Split View: AWS vs GCP vs Azure 완전 비교 2025: 서비스별 매핑, 비용, 자격증, 선택 기준 총정리

AWS vs GCP vs Azure Complete Comparison 2025: Service Mapping, Pricing, Certifications, and Decision Framework

• Introduction: Cloud Market 2025
• 1. Service Mapping: 30+ Services Compared 1:1
  • 1.1 Compute
  • 1.2 Storage
  • 1.3 Database
  • 1.4 AI/ML
  • 1.5 Networking
  • 1.6 Serverless and Events
  • 1.7 DevOps and Monitoring
• 2. Compute Deep Dive
  • 2.1 Virtual Machines: EC2 vs GCE vs Azure VM
  • 2.2 Serverless: Lambda vs Cloud Functions vs Azure Functions
• 3. Storage Deep Dive
  • 3.1 Object Storage: S3 vs GCS vs Blob Storage
  • 3.2 Block Storage
• 4. Database Deep Dive
  • 4.1 Relational DB: Aurora vs AlloyDB vs Azure SQL
  • 4.2 NoSQL: DynamoDB vs Firestore vs Cosmos DB
  • 4.3 Data Warehouse: Redshift vs BigQuery vs Synapse
• 5. AI/ML Services Comparison
  • 5.1 ML Platform: SageMaker vs Vertex AI vs Azure ML
  • 5.2 LLM/Generative AI Comparison
• 6. Kubernetes: EKS vs GKE vs AKS
  • 6.1 Managed Kubernetes Comparison
  • 6.2 Kubernetes Architecture Comparison
• 7. Pricing Comparison
  • 7.1 Pricing Model Comparison
  • 7.2 Free Tier Comparison
  • 7.3 Cost Optimization Strategies
• 8. Strengths and Weaknesses Analysis
  • 8.1 AWS
  • 8.2 GCP
  • 8.3 Azure
• 9. Certification Roadmap
  • 9.1 AWS Certifications
  • 9.2 GCP Certifications
  • 9.3 Azure Certifications
  • 9.4 Certification Comparison
• 10. Decision Framework
  • 10.1 Scenario-Based Recommendations
  • 10.2 Decision Flowchart
• 11. Multi-Cloud and Hybrid Strategy
  • 11.1 When to Consider Multi-Cloud
  • 11.2 Multi-Cloud Tech Stack
  • 11.3 Multi-Cloud Caveats
• 12. Interview Questions (15)
  • Basic
  • Intermediate
  • Advanced
• 13. Quiz
• References

Introduction: Cloud Market 2025

The global cloud market in 2025 has surpassed approximately $600 billion. The market share of the three major cloud vendors:

AWS still leads, but Azure is growing rapidly in the enterprise market, and GCP has established a strong position in data/AI/ML and Kubernetes.

Which cloud should you choose? This guide maps all three clouds service-by-service and provides a complete comparison of pricing, strengths/weaknesses, certifications, and decision frameworks.

1. Service Mapping: 30+ Services Compared 1:1

1.1 Compute

1.2 Storage

1.3 Database

1.4 AI/ML

1.5 Networking

1.6 Serverless and Events

1.7 DevOps and Monitoring

2. Compute Deep Dive

2.1 Virtual Machines: EC2 vs GCE vs Azure VM

Performance Comparison (4 vCPU, 16GB Memory):

AWS EC2 (m6i.xlarge):
- vCPU: 4, Memory: 16GB
- On-demand: ~$0.192/hr (~$140/mo)
- Network: up to 12.5 Gbps

GCP GCE (n2-standard-4):
- vCPU: 4, Memory: 16GB
- On-demand: ~$0.194/hr (~$142/mo)
- Custom Machine Types: independently adjust CPU/memory (GCP exclusive!)

Azure VM (D4s v5):
- vCPU: 4, Memory: 16GB
- On-demand: ~$0.192/hr (~$140/mo)
- Hybrid Benefit: up to 40% discount with existing Windows Server license

Key Differences:

• AWS: Most diverse instance types (500+). Graviton (ARM) processors offer best price-performance
• GCP: Custom machine types for flexible resource allocation. Sustained Use Discounts applied automatically
• Azure: Optimized for Windows workloads. Hybrid Benefit leverages existing licenses

2.2 Serverless: Lambda vs Cloud Functions vs Azure Functions

Serverless Function Comparison:

AWS Lambda:
- Max execution: 15 minutes
- Memory: 128MB - 10GB
- Concurrency: 1,000 (default, can be increased)
- Cold start: Medium (improved with SnapStart)
- Provisioned Concurrency supported

GCP Cloud Functions (2nd gen):
- Max execution: 60 minutes (2nd gen)
- Memory: 128MB - 32GB
- Concurrency: up to 1,000 requests per instance
- Cold start: Shortest
- Cloud Run-based scaling

Azure Functions:
- Max execution: Unlimited (Premium Plan)
- Memory: 1.5GB - 14GB
- Concurrency: 200 (default)
- Durable Functions: stateful workflows
- Excellent Visual Studio integration

3. Storage Deep Dive

3.1 Object Storage: S3 vs GCS vs Blob Storage

3.2 Block Storage

Block Storage Comparison:

AWS EBS (gp3):
- IOPS: 3,000 baseline (max 16,000)
- Throughput: 125 MB/s baseline (max 1,000 MB/s)
- Price: $0.08/GB/mo
- Snapshots: incremental backup

GCP Persistent Disk (pd-ssd):
- IOPS: proportional to disk size
- Throughput: proportional to disk size
- Price: $0.17/GB/mo (SSD)
- Feature: multi-instance attachment

Azure Managed Disks (Premium SSD v2):
- IOPS: up to 80,000
- Throughput: up to 1,200 MB/s
- Price: $0.132/GB/mo
- Feature: independent IOPS/throughput provisioning

4. Database Deep Dive

4.1 Relational DB: Aurora vs AlloyDB vs Azure SQL

4.2 NoSQL: DynamoDB vs Firestore vs Cosmos DB

NoSQL Comparison:

AWS DynamoDB:
- Model: Key-value + Document
- Latency: Single-digit ms
- Capacity: On-demand / Provisioned
- Global Tables: multi-region replication
- DAX: in-memory cache layer
- Strength: massive traffic handling, consistent performance

GCP Firestore:
- Model: Document (collections + documents)
- Latency: few ms
- Real-time sync: built-in (mobile/web)
- Offline support: native
- Strength: mobile/web app development, real-time features

Azure Cosmos DB:
- Model: Multi-model (document, key-value, graph, column)
- API: SQL, MongoDB, Cassandra, Gremlin, Table
- Global distribution: turnkey multi-region
- Consistency: 5 levels selectable
- Strength: multi-model flexibility, global distribution

4.3 Data Warehouse: Redshift vs BigQuery vs Synapse

BigQuery's overwhelming advantage: Serverless architecture requires no infrastructure management, petabyte-scale data queried in seconds, scan-based pricing makes costs predictable.

5. AI/ML Services Comparison

5.1 ML Platform: SageMaker vs Vertex AI vs Azure ML

5.2 LLM/Generative AI Comparison

LLM Service Comparison:

AWS Bedrock:
- Models: Claude (Anthropic), Llama, Titan, Mistral
- Strength: diverse model selection, model customization
- Fine-tuning: supported
- RAG: Knowledge Bases feature

GCP Vertex AI:
- Models: Gemini (Google), PaLM 2
- Strength: Google Search integration, Grounding
- Fine-tuning: supported
- Multimodal: Gemini image/video understanding

Azure OpenAI:
- Models: GPT-4o, GPT-4, DALL-E, Whisper
- Strength: exclusive OpenAI partnership
- Fine-tuning: supported
- Enterprise: strongest security/compliance

6. Kubernetes: EKS vs GKE vs AKS

6.1 Managed Kubernetes Comparison

Why GKE is the gold standard for Kubernetes: Google created Kubernetes, providing the fastest version support, Autopilot mode with full management, and superior networking performance.

6.2 Kubernetes Architecture Comparison

GKE Autopilot (Recommended):
+---------------------------------+
|  Google-managed Control Plane   |
|  + Automated node management    |
|  + Per-pod billing              |
|  + Auto security hardening      |
+----------+----------------------+
           |
     +-----v-----+
     |  Workloads |
     |  (Pods)    |
     +-----------+

EKS with Karpenter:
+---------------------------------+
|  AWS-managed Control Plane      |
|  ($73/mo)                       |
+----------+----------------------+
           |
  +--------v--------+
  |    Karpenter     |
  | (auto node       |
  |  provisioning)   |
  +--------+---------+
           |
     +-----v-----+
     |   Nodes    |
     |  + Pods    |
     +-----------+

AKS:
+---------------------------------+
|  Azure-managed Control Plane    |
|  (Free)                         |
+----------+----------------------+
           |
  +--------v--------+
  |   Node Pools     |
  | + Azure AD       |
  |   integration    |
  +------------------+

7. Pricing Comparison

7.1 Pricing Model Comparison

7.2 Free Tier Comparison

AWS Free Tier (12 months):
- EC2: t2.micro 750 hrs/mo
- S3: 5GB
- RDS: db.t2.micro 750 hrs/mo
- Lambda: 1M requests/mo (Always Free)
- DynamoDB: 25GB + 25 WCU/RCU (Always Free)

GCP Free Tier (12 months):
- GCE: e2-micro (Always Free, US regions)
- Cloud Storage: 5GB (Always Free)
- BigQuery: 1TB queries/mo (Always Free)
- Cloud Functions: 2M invocations/mo (Always Free)
- Firestore: 1GB storage (Always Free)

Azure Free Tier (12 months):
- VM: B1s 750 hrs/mo
- Blob: 5GB
- Azure SQL: 250GB S0 instance
- Functions: 1M requests/mo (Always Free)
- Cosmos DB: 1000 RU/s + 25GB (Always Free)

7.3 Cost Optimization Strategies

Common Cost Optimization Strategies:

1. Right-sizing
   - AWS: Compute Optimizer
   - GCP: Recommender
   - Azure: Advisor

2. Reserved Instances / Committed Use Discounts
   - Stable workloads: up to 72% savings with 1-3 year commitments

3. Spot/Preemptible Instances
   - Best for batch processing, CI/CD, non-critical workloads
   - Up to 90% cost savings

4. Auto-scaling Configuration
   - Automatic scale-down during off-hours

5. Storage Lifecycle Policies
   - Auto-transition old data to cheaper tiers

6. Cost Monitoring Tools
   - AWS: Cost Explorer, Budgets
   - GCP: Billing Reports, Budgets
   - Azure: Cost Management

8. Strengths and Weaknesses Analysis

8.1 AWS

Strengths:

• Service breadth: 200+ services, widest portfolio
• Market maturity: Oldest cloud (since 2006), largest community
• Global infrastructure: 33 regions, 100+ availability zones
• Partner ecosystem: Most ISV and SI partners
• Graviton: ARM-based processors with best price-performance ratio

Weaknesses:

• Complex pricing structure (especially inter-service data transfer costs)
• Console UX more complex than GCP
• EKS setup more complex than GKE
• Native data analytics weaker than BigQuery

8.2 GCP

Strengths:

• Data/AI: BigQuery, Vertex AI, TPU make it the strongest for analytics/ML
• Kubernetes: GKE Autopilot is the most mature managed K8s
• Network: Google's global network (owns submarine cables)
• Cost transparency: Automatic sustained use discounts, custom machine types
• Developer experience: Clean console UX, excellent documentation

Weaknesses:

• Narrower service range than AWS/Azure
• Enterprise support maturity lower than AWS/Azure
• Service deprecation history raises trust concerns
• Fewer availability zones in some regions

8.3 Azure

Strengths:

• Enterprise: Active Directory integration, Office 365/Teams integration
• Hybrid cloud: Azure Arc, Azure Stack for on-premises integration
• Compliance: Most diverse government/financial/healthcare certifications
• OpenAI partnership: Exclusive GPT-4o access
• Windows workloads: .NET, SQL Server optimization

Weaknesses:

• More frequent stability issues than AWS/GCP (especially 2024 major outages)
• Some services have weaker documentation than AWS/GCP
• Less cost-competitive for Linux workloads vs AWS/GCP
• Console is complex and slow

9. Certification Roadmap

9.1 AWS Certifications

AWS Certification Roadmap:

Entry: Cloud Practitioner (CLF-C02)
  |
Mid: Solutions Architect Associate (SAA-C03)  <- Most popular
  |
Mid: Developer Associate (DVA-C02)
  |
Mid: SysOps Administrator Associate
  |
Advanced: Solutions Architect Professional (SAP-C02)
  |
Advanced: DevOps Engineer Professional
  |
Specialty: Security / Database / ML Specialty

9.2 GCP Certifications

GCP Certification Roadmap:

Entry: Cloud Digital Leader
  |
Mid: Associate Cloud Engineer (ACE)  <- Most popular
  |
Advanced: Professional Cloud Architect (PCA)
  |
Advanced: Professional Data Engineer
  |
Advanced: Professional ML Engineer
  |
Specialty: Professional Cloud Security Engineer

9.3 Azure Certifications

Azure Certification Roadmap:

Entry: Azure Fundamentals (AZ-900)
  |
Mid: Azure Administrator (AZ-104)  <- Most popular
  |
Mid: Azure Developer (AZ-204)
  |
Advanced: Azure Solutions Architect Expert (AZ-305)
  |
Specialty: Azure Security Engineer (AZ-500)
  |
Specialty: Azure AI Engineer (AI-102)

9.4 Certification Comparison

10. Decision Framework

10.1 Scenario-Based Recommendations

Cloud Recommendations by Scenario:

Startup (Early Stage):
-> AWS: Widest free tier, credit programs, vast tutorials
-> GCP: $300 credits, Firebase free tier (great for mobile apps)

Data/AI-Focused Team:
-> GCP: BigQuery + Vertex AI + TPU combination is overwhelming
-> AWS: SageMaker + Bedrock is strong but more expensive

Enterprise (Existing MS Environment):
-> Azure: AD integration, Office 365, Hybrid Benefit are decisive
-> AWS: Suitable for enterprises without MS environment

Kubernetes-Centric Architecture:
-> GCP: GKE Autopilot provides the best K8s experience
-> AWS: EKS + Karpenter is good but more complex setup

Compliance-Heavy (Government/Finance/Healthcare):
-> Azure: Most compliance certifications
-> AWS: GovCloud for government workloads

Gaming/Media:
-> AWS: GameLift, IVS and specialized services
-> GCP: Google network performance + Game Servers

Cost-First:
-> GCP: Sustained use discounts + custom machine types
-> AWS: Graviton + Spot Instances combination

10.2 Decision Flowchart

Cloud Selection Decision:

Q1: Existing Microsoft environment (AD, Office 365)?
  - Yes -> Azure (Hybrid Benefit + AD integration)
  - No -> Q2

Q2: Primary workload is data analytics/AI?
  - Yes -> GCP (BigQuery + Vertex AI)
  - No -> Q3

Q3: Is Kubernetes your core infrastructure?
  - Yes -> GCP (GKE Autopilot)
  - No -> Q4

Q4: Is service diversity and ecosystem important?
  - Yes -> AWS (200+ services)
  - No -> Q5

Q5: Is budget the primary constraint?
  - Yes -> GCP (sustained use discounts) or AWS (Graviton)
  - No -> AWS (most stable, widest partner ecosystem)

11. Multi-Cloud and Hybrid Strategy

11.1 When to Consider Multi-Cloud

Scenarios Requiring Multi-Cloud:

1. Vendor Lock-in Prevention
   - Avoid dependency on a single cloud for core services
   - Achieve portability with Kubernetes + Terraform

2. Best-of-Breed Approach
   - AI/ML: GCP (BigQuery + Vertex AI)
   - Backend infrastructure: AWS (EC2 + Lambda)
   - Identity management: Azure (Active Directory)

3. Compliance
   - Data sovereignty: store data in specific regions only
   - Multiple clouds to meet regulatory requirements

4. Disaster Recovery (DR)
   - Failover to another cloud during outages
   - Active-Active or Active-Passive DR

11.2 Multi-Cloud Tech Stack

Multi-Cloud Technology Stack:

IaC (Infrastructure as Code):
- Terraform: all-cloud support, most popular
- Pulumi: write IaC in programming languages

Container Orchestration:
- Kubernetes: runs on all clouds
- Istio/Linkerd: service mesh for cross-cloud communication

CI/CD:
- GitHub Actions: cloud-independent
- GitLab CI: self-hostable

Monitoring:
- Datadog: multi-cloud integrated monitoring
- Grafana + Prometheus: open-source alternative

Service Mesh:
- Istio: cross-cloud traffic management
- Consul: service discovery + multi-cloud

11.3 Multi-Cloud Caveats

Multi-cloud significantly increases complexity. Consider:

• Team capability: need experts in 2-3 clouds
• Network costs: high inter-cloud data transfer costs
• Consistency: must unify security policies, IAM, and logging
• Do not adopt multi-cloud without at least 2 team agreements

Practical advice: For most organizations, one primary cloud + 1-2 services for special purposes is optimal. True multi-cloud is only necessary for large enterprises.

12. Interview Questions (15)

Basic

Q1: Briefly explain the key differences between AWS, GCP, and Azure.

AWS has the widest service portfolio (200+ services) and largest market share. GCP excels in data analytics (BigQuery), AI/ML (Vertex AI, TPU), and Kubernetes (GKE). Azure excels in Microsoft ecosystem integration (AD, Office 365) and enterprise/hybrid cloud.

Q2: What is serverless computing? Compare each cloud's offering.

Serverless lets you run code without managing servers. AWS Lambda (max 15 min, widest event sources), GCP Cloud Functions (max 60 min 2nd gen, Cloud Run-based), Azure Functions (Durable Functions for stateful workflows). Lambda is most mature; Cloud Functions has shortest cold starts.

Q3: What are the free tier differences across clouds?

AWS: EC2 t2.micro 750 hrs, S3 5GB, Lambda 1M requests Always Free. GCP: e2-micro Always Free (US), BigQuery 1TB/mo Always Free, Cloud Functions 2M invocations. Azure: B1s VM 750 hrs, 250GB SQL, Functions 1M requests. GCP has the widest Always Free scope.

Q4: What is vendor lock-in and how do you mitigate it?

Vendor lock-in means depending on a cloud's proprietary services, making migration difficult. Mitigation: 1) Use Kubernetes + Terraform, 2) Choose open standard technologies (PostgreSQL vs Aurora), 3) Introduce abstraction layers, 4) Container-based deployments.

Q5: Explain the difference between Reserved Instances and Spot Instances.

Reserved Instances offer up to 72% discount with 1-3 year commitments, suitable for stable workloads. Spot Instances use idle capacity at up to 90% discount but can be reclaimed anytime, suitable for batch processing, CI/CD, and interruptible workloads.

Intermediate

Q6: Why is BigQuery better than Redshift for data analytics?

BigQuery is fully serverless requiring no infrastructure management, with scan-based pricing for predictable costs. Storage and compute are separated for independent scaling, and ML models can be trained directly with SQL (BigQuery ML). Redshift requires pre-provisioned cluster sizing.

Q7: Why does GKE offer a superior Kubernetes experience over EKS/AKS?

Google created Kubernetes, providing fastest version support. GKE Autopilot fully automates node management with per-pod billing. Superior networking performance and managed Anthos Service Mesh are included. EKS has control plane costs and more complex setup.

Q8: Why is Azure strong in the enterprise market?

Active Directory integration provides seamless connectivity with existing identity management. Natural integration with Office 365, Teams, and Dynamics 365. Azure Arc and Azure Stack enable on-premises/cloud hybrid operations. Windows Server Hybrid Benefit reduces existing license costs.

Q9: Explain the pros and cons of multi-cloud strategy.

Pros: vendor lock-in prevention, best-of-breed services, enhanced DR, easier compliance. Cons: increased operational complexity, specialized staff needed, higher network costs, difficulty maintaining consistent security policies. One primary cloud is recommended for most organizations.

Q10: Describe five cloud cost optimization strategies.

1. Right-sizing: use Compute Optimizer/Recommender for proper instance sizing. 2) Reserved Instances/CUDs for stable workload discounts. 3) Spot/Preemptible for non-critical workload savings. 4) Auto-scaling for off-hours reduction. 5) Storage lifecycle policies for automatic tier transitions.

Advanced

Q11: Explain the architectural differences between Cosmos DB and DynamoDB.

DynamoDB is key-value + document model only, guaranteeing consistent single-digit ms latency. Global Tables enable multi-region replication. Cosmos DB supports multi-model (document, key-value, graph, column) with 5 consistency levels (Strong to Eventual). Turnkey global distribution is its strength.

Q12: Explain the cloud migration strategy (6Rs).

1. Rehost (Lift and Shift): migrate without changes. 2) Replatform: minimal optimization before migration. 3) Refactor: redesign as cloud native. 4) Repurchase: switch to SaaS. 5) Retain: keep on-premises. 6) Retire: decommission. Most start with Rehost and progressively Refactor.

Q13: How do you manage multi-cloud with Terraform?

Terraform defines infrastructure in HCL and supports AWS/GCP/Azure through Providers. State management uses remote backends (S3, GCS, etc.), Modules define reusable infrastructure components, and Workspaces separate environments (dev/staging/prod).

Q14: Compare the AI/ML ecosystem of each cloud.

AWS: SageMaker (ML platform) + Bedrock (LLM: Claude/Llama) -- widest framework support. GCP: Vertex AI + BigQuery ML + TPU + Gemini -- strongest data analytics and ML integration. Azure: Azure ML + Azure OpenAI (exclusive GPT-4o) -- strongest for enterprise generative AI.

Q15: Explain cloud-native application design principles.

Based on 12-Factor App principles: 1) Microservices architecture. 2) Container-based deployment. 3) CI/CD automation. 4) Infrastructure as Code. 5) Stateless design. 6) Built-in observability. 7) Auto-scaling. 8) Fault isolation (Circuit Breaker, etc.). 9) Declarative APIs. 10) Event-driven async processing.

13. Quiz

References

• AWS Official Documentation
• Google Cloud Official Documentation
• Azure Official Documentation
• Gartner Magic Quadrant for Cloud Infrastructure
• Flexera State of the Cloud Report
• CNCF Cloud Native Survey
• Google Cloud Architecture Framework
• AWS Well-Architected Framework
• Azure Architecture Center
• Terraform Documentation
• Kubernetes Documentation
• InfoQ Cloud Trends Report
• The New Stack - Cloud Native News
• Cloud Native Computing Foundation (CNCF)
• CloudHealth by VMware