Cloud Data Platform Architecture: AWS vs GCP vs Azure

TL;DR

Choosing cloud for Data Platform = One of most critical decisions. Each cloud has strengths:

Quick Recommendation:

• Startup/Scale-up with analytics focus → GCP (BigQuery best-in-class, simpler pricing)
• Enterprise với AWS ecosystem → AWS (most mature, broadest services)
• Microsoft shop (Office 365, Power BI) → Azure (tight integration)

Component Comparison:

Cost Comparison (same 10TB workload):

• AWS: ~$5,500/month (complex pricing)
• GCP: ~$4,200/month (simpler, transparent)
• Azure: ~$5,000/month (good for existing licenses)

Vietnamese Market:

• Startups: 60% choose GCP (BigQuery, ease of use)
• Enterprises: 30% AWS, 10% Azure

Case Studies:

• Vietnamese fintech: GCP BigQuery → 5x faster queries, 40% cost reduction vs Redshift
• E-commerce: AWS (existing infra) → comprehensive ecosystem
• Enterprise: Azure (Microsoft stack) → seamless Power BI integration

Bài này sẽ deep-dive từng cloud với architecture diagrams, pricing details, và decision framework.

1. Why Cloud-Native Data Platform?

1.1. On-Premise Pain Points

Traditional setup (Oracle, SQL Server on-premise):

Problems:
❌ Upfront cost: $500K+ hardware investment
❌ Provisioning time: 3-6 months to procure servers
❌ Over-provisioning: Buy for peak capacity (90% idle most of time)
❌ Maintenance: DBAs spend 40% time on patching, backups
❌ Scaling: Hard limits, can't handle traffic spikes
❌ Disaster recovery: Complex, expensive redundancy

Example: Vietnamese bank

• Bought $2M Oracle Exadata in 2015
• Used 30% capacity on average
• Takes 6 months to add capacity
• DR site costs another $1M

1.2. Cloud Advantages

✅ Scalability: Scale from 1GB to 1PB in minutes ✅ Pay-as-you-go: No upfront costs, pay for what you use ✅ Managed services: No server management, auto-patching ✅ Global: Deploy across multiple regions easily ✅ Innovation: New features released continuously ✅ Disaster recovery: Built-in redundancy, multi-region replication

Cost comparison:

On-Premise (3 years TCO):
  Hardware: $500K
  Software licenses: $300K
  Data center: $150K
  Staff (3 DBAs): $540K
  Total: $1.49M

Cloud (3 years, average $5K/month):
  Monthly cost: $5K × 36 = $180K
  Staff (1 engineer, less maintenance): $180K
  Total: $360K

Savings: $1.13M (76% reduction)

Note: Cloud cheaper for most workloads, BUT can get expensive if not optimized (see Bài 32: Cost Optimization upcoming).

2. AWS: The Mature Giant

2.1. Overview

Launched: 2006 (first cloud provider) Market share: 32% globally (Gartner 2024) Strengths: Broadest services (200+), most mature, largest community Weaknesses: Complex, steeper learning curve, pricing complicated

2.2. Data Platform Components

Architecture:

┌─────────────────────────────────────────────────────────┐
│              AWS Data Platform Architecture             │
├─────────────────────────────────────────────────────────┤
│                                                         │
│  Data Sources                                           │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐            │
│  │ RDS      │  │ APIs     │  │ SaaS     │            │
│  │(Postgres)│  │          │  │ (Shopify)│            │
│  └────┬─────┘  └────┬─────┘  └────┬─────┘            │
│       │             │              │                   │
│       └─────────────┴──────────────┘                   │
│                     │                                   │
│                     ▼                                   │
│  ┌─────────────────────────────────────────┐           │
│  │ Ingestion Layer                         │           │
│  │  - DMS (Database Migration Service)     │           │
│  │  - Kinesis (Streaming)                  │           │
│  │  - Lambda (Event-driven)                │           │
│  └─────────────────┬───────────────────────┘           │
│                    │                                    │
│                    ▼                                    │
│  ┌─────────────────────────────────────────┐           │
│  │ Data Lake (S3)                          │           │
│  │  - Raw zone: s3://raw/                  │           │
│  │  - Processed: s3://processed/           │           │
│  │  - Analytics: s3://analytics/           │           │
│  │  - Format: Parquet, Delta Lake          │           │
│  └─────────────────┬───────────────────────┘           │
│                    │                                    │
│        ┌───────────┼───────────┐                       │
│        │           │           │                       │
│        ▼           ▼           ▼                       │
│  ┌─────────┐ ┌──────────┐ ┌──────────┐               │
│  │ Glue    │ │ EMR      │ │ Lambda   │               │
│  │ (ETL)   │ │ (Spark)  │ │(Serverless)              │
│  └────┬────┘ └────┬─────┘ └────┬─────┘               │
│       │           │            │                       │
│       └───────────┴────────────┘                       │
│                   │                                     │
│                   ▼                                     │
│  ┌─────────────────────────────────────────┐           │
│  │ Data Warehouse (Redshift)               │           │
│  │  - Clusters: ra3.4xlarge nodes          │           │
│  │  - Auto-scaling                         │           │
│  │  - Redshift Spectrum (query S3)         │           │
│  └─────────────────┬───────────────────────┘           │
│                    │                                    │
│        ┌───────────┴───────────┐                       │
│        │                       │                       │
│        ▼                       ▼                       │
│  ┌─────────┐            ┌──────────┐                  │
│  │QuickSight│            │ Tableau  │                  │
│  │  (BI)   │            │(3rd party)│                  │
│  └─────────┘            └──────────┘                  │
│                                                         │
│  Orchestration: MWAA (Managed Airflow)                 │
│  Governance: Lake Formation, Glue Data Catalog         │
│  Security: IAM, KMS encryption                         │
└─────────────────────────────────────────────────────────┘

2.3. Key Services Deep-Dive

Amazon Redshift (Data Warehouse)

Pros:

• ✅ Columnar storage (fast analytics)
• ✅ Massively parallel processing (MPP)
• ✅ Redshift Spectrum: Query S3 directly (no loading)
• ✅ Auto-scaling: Add nodes on demand
• ✅ Mature ecosystem

Cons:

• ❌ Requires cluster management (not fully serverless)
• ❌ Pauses/resumes manually (or via scheduler)
• ❌ Complex pricing (compute + storage separate)
• ❌ Slower than BigQuery for ad-hoc queries

Pricing:

ra3.4xlarge node:
  - $3.26/hour = $2,347/month
  - 128GB RAM, 32 vCPUs
  - Managed storage: $0.024/GB/month

Example cluster (2 nodes):
  Compute: $2,347 × 2 = $4,694/month
  Storage (10TB): $10,000 × $0.024 = $240/month
  Total: ~$4,934/month

AWS Glue (ETL)

Pros:

• ✅ Serverless (no infrastructure)
• ✅ Auto-scaling
• ✅ Python/Spark based
• ✅ Data Catalog (metadata repository)

Cons:

• ❌ Learning curve (PySpark knowledge needed)
• ❌ Debugging difficult (limited logs)
• ❌ Cold starts (5-10 min for large jobs)

Pricing: $0.44 per DPU-hour (Data Processing Unit)

Amazon S3 (Data Lake)

Pros:

• ✅ 99.999999999% durability (11 nines)
• ✅ Unlimited scalability
• ✅ Lifecycle policies (auto-archive to Glacier)
• ✅ Versioning, encryption

Pricing:

Storage tiers:
  - S3 Standard: $0.023/GB/month (frequent access)
  - S3 Infrequent Access: $0.0125/GB/month
  - S3 Glacier: $0.004/GB/month (archive)

10TB example:
  Standard: 10,000 × $0.023 = $230/month
  With lifecycle (50% to IA after 30 days):
    5,000 × $0.023 + 5,000 × $0.0125 = $178/month

2.4. When to Choose AWS

✅ Already on AWS: Existing EC2, RDS → easier to stay ✅ Need broad services: AWS has most services (IoT, ML, etc.) ✅ Enterprise contracts: Discounts via EDP (Enterprise Discount Program) ✅ Hybrid cloud: AWS Outposts for on-premise integration ✅ Regulatory: Some Vietnamese banks require AWS (proven compliance)

3. GCP: The Analytics Powerhouse

3.1. Overview

Launched: 2008 (Google App Engine) Market share: 11% globally Strengths: Best analytics (BigQuery), simpler pricing, innovation Weaknesses: Smaller ecosystem than AWS, less enterprise features

3.2. Data Platform Components

Architecture:

┌─────────────────────────────────────────────────────────┐
│              GCP Data Platform Architecture             │
├─────────────────────────────────────────────────────────┤
│                                                         │
│  Data Sources                                           │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐            │
│  │Cloud SQL │  │ APIs     │  │ SaaS     │            │
│  │(Postgres)│  │          │  │          │            │
│  └────┬─────┘  └────┬─────┘  └────┬─────┘            │
│       │             │              │                   │
│       └─────────────┴──────────────┘                   │
│                     │                                   │
│                     ▼                                   │
│  ┌─────────────────────────────────────────┐           │
│  │ Ingestion Layer                         │           │
│  │  - Pub/Sub (Streaming)                  │           │
│  │  - Cloud Functions (Event-driven)       │           │
│  │  - Dataflow (Batch + Streaming)         │           │
│  └─────────────────┬───────────────────────┘           │
│                    │                                    │
│                    ▼                                    │
│  ┌─────────────────────────────────────────┐           │
│  │ Data Lake (GCS - Cloud Storage)         │           │
│  │  - Buckets: gs://raw/, gs://processed/  │           │
│  │  - Format: Parquet, Avro                │           │
│  │  - Lifecycle: Auto-archive after 30d    │           │
│  └─────────────────┬───────────────────────┘           │
│                    │                                    │
│                    ▼                                    │
│  ┌─────────────────────────────────────────┐           │
│  │ Transformation (Dataflow)               │           │
│  │  - Apache Beam (Python/Java)            │           │
│  │  - Auto-scaling workers                 │           │
│  │  - Batch + Streaming unified            │           │
│  └─────────────────┬───────────────────────┘           │
│                    │                                    │
│                    ▼                                    │
│  ┌─────────────────────────────────────────┐           │
│  │ Data Warehouse (BigQuery) ⭐            │           │
│  │  - Serverless (no clusters!)            │           │
│  │  - Slots auto-allocated                 │           │
│  │  - Petabyte-scale                       │           │
│  │  - ML built-in (BigQuery ML)            │           │
│  └─────────────────┬───────────────────────┘           │
│                    │                                    │
│        ┌───────────┴───────────┐                       │
│        │                       │                       │
│        ▼                       ▼                       │
│  ┌─────────┐            ┌──────────┐                  │
│  │ Looker  │            │ Metabase │                  │
│  │  (BI)   │            │(3rd party)│                  │
│  └─────────┘            └──────────┘                  │
│                                                         │
│  Orchestration: Cloud Composer (Managed Airflow)       │
│  Governance: Data Catalog, Dataplex                    │
│  Security: IAM, CMEK encryption                        │
└─────────────────────────────────────────────────────────┘

3.3. Key Services Deep-Dive

BigQuery (Data Warehouse) ⭐ Best-in-Class

Pros:

• ✅ Fully serverless: No clusters, no nodes, no management
• ✅ Blazing fast: Dremel architecture, 1 PB scan in seconds
• ✅ Transparent pricing: $5/TB scanned (on-demand)
• ✅ SQL standard: Easy to learn
• ✅ Built-in ML: BigQuery ML (train models in SQL)
• ✅ Auto-scaling: Infinitely scalable

Cons:

• ❌ Cost spiral if not careful (scan-based pricing)
• ❌ Limited procedural SQL (no loops, complex logic)
• ❌ Partitioning required for large tables

Pricing:

On-Demand:
  $5 per TB scanned
  Example: 100GB query = $0.50

  Monthly estimate:
    1TB scanned/day × 30 days = 30TB
    30TB × $5 = $150/month

Flat-Rate (Reserved slots):
  $2,000/month = 100 slots (fixed cost, unlimited queries)

  Break-even: 400TB scanned/month
  Good for: Heavy query workloads

Storage:
  Active: $0.02/GB/month
  Long-term (90+ days): $0.01/GB/month

  10TB storage: $200/month

Google Cloud Storage (Data Lake)

Pros:

• ✅ Simple pricing (clearer than S3)
• ✅ Global by default (no regions to choose for standard)
• ✅ Strong integration với BigQuery

Pricing:

Storage classes:
  - Standard: $0.020/GB/month
  - Nearline (30+ days): $0.010/GB/month
  - Coldline (90+ days): $0.004/GB/month
  - Archive (365+ days): $0.0012/GB/month

10TB example (Standard): $200/month

Dataflow (ETL)

Pros:

• ✅ Apache Beam (portable, run on Spark/Flink too)
• ✅ Unified batch + streaming
• ✅ Auto-scaling

Cons:

• ❌ Complex for simple ETL (overkill)
• ❌ Beam learning curve

Alternative: Many use dbt on BigQuery instead (simpler for SQL transformations)

3.4. When to Choose GCP

✅ Analytics-first: BigQuery là best data warehouse ✅ Startups: Simpler, faster to get started ✅ Data science heavy: Vertex AI excellent for ML ✅ Want simplicity: Less moving parts than AWS ✅ Budget transparency: Clearer pricing

Vietnamese market: 60% startups choose GCP (ease of use + BigQuery)

4. Azure: The Microsoft Ecosystem

4.1. Overview

Launched: 2010 (Windows Azure) Market share: 23% globally Strengths: Microsoft integration (Office 365, AD), Power BI, hybrid cloud Weaknesses: Catch-up mode (後起), less data-specific innovation

4.2. Data Platform Components

Architecture:

┌─────────────────────────────────────────────────────────┐
│             Azure Data Platform Architecture            │
├─────────────────────────────────────────────────────────┤
│                                                         │
│  Data Sources                                           │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐            │
│  │Azure SQL │  │ APIs     │  │ Dynamics │            │
│  │ Database │  │          │  │   365    │            │
│  └────┬─────┘  └────┬─────┘  └────┬─────┘            │
│       │             │              │                   │
│       └─────────────┴──────────────┘                   │
│                     │                                   │
│                     ▼                                   │
│  ┌─────────────────────────────────────────┐           │
│  │ Ingestion Layer                         │           │
│  │  - Event Hubs (Streaming)               │           │
│  │  - Azure Functions (Event-driven)       │           │
│  │  - Data Factory (Batch)                 │           │
│  └─────────────────┬───────────────────────┘           │
│                    │                                    │
│                    ▼                                    │
│  ┌─────────────────────────────────────────┐           │
│  │ Data Lake (ADLS Gen2)                   │           │
│  │  - Hierarchical namespace               │           │
│  │  - ACLs (POSIX-style)                   │           │
│  │  - Integration với Azure AD             │           │
│  └─────────────────┬───────────────────────┘           │
│                    │                                    │
│                    ▼                                    │
│  ┌─────────────────────────────────────────┐           │
│  │ Transformation (Data Factory/Synapse)   │           │
│  │  - Mapping Data Flows                   │           │
│  │  - Spark pools                          │           │
│  └─────────────────┬───────────────────────┘           │
│                    │                                    │
│                    ▼                                    │
│  ┌─────────────────────────────────────────┐           │
│  │ Data Warehouse (Synapse Analytics)      │           │
│  │  - Dedicated SQL pools                  │           │
│  │  - Serverless SQL pools                 │           │
│  │  - Spark pools (unified)                │           │
│  └─────────────────┬───────────────────────┘           │
│                    │                                    │
│        ┌───────────┴───────────┐                       │
│        │                       │                       │
│        ▼                       ▼                       │
│  ┌─────────┐            ┌──────────┐                  │
│  │Power BI │⭐          │ Tableau  │                  │
│  │  (BI)   │            │(3rd party)│                  │
│  └─────────┘            └──────────┘                  │
│                                                         │
│  Orchestration: Data Factory, Synapse Pipelines        │
│  Governance: Purview                                   │
│  Security: Azure AD, Key Vault                         │
└─────────────────────────────────────────────────────────┘

4.3. Key Services Deep-Dive

Azure Synapse Analytics (Data Warehouse)

Pros:

• ✅ Unified analytics (SQL + Spark in one place)
• ✅ Serverless SQL pools (pay-per-query like BigQuery)
• ✅ Tight Power BI integration
• ✅ Data Explorer (time-series analytics)

Cons:

• ❌ Complex (many components to learn)
• ❌ Performance varies (not as fast as BigQuery)
• ❌ Dedicated pools expensive

Pricing:

Serverless SQL Pool:
  $5 per TB scanned (same as BigQuery)

Dedicated SQL Pool (DW100c):
  $1.20/hour = $864/month
  (Small warehouse, 100 compute units)

DW500c (typical production):
  $6/hour = $4,320/month

ADLS Gen2 (Data Lake)

Pros:

• ✅ Hierarchical namespace (like filesystem)
• ✅ POSIX ACLs (fine-grained permissions)
• ✅ Cheaper than competitors

Pricing:

Hot tier: $0.0184/GB/month
Cool tier (30+ days): $0.01/GB/month
Archive: $0.00099/GB/month

10TB example: $184/month

Power BI ⭐ Best BI Tool

Pros:

• ✅ Best visualization tool (voted #1 by Gartner)
• ✅ Excel-like familiarity (easy for business users)
• ✅ Tight Office 365 integration
• ✅ Embedded analytics

Pricing:

Power BI Pro: $10/user/month
Power BI Premium: $20/user/month (more features)

For teams:
  10 users × $10 = $100/month (cheap!)

4.4. When to Choose Azure

✅ Microsoft shop: Office 365, Active Directory, .NET stack ✅ Power BI requirement: Best BI tool, tightly integrated ✅ Hybrid cloud: Azure Stack for on-premise ✅ Enterprise agreements: Existing Microsoft licenses (savings) ✅ Vietnam government: Some prefer Azure (Microsoft presence)

5. Side-by-Side Comparison

5.1. Feature Matrix

5.2. Cost Comparison (Same Workload)

Workload: E-commerce startup

• 10TB storage (data lake)
• 5TB active warehouse data
• 30TB query scan/month
• 50 DAG runs/day (ETL)

Note: GCP significantly cheaper for this analytics-heavy workload (BigQuery on-demand magic).

Different workload (less queries, more storage):

• 50TB storage
• 10TB warehouse
• 5TB query scan/month (light usage)

Takeaway: GCP wins for analytics, AWS competitive for storage-heavy, Azure good for Microsoft ecosystem.

6. Decision Framework

6.1. Choose AWS If...

✅ Already heavily invested in AWS ecosystem
✅ Need broadest service catalog (IoT, ML, etc.)
✅ Enterprise với EDP discount program
✅ Specific services only on AWS (e.g., Aurora)
✅ Team expertise in AWS
✅ Hybrid cloud với AWS Outposts
✅ Vietnamese enterprises (familiar, compliant)

6.2. Choose GCP If...

✅ Analytics is core use case (BigQuery best)
✅ Startup/scale-up (fast, simple)
✅ Want transparent pricing
✅ Heavy data science (Vertex AI excellent)
✅ Modern stack (dbt, Fivetran, Looker)
✅ Small team (less ops overhead)
✅ Cost-conscious (often cheapest)

6.3. Choose Azure If...

✅ Microsoft shop (Office 365, AD, .NET)
✅ Power BI is requirement (best BI tool)
✅ Existing Microsoft licenses (cost savings)
✅ Hybrid cloud với Azure Stack
✅ Enterprise agreements with Microsoft
✅ Government/regulated industries (Microsoft compliance)

6.4. Decision Tree

Start
  │
  ├─ Using Office 365 / Power BI? → YES → Azure
  │
  ├─ Analytics-first, small team? → YES → GCP
  │
  ├─ Already on AWS, large ecosystem? → YES → AWS
  │
  └─ Greenfield, no constraints?
       │
       ├─ Budget < $5K/month → GCP (often cheapest)
       ├─ Need ML/AI heavy → GCP (Vertex AI best)
       ├─ Need breadth of services → AWS
       └─ Team Microsoft-familiar → Azure

7. Real-World Case Studies

7.1. Vietnamese Fintech: GCP BigQuery

Company: Lending platform (500K users)

Before: AWS Redshift

• 2-node ra3.4xlarge cluster
• Cost: $5,000/month
• Query performance: 30-60 seconds for complex queries
• Maintenance: 20% engineer time on cluster management

Migration to GCP BigQuery:

• Serverless (no cluster management)
• Cost: $2,500/month (50% reduction)
• Query performance: 5-10 seconds (5x faster)
• Maintenance: 0% time (fully managed)

Why BigQuery won:

• No cluster sizing decisions
• Auto-scaling for traffic spikes
• BigQuery ML for churn prediction (no separate ML infra)
• Transparent pricing (easy to forecast)

Quote from CTO:

"BigQuery transformed our analytics. Queries that took minutes now take seconds. Engineers focus on insights, not infrastructure."

7.2. E-Commerce: Staying on AWS

Company: Fashion e-commerce (2M customers)

Stack: All AWS

• Redshift for data warehouse
• S3 for data lake
• EMR for Spark jobs
• QuickSight for BI

Why stayed on AWS:

• Existing infra: 50+ EC2 instances, RDS databases
• Team expertise: 5 engineers AWS-certified
• EDP discount: 20% discount via enterprise agreement
• Ecosystem: Tight integration với existing services
• Switching cost: 6 months + $200K migration effort

Performance: Good enough (queries < 1 minute acceptable)

Quote:

"We considered BigQuery, but migration cost + team retraining outweighed benefits. AWS meets our needs."

7.3. Enterprise: Azure for Microsoft Stack

Company: Manufacturing company (3,000 employees)

Before: On-premise SQL Server

• $500K upfront hardware
• 3 DBAs maintaining
• Limited scalability

Migration to Azure:

• Synapse Analytics for warehouse
• Power BI for BI (already used)
• Azure Data Factory for ETL

Why Azure:

• Power BI: Already company-wide (1000+ users)
• Active Directory: Seamless SSO
• Microsoft licenses: Enterprise agreement savings
• Hybrid: Some data still on-premise (Azure Stack)
• Training: Team familiar with Microsoft tools

Cost: $8,000/month (vs $1.5M TCO on-premise over 3 years)

Quote:

"Azure was natural choice. Power BI integration alone worth it - business users self-serve analytics now."

8. Multi-Cloud Strategy

8.1. When Multi-Cloud Makes Sense

Pros:

• ✅ Avoid vendor lock-in: Can switch if pricing/service changes
• ✅ Best-of-breed: Use BigQuery (GCP) + Power BI (Azure)
• ✅ Redundancy: Disaster recovery across clouds
• ✅ Geographic: Some regions only on certain clouds

Cons:

• ❌ Complexity: 2-3x operational overhead
• ❌ Cost: Data transfer between clouds expensive
• ❌ Expertise: Team needs to know multiple platforms
• ❌ Integration: Harder to connect services

8.2. Hybrid Example

Company: Scale-up using best of each

GCP:
  - BigQuery (data warehouse) ← Core analytics
  - Vertex AI (ML models)
  - Cloud Storage (data lake)

Azure:
  - Power BI (BI tool) ← Business users
  - Active Directory (SSO)

AWS:
  - EC2 (application servers) ← Legacy infra

Glue layer:
  - Fivetran: Sync data to BigQuery
  - Power BI: Connect to BigQuery (connector)
  - Terraform: Manage all 3 clouds

Cost: 10% higher than single-cloud, but flexibility worth it

8.3. Recommendation

For most Vietnamese companies: Single cloud (simpler)

• Startups → GCP
• Enterprises → AWS or Azure (depending on ecosystem)

Multi-cloud only if:

• Large enterprise (>500 employees)
• Specific requirements (certain service only on X cloud)
• Mergers/acquisitions (inherit different clouds)

9. Vietnam-Specific Considerations

9.1. Data Residency

Law: Decree 72/2013, Cybersecurity Law 24/2018

• Some data (banking, telecom) must stay in Vietnam

Cloud options:

• AWS: Singapore region (closest, 10ms latency)
• GCP: Singapore region
• Azure: Singapore region
• Note: No Vietnam regions yet (2025)

Workaround: Some enterprises use local hosting (Viettel Cloud, VNG Cloud) for regulated data, cloud for analytics.

9.2. Pricing (USD vs VND)

Cloud bills in USD → Exchange rate risk

Example:

• Contract in Jan 2023: $5,000/month = 115M VND (rate 23,000)
• By Dec 2023: $5,000/month = 125M VND (rate 25,000)
• 10M VND increase (8.7%) despite same usage

Mitigation:

• Use GCP committed use discounts (lock USD price)
• Budget in VND with 10% buffer
• Monitor exchange rates

9.3. Support

AWS:

• ✅ Vietnam office (Hanoi, HCMC)
• ✅ Vietnamese support team
• ✅ Local AWS user groups

GCP:

• ✅ Vietnam office (HCMC)
• ✅ Growing Vietnamese community
• ✅ Startup programs (credits)

Azure:

• ✅ Microsoft Vietnam (long presence)
• ✅ Strong enterprise support
• ✅ Power BI community

All 3 have good Vietnam presence - support not differentiator.

Kết Luận

No "best" cloud - depends on your needs.

Quick Recommendations:

Key Takeaways:

1. GCP: Best for analytics (BigQuery unmatched), simpler, often cheaper
2. AWS: Most mature, broadest services, enterprise-friendly
3. Azure: Best for Microsoft ecosystem, Power BI integration
4. Cost: GCP often cheapest for analytics workloads
5. Vietnam market: GCP popular with startups (60%), AWS with enterprises (30%)
6. Multi-cloud: Only for large enterprises (complexity cost)

Next Steps:

• ✅ Assess current infrastructure (on-premise vs cloud)
• ✅ Define requirements (analytics vs storage vs ML)
• ✅ Try free tiers (GCP $300, AWS 12 months free, Azure $200)
• ✅ Đọc Cost Optimization to control cloud spend (Bài 32 upcoming)
• ✅ Đọc Lakehouse Architecture for modern data lake (Bài 33)

Need help? Carptech has implemented Data Platforms on all 3 clouds for Vietnamese companies. Book consultation to discuss your cloud strategy.

Related Posts:

• Coming: Cost Optimization cho Data Platform (Bài 32)
• Coming: Lakehouse Architecture (Bài 33)
• Coming: Serverless Data Architecture (Bài 34)
• Data Platform 101 - Foundation concepts