Database Scaling Costs in 2025: When to Scale and What It Actually Costs

Database costs are the surprise that hits most startups between $10K and $100K MRR. Compute costs are predictable. Database costs scale non-linearly — storage, read replicas, and connection pooling all add up in unexpected ways.

Managed Database Pricing (2025)

PostgreSQL — AWS RDS Pricing

Add to these: Multi-AZ adds ~2x cost. Read replicas add the same cost per replica. Data transfer out (reads to application) billed additionally at $0.09/GB.

PlanetScale (MySQL-compatible, serverless)

Supabase (PostgreSQL managed)

Supabase Pro at $25/month is exceptional value for startups — includes auth, realtime, storage, and APIs in addition to PostgreSQL.

The Scaling Cost Curve

A typical web application's database costs at different user scales:

These are guidelines, not guarantees — actual costs depend heavily on query patterns, indexing, and whether you've optimized before scaling hardware.

The Optimization Sequence (Before Scaling Up)

Hardware scaling is expensive and permanent. Optimization is one-time work with permanent savings. Always exhaust optimization first:

1. Query optimization (0-cost, biggest impact)

Unindexed queries on large tables are the #1 database performance killer. Tools:

• EXPLAIN ANALYZE in PostgreSQL — shows query execution plan
• pg_stat_statements — identifies slowest queries
• AWS RDS Performance Insights — visual query performance

One missing index on a high-frequency query can cause 100x query time. Adding it costs nothing.

2. Connection pooling (0-cost, major impact)

Each database connection uses ~5-10 MB RAM. A Rails/Node app without pooling might open 100+ connections per instance.

PgBouncer or pgcat (connection pooler) can reduce required database connections by 10-20x — allowing a smaller instance to handle the same load.

3. Read replicas (moderate cost, major scalability)

For read-heavy workloads (most web apps), directing read queries to read replicas allows the primary to focus on writes. A single read replica effectively doubles your read capacity.

Cost: same as primary instance (≈ 2x total cost), but avoids moving to the next instance tier.

4. Caching layer (moderate cost, major impact)

Redis or Memcached for frequently-accessed data:

• Cache database query results for 60-300 seconds
• Eliminates repeated identical queries
• ElastiCache Redis: $15-200/month depending on size

For most applications, a proper caching layer reduces database load by 50-80%.

5. Vertical scaling (last resort)

Only scale instance size after: queries are optimized, connection pooling is in place, read replicas are directing read traffic, and caching is active.

Estimated Savings from Optimization

A company spending $800/month on RDS at 200K users with unoptimized queries:

Total: $650-1,300/month savings — often allowing a smaller (cheaper) primary instance.

Use the Cloud Cost Calculator to estimate database costs at different scale levels.