Cheap always-on Postgres in 2026: $5 to $25 options, and what each one leaves out

Your app is ready for real users. Your free database has a different idea: sleep after inactivity, pause until you open a dashboard, or expire after a trial. You are ready to pay a little to remove that problem. How little, and what are you giving up?

Always-on Postgres starts at $5 at several providers. Aiven Developer, PlanetScale single-node, Heroku Essential-0 and Layerbase Solo all have a $5 entry. Render starts at $6 for compute plus storage. Neon Launch at its smallest compute held awake costs about $19.35 for a 730-hour month before storage, and Supabase Pro starts at $25 with one Micro project covered. Railway's $5 is a billing minimum with usage credit, not a fixed database price.

I build Layerbase, which is in this comparison. Check the linked vendor documentation, including ours. There is no single winner: the useful question is which plan's missing feature you can live without. Jump to the comparison.

What does always-on Postgres actually mean?

Here it means your database does not intentionally stop its compute because it is idle. It does not mean zero downtime, high availability, an uptime SLA, or that every first query is equally fast.

There are three common ways to buy it:

• An always-running instance. Aiven Developer and PlanetScale single-node sell this directly. You still need to read their maintenance and failure-recovery terms.
• Serverless compute with sleep disabled. On Neon Launch, you can disable scale-to-zero and pay for every hour that compute remains active. The switch is per compute; check the branch your application uses.
• Reserved capacity in a shared pool. Layerbase lets you pin a database into a paid always-on pool. Paying for the plan alone does not pin every database in the account.

On Railway, leave Serverless sleep disabled for the database service. Railway's PostgreSQL template is unmanaged: you retain operational responsibilities that a managed database provider takes on.

If your database has already disappeared behind a dashboard action, start with sleep vs pause vs archive. This guide compares the paid step after Free, not another keep-alive workaround.

The always-on Postgres comparison

Checked October 20, 2026. Prices are USD per month before tax. Usage examples assume 730 hours. Prices marked “+” exclude the named charges; annual discounts and trial credits are excluded. This is a shortlist of entry tiers, not a claim to cover every host. RAM pools, shared instances and dedicated VMs are different products, and connection limits are not throughput benchmarks.

Sources for operational details: Aiven tier restrictions, Heroku plan limits, Render backups, Render storage billing, Crunchy Bridge pricing, Neon pooling, and Supabase compute specifications.

A connection ceiling is not a recommendation to open that many active queries. A 256 MB database with 100 allowed connections can run out of memory long before it runs out of connection slots. Size your application's pool across all web processes, workers and preview deployments.

For one small app and a development copy: start free, then pin one Postgres always-on for $5 when you launch.

What $5 buys, and what it leaves out

Aiven: more RAM, with explicit development-tier restrictions

Aiven includes a dedicated VM, daily backups and networking at $5. Its own positioning excludes production use and an uptime SLA. You can choose North America, Europe or Asia Pacific, but not a specific cloud region. That geographic choice is useful; calling it “no region choice” would be misleading.

The connection limit needs care. The PostgreSQL connection reference lists 15 for Developer; the service-pricing page says 20. Both were present when checked. Verify the provisioned service's setting and usable capacity before committing to either number. PostgreSQL pooling is excluded on this tier.

PlanetScale: a small node with an inexpensive HA path

The $5 PS-5 is 512 MiB with 1/16 vCPU. PlanetScale explicitly permits production workloads that do not require HA. That is a different statement from Aiven's development-only positioning.

The current catalog lists a three-node PS-5 HA configuration at $15 in its AWS US East example. Storage, backup storage and egress remain additional charges; prices vary by region. Buy the $5 node because its capacity and recovery model fit, not because “Postgres from $5” implies every production feature is included.

Layerbase: one pinned database and one that may sleep

Solo is $5/mo for two databases, with one kept always-on from a 768 MB pool. It includes 10 GB total storage, 100 concurrent connections and daily backups retained for seven days. The pool is a plan reservation, not a claim that each database receives a dedicated 768 MB machine.

The unpinned database sleeps after six idle hours and wakes on connection. Paid databases are never archived for inactivity. Shared hosting is US East, without customer-selected regional placement, and we do not advertise shared-plan HA or an uptime SLA. Bandwidth is unmetered, with a 25 GB per 24-hour Solo acceptable-use allowance, not unlimited sustained transfer.

If both databases must stay awake, price the next plan before buying. If you already use Layerbase Free, the upgrade keeps your existing database; there is no migration to another host.

Heroku: convenient if your app already lives there

Essential-0 is a real $5 option with 1 GB storage and 20 connections. The trade-off is operational: Heroku documents unannounced maintenance, no fork/follow support, and 99.5% expected monthly uptime on Essential. Enable and verify scheduled logical backups rather than assuming an untouched plan includes the schedule you want. See the Essential limitations.

What always-on costs on usage meters

Neon: about $19 at the minimum, not automatically $77

On Launch, disable scale-to-zero and hold the compute at 0.25 CU:

0.25 CU × 730 hours = 182.5 CU-hours
182.5 × $0.106 = $19.345, or about $19.35 in compute
1 GB stored × $0.35 = $0.35 in storage
Illustrative total: $19.70, before restore history and other usage

At 1 CU, the same calculation is 1 × 730 × $0.106 = $77.38 before storage. Both numbers are valid; only the first describes the smallest always-on compute. Autoscaling above 0.25 CU changes the result. These are estimates using Neon's published rates, not bills from a workload benchmark.

Our earlier free-tier comparison used the 1 CU example. It now shows both sizes. If branching and elastic compute matter, paying for Neon can be the sensible choice; see the longer Neon comparison before deciding to migrate.

Railway: the $5 credit is included, not added twice

Railway lists RAM at $10/GB-month, CPU at $20/vCPU-month, volumes at $0.15/GB-month and egress separately. For illustration, assume average usage of 0.5 GB RAM, 0.05 vCPU and 1 GB of volume:

Memory:  0.5 × $10   = $5.00
CPU:     0.05 × $20  = $1.00
Volume:  1 × $0.15   = $0.15
Usage subtotal:       $6.15
Hobby bill: max($5 minimum, $6.15 usage) = $6.15, before egress

Those resource levels are assumptions, not measured Postgres consumption. The credit also covers other services in the account, so an app can use it before the database does. Check Railway's current rates and Serverless behavior. A dollar cap and uninterrupted availability are different goals if exceeding the cap stops a service.

Regions and the next rung matter before you migrate

A cheap database across an ocean from your application can be an expensive latency decision. Check the database tier's locations, not just the provider's general region map. Aiven Developer offers broad geographic areas; Layerbase shared plans run in US East; PlanetScale's quoted catalog is an AWS US East example, with regional variations. For the other providers, confirm that the desired size is offered alongside your application before paying.

Plan the first upgrade too:

DigitalOcean changed the upgrade path. Starting October 15, 2026 for new accounts, new Standard Edition clusters cannot add standby or read-only nodes. The restriction reaches all accounts creating new clusters on November 30; existing clusters retain their capabilities. Its pricing page lists Advanced Edition starting at $130 per primary, with matching standby pricing, but get a quote for the eligible configuration rather than assuming the old $60 HA total. Read the plan-change notice alongside the price list.

Aiven's production upgrade also needs a cloud/region-specific quote. A higher price alone does not prove that PITR, HA and an SLA all come with the selected tier.

Backups, PITR and HA solve different problems

Daily backups help after data loss, but can leave work since the last backup outside the restore point. PITR adds a recovery timeline, useful when a migration deleted the wrong rows ten minutes ago. HA adds another node and failover to reduce interruption when a machine fails. It does not undo a bad transaction replicated to the standby.

This is why Render's included three-day PITR can matter more than a larger RAM number, and why a second database on the same shared plan is not a standby. Ask how a restore works, whether it creates a new endpoint, and whether you have tested the procedure. Our Postgres PITR explainer covers that distinction.

MySQL: the same question, fewer low-cost choices

Aiven Developer offers MySQL for $5 with the same 1 GB RAM / 8 GB storage outline and development-tier positioning. Do not transfer its PostgreSQL connection limit to MySQL without checking the engine-specific setting.

DigitalOcean MySQL starts at $15 for a 1 GiB Standard single node. Its October/November edition restrictions also apply to MySQL. PlanetScale's $5 offer is Postgres: the Vitess catalog starts at $39 for the listed three-node PS-10 configuration, before additional charges.

On Layerbase, MySQL requires Solo or higher. Pinning MySQL consumes all 768 MB of Solo's always-on pool. You can have a second database, but cannot keep it pinned alongside that MySQL on Solo. Pro is $15/mo and is the option to consider when both need to stay awake, subject to their combined pool draws.

MariaDB is available on Free and speaks the MySQL wire protocol. It is not identical to MySQL: check compatibility, especially SQL behavior, authentication and version-specific features. A smaller bill is not a reason to skip migration testing.

See MySQL hosting and the Solo limits, or start with MariaDB if it fits your application.

Other cheap options: separate the product from the headline

These are worth knowing about, but should not be quietly mixed into a table of ready-to-buy, pinned Postgres instances:

That $3 MVP-DB listing is one reason we do not call Layerbase “the cheapest.” Lower prices exist; the questions about recoverability, isolation, support and regions still apply.

Where Layerbase fits, and where I would choose something else

Layerbase Solo fits a small application whose main Postgres must stay awake, whose development copy may sleep, and whose app can run near US East. It gives you a flat price and a straightforward Free-to-Solo upgrade. Compare the plan limits.

I would choose something else in these cases:

• You require HA or a contractual uptime guarantee. Evaluate PlanetScale's HA topology or another provider's qualifying managed HA tier. Check the SLA separately; HA and an SLA are not synonyms.
• You need a named UK, EU or other non-US region. Layerbase shared hosting is US East. Eliminate it if that fails your placement requirement.
• You want auth, object storage, realtime and functions as one platform. Supabase Pro buys more than a database. Replacing only Postgres does not replace those services; see the Supabase comparison.
• You need autoscaling or many independent preview branches. Evaluate Neon against your workload instead of forcing it into a fixed-size comparison.
• You need a compliance commitment. Get the required documentation and contract from the provider; this entry-price table is not a compliance assessment.
• Your application uses SQLite at the edge. Evaluate Turso or D1 before converting it to Postgres.

For a switch, check the migration guide for your source. We have paths for Supabase, Neon, Render, Railway, PlanetScale, Aiven and DigitalOcean. Test the copy and plan the write cutover before changing your application's connection string.

Frequently asked questions

What's the cheapest way to keep a Postgres database always on?

Several managed options start at $5; MVP-DB advertises $3 with a different shared-service model. For an app you are launching, compare the complete bill and recovery terms. A $5 compute line with storage billed separately is not the same offer as a $5 flat plan.

Is $5 Postgres good enough for production?

Sometimes, for a small workload that can tolerate the tier's maintenance and recovery behavior. PlanetScale explicitly supports non-HA production use on single-node; Aiven says Developer is not for production. Test your workload and restores, and choose HA when your outage tolerance requires it.

Can I stop Neon Free from sleeping?

Neon Free cannot disable its five-minute scale-to-zero policy. Launch can; the smallest compute held awake for 730 hours is about $19.35 before storage and other usage. Keeping it awake through requests is not the same as purchasing an always-on guarantee.

Does Supabase Pro pause?

Paid projects are not subject to Free's inactivity-pausing rule. That does not eliminate maintenance or outages. Supabase's pausing documentation distinguishes the plans.

What happens to my second database on Solo?

Solo includes two databases but promises one always-on database. The one you do not pin sleeps after six idle hours and wakes on connection; it is not archived for inactivity. A pinned MySQL uses the entire Solo pool.

What is the cheapest always-on MySQL hosting here?

Aiven Developer and Layerbase Solo both start at $5. Aiven's development-tier restrictions still apply; Layerbase MySQL consumes the entire Solo pool when pinned. MariaDB can be another option if its differences from MySQL suit your app.

Do I need HA if I already have backups?

Backups help recover data; HA helps keep service available through a node failure. Neither replaces the other. Decide how long your app can be unavailable and how much recent data it can afford to lose, then test the plan against those requirements.

Are these prices based on a cold-start benchmark?

No. They are dated vendor prices, documented behavior and explicit billing calculations. We have not run a current, comparable connect-through-first-query benchmark across these hosts. If cold-start latency matters, test a new client connection after the relevant idle period, include retries and failures, and record region and database size. For Layerbase MySQL/MariaDB, allow a 30-second connection timeout and retries when unpinned; pinning avoids idle wake-up altogether.

Start with the database that has to stay awake

Choose the cheapest plan whose omitted feature you can actually live without. For one small database and a development copy in US East, start free on Layerbase, then pin one Postgres always-on for $5 when you launch.

Price review: October 20, 2026. Recheck the linked plan pages before purchase, especially Aiven's conflicting connection documentation and DigitalOcean's edition transition.