GCP Free Tier: What You Can Build Without Paying

"Is Google Cloud free?" has three answers, because there are three different offers, each with its own rules. Mixing them up is how learners end up either afraid to touch anything or surprised by a charge. Taken together they are generous enough to learn most of data engineering on the platform without paying.

Three offers, not one

The sandbox is BigQuery with no billing account at all. It allows 10 GiB of storage and 1 TiB of queries a month. Tables expire after 60 days, and streaming, DML statements and the Data Transfer Service are unavailable (Google Cloud: BigQuery sandbox). Nothing you do there can cost money.

The Free Trial gives new customers USD 300 of credit to spend within 90 days. Signing up requires a payment method, which Google uses to verify identity, and you are not billed during the trial. When the credit runs out or the 90 days pass, the account closes and your resources are stopped. You then have a 30-day grace period to upgrade before they are deleted (Google Cloud: free features and trial offer).

The Free Tier is permanent: a monthly allowance on about twenty products, counted per billing account. It needs a billing account, so in practice it starts with the trial and continues after you upgrade. Above the allowance you pay standard rates. Google can change the limits with 30 days' notice.

The limits that matter for data work

From the Free Tier table, the lines a data learner will use:

Two conditions are easy to miss. The Cloud Storage and Compute Engine allowances apply only in us-east1, us-west1 and us-central1; a bucket created in Mumbai is billed from the first byte. And the allowances are per billing account, not per project, so creating more projects does not multiply them.

There is also free data to practise on. BigQuery's public datasets cost nothing to store, and you pay only for the queries, which count against the same 1 TiB (Google Cloud: BigQuery public datasets).

Does a real learning project fit?

The script checks one plan against those limits: a year of trip data in BigQuery, a couple of gigabytes of files, a small event stream, a function triggered by file arrivals and a daily workflow. The usage figures are an example plan, not measurements.

# A planning calculator: one learner's monthly usage against the documented Free Tier limits.
# The limits are from Google's Free Tier table on 2 October 2026; the usage figures are an example
# plan, not measurements. Usage above a limit is billed at standard rates.
PLAN = [   # product, what is measured, planned monthly usage, free limit
    ("BigQuery",            "GiB of table storage",        4.8,     10),
    ("BigQuery",            "GiB scanned by queries",      120,     1024),
    ("Cloud Storage",       "GB stored, US regions",       2,       5),
    ("Cloud Storage",       "Class A operations (writes)", 600,     5_000),
    ("Cloud Storage",       "Class B operations (reads)",  9_000,   50_000),
    ("Pub/Sub",             "GiB of messages",             6,       10),
    ("Cloud Run functions", "invocations",                 90_000,  2_000_000),
    ("Cloud Run",           "requests",                    50_000,  2_000_000),
    ("Workflows",           "internal steps",              600,     5_000),
    ("Logging",             "GiB of log data",             2,       50),
]

print(f"{'product':20} {'measure':28} {'planned':>10} {'free limit':>11} {'used':>6}")
for product, measure, planned, limit in PLAN:
    print(f"{product:20} {measure:28} {planned:10,.1f} {limit:11,} {planned / limit:6.0%}")

over = [product for product, _, planned, limit in PLAN if planned > limit]
print(f"\nlines over their limit: {len(over)}")

# How far the same plan could grow before the first limit is reached.
factor, tightest = min((limit / planned, f"{product}, {measure}") for product, measure, planned, limit in PLAN)
print(f"first limit reached at {factor:.1f} times this usage: {tightest}")

product              measure                         planned  free limit   used
BigQuery             GiB of table storage                4.8          10    48%
BigQuery             GiB scanned by queries            120.0       1,024    12%
Cloud Storage        GB stored, US regions               2.0           5    40%
Cloud Storage        Class A operations (writes)       600.0       5,000    12%
Cloud Storage        Class B operations (reads)      9,000.0      50,000    18%
Pub/Sub              GiB of messages                     6.0          10    60%
Cloud Run functions  invocations                    90,000.0   2,000,000     4%
Cloud Run            requests                       50,000.0   2,000,000     2%
Workflows            internal steps                    600.0       5,000    12%
Logging              GiB of log data                     2.0          50     4%

lines over their limit: 0
first limit reached at 1.7 times this usage: Pub/Sub, GiB of messages

Every line fits, most with a wide margin. The plan could grow by about 1.7 times before the first limit is reached, and that limit is Pub/Sub, not BigQuery. Storage and queries, which worry beginners most, are the least of it: 120 GiB of scanning is 12% of the monthly allowance.

So a complete batch pipeline, with event-driven loading and a little streaming, can be built and run continuously for nothing.

What is not covered

The Free Tier table has no line for several services that data engineers use. If a product is not listed, check its pricing page, and assume it is billed:

• Managed Airflow environments. An environment runs, and bills, around the clock.
• Dataflow jobs, which bill for worker time.
• Spark clusters and serverless Spark jobs.
• Cloud SQL instances.
• Model training and serving.

This is what the Free Trial credit is for. Use the USD 300 deliberately, on the services the Free Tier does not cover: run a Dataflow pipeline for an afternoon, keep an Airflow environment for a week, and delete each when you have learned what you came for.

Some products have their own trials: a 90-day Spanner instance, a 30-day AlloyDB cluster and a 30-day Looker instance.

An order that costs nothing

1. Start in the sandbox. Learn BigQuery SQL on public datasets with no card.
2. Start the Free Trial when you need more: loading your own files, DML, streaming.
3. Create everything in a US free-tier region while learning.
4. Set a budget alert on day one, before creating anything. See cost controls and budgets.
5. Spend the credit on the paid services, one at a time, and delete each afterwards.
6. Before day 90, decide. Upgrade to keep your resources, or export what you want to keep. After the grace period they are gone.

How people get charged anyway

• Leaving something running. An idle cluster, an Airflow environment or a VM larger than the free one bills by the hour whether you use it or not.
• Choosing a nearby region. Sensible for latency and outside the free allowance for storage and the free VM.
• Upgrading and forgetting. Once the account is paid, anything above the Free Tier is billed, and the trial's safety net is gone.
• A query on a huge public table with every column selected. One careless statement can use a large share of the month's 1 TiB. Read the estimate before running. See BigQuery cost optimization.
• Assuming a product is free because another is.

None of these is a trap. They are the ordinary consequence of resources that stay on until you turn them off, and learning to turn them off is part of the skill.

For what to learn at each step, see the GCP roadmap. For projects that fit inside these limits, see GCP data engineering projects.

Hands-on cloud practice is part of the Data Engineering course.

Sources

• Free Google Cloud features and trial offer (Google Cloud Documentation) — Google Cloud
• Try BigQuery using the sandbox (Google Cloud Documentation) — Google Cloud
• BigQuery public datasets (Google Cloud Documentation) — Google Cloud