Introduction to Google Cloud Platform (GCP)
What Google Cloud is, how projects and billing accounts organize it, its core services, where it is strong and weak, and how to start safely.
Google Cloud is the public cloud built on the infrastructure Google uses for Search, Gmail and YouTube. Its individual services will look familiar if you know AWS or Azure: virtual machines, object storage, managed databases, Kubernetes. What sets it apart is mostly in the details: a global private network, networks that span regions, a serverless data warehouse at the center of its analytics story, and a resource model built around projects.
This article gives you the map. It covers how Google Cloud is organized, which services make up the core, where the platform is strong and where it is weaker, how current product names line up with the older ones in many tutorials, and how to create and clean up your first project from the command line. The rest of the Exploring Google Cloud Platform (GCP) series goes deeper on each piece.
Google Cloud, GCP and what is in scope
Google uses “Google Cloud” for its whole cloud business, including Google Workspace and its AI products for businesses. “Google Cloud Platform”, or GCP, is the older name for the infrastructure and developer services: compute, storage, networking, databases, data and AI. You’ll see both in documentation and job ads. This article, like the rest of the series, uses Google Cloud and GCP to mean those infrastructure services.
Google Cloud runs in regions spread around the world, each split into three or more zones. GCP Global Infrastructure explains regions, zones and multi-regions, and which resources are tied to each.
How Google Cloud is organized
Everything you create in Google Cloud belongs to a project. Projects are the most important concept to learn early, because almost everything hangs off them: which APIs are turned on, who has access, which quotas apply and which billing account pays.
flowchart TD
O["Organization: example.com"] --> F1["Folder: Production"]
O --> F2["Folder: Sandbox"]
F1 --> P1["Project: shop-prod"]
F1 --> P2["Project: data-prod"]
F2 --> P3["Project: alice-experiments"]
P1 --> R1["Compute Engine VM"]
P1 --> R2["Cloud Run service"]
P2 --> R3["BigQuery dataset"]
P3 --> R4["Cloud Storage bucket"]
B["Cloud Billing account"] -.->|pays for| P1
B -.->|pays for| P2
B -.->|pays for| P3
From the top down:
- Organization. The root node, tied to a Google Workspace or Cloud Identity domain. Personal accounts with a Gmail address don’t have one, and their projects sit without a parent. Companies almost always do.
- Folders. Optional groupings of projects, often by department or environment. IAM policies set on a folder are inherited by every project inside it.
- Projects. The container for resources, and the boundary for APIs, quotas and most permissions.
- Resources. VMs, buckets, datasets, Cloud Run services and everything else you pay for.
Billing sits alongside the hierarchy rather than inside it. A Cloud Billing account holds the payment method, and each project is linked to one billing account at a time. One billing account can pay for many projects, which is how most organizations work. GCP Free Tier and Billing Overview covers billing accounts, budgets and alerts in detail.
Three identifiers for one project
Every project has three identifiers, and mixing them up is a common source of confusion:
| Identifier | Example | Can you change it? | Where you use it |
|---|---|---|---|
| Project name | Shop Production | Yes | Display only |
| Project ID | shop-prod-4821 | No, and it is globally unique | gcloud, APIs, resource names |
| Project number | 123456789012 | No, assigned by Google | Some APIs and default service account names |
Choose project IDs carefully. A project ID can’t be changed later, and once used, it can’t be reused, even after the project is deleted.
APIs are turned on per project
A new project can’t create VMs or buckets until the matching API is enabled. The console usually prompts you to enable it the first time you open a product page; from the command line you do it with gcloud services enable. This feels like friction at first, but it means a project only exposes the services it actually uses.
Access control follows the hierarchy
Google Cloud’s Identity and Access Management (IAM) grants roles to principals, such as users, groups and service accounts, at a level of the hierarchy. A role granted on a folder applies to every project in it; a role granted on a project applies to every resource in that project. GCP Identity and Access Management explains roles, service accounts and how to keep grants narrow.
A good default is one project per application per environment, such as shop-dev and shop-prod. Deleting a project is also the cleanest way to remove an experiment, because it removes everything inside it.
You want to rename a project from "Test" to "Shop Staging". Which identifier can you change?
Correct! The project name is a display label you can change; the project ID and number are permanent.
The core services
Google Cloud has well over a hundred products, but most applications use a small set of them. The table maps common needs to the Google Cloud service and its nearest equivalent on AWS and Azure, which helps if you already know one of those.
| Need | Google Cloud | AWS | Azure |
|---|---|---|---|
| Virtual machines | Compute Engine | EC2 | Azure Virtual Machines |
| Managed Kubernetes | Google Kubernetes Engine (GKE) | EKS | AKS |
| Serverless containers | Cloud Run | ECS on Fargate | Azure Container Apps |
| Functions | Cloud Run functions | Lambda | Azure Functions |
| Object storage | Cloud Storage | S3 | Azure Blob Storage |
| Private network | VPC (global) | VPC (regional) | Azure Virtual Network |
| Managed relational database | Cloud SQL, AlloyDB | RDS, Aurora | Azure SQL Database, Azure Database for PostgreSQL |
| Globally distributed database | Spanner | Aurora DSQL | Azure Cosmos DB |
| Data warehouse | BigQuery | Redshift | Microsoft Fabric |
| Container images and packages | Artifact Registry | ECR, CodeArtifact | Azure Container Registry |
Compute
Compute Engine gives you Linux or Windows VMs where you manage the operating system. You choose a machine series, such as general-purpose E2, N4 or C4, compute-optimized, memory-optimized or accelerator-optimized, and can define custom machine types with the exact vCPU and memory you need. Exploring GCP Compute Engine Instances explains the families, and Launching Your First GCP Compute Engine Instance walks through creating one.
Above VMs, Google Cloud offers managed ways to run code:
- Cloud Run runs containers, or source code it builds into a container, without servers to manage. It scales with requests and down to zero when idle, and it also runs jobs that execute to completion.
- Cloud Run functions, previously Cloud Functions, runs event-driven functions. Since 2024 it is part of Cloud Run, so functions and services share the same platform.
- Google Kubernetes Engine (GKE) is managed Kubernetes. Autopilot mode manages the nodes for you and bills for the resources your pods request; Standard mode gives you control over node pools.
- App Engine is Google’s original platform service. It still runs, but new web applications usually start on Cloud Run.
flowchart TD
Q["What are you running?"] --> E{"Short code reacting to events?"}
E -->|Yes| F["Cloud Run functions"]
E -->|No| C{"A container or web app?"}
C -->|"Yes, stateless HTTP or jobs"| CR["Cloud Run"]
C -->|"Yes, and you need Kubernetes APIs"| GKE["GKE Autopilot or Standard"]
C -->|"No, you need OS control"| GCE["Compute Engine"]
Moving down that chart toward Compute Engine gives you more control and more work: patching, scaling and monitoring become yours. Cloud Service Models: IaaS, PaaS, SaaS explains that spectrum in general terms.
Storage
Cloud Storage holds objects in buckets, with Standard, Nearline, Coldline and Archive storage classes for data accessed less often. A bucket’s location can be a region, a dual-region or a multi-region such as US or EU. VM disks are separate: Persistent Disk and the newer Hyperdisk provide block storage for Compute Engine, and Filestore provides managed NFS file shares.
Networking
Networking is where Google Cloud differs most from the other two providers. A VPC network is global: one network can have subnets in Iowa, London and Tokyo, and VMs in all three talk over internal IP addresses without peering or VPNs. Firewall rules and policies apply across the network. Cloud Load Balancing offers a global external Application Load Balancer that gives your service a single anycast IP address worldwide, and Cloud CDN and Cloud DNS sit alongside it.
Databases
Cloud SQL runs managed MySQL, PostgreSQL and SQL Server. AlloyDB is a PostgreSQL-compatible database built for higher performance and analytics on live data. Spanner is a relational database that scales horizontally and keeps strong consistency across regions. Firestore is a serverless document database popular with mobile and web apps, and Bigtable is a wide-column store for very high-throughput workloads such as time series.
You have VMs in us-central1 and europe-west2 in the same Google Cloud VPC network. What do you need for them to talk over internal IP addresses?
Correct! A Google Cloud VPC network is global, so subnets in different regions are already connected; firewall rules decide what traffic is allowed.
Data and AI
Data and AI are the areas most often named as reasons to choose Google Cloud.
BigQuery is a serverless data warehouse: you load or stream data into tables and run SQL without sizing or managing a cluster. You pay either for the data your queries scan or for reserved compute capacity, so learning to partition tables and select only the columns you need matters from day one. Around it sit Pub/Sub for messaging and event ingestion, Dataflow for batch and streaming pipelines based on Apache Beam, Dataproc for managed Spark and Hadoop, and Looker and Looker Studio for dashboards and reporting.
For AI, Google’s platform was called Vertex AI until April 2026, when Google renamed it Gemini Enterprise Agent Platform. It covers access to Gemini and partner models, tools for building and running agents, and the machine learning features that were in Vertex AI, such as training, pipelines and model deployment. You’ll still see the Vertex AI name in API names, client libraries and older documentation. Cloud Services for Generative AI: GCP Offerings covers this area in more depth.
Google also designs its own AI accelerators, Tensor Processing Units (TPUs), which you can rent alongside NVIDIA GPUs. TPUs work with JAX, PyTorch and TensorFlow, and they are the hardware Google trains its own Gemini models on.
Where Google Cloud is strong
The network. Google runs a private backbone between its regions and edge locations. With the default Premium network tier, traffic enters Google’s network close to the user and stays on it, which tends to give lower and more predictable latency than routing over the public internet. Global VPCs and global load balancing make multi-region designs simpler than on providers where networks are regional.
Data and analytics. BigQuery set the pattern for serverless warehouses, and Google’s data services connect to it closely. Teams with large analytics workloads often choose Google Cloud for this reason alone.
Kubernetes. Google created Kubernetes, and GKE is one of the most mature managed Kubernetes services. Fleet management features that used to need the separate Anthos product, later called GKE Enterprise, have been part of GKE itself since September 2025.
Pricing mechanics. Compute Engine bills per second after the first minute. Sustained use discounts apply automatically when certain older machine series, such as N1 and N2, run for a large part of the month; newer series such as E2 and C3 don’t receive them. Spot VMs offer deep discounts for workloads that tolerate being stopped at short notice; they replace the older preemptible VMs, which stopped after 24 hours. Committed use discounts reduce prices in exchange for a one-year or three-year commitment, either to specific resources or to a level of spend. Pricing Models compares these options across the three providers.
Where it is weaker
A smaller ecosystem. AWS has the largest catalog of services, the largest marketplace and the most third-party tutorials and integrations, and Azure has the strongest tie-in with Microsoft software. On Google Cloud you’ll sometimes find fewer partner products or community answers for niche problems.
Microsoft-centric organizations. If your identities live in Microsoft Entra ID and your workloads are Windows Server and SQL Server, Azure’s licensing and identity integration usually make it the easier fit. Google Cloud can federate with Entra ID and run Windows VMs, but it takes more setup.
Product retirements. Google has closed products that customers built on. Cloud IoT Core was retired in August 2023, and Cloud Source Repositories stopped accepting new customers in June 2024, with Secure Source Manager as the replacement. Google publishes deprecation notices and timelines, but check a product’s status before building something long-lived on it.
Hybrid hardware. Google’s answer to AWS Outposts and Azure Local is Google Distributed Cloud, which runs Google Cloud services on hardware in your own data center or at the edge, including air-gapped deployments. It is aimed at large or regulated organizations rather than small teams.
None of this makes Google Cloud the right or wrong answer by itself. AWS vs. GCP vs. Azure: Core Differences compares the three in more detail, and Choosing a Cloud Provider covers how to make the decision.
Which Compute Engine discount applies automatically, with no commitment, to eligible machine series that run for much of the month?
Correct! Sustained use discounts are applied automatically to eligible machine series; committed use discounts need a commitment, and Spot pricing needs you to choose Spot VMs.
Old names you will still see
Google renames and replaces products regularly, and older tutorials, including an earlier version of this article, use names that no longer appear in the console. If a guide mentions one of these, look for the current name.
| Older name | Current name or replacement |
|---|---|
| Cloud Functions | Cloud Run functions |
| AI Platform, Vertex AI | Gemini Enterprise Agent Platform |
| Anthos, GKE Enterprise | Features now part of GKE |
| Container Registry | Artifact Registry |
| Cloud Source Repositories | Secure Source Manager (closed to new customers) |
| Stackdriver | Google Cloud Observability (Cloud Monitoring, Cloud Logging) |
| Data Studio | Looker Studio |
| Deployment Manager | Infrastructure Manager |
| Duet AI for Google Cloud | Gemini for Google Cloud |
| Preemptible VMs | Spot VMs |
| Cloud IoT Core | Retired in 2023; use a partner IoT platform |
Getting started
New customers can start with the Free Trial: $300 in credits to use within 90 days. Separately, the Free Tier gives every billing account monthly allowances on selected products, such as one e2-micro VM in certain US regions and a quota of Cloud Run requests. GCP Free Tier and Billing Overview explains how the two interact, and Setting Up Free Accounts on AWS, GCP, and Azure walks through sign-up. Set a budget with an email alert before you build anything.
The Google Cloud console is the easiest place to look around. The gcloud CLI shows the structure more clearly, and Cloud Shell, which opens from the console toolbar, already has it installed and signed in. This sequence creates a project, links it to billing, enables an API and deletes the project again:
# Sign in and pick defaults (skip in Cloud Shell)
gcloud init
# Create a project; the ID must be globally unique
gcloud projects create intro-sandbox-4821 --name="Intro sandbox"
# Make it the active project for later commands
gcloud config set project intro-sandbox-4821
# Find your billing account ID and link the project to it
gcloud billing accounts list
gcloud billing projects link intro-sandbox-4821 \
--billing-account=XXXXXX-XXXXXX-XXXXXX
# Enable the APIs you need, then list what is enabled
gcloud services enable compute.googleapis.com run.googleapis.com
gcloud services list --enabled
# Delete the project and everything in it when you're done
gcloud projects delete intro-sandbox-4821
A deleted project is shut down right away and billing stops, but it stays recoverable for about 30 days before it is removed for good, and its ID is never available again. Running gcloud config list before creating resources shows which project is active, so they land where you expect.
Further reading
- Google Cloud documentation is the starting point for every product.
- Resource hierarchy explains organizations, folders and projects.
- Google Cloud products lists every service with a short description.
- Google Cloud Architecture Center has reference architectures and design guidance.
- gcloud CLI reference lists every command.
- Google Cloud pricing calculator estimates costs before you deploy.
- GCP Support and Documentation Resources covers support plans and where to get help.
Key takeaways
- Google Cloud organizes resources as organization, folders, projects and resources; projects are the boundary for APIs, quotas, billing and most permissions.
- Every project has a changeable name, a permanent globally unique ID and a project number; choose IDs with care.
- The core services map closely to AWS and Azure; choose compute by how much of the platform you want to manage, from Cloud Run functions to Compute Engine.
- Google Cloud’s strengths are its global network and VPCs, BigQuery and its data services, GKE, and pricing mechanics such as per-second billing and automatic sustained use discounts.
- Its weaker points are a smaller ecosystem, a less direct fit for Microsoft-centric shops, and a history of retiring products, so check a service’s status before building on it.
- Many tutorials use old names: Cloud Functions is Cloud Run functions, Vertex AI is Gemini Enterprise Agent Platform, and Anthos features are now part of GKE.
Part 1 of 7 · Exploring Google Cloud Platform (GCP)