Google Cloud Platform (GCP, these days often just Google Cloud) is Google’s public cloud: virtual machines, container hosting, databases, storage and data analytics services, usually billed per second rather than per hour. At smaller companies, the Google name often comes up only because of Google Workspace (email, Drive, calendar), yet the two products are separate worlds: Workspace is a ready-made office service, while GCP is a set of building blocks from which you have to assemble the infrastructure yourself.
In this article we look at when Google Cloud Platform is a good choice for a small or medium-sized business, when it isn’t, and what to watch out for if you go ahead.
What does Google Cloud Platform offer?
Of GCP’s more than a hundred services, a smaller company usually needs only a few. The most important ones, with their AWS equivalents:
| GCP service | What is it for? | AWS equivalent |
|---|---|---|
| Compute Engine | Virtual machines (Linux or Windows) | EC2 |
| Cloud Run | Serverless container hosting, billed per request | App Runner, Fargate |
| Google Kubernetes Engine (GKE) | Managed Kubernetes | EKS |
| Cloud SQL | Managed MySQL, PostgreSQL and SQL Server | RDS |
| Cloud Storage | Object storage for files and backups | S3 |
| BigQuery | Data warehouse for querying large amounts of data | Redshift, Athena |
The project is the basic unit of GCP: every resource belongs to a project, and projects belong to a billing account. If you give every application and environment (test, production) its own project, both costs and permissions stay easy to follow.
When is Google Cloud Platform a good choice?
For containerised web applications. Cloud Run is one of GCP’s strongest services: you upload a Docker image, and the platform starts it, scales it, puts HTTPS in front of it, and you only pay for the time spent serving requests. A low-traffic internal application or API can cost as little as a few euros a month this way, because when there are no requests, no instance runs.
If you want to use Kubernetes. Kubernetes was born at Google, and GKE is still one of the most convenient managed versions. In Autopilot mode you don’t even have to deal with the nodes: on top of the cluster management fee, you pay for the CPU and memory requested by the pods.
For data analytics. In BigQuery you can query terabyte-sized datasets with SQL, without your own database server. If the company already works with Google Analytics or Google Ads data, these can be exported directly to BigQuery.
If you are in the Google ecosystem. If the company uses Workspace, its users and groups also appear in GCP’s permission management (IAM), so you don’t have to maintain separate accounts.
If you want transparent pricing. GCP bills most resources to the second, and for certain machine types it gives an automatic discount if a machine runs for most of the month (sustained use discount). Compared with AWS, there are fewer hidden settings and fewer region-specific quirks.
When is GCP not the right answer?
- For a single traditional server. If you want to run a web shop or a few WordPress sites on one machine, the same performance is usually cheaper at a VPS provider or at OVHcloud, and the fee is fixed. We wrote about how a VPS works in more detail on the blog of aWh, our hosting brand.
- In a heavily Microsoft-based environment. If everything is built on Active Directory, Microsoft 365 and Windows servers, there is less friction with Azure.
- If you need a fixed monthly fee. In the cloud, the bill depends on usage. You have to pay for outbound data traffic (egress), for example, and a misconfigured service or a sudden rise in traffic can cause a surprise.
- If nobody is going to run it. GCP is not a “switch it on and you’re done” service: someone has to set up and maintain the network, the firewall rules, the backups and the updates.
Which region should you choose?
There is no GCP region in Hungary; the nearest ones are:
| Region | Location |
|---|---|
europe-central2 |
Warsaw |
europe-west3 |
Frankfurt |
europe-west10 |
Berlin |
europe-west8 |
Milan |
From a data protection point of view, it’s also simpler if the data stays within the EU; you can enforce this by choosing the region and with an organization policy, so that nobody creates resources anywhere else. Keep in mind that not every service and machine type is available in every region, and prices differ by region: Frankfurt is usually more expensive than the US regions.
How much does it cost?
A new account gets $300 in credit, which you can use within 90 days, and Google doesn’t charge your card during the trial period. There is also a permanent free tier, for example one e2-micro virtual machine and 30 GB of standard disk per month, or two million requests a month in Cloud Run. The free e2-micro, however, is only available in three US regions (us-west1, us-central1, us-east1), so it’s not suitable for a production system serving customers in Hungary – but it’s fine for learning and experimenting. The exact terms are on the Google Cloud free features page.
In production use, the main cost factors are:
- Compute capacity: the size and running time of the virtual machines. A one- or three-year commitment (committed use discount) gets you a significant discount, and preemptible Spot VMs are much cheaper for tasks that can tolerate being stopped.
- Storage: disks, snapshots, object storage.
- Network: outbound traffic, public IP addresses and load balancers.
- Managed services: a Cloud SQL instance is more expensive than the same database on your own virtual machine, but in return Google takes care of backups, updates and failover.
A welcome development is that in January 2024 Google eliminated the data transfer fee for those who move off GCP completely. This doesn’t apply to everyday traffic, but it reduces the risk of vendor lock-in.
How to get started
For a smaller company’s GCP environment, we recommend these steps before the first application:
- Organisation and projects. If you have a company Workspace or Cloud Identity domain, create the organisation, and put the projects under it rather than tying them to a personal account.
- Budget and alerts. Set up a budget with alerts. Important: an alert doesn’t stop the services, it only notifies you.
- Permissions. Nobody should work with the Owner role day to day, and every account should have two-step verification.
- Region rule. Use an organization policy to fix which regions resources can be created in.
- Infrastructure as code. It’s worth describing the resources in Terraform from the very beginning, so that the environment is reproducible and documented.
With the command-line tool (gcloud), a Debian server in Warsaw and a monthly budget can be created like this:
gcloud compute instances create web-1 \
--zone=europe-central2-a \
--machine-type=e2-small \
--image-family=debian-12 \
--image-project=debian-cloud
gcloud billing budgets create \
--billing-account=XXXXXX-XXXXXX-XXXXXX \
--display-name="Monthly budget" \
--budget-amount=100EUR \
--threshold-rule=percent=0.5 \
--threshold-rule=percent=0.9
The currency of the budget has to match the currency of the billing account.
GCP, AWS or OVHcloud?
There is no overall winner. AWS has the widest range and is the most widespread, so it’s the one with the most specialists and guides – but its pricing is the most complex: according to the AWS announcement, for example, since February 2024 you have to pay a separate hourly fee for every public IPv4 address. GCP is strong for container and data-centred workloads, and it’s simpler to manage. OVHcloud is a European provider with predictable fees and cheap data traffic, but fewer managed services. For smaller companies, the right solution is often to run traditional servers at a provider with predictable pricing, and to move only those parts to GCP or AWS where a managed service gives a real advantage.
If you already use Google Workspace, our article on Workspace migration may also be useful. We help with designing, building and running GCP, AWS and OVHcloud environments; you can read about this on the Cloud and DevOps page.