GCP vs AWS — Service Comparison

This is the service-mapping lookup table. For the conceptual mental-model shifts an AWS engineer needs (resource hierarchy, additive IAM, global VPC, pricing model differences), see the companion doc from-aws.md.

Core Philosophy Difference

graph LR
    subgraph AWS["AWS — breadth first"]
        A1["150+ services<br/>every use case covered<br/>built over 20 years<br/>most mature ecosystem<br/>most enterprise adoption"]
    end
    subgraph GCP["GCP — engineering first"]
        G1["70+ services<br/>built on Google's own infra<br/>BigQuery, Bigtable, Spanner<br/>best Kubernetes (invented it)<br/>best data/ML platform"]
    end

Service-by-Service Mapping

Compute

Storage and Databases

Networking

Key networking difference:

graph LR
    subgraph AWS["AWS VPC — regional"]
        AVPC1["VPC us-east-1<br/>10.0.0.0/16"] 
        AVPC2["VPC eu-west-1<br/>10.1.0.0/16"]
        AVPC1 -->|"VPC Peering or<br/>Transit Gateway"| AVPC2
        NOTE_A["Separate VPCs per region<br/>explicit peering required"]
    end
    subgraph GCP["GCP VPC — global"]
        GVPC["Single VPC<br/>spans ALL regions"]
        US["Subnet: us-central1<br/>10.0.1.0/24"]
        EU["Subnet: europe-west1<br/>10.0.2.0/24"]
        GVPC --> US & EU
        NOTE_G["One VPC, subnets in each region<br/>automatic routing between regions"]
    end

Observability

AI/ML

GCP TPUs: Google's Tensor Processing Units give 10-100× better price/performance for LLM training vs GPUs on any cloud. If you're training large models, GCP is often cheaper.

GCP Unique Advantages

1. Global VPC

Single VPC spanning all regions — no peering needed. Add a subnet in Tokyo, your London VM can reach it without configuration.

2. BigQuery Pricing

BigQuery: $5/TB scanned, $0.02/GB storage. AWS Redshift: $0.25/hour for smallest cluster (even when idle). For sporadic analytics, BigQuery is often 10× cheaper.

3. Sustained Use Discounts

GCP automatically gives discounts for VMs running >25% of the month — no upfront commitment required. AWS requires Reserved Instances (1-3 year commitment) for equivalent discounts.

4. GKE Quality

GKE gets Kubernetes features first (it's the reference implementation). Autopilot mode — true serverless Kubernetes — doesn't exist on AWS/Azure.

5. Spanner — True Global ACID

AWS Aurora Global has replication lag (seconds). Spanner provides globally consistent transactions with ~10ms latency using TrueTime (atomic clocks).

AWS Unique Advantages

1. Service breadth

AWS has services GCP doesn't: AppSync (GraphQL), Kinesis Data Firehose simplicity, CodePipeline, many enterprise integrations.

2. Enterprise ecosystem

AWS has the most ISV integrations, compliance certifications, and enterprise support.

3. Mature marketplace

AWS Marketplace has thousands of third-party products pre-configured.

4. On-premises hybrid

AWS Outposts, AWS Local Zones — better on-prem story than GCP Distributed Cloud.

When to Choose Which

Choose AWS when:
  - Your team already knows AWS
  - You need maximum service breadth
  - Enterprise compliance requirements (AWS has most certifications)
  - Strong on-prem hybrid requirements

Choose GCP when:
  - Analytics-heavy workload (BigQuery is hard to beat)
  - K8s-native platform (GKE Autopilot, Workload Identity)
  - Training large ML models (TPUs)
  - Multi-region global application (global VPC simplicity)
  - Cost optimization for data warehousing