Is Google Cloud Cheaper Than AWS? A Per-Service Cost Breakdown (2026)

Key Points:

• Short answer: Google Cloud is usually a few percent to roughly 10-15% cheaper than AWS on on-demand general-purpose compute, because sustained-use discounts apply automatically with no commitment. AWS matches or beats that once you commit through Savings Plans or Reserved Instances, or move to Graviton ARM.
• There is no single winner - it flips per service. Google Cloud tends to win on VM list price, custom machine types, automatic discounts, and BigQuery analytics. AWS tends to win on committed-discount depth, ARM price-performance, spot capacity breadth, and storage tier variety.
• Egress and support are where headline comparisons break. Both clouds bill support as a percentage of monthly spend and both charge for internet and cross-region traffic, which routinely dwarfs a 10% VM price delta. Both now waive egress fees if you fully exit the cloud, under EU Data Act pressure.
• Managed Kubernetes is close to a tie on the control plane. Amazon EKS and GKE Standard both charge $0.10 per cluster per hour, so the node layer decides the bill. GKE Autopilot bills requested pod resources instead of whole nodes, which wins on spiky workloads and loses on densely packed clusters.
• Migrating clouds to chase a 5-15% list-price gap is almost always a bad trade. Right-sizing clusters and shutting down idle non-production environments saves more, faster. If you do want both clouds, an internal developer platform like Qovery deploys the same app into your own AWS, GCP, Azure, or Scaleway account - or your existing Kubernetes cluster - so the cloud bill and your committed-use discounts stay in your name.

Most "which cloud is cheaper" comparisons are stale within a quarter, and almost none of them survive contact with a real invoice. When I talk to CTOs about their cloud bill, the 10% list-price difference they agonized over never turns out to be the line item that hurt. Egress did. Support did. A cluster running at 8% CPU did.

So here is the honest, per-service version, with real list prices and a date attached to every number.

All prices below were verified as of August 2026. Cloud list prices change often, so re-check the official pricing pages linked throughout before you commit a dollar.

Is Google Cloud cheaper than AWS? The short answer

Google Cloud is usually a few percent to about 15% cheaper than AWS on on-demand general-purpose compute, because sustained-use discounts apply automatically with no commitment. Once you commit for one or three years, or move to Graviton, AWS often wins.

Three variables decide it for your team:

• Workload shape. Steady, always-on compute favors Google Cloud's automatic discounts. Bursty, unpredictable compute favors whoever has deeper spot capacity, which is AWS.
• Commitment appetite. If you can commit a large slice of your fleet for three years, AWS gives you the deepest discount. If you cannot, Google Cloud discounts you for doing nothing.
• Egress volume. Move a lot of data out or across regions and the network bill can erase any VM saving on either cloud.

Redo the math per region. us-east-1 versus us-central1 is different arithmetic, and Europe and APAC regions widen the spread. And remember that headline VM prices exclude support, egress, load balancers, NAT, and managed-service premiums, which is where most of the real delta lives.

This article is the cost-only companion to our broader GCP vs AWS for startups guide. I will not re-argue hiring, ecosystem, or AI stacks here. This is about the invoice.

How do GCP and AWS compare on compute (VM) pricing in 2026?

On like-for-like general-purpose VMs, Google Compute Engine list prices sit at or slightly below comparable EC2 instances, and Google then applies sustained-use discounts automatically for instances that run most of the month. AWS has no automatic equivalent - every AWS discount requires a commitment or a spot bid.

Take a concrete 4 vCPU / 16 GB pair. Google's n2-standard-4 runs about $0.19/hour in us-central1 (Compute Engine pricing), and the cost-optimized e2-standard-4 drops to about $0.13/hour. On AWS, m7i.xlarge is about $0.20/hour and the older m5.xlarge about $0.19/hour in us-east-1 (EC2 pricing). Azure's D4as v5 lands around $0.17/hour in East US (Azure VM pricing). The list-price winner is Google Cloud, and it widens if you use the E2 family.

A few mechanics that move the number more than list price does:

• Sustained-use discounts are automatic. Run an eligible instance for enough of the month and Google discounts it with no contract: up to 30% on N1, up to 20% on N2, N2D, and C2 (Google Cloud). Newer families use committed discounts instead.
• Both clouds bill per second. EC2 bills per second with a 60-second minimum on Linux (AWS); Compute Engine bills per second after a one-minute minimum (Google Cloud). Short-lived environments cost almost nothing when stopped promptly.
• Custom machine types are a real lever on Google Cloud. You can pay for the exact vCPU-to-RAM ratio instead of rounding up to the next instance size, at roughly a 5% premium over predefined shapes (Google Cloud).
• Spot goes to AWS on depth, to Google on simplicity. EC2 Spot saves up to 90% and has more instance families and deeper capacity (AWS). Google Cloud Spot VMs save 60% to 91% with no bidding and steadier pricing (Google Cloud).

Here is the AWS counter-argument, and it is a strong one. ARM changes the ranking. AWS Graviton claims up to 40% better price-performance than comparable x86 instances (AWS). Google's answer, Axion (C4A), claims up to 65% better price-performance than comparable current-generation x86, on a September 2024 comparison basis (Google Cloud). If your containers build for ARM, this is a bigger saving than the x86 list-price gap on either side.

One caveat: GPU and AI accelerator pricing follows different rules and is routinely negotiated. Do not extrapolate from CPU list prices.

Prices as of August 2026, US regions. Sources: EC2, Compute Engine, Azure VMs; discounts: SUD, Savings Plans, CUDs, Graviton, Axion.

Which saves more: Google Cloud sustained-use and CUDs or AWS Savings Plans and Reserved Instances?

AWS offers the deeper maximum discount - up to 72% for three-year commitments - while Google Cloud gives you a meaningful discount for doing nothing at all. Which is cheaper depends on how much of your fleet you can confidently commit for years.

The three models work differently:

• Google sustained-use discounts are automatic, need no commitment, and apply per instance family per region, up to 30% (Google Cloud).
• Google committed use discounts come in two flavors. Resource-based CUDs reach up to 55% on general families and up to 70% on memory-optimized ones. Flexible, spend-based CUDs are about 28% for one year and 46% for three, and they float across machine families (Google Cloud).
• AWS commitments run deeper but demand more. Compute Savings Plans reach up to 66% and stay flexible across instance family, size, region, and OS; EC2 Instance Savings Plans and Standard Reserved Instances reach up to 72% but lock you to a family and region; Convertible Reserved Instances reach up to 66% and can be exchanged (AWS Savings Plans, AWS Reserved Instances).

The real-world catch is not the headline percentage, it is coverage. Teams under-commit and leave discounts on the table, or over-commit and pay for capacity they stopped using. Flexera's 2026 survey puts wasted cloud spend at about 29%, and 17% of organizations reported exceeding their public cloud budget in the past year (Flexera). A 46% discount on capacity you no longer use is still money burned.

Azure is the third data point, and it has one genuinely unique lever. Azure Reservations reach up to 72%, and Azure Hybrid Benefit adds up to 80% for Windows Server and up to 85% for SQL Server if you already own the licenses (Azure Hybrid Benefit). For a Windows and SQL estate, that is a real Azure-only advantage.

My practical rule: commit to your steady-state floor on whichever cloud you run, and run the volatile layer on spot or on-demand. And know that marketplace and enterprise-agreement spend can count toward commitments, which quietly distorts every published comparison of large accounts.

As of August 2026. Sources: SUD, CUDs, AWS Savings Plans, AWS RIs, Azure Hybrid Benefit.

Is GKE cheaper than EKS for managed Kubernetes?

GKE Standard and Amazon EKS both charge $0.10 per cluster per hour for the control plane, so managed Kubernetes cost is decided at the node layer, not the control plane (EKS pricing, GKE pricing). The one structural difference is GKE Autopilot, which bills requested pod CPU, memory, and storage instead of whole nodes.

Autopilot charges roughly $0.0445 per vCPU-hour, $0.0049 per GB-hour, and $0.0001389 per GB-hour of ephemeral storage for general-purpose pods in us-central1 (GKE pricing). That wins for spiky or poorly packed workloads, because you pay for requests rather than idle node headroom. It loses on dense clusters you already bin-pack well, where whole-node billing is cheaper.

Control-plane fees are close, but the fine print differs:

• EKS: $0.10 per cluster per hour, jumping to $0.60 per cluster per hour once a Kubernetes version enters extended support (months 15 to 26 after release) (AWS). Staying current avoids that surcharge.
• GKE: $0.10 per cluster per hour, with a $74.40 per month credit that covers one zonal or Autopilot cluster (Google Cloud).
• AKS: a Free tier with no control-plane charge and no financially backed SLA, or a Standard tier at about $0.10 per hour with an uptime SLA (Azure).

Then come the hidden line items that decide the actual bill. On EKS: NAT Gateway hourly plus per-GB processing, load balancers, EBS volumes, and cross-AZ traffic at $0.01/GB in each direction. On GKE: Cloud NAT, load balancing, persistent-disk classes, and inter-zone egress. These network and storage charges routinely exceed the control-plane fee.

The number that should reset your priorities: CAST AI's 2026 benchmark of tens of thousands of production clusters found average CPU utilization of just 8% and memory utilization of 20% before optimization (CAST AI). Utilization beats list price every time. A right-sized EKS cluster is cheaper than an oversized GKE cluster, and the reverse is equally true. Whichever cloud is cheaper for your specific node shape usually wins the cluster.

As of August 2026. Sources: EKS, GKE, AKS, utilization: CAST AI 2026.

What do storage, databases, and egress cost on GCP vs AWS?

Standard object storage list prices on Amazon S3 and Google Cloud Storage are within a few percent of each other, so the difference lands in retrieval fees, tiering rules, and egress. Egress is the line item that blows up bills on both clouds, and cross-AZ traffic on AWS is the single most commonly missed charge.

Object storage is nearly a wash: S3 Standard is $0.023/GB-month in us-east-1 (AWS), Google Cloud Storage Standard is $0.020/GB-month (Google Cloud), and Azure Blob Hot is $0.018/GB-month (Azure). Archive tiers separate them on behavior more than price: S3 Glacier Deep Archive and Azure Archive both sit near $0.00099/GB-month with multi-hour retrieval, while Google Cloud Archive is $0.0012/GB-month but retrieves in milliseconds for a per-GB fee. Watch the minimum-storage-duration penalties on all three.

Block storage is where AWS reads simpler. EBS gp3 is $0.08/GB-month and includes a free baseline of 3,000 IOPS and 125 MB/s (AWS). Google Compute Engine pd-balanced is about $0.10/GB-month and pd-ssd about $0.17/GB-month, with IOPS and throughput scaling with capacity rather than a fixed free baseline (Google Cloud).

Managed Postgres prices as instance plus storage plus I/O on every cloud. A small production RDS PostgreSQL instance (db.m6g.large, 2 vCPU / 8 GB, single-AZ, us-east-1) is about $0.159/hour, roughly $116/month for compute (AWS); Multi-AZ roughly doubles that. Aurora PostgreSQL on a comparable db.r6g.large is about $0.276/hour, and its I/O-Optimized tier costs more per instance but drops per-request I/O charges entirely, which changed the math for write-heavy workloads (AWS). Google prices Cloud SQL and AlloyDB per vCPU and per GB rather than per instance shape - Cloud SQL Enterprise is about $0.0413/vCPU-hour plus $0.007/GB-hour (Google Cloud), which works out slightly below the equivalent RDS instance for a basic 2 vCPU / 8 GB configuration.

Now the part that eats budgets. Internet egress is $0.09/GB on AWS after a 100 GB monthly free allowance (AWS), and $0.12/GB on Google Cloud's Premium Tier or about $0.085/GB on Standard Tier (Google Cloud). On top of that, AWS charges $0.01/GB in each direction for cross-AZ traffic inside a region, which Google prices differently as inter-zone egress. A chatty microservice mesh spread across availability zones can quietly out-cost your compute.

Two things worth knowing:

• Both clouds now waive egress when you fully leave. Under EU Data Act pressure, AWS published "Free data transfer out to internet when moving out of AWS" and Google published "Eliminating data transfer fees when migrating off Google Cloud". Both require support approval and apply only to full exits (AWS gives 90 days, Google 60).
• Free tiers diverge. AWS moved new accounts to a credits-based plan in 2025, up to $200 in credits over a six-month window plus 30-plus always-free services (AWS). Google keeps a $300 credit for 90 days plus 20-plus always-free products (Google Cloud). Azure offers $200 for 30 days plus a set of 12-month and always-free services (Azure).

As of August 2026, US regions. Sources: S3, GCS, Azure Blob, EBS, GCE disks, AWS data transfer, GCP network, RDS, Cloud SQL.

When is Google Cloud actually cheaper, and when is AWS cheaper?

Pick Google Cloud when you run steady-state compute you will not commit to, want discounts without FinOps headcount, or run heavy analytics on BigQuery. Pick AWS when you can commit at scale, your workloads run well on Graviton ARM, you lean on deep spot capacity, or you are already invested in AWS-native services. Pick Azure when you have a Windows and SQL estate to bring under Hybrid Benefit.

Analytics is a genuine Google Cloud edge for some teams. BigQuery is $6.25 per TiB scanned on demand with the first TiB free each month (Google Cloud). On AWS, Amazon Athena is $5.00 per TB scanned (AWS) and Redshift Serverless is $0.375 per RPU-hour (AWS). Which is cheaper depends entirely on query patterns, but for exploratory, scan-heavy analytics BigQuery's model is hard to beat operationally.

Support distorts everything, because both clouds bill it as a percentage of monthly spend. AWS Business Support starts at the greater of $29/month or 9% of spend, tiering down to 3% at scale, and Enterprise Support starts at $5,000/month (AWS). Google Cloud Customer Care mirrors this: Standard is the greater of $29/month or 3%, Enhanced starts at $100/month on a 10-7-5-3% scale, and Premium starts at $15,000/month (Google Cloud). On a large account, support alone can exceed the entire VM-price gap between the two clouds.

And published list prices understate what big accounts actually pay. Enterprise agreements, private pricing, and marketplace spend counting toward commitments all bend the real number down in ways no comparison table captures.

Here is the credibility point, stated plainly: for most teams, migrating clouds to chase a 5-15% list-price gap is a bad trade. The migration cost nobody prices - rewriting Terraform, CI/CD, IAM, networking, and observability per cloud, plus retraining the team - usually dwarfs the saving. Right-sizing clusters and shutting down idle non-production environments saves more, faster, on the cloud you already run. Given the market is dominated by AWS at 28%, Azure at 20%, and Google Cloud at 15% (Synergy Research Group), most teams also have more existing skills and tooling on AWS, which is a real switching cost.

The order of operations is not "switch clouds." It is: fix utilization first, then right-size commitments, then fix egress architecture, and only then consider changing providers.

As of August 2026. Sources: BigQuery, Athena, Redshift, AWS support, Google support, market share.

Can you run on both AWS and Google Cloud without doubling your platform work?

Yes, if your applications are containerized and your deployment workflow is cloud-agnostic, you can place each workload where it is cheapest and still keep committed-use discounts on both clouds. The blocker is rarely the application. It is rebuilding the platform layer once per cloud.

What actually has to be rebuilt per cloud is the unglamorous part: cluster provisioning, VPC and networking, IAM, CI/CD wiring, secrets management, observability, and environment lifecycle. Kubernetes ports your compute cleanly, but managed databases, queues, and IAM do not. That platform tax, not the app, is what makes multi-cloud expensive.

This is where an internal developer platform earns its place, and where Qovery fits. Qovery deploys the same application into your own AWS, GCP, Azure, or Scaleway account, or your existing Kubernetes cluster, with one git-push workflow, so the platform layer is built once instead of once per cloud.

Lead with the economics, because they are the point. Qovery runs BYOC (bring your own cloud) inside your own accounts, so the cloud bill stays in your name. Your Savings Plans, your committed-use discounts, and any negotiated enterprise pricing keep applying, and nothing is resold to you with a margin. That is the opposite of a managed PaaS that hosts your workload on its account and marks up the infrastructure.

The cost levers that work identically on any of those clouds are concrete:

• Environment auto-stop for non-production, so staging and sandbox stop billing when nobody is using them.
• Ephemeral preview environments per pull request, destroyed on merge, which is why they cost almost nothing.
• Managed cluster upgrades, so you stay current and avoid the EKS extended-support surcharge of $0.60 per cluster per hour.
• Per-environment RBAC and databases backed by managed cloud services, such as an RDS instance on AWS.

Be fair about the alternatives. Heroku and similar managed PaaS are genuinely simpler to start with, but they run on someone else's account at a markup and your discounts do not apply. StackIt and other European providers matter for sovereignty and data residency rather than raw price. Raw Terraform gives you full control, but you maintain a separate stack per cloud forever. And multi-cloud is a bad idea for small teams, single-region apps, and anything with heavy cross-cloud data transfer, where egress will eat the savings you were chasing. If lock-in is your worry, we wrote separately about the high cost of vendor lock-in.

FAQs

Is Google Cloud cheaper than AWS overall in 2026?

Usually a little, on on-demand general-purpose compute. A 4 vCPU / 16 GB VM is about $0.19/hour on Google Cloud (n2-standard-4) versus about $0.20/hour on AWS (m7i.xlarge) as of August 2026, and Google adds automatic sustained-use discounts up to 30% with no commitment. AWS closes and often reverses that gap once you commit (up to 72%) or move to Graviton ARM.

Is GKE cheaper than Amazon EKS?

The control plane is a tie: both GKE Standard and EKS charge $0.10 per cluster per hour as of August 2026, and GKE credits $74.40/month for one cluster. The node layer decides the bill, so whichever cloud is cheaper for your instance shape usually wins. GKE Autopilot bills requested pod resources instead of whole nodes, which is cheaper for spiky workloads and more expensive for clusters you already bin-pack tightly.

Which cloud has the cheapest data egress: AWS, Google Cloud, or Azure?

For standard internet egress, AWS and Azure list around $0.09/GB and $0.087/GB after a 100 GB monthly free tier, while Google Cloud Premium Tier is $0.12/GB (or about $0.085/GB on Standard Tier), as of August 2026. Watch AWS cross-AZ traffic at $0.01/GB each direction, which is the most commonly missed charge. All three now waive egress entirely if you fully migrate off the cloud.

Do Google Cloud sustained-use discounts beat AWS Savings Plans?

Not on maximum depth. Sustained-use discounts top out around 30% and apply automatically with zero commitment, while AWS Savings Plans and Reserved Instances reach up to 72% but require a one or three-year commitment. If you cannot commit much of your fleet, Google wins; if you can commit at scale, AWS wins.

Is Google Cloud's free tier better than the AWS free tier?

Google Cloud gives a larger upfront credit: $300 for 90 days plus 20-plus always-free products, versus AWS's up to $200 in credits over a six-month window plus 30-plus always-free services, as of August 2026. Azure sits between them with $200 for 30 days. For a short evaluation Google's credit stretches furthest; for long-running always-free services the catalogs are comparable.

Is it worth migrating from AWS to Google Cloud just to save money?

Almost never for a 5-15% list-price gap. Rewriting Terraform, CI/CD, IAM, networking, and observability per cloud, plus retraining, usually costs more than the saving. Right-sizing clusters (the average runs at 8% CPU and 20% memory per CAST AI's 2026 data) and shutting down idle non-production environments saves more, faster, on the cloud you already run.

Fix utilization first, right-size your commitments second, and treat a cloud migration as the expensive last resort it usually is. If you want to run one cloud well, or both without doubling the platform work, you can try Qovery free or come argue the numbers with us in the Qovery Discord community.