How to Choose a Cloud Database: AWS vs Azure vs GCP (2026)
Compare managed database options across cloud providers. Covers relational, NoSQL, in-memory, and data warehouse services with pricing considerations.
TL;DR
- • For relational (MySQL/Postgres): AWS RDS and GCP Cloud SQL are most mature; DigitalOcean, Linode, and Vultr offer managed databases from $15/mo.
- • For NoSQL: DynamoDB (AWS) for throughput, Cosmos DB (Azure) for multi-model/global, Firestore (GCP) for generous free tier.
- • For cost-sensitive projects, managed databases on smaller providers are 50-80% cheaper than Big 3 equivalents with the same engine.
Choosing the right cloud database service is one of the most impactful architectural decisions you'll make. Beyond the Big 3 (AWS, Azure, GCP), smaller providers now offer fully managed MySQL and PostgreSQL databases at a fraction of the cost. This guide compares managed database options across all 8 providers to help you choose based on requirements, cost, and operational complexity.
Side-by-Side Comparison
| Feature | AWS | Azure | GCP | DO | Hetzner | Linode | OCI | Vultr |
|---|---|---|---|---|---|---|---|---|
| Relational (MySQL/PostgreSQL) | Amazon RDS | Azure SQL Database / Flexible Server | Cloud SQL | Managed Databases | — | Managed Databases | MySQL/PostgreSQL DB System | Managed Databases |
| Relational (proprietary/enhanced) | Amazon Aurora | Azure SQL Managed Instance | AlloyDB | — | — | — | Autonomous Database | — |
| NoSQL document | DynamoDB | Cosmos DB | Firestore | — | — | — | NoSQL Database Cloud | — |
| In-memory cache | ElastiCache (Redis/Memcached) | Azure Cache for Redis | Memorystore | Managed Redis | — | — | — | Managed Caching (Redis) |
| Data warehouse | Redshift | Synapse Analytics | BigQuery | — | — | — | Autonomous Data Warehouse | — |
| Pricing model | Per-hour instance + storage | DTU or vCore-based | Per-hour instance + storage | Fixed monthly + storage | — | Fixed monthly + storage | Per-hour OCPU + storage | Fixed monthly + storage |
| Serverless option | Aurora Serverless, DynamoDB | Azure SQL Serverless, Cosmos DB | BigQuery, Firestore, AlloyDB Omni | — | — | — | Autonomous DB Serverless | — |
| Cheapest entry price | db.t4g.micro ~$12/mo | Basic B1 ~$5/mo | db-f1-micro ~$9/mo | $15/mo (1 vCPU, 1 GB) | — | $15/mo (1 vCPU, 1 GB) | Always Free (1 OCPU, 20 GB) | $15/mo (1 vCPU, 1 GB) |
Relational Databases: The Full Landscape
For traditional relational workloads (MySQL, PostgreSQL), options range from Big 3 managed services to budget-friendly alternatives. AWS RDS supports 6 engines with the broadest instance selection. Azure SQL Database offers DTU-based (bundled) or vCore-based (flexible) pricing. GCP Cloud SQL provides straightforward per-second billing.
Smaller providers offer managed MySQL and PostgreSQL at significantly lower prices. DigitalOcean Managed Databases start at $15/mo for 1 vCPU / 1 GB with daily backups and automatic failover. Linode Managed Databases start at $15/mo with similar specs. Vultr Managed Databases also start at $15/mo. These services handle backups, updates, and high availability.
For PostgreSQL specifically, AWS Aurora and GCP AlloyDB offer enhanced performance at premium pricing. Oracle Autonomous Database provides self-tuning capabilities.
Example: A production PostgreSQL database with 4 vCPU, 16 GB RAM, 100 GB storage: AWS RDS db.m5.xlarge: ~$280/mo. GCP Cloud SQL: ~$260/mo. DigitalOcean: $120/mo. Vultr: $120/mo. Linode: $120/mo. The smaller providers are 55-57% cheaper for equivalent specs, though they offer fewer enterprise features (read replicas may cost extra, fewer monitoring tools).
NoSQL Databases: DynamoDB vs Cosmos DB vs Firestore
For NoSQL workloads, the pricing models differ dramatically. DynamoDB charges per read/write capacity unit (provisioned or on-demand), making cost proportional to throughput. Cosmos DB charges per Request Unit (RU) and offers 5 consistency levels — the only service to do so. Firestore charges per document read/write/delete plus storage.
For small workloads, Firestore's generous free tier (50K reads, 20K writes/day) makes it the cheapest — many small apps run for free. For high-throughput workloads, DynamoDB provisioned mode with auto-scaling is typically 60-80% cheaper than on-demand mode. For global distribution, Cosmos DB's multi-region active-active replication is unmatched but expensive.
Smaller providers generally don't offer managed NoSQL databases. If you need NoSQL on a smaller provider, self-host MongoDB or Redis on a compute instance.
Data Warehouses: Redshift vs Synapse vs BigQuery
Data warehouse pricing varies the most across providers. Redshift uses provisioned clusters (RA3 nodes) with separate compute and storage pricing. Synapse Analytics offers dedicated SQL pools and serverless on-demand queries billed per TB scanned. BigQuery is fully serverless with on-demand ($6.25/TB scanned) or flat-rate slots.
Oracle Autonomous Data Warehouse is an underrated option — with Always Free tier (1 OCPU, 20 GB), you can run small analytics workloads at zero cost. For larger workloads, Oracle's pricing is competitive with Redshift.
For sporadic queries, BigQuery on-demand or Synapse serverless win on cost. For predictable analytics workloads, Redshift RA3 or Synapse dedicated pools are cost-effective.
Example: Querying 1 TB of data once per day: BigQuery on-demand: ~$190/mo ($6.25/TB × 30). Redshift dc2.large (2 nodes): ~$380/mo but unlimited queries. Oracle ADW: Free tier covers small datasets.
Managed Databases on Smaller Providers
DigitalOcean, Linode, and Vultr have made managed databases accessible and affordable. All three offer managed MySQL and PostgreSQL with automatic backups, encryption, and maintenance. DigitalOcean also offers managed Redis and MongoDB.
DigitalOcean Managed Databases: MySQL, PostgreSQL, Redis, MongoDB. Plans from $15/mo (1 vCPU, 1 GB) to $1,680/mo (32 vCPU, 128 GB). Standby nodes and read replicas available. Connection pooling included for PostgreSQL.
Linode Managed Databases: MySQL, PostgreSQL. Plans from $15/mo. High availability with automatic failover. Daily backups included.
Vultr Managed Databases: MySQL, PostgreSQL, Redis (caching). Plans from $15/mo (1 vCPU, 1 GB). Automated backups, point-in-time recovery.
These services lack some Big 3 features (no serverless, limited monitoring, fewer engine options), but for straightforward relational workloads, they're excellent value.
How to Choose
Start with your workload requirements:
1. Data model — relational (SQL) or document/key-value (NoSQL)?
2. Scale — how much data and throughput?
3. Access pattern — OLTP (many small transactions) or OLAP (few large queries)?
4. Serverless need — do you want zero ops?
5. Budget — are Big 3 premium features worth the cost?
For OLTP relational workloads under $500/mo budget: DigitalOcean, Linode, or Vultr Managed Databases. For OLTP needing global replication or advanced features: AWS RDS/Aurora or GCP Cloud SQL. For NoSQL: DynamoDB for throughput, Cosmos DB for global, Firestore for simplicity. For analytics: BigQuery for serverless, Redshift for predictable workloads, Oracle ADW for free tier. Always prototype with your actual query patterns to compare costs accurately.
Frequently Asked Questions
Which cloud database is cheapest for small applications?
Oracle Autonomous Database has a free tier (1 OCPU, 20 GB) that never expires. For managed databases, DigitalOcean, Linode, and Vultr all offer MySQL/PostgreSQL from $15/mo with backups and maintenance included. GCP Firestore's free tier covers many small NoSQL apps. Azure SQL Basic tier is $5/mo for very small relational workloads.
Should I use a serverless database?
Serverless databases (Aurora Serverless, Cosmos DB Serverless, BigQuery, Firestore) are ideal for variable workloads with quiet periods. They eliminate idle costs but can be more expensive during sustained high usage. Use serverless for dev/test and unpredictable workloads; use provisioned for steady production workloads.
How can I reduce my cloud database costs?
1) Consider a smaller provider for standard MySQL/PostgreSQL workloads (save 50-80%). 2) Use Reserved Instances on Big 3 for predictable workloads (save 40-60%). 3) Right-size your instances based on actual CPU/memory usage. 4) Use read replicas instead of scaling up. 5) Archive old data to cheaper storage. 6) Optimize queries to reduce I/O.
Are managed databases on DigitalOcean/Linode/Vultr production-ready?
Yes. All three offer automatic backups, encryption at rest, high availability with failover, and connection pooling. They use the same MySQL and PostgreSQL engines as the Big 3. They're widely used in production for web applications, SaaS products, and APIs. The main limitations are fewer engine options, less granular monitoring, and fewer advanced features (like zero-downtime migration or global replication).