Google Cloud Platform (GCP) is the third-largest cloud provider. It has genuine strengths in data analytics (BigQuery is widely considered best-in-class), Kubernetes (Google invented it), AI and machine learning (Vertex AI, Gemini), and global networking via Google's private fiber backbone. If Cartara flagged something referencing GCP service names, Firebase SDKs, or BigQuery queries, this guide explains what those services do and when GCP makes sense.

Why Teams Choose GCP

GCP is often the right choice when:

• Data and ML are central to your product — BigQuery and Vertex AI are genuinely best-in-class.
• You want the most mature managed Kubernetes offering (GKE).
• You're building on Google Workspace, YouTube APIs, or Google Maps.
• You want Gemini models natively integrated with your cloud infrastructure.
• Low latency globally matters — Google's premium network tier routes traffic over its private backbone rather than the public internet.

Compute

Compute Engine is IaaS — virtual machines you configure and manage. GCP offers sustained use discounts automatically (no commitment needed), which makes it more cost-effective than AWS EC2 for workloads that run most of the month.

Google Kubernetes Engine (GKE) is the most mature managed Kubernetes offering — Google invented Kubernetes and open-sourced it in 2014. GKE Autopilot mode is fully serverless: you pay per Pod, not per node, and GCP handles all node management.

Cloud Run is serverless containers. Deploy any container image and it scales from zero to thousands automatically, billed only when it's handling requests. No Kubernetes knowledge required — just a container and Cloud Run handles the rest. This is one of the cleanest serverless container experiences across all cloud providers.

Cloud Functions is event-driven serverless compute. Write a function; trigger it from HTTP, Pub/Sub, Firestore, Cloud Storage, or other events. The second generation is built on Cloud Run and supports higher limits.

App Engine is Google's original PaaS, now largely superseded by Cloud Run for new workloads. Still used for apps that need to scale to zero on the Standard environment.

Storage

Cloud Storage (GCS) is GCP's object storage — equivalent to AWS S3. Storage classes include Standard (hot data), Nearline (~once/month access), Coldline (~once/quarter), and Archive (long-term, rarely accessed). A FUSE adapter lets you mount GCS buckets as a filesystem.

Persistent Disk is block storage for Compute Engine VMs — SSDs and HDDs at various performance tiers.

Filestore is managed NFS file shares at different performance tiers, integrating with GKE for persistent volumes.

Networking

Virtual Private Cloud (VPC) in GCP is global by default — one VPC spans all regions, with subnets being regional. This simplifies multi-region architectures compared to AWS where VPCs are regional.

Cloud Load Balancing uses a single global anycast IP for all load balancers. The Global External Application Load Balancer routes traffic across Google's network with no warm-up time needed — it handles traffic spikes instantly, which is a meaningful advantage over AWS and Azure.

Cloud CDN is integrated directly with the global load balancer, caching content at Google's edge locations worldwide.

Cloud Armor is DDoS protection and WAF — uses ML to detect and mitigate attacks. Policies attach to load balancers.

Apigee is Google's enterprise API management platform. Full API lifecycle management — design, security, analytics, developer portal, monetization. More powerful than AWS API Gateway, used for large-scale API programs.

Databases

Cloud SQL is managed relational databases — PostgreSQL, MySQL, and SQL Server — with automatic backups, point-in-time recovery, and read replicas.

AlloyDB for PostgreSQL is Google's proprietary PostgreSQL-compatible database. Claims significantly faster performance than standard PostgreSQL for both transactional and analytical workloads. High availability with under 60-second failover.

Cloud Spanner is a globally distributed relational database — horizontal scaling with strong consistency, something most distributed databases sacrifice. It combines SQL and ACID transactions with the scale of NoSQL. Expensive, but genuinely unique. Used by Google internally for YouTube and Google Ads scale.

Firestore is a fully managed serverless NoSQL document database. Two modes: Native mode has real-time listeners, offline sync, and mobile/web SDKs (ideal for apps). Datastore mode is for high-volume server-side workloads.

Bigtable is a managed wide-column NoSQL database — the same technology that inspired Apache HBase and Cassandra. Petabyte-scale, millisecond latency. Used internally by Google for Search, Maps, and Gmail. Best for time-series, IoT telemetry, and financial data at extreme scale.

Memorystore is managed Redis and Memcached for in-memory caching.

AI and Machine Learning

Vertex AI is Google's unified ML platform. Covers the full lifecycle: data labeling, training, hyperparameter tuning, MLflow-compatible experiment tracking, model registry, and inference endpoints. It hosts Google's foundation models:

• Gemini — Google's multimodal model family (handles text, image, audio, video, and code)
• Imagen — image generation
• Partner models including Claude (Anthropic) and Llama (Meta)

Gemini API / AI Studio provides direct API access to Gemini models. AI Studio is the playground for development and testing.

Cloud Vision AI analyzes images: object detection, OCR, face detection, safe search, and landmark identification.

Cloud Speech-to-Text / Text-to-Speech provides speech APIs supporting 125+ languages with real-time streaming.

Document AI extracts structured data from documents — invoices, forms, contracts — going beyond basic OCR to understand document structure.

Security and Identity

Cloud IAM is the core permission system. Service accounts are the primary mechanism for service-to-service authentication — each service runs as a service account with specific permissions rather than sharing credentials.

Secret Manager stores and manages secrets with versioning and automatic rotation.

Cloud KMS manages encryption keys. Customer-managed encryption keys (CMEK) let you encrypt data with keys you control.

Binary Authorization enforces deploy-time policy for containers: only signed, verified container images can run in GKE or Cloud Run. Useful for supply chain security.

Security Command Center is the centralized security and risk dashboard — asset inventory, vulnerability detection, threat detection, and compliance monitoring.

Developer and DevOps Tools

Cloud Build is managed CI/CD. Trigger builds on Git push, run Docker builds, and execute custom build steps. Fast parallel execution.

Artifact Registry stores container images (OCI), npm packages, Python packages, Maven artifacts, and more. The successor to Container Registry.

Cloud Deploy is managed continuous delivery for GKE and Cloud Run, with approval gates and rollback support.

Cloud Monitoring collects metrics, uptime checks, and alerts from GKE, Cloud Run, and Compute Engine. Custom metrics via OpenTelemetry.

Cloud Logging is managed log aggregation with Log Explorer for querying. Log Analytics uses BigQuery under the hood for advanced analysis.

Cloud Trace provides distributed tracing — latency analysis across microservices, integrating with OpenTelemetry.

Data and Analytics

BigQuery is GCP's flagship managed data warehouse and the one product that genuinely differentiates GCP for data-heavy teams. Serverless — no cluster to manage, you pay per query (per TB scanned) or on a flat rate. Key capabilities:

• BigQuery ML — run ML models directly in SQL without exporting data
• BigQuery Omni — query data in AWS S3 or Azure Blob without moving it
• BigQuery Vector Search — semantic similarity search for AI applications
• Considered best-in-class for ad-hoc analytics at scale

Dataflow is managed Apache Beam for stream and batch data processing. GCP's ETL engine — unified programming model for both streaming and batch workloads.

Dataproc is managed Spark and Hadoop clusters with faster startup than AWS EMR. Dataproc Serverless runs Spark jobs without managing clusters.

Pub/Sub is GCP's managed messaging service — publishers push to topics, subscribers pull or receive push delivery. The equivalent of Kinesis Data Streams or Kafka.

Looker is Google's BI and data modeling platform (acquired 2019). LookML defines a semantic layer over your warehouse. Looker Studio (formerly Data Studio) is the free dashboarding tool.

Firebase

Firebase is Google's mobile and web development platform, built on top of GCP. It abstracts the cloud for app developers — authentication, database, functions, and hosting in one SDK.

If you're building a mobile app and want to avoid managing a backend, Firebase gives you auth, database, storage, and functions without touching a VM.

What You'll See in Your Code

A Cloud Run deployment using the gcloud CLI:

```bash

Build and push a container

gcloud builds submit --tag gcr.io/my-project/my-api

Deploy to Cloud Run

gcloud run deploy my-api
--image gcr.io/my-project/my-api
--platform managed
--region us-central1
--allow-unauthenticated
--min-instances 0
--max-instances 10
```

A BigQuery query that runs ML directly on warehouse data:

```sql
-- BigQuery ML: train a model without leaving SQL
CREATE OR REPLACE MODEL my_dataset.churn_model
OPTIONS(model_type='logistic_reg', input_label_cols=['churned']) AS
SELECT
days_since_last_login,
total_purchases,
support_tickets,
churned
FROM my_dataset.user_features;
```

Related Concepts

• AWS Services
• Azure Services
• Managed Platforms
• Data and Analytics Platforms
• Containerisation Basics
• Database Fundamentals