without doubling the platform work?

Yes, with a platform layer that deploys one application definition to both clouds. Qovery deploys and operates apps on Kubernetes in your own AWS and Google Cloud accounts, including cluster upgrades, managed databases and preview environments, so a parallel run during migration does not need a second platform team. Private networking between the two clouds is configured separately, for example with a VPN or Tailscale.

Mélanie Dallé
About the author
Mélanie Dallé

Melanie leads content at Qovery. She covers platform engineering trends, Kubernetes operations, FinOps, and the tools that help engineering teams ship faster.

Next step

Run your app on GCP, AWS, or both, without a DevOps hire.

Qovery deploys and operates your applications in your own Google Cloud, AWS, Azure or Scaleway account, or on your existing Kubernetes cluster, with git-push deployments, preview environments and managed cluster upgrades.