Postgres Hosting Compared: Supabase, Neon, Railway, RDS

The best Postgres hosting for most new apps is Supabase or Neon: Supabase when you want auth, storage and a fixed monthly plan around the database, Neon when you want only Postgres with scale to zero and cheap branches. Railway suits apps already deployed there, and AWS RDS is the pick once you need AWS networking, compliance controls or reserved pricing.

Prices and limits checked October 2026 on each provider's official pages: Supabase pricing, Neon pricing, Railway pricing and Amazon RDS for PostgreSQL pricing (US East, N. Virginia). Workload costs below are illustrative calculations from those published rates, not measurements of a running app.

Most comparisons stop at the free tier. That is the wrong place to decide, because the free tier is the one stage every provider handles well. The real differences show up when a database must stay on all day, keep backups, accept connections from serverless functions and give each pull request its own copy of the data. This page prices those exact things. If you are still choosing the rest of the stack, start with our guide to building an AI SaaS, which places the database in the full launch order.

Postgres hosting providers compared side by side

All four run standard PostgreSQL, so your schema, queries and extensions such as pgvector move with you. What differs is how you pay, how connections are handled, how backups work and whether you can branch the database. The table summarises the published terms.

Sources, checked October 2026: provider pricing pages linked above; Railway backups and point-in-time recovery docs; AWS backup retention docs; AWS Free Tier.

Is there good free Postgres hosting?

Yes, for building and testing. The free tiers differ in what happens when you stop using them, and that matters more than the storage number.

• Neon Free gives up to 100 projects, each with 100 CU-hours of compute a month and 1 GB of storage, with compute sizes up to 2 CU. Compute scales to zero after 5 minutes of inactivity, so an idle project uses nothing. Neon's own example: at the default 0.25 CU, 100 CU-hours covers about 400 hours of active time.
• Supabase Free includes a 500 MB database, shared CPU with 500 MB of RAM, 5 GB of egress and the rest of the Supabase platform (auth for 50,000 monthly active users, 1 GB of file storage). It is limited to two active projects, has no automatic backups, and pauses projects after one week of inactivity.
• Railway starts with a 30-day trial that includes a one-time $5 credit, then a free plan with up to 0.5 GB of RAM per service and 0.5 GB of volume storage. It is enough to try the Postgres template, not to run a product.
• AWS gives new accounts up to $200 in Free Tier credits over six months, which can cover a small RDS instance while you learn the console.

What each provider costs for a realistic workload

Headline prices do not compare cleanly: Supabase sells a plan plus a compute size, Neon bills compute by the hour it runs, Railway bills per second for the memory and CPU the container uses, and RDS bills an instance by the hour plus provisioned storage. To make them comparable, the table prices three stages from each provider's published rates (checked October 2026), using 730 hours a month.

Illustrative calculations, not measurements. Supabase: $25 Pro plus compute ($15 Small, $110 Large) minus the $10 compute credit, plus $0.125 per GB of disk over 8 GB. Neon Launch: $0.106 per CU-hour and $0.35 per GB-month; the launched-app row uses Neon's published low-load example (190 CU-hours, 5 GB), and the growing row assumes a fixed 2 CU (about 8 GB RAM) all month with autoscaling off. Railway: $0.00000386 per GB of RAM per second, $0.00000772 per vCPU per second and $0.00000006 per GB of volume per second, with the stated usage assumptions. RDS us-east-1 on-demand: db.t4g.micro $0.016/hour, db.t4g.small $0.032 (Multi-AZ $0.065), db.m7g.large $0.168 (Multi-AZ $0.337), gp3 storage at the 20 GB minimum or 50 GB, $0.115 per GB-month (Multi-AZ $0.23). Egress, history storage and support plans are excluded.

Three things stand out. First, at the launched-app stage everyone lands in a similar band, so price should not be the deciding factor there; features should. Second, Railway looks cheapest on paper because it bills what the container actually uses, but that number moves with load, and the assumptions above are yours to replace with real metrics. Third, high availability roughly doubles the RDS line because Multi-AZ runs a standby; Neon's storage is multi-AZ by default, Supabase reserves uptime SLAs for Enterprise, and Railway offers a Postgres high-availability conversion on its own pricing terms.

Supabase: Postgres plus the backend around it

Supabase gives each project a dedicated Postgres instance and builds auth, file storage, realtime, edge functions and an auto-generated API around it. Pro starts at $25 a month and includes $10 of compute credit, enough for one Micro instance (1 GB RAM, 60 direct and 200 pooled connections). Each larger size raises both: Small is $15 (90 direct, 400 pooled), Medium $60, Large $110 on 2 dedicated vCPUs and 8 GB (160 direct, 800 pooled). Pro also includes 8 GB of disk per project, 250 GB of egress, daily backups kept 7 days, and a spend cap that is on by default (all checked October 2026 on the Supabase pricing page).

The costs that surprise people are add-ons: point-in-time recovery is $100 a month per 7 days of retention, branching is $0.01344 per branch-hour (about $9.81 for a branch left running all month), and each additional project starts at $10 a month for compute. Pick Supabase when the extras replace services you would otherwise wire together. Our Supabase full-stack tutorial shows that setup end to end.

Neon: serverless Postgres with branching

Neon separates storage from compute, so the database can scale to zero when idle and autoscale under load. On the Launch plan you pay $0.106 per CU-hour (one CU is about 4 GB of RAM) and $0.35 per GB-month of storage, with no monthly minimum, sizes up to 16 CU, 10 branches per project included, and instant restore up to 7 days. History storage for that restore window is billed at $0.20 per GB-month of retained changes. Extra branches cost about $0.002 per branch-hour. Scale raises the CU-hour rate to $0.222 and adds SLAs, HIPAA and private networking (checked October 2026 on the Neon pricing page).

Two settings control the bill: a maximum autoscaling size, which acts as a cost ceiling, and scale to zero on every non-production branch. Neon notes that an always-on database at the 0.25 CU minimum uses about 182.5 CU-hours a month, which is the floor for a small production app. Pick Neon for serverless apps, per-pull-request preview databases and many small databases.

Railway: Postgres next to your app servers

Railway deploys Postgres as a service from its SSL-enabled image based on the official Postgres image, inside the same project as your app. Billing is per second for the RAM and CPU the service uses, plus volume storage and $0.05 per GB of egress. Hobby has a $5 minimum that includes $5 of usage; Pro has a $20 minimum with $20 included and higher limits (checked October 2026 on the Railway pricing page).

Backups are a volume feature: daily backups are kept 6 days, weekly 27 days and monthly 89 days, billed at volume rates for the incremental data, and Railway's docs describe them as still under development. Point-in-time recovery is available for Postgres through a storage bucket with a roughly four-week window. The trade-off is that Railway's database support scope is narrower than a dedicated database host: change the template's image or start command and the database becomes yours to manage. Pick Railway when your API, workers and cron jobs already live there and private networking between them matters. If you self-host automation tools on the same project, our n8n self-hosting guide covers the Postgres side.

AWS RDS for PostgreSQL: the grown-up default

RDS bills an instance by the hour and storage by the GB you provision. In US East, a db.t4g.micro is $0.016 an hour and a db.m7g.large (2 vCPU, 8 GiB) is $0.168, with Multi-AZ roughly double. gp3 storage is $0.115 per GB-month with a 20 GiB minimum and 3,000 IOPS baseline, and backup storage is free up to 100% of your provisioned database storage (checked October 2026 on the RDS for PostgreSQL pricing page). Automated backups support retention of up to 35 days. T4g and T3 instances run in Unlimited mode, so sustained CPU above baseline adds $0.075 per vCPU-hour.

There is no branching, no scale to zero and more to configure: VPC, security groups, parameter groups and IAM. In return you get the full AWS toolbox, reserved instances and Database Savings Plans for steady load, and a database that sits beside the rest of your AWS infrastructure. Pick RDS when a customer contract, compliance review or existing AWS footprint asks for it.

How do you keep Postgres hosting cheap?

The biggest lever is usually not the provider but idle compute. A preview database left running all month, or a dev project that never scales down, can cost as much as production. If you also call model APIs from the same app, our Claude API pricing breakdown covers the other large line on an AI product's bill.

Our picks by stage

Building an AI product on top of this stack is what our AI SaaS Builder program covers, from the database schema to auth, billing and deployment.

Moving between providers later

Because all four are standard Postgres, leaving is a data move, not a rewrite. Use pg_dump and pg_restore for small databases with a short maintenance window, or logical replication to keep a new database in sync until you switch the connection string. The lock-in is in the extras: Supabase auth users and storage buckets, Neon branch workflows in CI, Railway private networking, AWS IAM and VPC rules. Keep migrations in your repository and secrets in environment variables, and any of these providers can be swapped in a weekend.

Postgres hosting: FAQ