Google Cloud: Projects, Data and the Parts That Stand Out

GCP is organised around projects rather than accounts, and its strongest products are in data and machine learning. Both facts shape how you should use it.

1 September 2026 4 min read

Google Cloud is organised differently from its competitors, and the difference is not cosmetic. Understanding projects and billing accounts first will save you a lot of confusion later.

Projects and billing accounts

In GCP, the project is the fundamental container. Every resource belongs to exactly one project. APIs are enabled per project. IAM policies attach per project. Quotas apply per project. Deleting a project deletes everything inside it.

Separately, a billing account is a payment instrument. One billing account can fund many projects, and a project without a linked billing account can only use free-tier resources. Above both sits an optional organisation node with folders, for companies that want policy inheritance.

Two practical consequences: use one project per environment or per application rather than piling everything into one, and check which billing account a project is linked to before you assume the credit you were given applies to it.

Where Google Cloud is genuinely ahead

Every provider markets everything. In practice GCP’s distinctive strengths are concentrated:

BigQuery — a serverless data warehouse where you write SQL and pay for bytes scanned or a flat slot commitment. No cluster to size. For analytical workloads it remains the reference product other clouds are measured against.
Google Kubernetes Engine — Kubernetes came out of Google, and GKE is the most mature managed offering, particularly in Autopilot mode where nodes are abstracted away.
Vertex AI — a single surface for training, tuning, deploying and monitoring models, with access to Google’s own foundation models.
Networking — Google runs its own global private backbone; cross-region traffic tends to be fast and consistent.

For plain virtual machines and object storage, GCP is competent and comparable rather than distinctive.

Compute options

Product	Use it for	Billing shape
Compute Engine	Virtual machines, full control	Per second, with sustained-use discounts
Cloud Run	Containers that scale to zero	Per request and per resource-second
GKE	Kubernetes workloads	Per node, or per pod in Autopilot
Cloud Functions	Small event-driven functions	Per invocation and duration
App Engine	Managed application platform	Per instance hour

Cloud Run deserves particular attention. It runs any container that listens on a port, scales to zero when idle, and costs nothing while it is not being used. For internal tools and low-traffic APIs it is frequently the cheapest sensible option on any cloud.

How discounts work here

GCP applies two kinds automatically or by commitment:

Sustained use discounts apply automatically to Compute Engine instances that run for a large share of the month. No action required.
Committed use discounts are one- or three-year commitments to a level of vCPU and memory, in exchange for a substantial reduction. These are commitments to spend, not to specific machines, so they survive resizing.

Egress is charged per gigabyte with the usual regional variation, and BigQuery is billed on bytes scanned unless you buy slots. A poorly written query over a large table is the classic way to produce an unexpected BigQuery bill; partition and cluster your tables, and select the columns you need rather than everything.

Free tier and credits

Google offers new billing accounts a trial credit valid for a fixed period, plus an always-free tier that includes a small Compute Engine instance, a monthly Cloud Run allowance and a BigQuery query allowance. The trial credit expires; the always-free allowances do not.

Credit is attached to the billing account, so it applies to every project linked to it. The GCP configurations we list are described by their credit balance and prepayment status for that reason — see the Buy Google Cloud Account page.

Securing a GCP account

Change the Google account password and enable 2-Step Verification.
Open IAM & Admin → IAM and review every principal on each project. Remove unknown accounts.
Open IAM & Admin → Service Accounts. Service accounts with downloaded JSON keys are long-lived credentials; delete keys you did not create.
Check APIs & Services → Credentials for stray API keys and OAuth clients.
Review Billing → Account management to confirm who has billing administrator rights.
Set a budget with alert thresholds under Billing → Budgets & alerts.
Enable audit logging so there is a record of API activity.

Service account keys are the item to be strictest about. Unlike a password, a downloaded key keeps working until it is explicitly revoked.

Common questions

Why is my project not letting me create resources?

Usually one of three things: no billing account linked, the relevant API not enabled for that project, or a quota of zero for the resource in that region. The error message names which.

What is the difference between Cloud Run and Cloud Functions?

Cloud Run runs a container you build and can serve any language or framework. Cloud Functions runs a single function in a supported runtime. Cloud Run is the more general tool and has largely absorbed the use cases.

Is BigQuery expensive?

It depends entirely on query discipline. On-demand pricing charges for bytes scanned, so partitioned tables and narrow column selection keep it cheap. Unpartitioned SELECT * over a large table is where the bills come from.

Can I move a project to a different billing account?

Yes, if you have the right permissions on both. This is common when a project outgrows a trial credit and moves onto an organisation’s main account.

Where to go next

For the platform decision, AWS versus Google Cloud and Azure versus Google Cloud cover the trade-offs. Account configurations and their credit balances are on the Google Cloud account page.

Google Cloud