Postgres on a €6 VPS vs $25 Managed: The Honest Math
€5.99 vs $25 — plus your hours the number that matters here
By Stan K.
A 4 GB Hetzner CX23 costs €5.99; Supabase Pro costs $25. But the VPS price excludes backups, monitoring and 2–4 hours a month of you. The full comparison — and the exit costs that make the choice about more than money.
The raw infrastructure comparison is brutal and simple: a Hetzner CX23 (2 vCPU, 4 GB RAM, 40 GB NVMe, 20 TB traffic) is €5.99/month. Supabase Pro is $25/month (including $10 of compute credits). Managed Postgres elsewhere starts at $25–60 for comparable RAM. Self-hosting saves 50–80% on the sticker, every month, at every scale.
Then the fine print arrives — on both sides.
The VPS column, itemized
What “Postgres on a VPS” actually costs when done properly:
| Item | Cost |
|---|---|
| VPS (4 GB production floor) | €5.99/month |
| Off-site WAL + snapshots (pgBackRest/WAL-G → R2 or B2, standard supported config) | $1–3/month |
| Monitoring (postgres_exporter + Prometheus/Grafana, or free uptime tiers) | $0–5/month |
| Admin time: OS patches, minor upgrades, VACUUM tuning, restore drills | 2–4 hours/month |
The first three lines are cheap. The fourth is the real price: at any professional rate, two hours a month of database admin is $150–300/month of time — which erases the savings entirely unless you already manage Linux boxes for a living (we do — which is exactly why this math works for us and not for everyone).
The other exclusions from the VPS column: failover is manual (minutes-to-hours of downtime versus managed Multi-AZ’s sub-minute), upgrades are your weekends, and compliance (SOC2/HIPAA) is your audit.
The managed column, itemized
The $25 Supabase Pro includes: always-on (no 7-day idle pause), 250 GB egress then $0.09/GB, automated backups, a pooler, and — the underrated feature — Postgres itself. Standard Postgres means the exit is a pg_dump: any Postgres anywhere takes the data with zero schema rewrite. The documented migration cost from Supabase is measured in days ($500–3K of work); the documented migration cost out of Firestore is 2–6 weeks ($5–25K) of flattening NoSQL into relational and rewriting every client query.
pg_dump: days. Firestore rewrite: weeks.The exit cost is part of the price
Supabase → RDS/self-hosted = export and go. Firestore → anywhere = re-architecture. When choosing a backend, price the leaving, not just the staying — vendor-friendly entry pricing is repaid at exit with interest.
The actual decision rules
Self-host wins when: you have sysadmin fluency, traffic is steady and predictable, storage/egress needs are large (50 GB+ where managed egress fees start compounding), or you want app and database on the same machine for sub-millisecond local-socket latency. Setups like ours — collectors and dashboards with bursty write jobs — sit nicely on a VPS with WAL streaming to a $6.95/TB bucket.
Managed wins when: you have zero engineering bandwidth for maintenance, you need Multi-AZ failover, compliance asks questions, your app is serverless with spiky connection patterns, or — honestly — the $19/month delta is cheaper than your time. For a solo builder shipping product, $25 to never think about the database is frequently the cheaper option, and pretending otherwise is the trap.
Neither wins when the answer is SQLite. Before this whole comparison: if one process can own the data, a single SQLite file eliminates the database server question entirely — our platforms ran that way for years.
The checklist if you go VPS
- pgBackRest or WAL-G to object storage from day one — a VPS database with no off-site WAL is a data-loss plan, not a cost saving
- Restore drill once a quarter (the restore-economics article generalizes: untested restores are superstition)
- Monitoring before you need it — disk, connections, replication lag
- Know your maintenance window and own it
The full per-provider pricing — all six managed providers at the same fixed workload, plus the self-hosted table — is in the Databases hub.