GCP Free Tier Explained: Free Trial, Always Free, and What Can Still Charge You

What the GCP free tier means in simple terms

Google Cloud has two separate free offers. The first is a 90-day free trial: new accounts get $300 in credits that can be spent on almost anything in GCP. The second is Always Free: a permanent set of monthly limits on specific services that never expire, even after the trial is long over.

The confusion comes from how they interact. During the trial, both are active at the same time. Usage charges draw from the $300 credits first. When the trial ends, the credits disappear and only the Always Free limits remain. Stay within those limits and you pay nothing. Go above them and you are charged at standard rates.

The other source of confusion: not every GCP service has an Always Free tier. Cloud SQL, Dataflow, Pub/Sub, and many others are not in the Always Free programme. They consume trial credits during the trial period, then charge your card afterward. By the end of this page, you will know exactly which is which.

How the GCP free tier works, step by step

1. You create a Google account and sign up for Google Cloud.

2. You add a payment method. GCP requires this to verify your identity, but your card is not charged during the trial.

3. If you are a new billing account, you receive $300 in credits. The 90-day clock starts at the moment you create the billing account, not when you first use a service.

4. You create a project and link it to your billing account. Without a linked billing account, most GCP services refuse to create resources. See GCP Projects for how projects, billing accounts, and resources fit together.

5. During the trial, usage is deducted from the $300 credits. Always Free limits still exist in parallel, but since the trial credit covers everything, you may not notice the quotas yet.

6. When the trial ends (90 days or $300 spent, whichever comes first), your account is paused. Resources stop. Data is retained temporarily.

7. To continue, you must manually upgrade to a paid account. Google does not charge you automatically. You have to opt in.

8. After upgrading, Always Free limits apply permanently. Stay within them and pay nothing. Exceed them and you are charged at standard GCP pay-as-you-go rates.

Free trial vs Always Free: the key differences

The practical implication: the trial is your window to explore anything, including services with no permanent free tier. After the trial, you are limited to specific services and specific quotas. Cloud SQL, larger VMs, and most managed services cost money on a paid account.

What is included in the Always Free tier

These quotas reset every calendar month and apply to all accounts with billing enabled (not just new ones). All quotas are shared across all projects linked to one billing account, not allocated per project.

Compute Engine

One e2-micro VM per month, but only in three US regions: us-west1, us-central1, or us-east1. This includes 30 GB of standard persistent disk storage and 1 GB of network egress from North America per month. An e2-micro is a small shared-CPU instance. It is enough for a personal project, a simple web server, or a learning environment, but it will feel slow for anything heavier.

Region matters here. An e2-micro in europe-west1 or asia-east1 is not free. See Regions and Zones for how GCP organises its infrastructure geographically. For a full overview of machine types and sizing, see Compute Engine Overview.

Cloud Storage

5 GB of Regional storage per month in the US regions (us-east1, us-west1, us-central1). Also includes 5,000 Class A operations (writes, object changes) and 50,000 Class B operations (reads, metadata) per month, plus 1 GB of egress to non-Google destinations. For most beginner use cases like static files, small uploads, and simple data storage, this is adequate.

Cloud Run

2 million requests per month, 360,000 GB-seconds of memory, and 180,000 vCPU-seconds of compute per month. Cloud Run is one of the best services for beginners precisely because it only charges while handling requests. Idle services cost nothing. The free tier covers a lot of traffic for a small API or web app. Read more in the Cloud Run Overview.

Cloud Functions

2 million invocations per month, 400,000 GB-seconds and 200,000 GHz-seconds of compute, plus 5 GB of network egress per month. Like Cloud Run, Cloud Functions only charges when actively running. There is no cost for idle time. See Event-Driven Compute with Cloud Functions for more.

BigQuery

10 GB of storage and 1 TB of query processing per month. The 1 TB query allowance is more generous than it sounds for learning. A typical query on a small dataset uses megabytes, not terabytes. Beginners rarely approach the 1 TB limit when working with sample data. For details on how BigQuery charges work, see BigQuery Pricing Explained.

What is not free: services that commonly charge beginners

The following services and situations fall outside Always Free and will charge your billing account once the trial ends, or if you have already upgraded to paid billing:

Cloud SQL

No Always Free tier. A Cloud SQL instance charges by the hour whether or not it has any connections. A db-f1-micro instance in us-central1 costs roughly $7–10 per month just to exist.

e2-micro VMs outside eligible US regions

The free Compute Engine instance only applies in three specific US regions. An e2-micro in europe-west1, asia-east1, or anywhere outside those three US regions bills at the full on-demand rate (approximately $6–8 per month).

Larger VM machine types

Only e2-micro qualifies for the Always Free compute tier. An e2-small, n1-standard-1, or any other machine type charges immediately, even in a free-tier US region. See Machine Types Explained for what the different options cost.

Network egress beyond free limits

Moving data out of GCP to the internet costs money above the small free allowances. Egress between regions also costs money. Keeping resources in the same region minimises these charges. For a full breakdown, see Network Egress Costs.

Unattached static IP addresses

Reserve an external static IP address, delete the VM it was attached to, and forget to release the IP. GCP charges a small hourly fee for reserved but unused external IP addresses. Individually small, but they accumulate invisibly across projects.

Services with no Always Free tier at all

Many GCP services have no free quota: Pub/Sub, Dataflow, Memorystore, Cloud Spanner, GKE cluster nodes, Cloud Armor, and others. Using them on a paid billing account charges from the first unit of usage. Always check a service’s pricing page before starting a tutorial that uses an unfamiliar service.

When the free tier is enough

The GCP free tier covers a lot of real beginner use cases:

• Learning the Cloud Console. Exploring the UI, navigating services, and reading dashboards is entirely free. See the Cloud Console Guide.

• Deploying a small Cloud Run service. A containerised API or web app handling a few thousand requests per day sits well within the 2 million free monthly requests.

• Running BigQuery on sample datasets. Google provides public datasets you can query instantly. A few dozen queries on small data uses megabytes, not terabytes.

• Experimenting with an e2-micro VM in us-central1. Set up a small Linux server, practice gcloud CLI commands, or run a simple web server.

• Trying Cloud Functions for small event-driven tasks. 2 million invocations per month covers most hobby project needs easily.

• Building beginner portfolio projects. A static frontend on Cloud Storage, a small Cloud Run backend, and BigQuery for data analysis can all stay within the free tier if usage stays modest.

When the free tier is not enough

The free tier is designed for learning and small experiments, not production. These scenarios will require paid usage:

• Any always-on database. Cloud SQL, Firestore above free limits, and Memorystore charge continuously whether or not they are receiving traffic.

• 24/7 VMs beyond the single e2-micro. Anything running continuously outside the one free e2-micro in a US region accrues hourly charges.

• Multi-region or high-availability architectures. Replicating across regions, load balancers, and premium networking all have costs beyond the free allowances.

• Larger compute or memory requirements. Anything heavier than e2-micro on Compute Engine, or Cloud Run services that exceed the free compute quotas under high traffic.

• Team environments with multiple projects. Always Free quotas are per billing account. Spreading work across projects does not multiply your free allowances.

• Real traffic at scale. 2 million Cloud Run requests sounds like a lot, but a public API with real users can exceed that in days.

How to avoid charges while learning

These steps cover the situations responsible for almost all beginner surprise bills:

1. Set a billing budget and alert before creating any resources. Go to Billing > Budgets & alerts in the Cloud Console. Create a budget for $1–5 per month with alert thresholds at 50% and 100%. Any charge above $1 is a signal that something has escaped the free tier. Full guide at Billing Budgets and Alerts.

2. Always create resources in eligible US regions. Use us-central1 as your default region while learning. All the major Always Free services have their free quotas in this region.

3. Prefer Cloud Run and Cloud Functions over Compute Engine VMs. Serverless services only cost money while handling requests. A VM costs money whether idle or busy. For a beginner HTTP API or background task, Cloud Run is almost always cheaper and safer.

4. Delete resources immediately after finishing a tutorial. Do not leave Cloud SQL instances, VMs, or GKE clusters running “just in case.” Recreate from scratch next time. The tutorial will still be there.

5. Check that resources are actually gone after deleting them. Use these commands to confirm nothing is left running:

# Check for running VMs
gcloud compute instances list

# Check for Cloud SQL instances
gcloud sql instances list

# Check for reserved static IPs (unattached ones cost money)
gcloud compute addresses list

# Check which billing account is linked to your project
gcloud billing projects describe PROJECT_ID

6. Review the billing dashboard once a week. The billing page in the Console shows running costs per service. If something unexpected appears, you catch it early. See Cost Breakdown Reports for how to read what is charging and why.

A safe first week on GCP

If you want to start learning without worrying about charges, this order works well:

Safest services to start with

• BigQuery with public datasets. Google provides public data you can query in seconds. Zero cost for the first 1 TB per month.
• Cloud Run with a simple container. Deploy a small “Hello, world” app. Only charges when receiving requests. Delete it when done.
• Cloud Shell. A free browser-based terminal with gcloud pre-installed. No VM required. See Cloud Shell.
• Cloud Console exploration. Browsing the UI, reading service pages, and checking dashboards costs nothing.

Riskiest services from a cost perspective

• Cloud SQL. No Always Free tier. Charges per hour from the moment it exists.
• GKE clusters. A cluster with nodes charges continuously for VM capacity. Not a “try and forget” resource.
• Compute Engine VMs in non-US regions. Easy to create in the wrong place, easy to forget about.
• Any service with no Always Free tier on a paid billing account after the trial ends.

A simple first-week learning path

1. Sign up, create a project, link billing (see GCP Projects)
2. Set a $5 billing budget alert in Billing > Budgets & alerts
3. Explore the Cloud Console for 30 minutes and navigate the major service menus
4. Run a BigQuery query on a public dataset (zero cost)
5. Deploy a “Hello, world” Cloud Run service using Cloud Shell
6. Delete the Cloud Run service after testing
7. Check the billing dashboard and confirm $0 charged

GCP free tier vs AWS free tier

Both providers offer a mix of trial credits and permanent free tiers, but the structure differs. AWS gives 12 months of specific service free tiers for new accounts, plus some permanently free services. GCP gives a 90-day credit with a higher dollar value ($300) that can be spent on any service, plus a permanent Always Free set.

For beginners choosing where to start: GCP’s $300 credit is more flexible because it applies to any service. AWS’s 12-month window is longer but tied to specific service limits. Both carry the same fundamental risk: leaving resources running after experiments.

The safest learning approach is the same on either platform. Prefer serverless services (Cloud Run on GCP, Lambda on AWS) over always-on VMs and databases, and delete resources as soon as you are done with them.