Best Postgres hosting for startups
Best Postgres hosting for startups
The best Postgres hosting for startups in 2026 is Neon for serverless preview-per-PR teams, Supabase if you also want auth and realtime, Render Postgres for the cheapest "just works" production box, and AWS RDS or Aurora once you cross 100 GB of hot data. Everything else is a niche bet you should make on purpose.
This post compares 11 hosts honestly: Neon, Supabase, Render Postgres, AWS RDS, Aurora, Aurora Serverless v2, Google Cloud SQL, DigitalOcean Managed Postgres, Crunchy Bridge, Tembo, and Xata. Real 2026 pricing, the trade-offs each one hides, and a decision matrix at the end.
The 2-minute verdict
If you stop reading here, this is the table.
| Stage | Workload | Pick |
|---|---|---|
| Pre-revenue MVP | Anything | Neon Free or Supabase Free |
| 1-50 paying users | Boring OLTP SaaS | Render Starter ($7/mo) |
| 50-1k users | OLTP + auth + realtime | Supabase Pro ($25/mo) |
| 1k-10k users | OLTP, lots of branches | Neon Pro ($25/mo and up) |
| 10k+ users | Hot data >100 GB | RDS db.m6g + read replica |
| Mission-critical | Fast failover, multi-AZ | Aurora or Aurora Serverless v2 |
| GCP-native | Anything | Cloud SQL |
| Postgres-purist | OLTP at any scale | Crunchy Bridge |
| Vector-heavy | Search + embeddings | pgvector on Neon, or Tembo VectorDB stack |
| Branchy + search | Content apps | Xata |
The rest of this post is the math behind those picks, the costs nobody publishes, and the failure modes each provider hides in its docs.
What Postgres hosting actually costs in 2026
The pricing pages all use different units, so here is a normalized table. All numbers pulled from public pricing pages in May 2026.
| Provider | Smallest paid plan | Per-GB-month storage | Compute baseline |
|---|---|---|---|
| Neon | $19/mo Launch | ~$0.35 above included | scale-to-zero, ~$0.16/CU-hr active |
| Supabase | $25/mo Pro | included to 8 GB, $0.125 after | 2-core shared baseline |
| Render Postgres | $7/mo Starter (1 GB) | $0.30 on Standard | 0.5 CPU on Starter |
| AWS RDS (db.t4g.medium) | ~$60/mo | $0.115 gp3 SSD | 2 vCPU / 4 GB |
| Aurora (db.t4g.medium) | ~$210/mo base | $0.10 + $0.20 per million I/Os | 2 vCPU / 4 GB |
| Aurora Serverless v2 | 0.5 ACU floor | $0.10 | $0.12 per ACU-hour |
| Google Cloud SQL (db-f1-micro shared) | ~$8/mo | $0.17 SSD | shared vCPU |
| DigitalOcean Managed | $15/mo (1 GB) | included on plan | 1 vCPU / 1 GB |
| Crunchy Bridge | $10/mo Hobby | included to 10 GB | 1 vCPU |
| Tembo | $0 Hobby, ~$50/mo Pro | varies by stack | usage-based |
| Xata | $0 free (15 GB), ~$20/mo | included to 15 GB | branch-based |
The number that matters most is not the headline plan price. It is the cost behavior under spike. Imagine your launch hits Hacker News and you get 50 million queries in a day, with database writes spiking 4x. Here is how each host behaves.
- Neon Pro auto-scales compute up to your plan limit; the extra compute is metered hourly. Expect $40-80 in extra spend that day.
- Supabase Pro runs on a fixed compute size; if you outgrow it, queries queue and you upgrade plan ($25 to $99). Spike is "service degrades" not "bill explodes."
- Render Standard is fixed-size; a 50M query day on a 1 vCPU plan will throttle before your wallet does.
- AWS RDS gp3 scales storage IOPS automatically up to provisioned ceiling; spike costs are mostly bandwidth and snapshot.
- Aurora charges $0.20 per million I/Os, so a 50M query day with average 4 reads per query is roughly $40 in I/O on top of compute.
- Aurora Serverless v2 silently scales ACUs up; bill goes from $43/mo (0.5 ACU) to $200+ on a hot day. Smooth, expensive.
This is the cost story most "best Postgres" posts skip.
Neon, Supabase, and the serverless-Postgres tier
These two are where most 2026 startups should start.
Neon is plain Postgres with three killer features: scale-to-zero on idle, instant database branching per pull request, and read replicas without re-architecting your app. The free tier (one project, 0.5 GB) is fine for prototypes. Pro at $25/mo (8 GB included) is the sweet spot for a side project that started getting users. We have a longer breakdown in our Neon Postgres review if you want the deep dive.
Where Neon breaks: a database that gets traffic every 5 minutes never sleeps, so scale-to-zero only saves money on truly idle stuff like preview branches and side projects. And the cold-start (~500 ms to wake) will surprise you in production if you ignore it.
Supabase is Postgres plus a batteries-included stack: auth, row-level security tooling, realtime subscriptions, storage, and auto-generated REST and GraphQL APIs. Pro is $25/mo with 8 GB. The thing nobody tells you: once your app depends on Supabase Auth and Supabase Realtime, switching off Supabase is a rewrite, not a migration. That is fine if you love the stack. It is a problem if you discover Supabase's pgvector ergonomics at 10M embeddings.
Pick Neon if you want pure Postgres and CI-friendly branching. Pick Supabase if you want to ship faster by skipping auth and realtime. Both are excellent. Both have failure modes worth knowing before you tattoo them on your stack.
Render and DigitalOcean: the boring-box tier
These hosts give you a Postgres that runs in a region, near your app server, with predictable monthly pricing and no scale-to-zero magic.
Render Postgres at $7/mo is the cheapest credible production database in 2026. Starter gives you 1 GB and 0.5 CPU; Standard at $20/mo gives you 4 GB and 1 CPU; from there you scale by plan. Render also runs your app, your background workers, and your cron jobs, so the Postgres is in the same VPC as everything else. The catch: no read replicas on Starter, smaller region list than AWS, and the Standard plan caps at fairly modest IOPS.
DigitalOcean Managed Postgres starts at $15/mo for 1 GB / 1 vCPU. The pricing is bundled (storage, bandwidth, daily backups all included), which is a big deal when you compare against AWS pricing pages. DigitalOcean's weakness in 2026 is region count: only 8 regions, no native multi-region writes, and slower roll-out of new Postgres major versions.
If you read our Fly.io review for production workloads, the DO trade-offs will sound familiar: predictable, regional, lacks the global-edge story.
AWS RDS, Aurora, and Aurora Serverless v2
If you are on AWS already, ignore the boutique providers and use the AWS native option. The question is which one.
RDS is what 80% of AWS-native startups should pick. A db.t4g.medium (2 vCPU, 4 GB) runs about $60/mo plus storage at $0.115/GB-month for gp3. Multi-AZ doubles the cost. RDS has a known quirk: minor version upgrades require maintenance windows, and major version upgrades sometimes do not happen for months after Postgres ships them.
Aurora is RDS's premium sibling. It rewrites Postgres storage to use a distributed log, which gives you sub-30-second failover, 6-way replication across AZs, and read replicas that share storage (cheap). The base is ~$210/mo for the smallest production setup, and you pay $0.20 per million I/Os on top. Aurora is the right answer at roughly 100 GB of hot data and high availability requirements. Below that it is overkill.
Aurora Serverless v2 is Aurora that scales compute in 0.5 ACU increments, with a 0.5 ACU floor (about $43/mo idle). It does not scale to zero. It is great for spiky workloads (campaign launches, marketplace flash demand). It is a tax on steady-state workloads where a fixed db.t4g.large would be cheaper.
The honest take: most startups overpay AWS by picking Aurora when RDS would do.
Google Cloud SQL, Crunchy Data, Tembo, Xata
The last four are smaller in market share but right for specific cases.
Google Cloud SQL is the GCP-native answer. Shared-core db-f1-micro is about $8/mo (no SLA), and standard tiers start around $50/mo. If your app runs on Cloud Run or GKE, do not fight gravity. The downside: Cloud SQL is slower than RDS to roll out new Postgres major versions and has weaker tooling around custom extensions.
Crunchy Bridge is "Postgres by people who only do Postgres." Hobby is $10/mo (10 GB included), Production starts around $135/mo for a high-availability cluster. They publish their internals, support every major extension out of the gate, and give you full superuser. If your team is Postgres-pilled (DBA in the founding team, lots of custom extensions), Crunchy is the most adult option on this list.
Tembo ships pre-built extension stacks: a "VectorDB stack" with pgvector, pgvectorscale, and pg_search; a "Message Queue stack" with PGMQ; a "Geospatial stack" with PostGIS. Hobby is free, Pro starts around $50/mo. Tembo is the right pick if you want a single Postgres database to do work that would normally require multiple services. Watch out: it is newer and the operational track record is shorter than AWS or Crunchy.
Xata is a Postgres derivative with branching, search, and an opinionated ORM-like SDK. Free tier gives you 15 GB and three branches; paid plans start around $20/mo. Xata is the right pick for content-heavy apps that need full-text search and faceted filters with no extra service. It is the wrong pick if you want raw Postgres SQL access without their abstractions in your way.
For pure vector workloads, our vector database comparison has the long take. The short version: stay on Postgres until you cross roughly 50 million vectors.
The decision matrix: stage by workload
Map your situation to the row, then the column.
| Workload | Pre-revenue | Early traction (1-50 paying) | Scale (1k+ users) |
|---|---|---|---|
| OLTP startup | Neon Free or Supabase Free | Render Standard or Supabase Pro | RDS multi-AZ or Crunchy Production |
| Analytics-heavy | Supabase Free + read replica plans | Neon Pro w/ replica | Aurora + Redshift sidecar |
| Multi-region | Skip; single region | Cloud SQL HA + read replicas | Aurora Global Database, or CockroachDB |
| Vector-search | Neon + pgvector | Supabase + pgvector | Tembo VectorDB stack, or Aurora + Pinecone sidecar |
If your row is "we do not know yet," default to Neon Free or Supabase Free and re-evaluate at 100 paying users. Re-platforming a 1 GB Postgres takes an afternoon. Re-platforming a 200 GB Postgres takes a week.
What to do this week
- Open your current Postgres bill. Note the line items.
- Add up your storage size (in GB) and your average queries-per-day from your APM or
pg_stat_statements. If you do not have those numbers, that is the first problem to solve. - Plug those numbers into the cost table at the top of this post. If your current host is more than 30% off the cheapest credible alternative for your shape, plan a migration.
- If you are already on Neon or Supabase and your hot data is over 50 GB, run the math on RDS. The break-even crosses around there.
- If you do not have anyone in-house who has migrated Postgres before, that is the moment to bring in help.
That last one is where Cadence fits. We have engineers in our 12,800-engineer pool who have done Neon-to-RDS, Supabase-to-Aurora, and RDS-to-Crunchy migrations. The median time-to-first-commit on a database migration project on Cadence is 27 hours from booking. Every engineer on Cadence is AI-native by default (vetted on Cursor, Claude Code, and Copilot fluency), so the migration script gets written in an afternoon, not a sprint.
If you are choosing your first production Postgres host this week, the fastest sanity check is a 30-minute review with someone who has shipped on three of the providers above. Cadence shortlists 4 vetted engineers in 2 minutes with a 48-hour free trial, so you can audit your hosting choice before you commit to a year of bills.
Picking the right Postgres host on day one matters less than picking one you can leave on day 1,000. Optimize for that.
FAQ
What is the cheapest Postgres hosting for startups?
Render Postgres at $7/month for a 1 GB Starter is the cheapest credible production database in 2026. If you can tolerate occasional cold starts on idle, Neon's free tier (0.5 GB) is fine for prototypes and side projects.
Is Neon or Supabase better for a new SaaS?
Neon if you only need Postgres and want one preview database per pull request. Supabase if you want auth, realtime, storage, and Postgres in the same dashboard. Both are $25/mo at the Pro tier; the choice is about what you want bundled.
When should a startup move off Neon or Supabase?
Around 100 GB of hot data, or when query volume passes a few hundred million per month. At that scale, the per-GB price on RDS or Aurora becomes meaningfully cheaper, and you start needing read replicas and tighter performance tuning than the serverless tier supports cleanly.
Should startups use AWS Aurora?
Only if you already run on AWS and need fast failover, six-way replication, or read replicas with shared storage. Aurora's $200+ monthly base is wasted on a side project. Use RDS until you outgrow it.
Do I need a separate vector database?
No, not under 50 million vectors. pgvector on Neon, Supabase, or RDS handles most embedding workloads cleanly. Move to Tembo's VectorDB stack, Pinecone, or Turbopuffer once recall or latency starts hurting at scale.