Managed Postgres pricing · prices checked Oct 18, 2026
The cheapest managed Postgres depends on how many hours a day your database is busy.
We priced 17 managed Postgres services for four workloads, from a side project to a production database that has to survive a data-center outage. A side project runs free on Neon. Keep a small app running all day and Xata is cheapest, at about $75 a month. A bigger app that's busy about 6 hours a day costs about $49 on Xata. Require automatic failover, a week of backups and HIPAA, and Cloud SQL comes out cheapest at $434.
17 services · 4 workloads
Checked Oct 18, 2026
Every row names the plan and links its source
Free to reuse (CC-BY-4.0)
Who's cheapest changes with the workload
Monthly cost of the same four workloads on six services. Large dots mark the cheapest service that meets each workload's requirements; faded dots fail one.
Log scale. A × after a label means that service doesn't meet the Production HA requirements at its listed price.
Why the cheapest option keeps changing
Managed Postgres services charge in one of three ways. Serverless ones (Neon, Xata, Lakebase and Aurora Serverless) bill for the hours the database is busy and little or nothing while it sits idle. Fixed-size services bill the same amount every hour of the month, busy or not. Backend platforms like Supabase charge a plan fee, and the database is one part of what you get.
That's why a side project is nearly free on a serverless service: Neon costs nothing for a database that's busy 3 hours a day, and 12 of the 17 services come in under $20 a month. Keep a database on all day and the advantage reverses. Neon ranks No. 1 for the side project and No. 12 of 17 for an always-on small app, at $160 a month.
For that always-on app, the cheapest is Xata at $75, a few dollars ahead of Ubicloud at $78. Both are newer than the big clouds, though: Xata's managed service came out of a private beta in 2025 and Ubicloud publishes its prices for Germany and says its US prices are comparable. A few dollars a month may not outweigh a longer track record.
Production is where most of the list drops out. Requiring automatic failover, at least seven days of point-in-time recovery and HIPAA eligibility at the listed price rules out 10 of 17 services, including Xata, Ubicloud, TigerData and Supabase. Of those left, Cloud SQL is cheapest at $434, then RDS at $514. Neon qualifies in its own way: it keeps no standby server, so if a data-center zone fails, its database restarts elsewhere, which Neon says takes one to ten minutes. The others keep a standby running and switch over to it.
Find your workload
Pick the workload closest to yours. Prices are monthly, US East, on-demand list prices, checked Oct 18, 2026; data transfer is excluded.
A side project that sits idle most of the day. What's the cheapest place to keep it?
A small production app that's in use around the clock, with no failover requirement.
A bigger app with real traffic for a few hours a day. Does paying only for busy hours save money?
A production database that must survive a data-center (zone) outage, keep at least seven days of point-in-time backups and be HIPAA-eligible on the plan you pay for. Services that can't offer all three at their listed price are shown separately.
| # | Service and plan priced | Per month | Breakdown | What to know |
|---|---|---|---|---|
| 1 | Neoncheapest Databricks · serverlessacq. Databricks 2025 Free plan (0.25 × (1 vCPU + 4 GB) for 3 h/day) neon.com · checked Oct 18, 2026 | $0 | Compute $0 Storage $0 | Compute is billed only while the database is busy, so idle time costs nothing — but an always-on database pays for every hour. An open connection or a polling job keeps the compute awake. |
| 2 | Xata Xata · serverless 1 GB (≤2 vCPU) · 3 h/day xata.io · checked Oct 18, 2026 | $1.36 | Compute $1.08 Storage $0.28 | A newer platform (in production since a May 2025 private beta) that bills each size by the hour it's active and scales to zero, waking in about a second. Cheapest per vCPU-hour in this comparison. |
| 3 | Heroku Postgres Salesforce · flat plansSalesforce Essential-0 elements.heroku.com · checked Oct 18, 2026 | $5.00 | Compute $5.00 Storage $0 | Essential plans share CPU and memory with other customers and cap storage at the plan size (1 GB on Essential-0). |
| 4 | Databricks Lakebase Databricks · serverlessbuilt on Neon (Databricks) Pay as you go: 0.25 × (1 vCPU + 4 GB) for 3 h/day docs.databricks.com · checked Oct 18, 2026 | $5.35 | Compute $5.00 Storage $0.35 | Neon's engine sold inside Databricks; the case for it is having your data platform there already. Rates are Databricks' published 'indicative' prices. |
| 5 | Aurora Serverless v2 (PostgreSQL) Amazon Web Services · serverlessPostgres-compatible Pay as you go: 0.25 × (1 vCPU + 4 GB) for 3 h/day aws.amazon.com · checked Oct 18, 2026 | $5.52 | Compute $5.40 Storage $0.10 I/O (est.) $0.02 | Capacity scales with load and, with auto-pause, down to zero — but each ACU-hour costs more than the same capacity as a fixed instance, so it only pays off when the database is mostly idle. |
| 6 | Amazon RDS for PostgreSQL Amazon Web Services · fixed instance db.t4g.micro · 2 vCPU / 1 GB aws.amazon.com · checked Oct 18, 2026 | $12 | Compute $12 Storage $0.12 | The t4g.micro is a burstable instance with 1 GB of memory — fine for a side project, and it runs (and bills) all month. Add $3.65 a month if it has a public IPv4 address. |
| 7 | Azure Database for PostgreSQL (Flexible Server) Microsoft Azure · fixed instance B1ms (1 vCore, 2 GiB, burstable) azure.microsoft.com · checked Oct 18, 2026 | $13 | Compute $12 Storage $0.13 | The B1ms burstable server is the cheap entry point; it builds up CPU credits while idle and spends them under load. |
| 8 | DigitalOcean Managed Postgres DigitalOcean · flat plans Standard 1 GiB · 1 vCPU / 1 GB docs.digitalocean.com · checked Oct 18, 2026 | $15 | Compute $15 Storage $0 | Simple flat plans from $15 a month. Since October 15, 2026, new clusters above 4 GiB of memory, or with a standby, need the Advanced Edition (from $130); existing accounts move on November 30. |
| 9 | PlanetScale Postgres PlanetScale · flat plans PS-20 · 0.25 vCPU / 2 GB planetscale.com · checked Oct 18, 2026 | $17 | Compute $17 Storage $0 | Flat monthly plans priced on ARM machines (x86 costs 20–25% more), with 8 GB of memory per vCPU. A single-node plan has no replicas; production clusters run three nodes at about three times the price. |
| 10 | Crunchy Bridge Snowflake · fixed instanceacq. Snowflake 2025 Hobby-1 · 2 vCPU / 1 GB crunchydata.com · checked Oct 18, 2026 | $18 | Compute $18 Storage $0.10 | Postgres run by long-time Postgres contributors; Snowflake bought the company in June 2025 and now sells it as Snowflake Postgres too. HA pricing isn't published — we priced it as a second instance of the same size. |
| 11 | Render Postgres Render · flat plans basic-0.5c-1g · 0.5 vCPU / 1 GB render.com · checked Oct 18, 2026 | $19 | Compute $19 Storage $0.30 | Flat instance prices plus $0.30 per GB of storage. Point-in-time recovery depends on the workspace plan: 3 days on the free Hobby workspace, 7 on Pro ($25 a month per workspace, not included here). |
| 12 | Ubicloud PostgreSQL Ubicloud · flat plans hobby-1 · 1 vCPU / 2 GB ubicloud.com · checked Oct 18, 2026 | $20 | Compute $20 Storage $0 | Open-source cloud with low flat prices and storage bundled into each size. Prices are published for Germany; Ubicloud says its US prices are comparable. |
| 13 | Supabase Supabase · backend platform Pro + Micro · 2 vCPU / 1 GB supabase.com · checked Oct 18, 2026 | $25 | Compute $25 Storage $0 | The free plan caps the database at 500 MB, so a 1 GB project needs Pro ($25). Free projects also pause after a week without activity. |
| 14 | Google Cloud SQL (PostgreSQL) Google Cloud · fixed instance db-g1-small (shared core, 1.7 GB) cloud.google.com · checked Oct 18, 2026 | $26 | Compute $26 Storage $0.17 | The cheapest shape is a shared-core instance (db-g1-small); Google doesn't cover shared-core instances under its SLA. |
| 15 | TigerData (ex-Timescale) TigerData · flat plansformerly Timescale Smallest plan (from $30) tigerdata.com · checked Oct 18, 2026 | $30 | Compute $13 Storage $0.09 | Postgres with the TimescaleDB extension for time-series and analytics. TigerData doesn't publish per-size prices (plans start at $30), so these totals are our estimate — treat them as rough. |
| 16 | Fly.io Managed Postgres Fly.io · flat plans Basic (shared-2x, 1 GB) docs.fly.io · checked Oct 18, 2026 | $38 | Compute $38 Storage $0.28 | Every plan includes HA, so the entry price is higher than single-node plans elsewhere. The jump from 2 to 4 performance CPUs is large ($282 to $962). |
| 17 | Amazon Aurora (PostgreSQL) Amazon Web Services · fixed instancePostgres-compatible db.t4g.medium · 2 vCPU / 4 GB aws.amazon.com · checked Oct 18, 2026 | $53 | Compute $53 Storage $0.10 I/O (est.) $0.02 | The smallest Aurora instance (t4g.medium) is a burstable class meant for light use; Aurora Serverless v2 is the cheaper Aurora option for an idle database. |
| 1 | Xatacheapest Xata · serverless 2 vCPU / 8 GB · 24 h/day xata.io · checked Oct 18, 2026 | $75 | Compute $69 Storage $5.60 | A newer platform (in production since a May 2025 private beta) that bills each size by the hour it's active and scales to zero, waking in about a second. Cheapest per vCPU-hour in this comparison. |
| 2 | Ubicloud PostgreSQL Ubicloud · flat plans standard-2 · 2 vCPU / 8 GB ubicloud.com · checked Oct 18, 2026 | $78 | Compute $78 Storage $0 | Open-source cloud with low flat prices and storage bundled into each size. Prices are published for Germany; Ubicloud says its US prices are comparable. |
| 3 | PlanetScale Postgres PlanetScale · flat plans PS-160 · 2 vCPU / 16 GB planetscale.com · checked Oct 18, 2026 | $97 | Compute $96 Storage $1.25 | Flat monthly plans priced on ARM machines (x86 costs 20–25% more), with 8 GB of memory per vCPU. A single-node plan has no replicas; production clusters run three nodes at about three times the price. |
| 4 | TigerData (ex-Timescale) TigerData · flat plansformerly Timescale 2 vCPU / 8 GB tigerdata.com · checked Oct 18, 2026 | $102 | Compute $100 Storage $1.80 | Postgres with the TimescaleDB extension for time-series and analytics. TigerData doesn't publish per-size prices (plans start at $30), so these totals are our estimate — treat them as rough. |
| 5 | Google Cloud SQL (PostgreSQL) Google Cloud · fixed instance 2 vCPU / 8 GB cloud.google.com · checked Oct 18, 2026 | $103 $51 with a commitment | Compute $100 Storage $3.40 | Plain Postgres priced by the vCPU and gigabyte, with no idle discount — it bills all month. HA keeps a full standby, which doubles compute and storage. One- and three-year commitments cut compute by 25% and 52%. |
| 6 | Render Postgres Render · flat plans 2c-8g · 2 vCPU / 8 GB render.com · checked Oct 18, 2026 | $106 | Compute $100 Storage $6.00 | Flat instance prices plus $0.30 per GB of storage. Point-in-time recovery depends on the workspace plan: 3 days on the free Hobby workspace, 7 on Pro ($25 a month per workspace, not included here). |
| 7 | Amazon RDS for PostgreSQL Amazon Web Services · fixed instance db.m7g.large · 2 vCPU / 8 GB aws.amazon.com · checked Oct 18, 2026 | $125 $67 with a commitment | Compute $123 Storage $2.30 | Plain Postgres on AWS with no surprises in the engine. Multi-AZ doubles both the instance and the storage bill. Backups up to the size of your database are free; each public IPv4 address adds $3.65 a month. |
| 8 | Supabase Supabase · backend platform Pro + Large · 2 vCPU / 8 GB supabase.com · checked Oct 18, 2026 | $127 | Compute $125 Storage $1.50 | You're paying for a backend — auth, file storage, realtime — with Postgres inside. The Pro plan is $25 and includes $10 of compute; bigger instances are add-ons. |
| 9 | DigitalOcean Managed Postgres DigitalOcean · flat plans Advanced 8 GiB · 2 vCPU / 8 GB docs.digitalocean.com · checked Oct 18, 2026 | $132 | Compute $130 Storage $2.30 | Simple flat plans from $15 a month. Since October 15, 2026, new clusters above 4 GiB of memory, or with a standby, need the Advanced Edition (from $130); existing accounts move on November 30. |
| 10 | Azure Database for PostgreSQL (Flexible Server) Microsoft Azure · fixed instance 2 vCPU / 8 GB azure.microsoft.com · checked Oct 18, 2026 | $133 $55 with a commitment | Compute $130 Storage $2.60 | General Purpose servers price at about $65 per vCore a month; zone-redundant HA runs a second server and storage copy. Three-year reserved capacity cuts compute by up to 60%. |
| 11 | Crunchy Bridge Snowflake · fixed instanceacq. Snowflake 2025 Standard-8 · 2 vCPU / 8 GB crunchydata.com · checked Oct 18, 2026 | $142 | Compute $140 Storage $2.00 | Postgres run by long-time Postgres contributors; Snowflake bought the company in June 2025 and now sells it as Snowflake Postgres too. HA pricing isn't published — we priced it as a second instance of the same size. |
| 12 | Neon Databricks · serverlessacq. Databricks 2025 Pay as you go: 2 × (1 vCPU + 4 GB) for 24 h/day neon.com · checked Oct 18, 2026 | $160 | Compute $153 Storage $7.00 | Compute is billed only while the database is busy, so idle time costs nothing — but an always-on database pays for every hour. An open connection or a polling job keeps the compute awake. |
| 13 | Heroku Postgres Salesforce · flat plansSalesforce Standard-2 · 8 GB elements.heroku.com · checked Oct 18, 2026 | $200 | Compute $200 Storage $0 | Plans are sized by memory, with storage bundled up to a cap; Heroku doesn't publish vCPU counts, so we matched on memory alone. HA starts at the Premium tier. |
| 14 | Amazon Aurora (PostgreSQL) Amazon Web Services · fixed instancePostgres-compatible db.r7g.large · 2 vCPU / 16 GB aws.amazon.com · checked Oct 18, 2026 | $204 | Compute $201 Storage $2.00 I/O (est.) $0.40 | Aurora's instance classes come with 8 GB of memory per vCPU, so a 4-vCPU need buys 32 GB whether you want it or not. Reads and writes are billed per request on the Standard tier; I/O-Optimized drops that charge for a higher instance price. Each public IPv4 address adds $3.65 a month. |
| 15 | Fly.io Managed Postgres Fly.io · flat plans Launch (performance-2x, 8 GB) docs.fly.io · checked Oct 18, 2026 | $288 | Compute $282 Storage $5.60 | Every plan includes HA, so the entry price is higher than single-node plans elsewhere. The jump from 2 to 4 performance CPUs is large ($282 to $962). |
| 16 | Databricks Lakebase Databricks · serverlessbuilt on Neon (Databricks) Pay as you go: 2 × (1 vCPU + 4 GB) for 24 h/day docs.databricks.com · checked Oct 18, 2026 | $327 | Compute $320 Storage $7.00 | Neon's engine sold inside Databricks; the case for it is having your data platform there already. Rates are Databricks' published 'indicative' prices. |
| 17 | Aurora Serverless v2 (PostgreSQL) Amazon Web Services · serverlessPostgres-compatible Pay as you go: 2 × (1 vCPU + 4 GB) for 24 h/day aws.amazon.com · checked Oct 18, 2026 | $348 | Compute $346 Storage $2.00 I/O (est.) $0.40 | Capacity scales with load and, with auto-pause, down to zero — but each ACU-hour costs more than the same capacity as a fixed instance, so it only pays off when the database is mostly idle. |
| 1 | Xatacheapest Xata · serverless 4 vCPU / 16 GB · 6 h/day xata.io · checked Oct 18, 2026 | $49 | Compute $35 Storage $14 | A newer platform (in production since a May 2025 private beta) that bills each size by the hour it's active and scales to zero, waking in about a second. Cheapest per vCPU-hour in this comparison. |
| 2 | Neon Databricks · serverlessacq. Databricks 2025 Pay as you go: 4 × (1 vCPU + 4 GB) for 6 h/day neon.com · checked Oct 18, 2026 | $94 | Compute $76 Storage $18 | Compute is billed only while the database is busy, so idle time costs nothing — but an always-on database pays for every hour. An open connection or a polling job keeps the compute awake. |
| 3 | Ubicloud PostgreSQL Ubicloud · flat plans standard-4 · 4 vCPU / 16 GB ubicloud.com · checked Oct 18, 2026 | $156 | Compute $156 Storage $0 | Open-source cloud with low flat prices and storage bundled into each size. Prices are published for Germany; Ubicloud says its US prices are comparable. |
| 4 | Databricks Lakebase Databricks · serverlessbuilt on Neon (Databricks) Pay as you go: 4 × (1 vCPU + 4 GB) for 6 h/day docs.databricks.com · checked Oct 18, 2026 | $177 | Compute $160 Storage $18 | Neon's engine sold inside Databricks; the case for it is having your data platform there already. Rates are Databricks' published 'indicative' prices. |
| 5 | Aurora Serverless v2 (PostgreSQL) Amazon Web Services · serverlessPostgres-compatible Pay as you go: 4 × (1 vCPU + 4 GB) for 6 h/day aws.amazon.com · checked Oct 18, 2026 | $179 | Compute $173 Storage $5.00 I/O (est.) $1.00 | Capacity scales with load and, with auto-pause, down to zero — but each ACU-hour costs more than the same capacity as a fixed instance, so it only pays off when the database is mostly idle. |
| 6 | PlanetScale Postgres PlanetScale · flat plans PS-320 · 4 vCPU / 32 GB planetscale.com · checked Oct 18, 2026 | $195 | Compute $190 Storage $5.00 | Flat monthly plans priced on ARM machines (x86 costs 20–25% more), with 8 GB of memory per vCPU. A single-node plan has no replicas; production clusters run three nodes at about three times the price. |
| 7 | TigerData (ex-Timescale) TigerData · flat plansformerly Timescale 4 vCPU / 16 GB tigerdata.com · checked Oct 18, 2026 | $205 | Compute $200 Storage $4.50 | Postgres with the TimescaleDB extension for time-series and analytics. TigerData doesn't publish per-size prices (plans start at $30), so these totals are our estimate — treat them as rough. |
| 8 | Google Cloud SQL (PostgreSQL) Google Cloud · fixed instance 4 vCPU / 16 GB cloud.google.com · checked Oct 18, 2026 | $209 $105 with a commitment | Compute $200 Storage $8.50 | Plain Postgres priced by the vCPU and gigabyte, with no idle discount — it bills all month. HA keeps a full standby, which doubles compute and storage. One- and three-year commitments cut compute by 25% and 52%. |
| 9 | Render Postgres Render · flat plans 4c-16g · 4 vCPU / 16 GB render.com · checked Oct 18, 2026 | $215 | Compute $200 Storage $15 | Flat instance prices plus $0.30 per GB of storage. Point-in-time recovery depends on the workspace plan: 3 days on the free Hobby workspace, 7 on Pro ($25 a month per workspace, not included here). |
| 10 | Supabase Supabase · backend platform Pro + XL · 4 vCPU / 16 GB supabase.com · checked Oct 18, 2026 | $230 | Compute $225 Storage $5.25 | You're paying for a backend — auth, file storage, realtime — with Postgres inside. The Pro plan is $25 and includes $10 of compute; bigger instances are add-ons. |
| 11 | Amazon RDS for PostgreSQL Amazon Web Services · fixed instance db.m7g.xlarge · 4 vCPU / 16 GB aws.amazon.com · checked Oct 18, 2026 | $251 $136 with a commitment | Compute $245 Storage $5.75 | Plain Postgres on AWS with no surprises in the engine. Multi-AZ doubles both the instance and the storage bill. Backups up to the size of your database are free; each public IPv4 address adds $3.65 a month. |
| 12 | DigitalOcean Managed Postgres DigitalOcean · flat plans Advanced 16 GiB · 4 vCPU / 16 GB docs.digitalocean.com · checked Oct 18, 2026 | $266 | Compute $260 Storage $5.75 | Simple flat plans from $15 a month. Since October 15, 2026, new clusters above 4 GiB of memory, or with a standby, need the Advanced Edition (from $130); existing accounts move on November 30. |
| 13 | Azure Database for PostgreSQL (Flexible Server) Microsoft Azure · fixed instance 4 vCPU / 16 GB azure.microsoft.com · checked Oct 18, 2026 | $267 $111 with a commitment | Compute $260 Storage $6.50 | General Purpose servers price at about $65 per vCore a month; zone-redundant HA runs a second server and storage copy. Three-year reserved capacity cuts compute by up to 60%. |
| 14 | Crunchy Bridge Snowflake · fixed instanceacq. Snowflake 2025 Standard-16 · 4 vCPU / 16 GB crunchydata.com · checked Oct 18, 2026 | $285 | Compute $280 Storage $5.00 | Postgres run by long-time Postgres contributors; Snowflake bought the company in June 2025 and now sells it as Snowflake Postgres too. HA pricing isn't published — we priced it as a second instance of the same size. |
| 15 | Heroku Postgres Salesforce · flat plansSalesforce Standard-3 · 15 GB elements.heroku.com · checked Oct 18, 2026 | $400 | Compute $400 Storage $0 | Plans are sized by memory, with storage bundled up to a cap; Heroku doesn't publish vCPU counts, so we matched on memory alone. HA starts at the Premium tier. |
| 16 | Amazon Aurora (PostgreSQL) Amazon Web Services · fixed instancePostgres-compatible db.r7g.xlarge · 4 vCPU / 32 GB aws.amazon.com · checked Oct 18, 2026 | $410 | Compute $404 Storage $5.00 I/O (est.) $1.00 | Aurora's instance classes come with 8 GB of memory per vCPU, so a 4-vCPU need buys 32 GB whether you want it or not. Reads and writes are billed per request on the Standard tier; I/O-Optimized drops that charge for a higher instance price. Each public IPv4 address adds $3.65 a month. |
| 17 | Fly.io Managed Postgres Fly.io · flat plans Scale (performance-4x, 32 GB) docs.fly.io · checked Oct 18, 2026 | $976 | Compute $962 Storage $14 | Every plan includes HA, so the entry price is higher than single-node plans elsewhere. The jump from 2 to 4 performance CPUs is large ($282 to $962). |
| 1 | Google Cloud SQL (PostgreSQL)cheapest Google Cloud · fixed instance 4 vCPU / 16 GB + standby cloud.google.com · checked Oct 18, 2026 | $434 $217 with a commitment | Compute $200 Storage $17 Standby $217 | Plain Postgres priced by the vCPU and gigabyte, with no idle discount — it bills all month. HA keeps a full standby, which doubles compute and storage. One- and three-year commitments cut compute by 25% and 52%. If a zone fails: Automatic: the regional standby takes over. |
| 2 | Amazon RDS for PostgreSQL Amazon Web Services · fixed instance db.m7g.xlarge · 4 vCPU / 16 GB · + standby aws.amazon.com · checked Oct 18, 2026 | $514 $278 with a commitment | Compute $245 Storage $12 Standby $257 | Plain Postgres on AWS with no surprises in the engine. Multi-AZ doubles both the instance and the storage bill. Backups up to the size of your database are free; each public IPv4 address adds $3.65 a month. If a zone fails: Automatic: the standby in another zone takes over (Multi-AZ). |
| 3 | Azure Database for PostgreSQL (Flexible Server) Microsoft Azure · fixed instance 4 vCPU / 16 GB + standby azure.microsoft.com · checked Oct 18, 2026 | $546 $226 with a commitment | Compute $260 Storage $13 Standby $273 | General Purpose servers price at about $65 per vCore a month; zone-redundant HA runs a second server and storage copy. Three-year reserved capacity cuts compute by up to 60%. If a zone fails: Automatic: the zone-redundant standby takes over. |
| 4 | Neon Databricks · serverlessacq. Databricks 2025 Scale plan: 4 × (1 vCPU + 4 GB) for 24 h/day · HA by restart (no standby) neon.com · checked Oct 18, 2026 | $674 | Compute $639 Storage $35 Standby none | HA costs nothing extra because there's no standby to pay for; the trade-off is the restart time below. HIPAA requires the Scale plan, which bills compute at $0.222 an hour instead of $0.106. Keeping a restore window adds $0.20 per GB-month of changed data, which isn't included here. If a zone fails: No standby: storage is replicated across 3 zones and the compute is restarted in a healthy zone — 1–10 minutes after a zone failure (Neon docs). |
| 5 | Amazon Aurora (PostgreSQL) Amazon Web Services · fixed instancePostgres-compatible db.r7g.xlarge · 4 vCPU / 32 GB · + standby aws.amazon.com · checked Oct 18, 2026 | $819 | Compute $404 Storage $10 Standby $404 I/O (est.) $2.00 | Aurora's instance classes come with 8 GB of memory per vCPU, so a 4-vCPU need buys 32 GB whether you want it or not. Reads and writes are billed per request on the Standard tier; I/O-Optimized drops that charge for a higher instance price. Each public IPv4 address adds $3.65 a month. If a zone fails: Automatic: a reader in another zone is promoted. |
| 6 | Render Postgres Render · flat plans 4c-16g · 4 vCPU / 16 GB · + standby · Scale workspace $499 + 20% HIPAA surcharge render.com · checked Oct 18, 2026 | $1,051 | Compute $200 Storage $30 Standby $230 Compliance $591 | HIPAA needs the Scale workspace ($499 a month) and adds 20% to every charge; both are included in this total. If a zone fails: Automatic: the standby takes over a few seconds after 30 seconds of primary downtime (Render docs). |
| 7 | Aurora Serverless v2 (PostgreSQL) Amazon Web Services · serverlessPostgres-compatible Pay as you go: 4 × (1 vCPU + 4 GB) for 24 h/day + standby aws.amazon.com · checked Oct 18, 2026 | $1,394 | Compute $691 Storage $10 Standby $691 I/O (est.) $2.00 | A failover-ready reader follows the writer's size, so HA doubles the compute bill. If a zone fails: Automatic: a reader in another zone, sized to follow the writer, is promoted. |
| Doesn't meet this workload's requirements at the listed price | ||||
| Xata Xata · serverless 4 vCPU / 16 GB · 24 h/day · no HA plan xata.io · checked Oct 18, 2026 | $166 | Compute $138 Storage $28 Standby — | Doesn't qualify: HA not offered at list price; PITR window not published; HIPAA not at list price HA isn't listed on the managed pricing page, and a BAA is offered only with bring-your-own-cloud deployments. If a zone fails: Not published for the managed service. | |
| Ubicloud PostgreSQL Ubicloud · flat plans standard-4 · 4 vCPU / 16 GB · + standby ubicloud.com · checked Oct 18, 2026 | $312 | Compute $156 Storage $0 Standby $156 | Doesn't qualify: no HIPAA Ubicloud has no HIPAA offering. If a zone fails: Automatic: one or two standbys; recovery usually takes a few minutes (Ubicloud docs). | |
| TigerData (ex-Timescale) TigerData · flat plansformerly Timescale 4 vCPU / 16 GB + standby tigerdata.com · checked Oct 18, 2026 | $418 | Compute $200 Storage $9.00 Standby $209 | Doesn't qualify: HIPAA not at list price HIPAA is available only on the Enterprise plan, at a price TigerData doesn't publish. If a zone fails: Automatic: an HA replica takes over (time not published). | |
| Supabase Supabase · backend platform Pro + XL · 4 vCPU / 16 GB · + standby supabase.com · checked Oct 18, 2026 | $459 | Compute $225 Storage $12 Standby $223 | Doesn't qualify: no automatic failover; PITR add-on only; HIPAA not at list price Point-in-time recovery is a paid add-on ($100 a month per 7 days), HIPAA needs the Team plan ($599) plus an add-on, and Pro has no automatic failover. If a zone fails: No automatic failover on Pro: read replicas serve reads but don't take over. | |
| DigitalOcean Managed Postgres DigitalOcean · flat plans Advanced 16 GiB · 4 vCPU / 16 GB · + standby docs.digitalocean.com · checked Oct 18, 2026 | $543 | Compute $260 Storage $12 Standby $272 | Doesn't qualify: no HIPAA DigitalOcean's managed databases are not on its list of HIPAA-covered products. If a zone fails: Automatic: a standby node takes over (Advanced Edition). | |
| Crunchy Bridge Snowflake · fixed instanceacq. Snowflake 2025 Standard-16 · 4 vCPU / 16 GB · + standby crunchydata.com · checked Oct 18, 2026 | $580 | Compute $280 Storage $10 Standby $290 | Doesn't qualify: PITR window not published Point-in-time restore is included, but Crunchy doesn't publish how many days it covers, so it can't be counted as meeting the seven-day requirement. If a zone fails: Automatic: a standby replica takes over (time not published). | |
| PlanetScale Postgres PlanetScale · flat plans PS-320 · 4 vCPU / 32 GB · + standby planetscale.com · checked Oct 18, 2026 | $606 | Compute $190 Storage $11 Standby $405 | Doesn't qualify: HIPAA not at list price A BAA for HIPAA is only available with Business support, Enterprise, or startup pricing — not at list price. If a zone fails: Automatic: one of two replicas in other zones takes over. | |
| Heroku Postgres Salesforce · flat plansSalesforce Standard-3 · 15 GB · + standby elements.heroku.com · checked Oct 18, 2026 | $750 | Compute $400 Storage $0 Standby $350 | Doesn't qualify: HIPAA not at list price Priced on Premium-3 (HA included, 15 GB of memory, just under the 16 GB target). HIPAA needs a Shield plan, which only verified Heroku Teams and Enterprise customers can buy. If a zone fails: Automatic: a synchronized standby takes over (Premium and above). | |
| Fly.io Managed Postgres Fly.io · flat plans Scale (performance-4x, 32 GB) · HA included docs.fly.io · checked Oct 18, 2026 | $990 | Compute $962 Storage $28 Standby included | Doesn't qualify: PITR window not published; HIPAA not at list price Fly.io doesn't publish a point-in-time recovery window or HIPAA terms for Managed Postgres. If a zone fails: Automatic: included on every plan (node count and timing not published). | |
| Databricks Lakebase Databricks · serverlessbuilt on Neon (Databricks) Pay as you go: 4 × (1 vCPU + 4 GB) for 24 h/day + standby docs.databricks.com · checked Oct 18, 2026 | $1,314 | Compute $639 Storage $35 Standby $639 | Doesn't qualify: HIPAA not at list price HA keeps one to three secondaries running in other zones and turns off scale-to-zero, so the standby bills like a second database. HIPAA requires Databricks' compliance security profile, an enterprise add-on. If a zone fails: Automatic: a running secondary in another zone is promoted; apps reconnect within seconds (Databricks docs). | |
Each total is the smallest published plan that fits the workload, with the exact plan named. "With a commitment" shows the vendor's published one- or three-year discount where one exists. How we matched plans, what counts as qualifying, and how sure we are of each number are explained in How we did this.
What developers report
List prices don't show how a bill behaves. These are firsthand accounts and vendor statements, linked so you can read them in full.
“At $250/month for 4 databases without any replicas, we were paying premium prices for subpar reliability.”
Commenters traced much of the bill to autoscaling, and one replied that Neon would be "lowering prices dramatically." The author wrote that they were charged 16 cents per compute-hour; Neon's Launch plan now charges $0.106.
“There's no middle ground for solo builders who've crossed the free cap but don't need enterprise headroom.”
Their app went over the free plan's 5 GB monthly egress allowance and started failing with errors instead of being billed, so they moved to the $25 Pro plan.
“It's weird that pricing is based on CPU but it's never defined.”
The complaint behind this comparison: every service sells compute in its own unit (CU, ACU, plan size), so we convert each one to vCPUs and memory before comparing.
“Up to 40% cost savings for I/O-intensive applications.”
AWS introduced Aurora I/O-Optimized because per-request I/O fees could dominate some bills. Our Aurora rows use the Standard tier with a small I/O estimate; a write-heavy app can pay much more.
Who owns what
Several of these services changed hands in the last year and a half, which can change their roadmaps and prices.
- May 2025Databricks agreed to buy Neon, reportedly for about $1 billion. Its engine also powers Databricks Lakebase. Source
- June 2025Snowflake agreed to buy Crunchy Data, reportedly for about $250 million; it now also sells it as Snowflake Postgres. Source
- June 2025Timescale renamed itself TigerData. The TimescaleDB extension kept its name. Source
- Sept 2025PlanetScale, known for MySQL, made its Postgres service generally available. Source
- Aug 2026Databricks agreed to buy Electric, maker of the PGlite embedded Postgres. Source
How we did this
What we priced
Each service's own published prices for the US East region, on demand, checked Oct 18, 2026. Fixed plans are monthly list prices; services billed by the hour are priced on a 30-day month. Data transfer, backups beyond what's included, and support plans are excluded unless a row says otherwise.
| Workload | Size | Data | Busy | Requirements |
|---|---|---|---|---|
| Hobby / spiky dev | 0.25 vCPU · 1 GB | 1 GB | 3 h/day | None |
| Steady small SaaS | 2 vCPU · 8 GB | 20 GB | 24 h/day | None |
| Bursty app | 4 vCPU · 16 GB | 50 GB | 6 h/day | None |
| Production HA | 4 vCPU · 16 GB | 100 GB | 24 h/day | Automatic failover, ≥7 days point-in-time recovery, HIPAA |
How we matched plans
For services sold in fixed sizes, we priced the smallest published plan with at least the workload's vCPUs and within 10% of its memory. Shared or burstable CPUs count only for workloads busy less than 12 hours a day without failover, because they slow down under sustained load. Heroku doesn't publish vCPU counts, so its plans are matched on memory alone. Serverless services are billed for the busy hours only, converted to the same vCPU-and-memory size.
What counts as qualifying
The Production HA workload requires automatic failover across zones, at least seven days of point-in-time recovery and HIPAA eligibility on the plan priced. If any of the three needs a sales contract, a support upgrade or an unpriced add-on, the service is shown separately rather than ranked. Where HIPAA is available at a published price, as on Neon's Scale plan or Render's Scale workspace, that's the price we used. A recovery window a vendor doesn't publish doesn't count either.
How we check the math
Before anything is published, the model recalculates prices that vendors publish directly, such as RDS's db.m7g.large at $122.64 a month or PlanetScale's three-node PS-320 at $570. If any calculation drifts from the vendor's number, nothing is published. Each row is also tagged with where its price came from: a published price list, vendor rates cross-checked against independent price trackers, or our own estimate where a vendor doesn't publish per-size prices (TigerData, and the HA prices of Crunchy Bridge and Ubicloud).
Sources
| Service | Price source | Checked | Basis | Confidence |
|---|---|---|---|---|
| Amazon Aurora (PostgreSQL) | aws.amazon.com | Oct 18, 2026 | vendor rates, cross-checked | medium |
| Amazon RDS for PostgreSQL | aws.amazon.com | Oct 18, 2026 | vendor rates, cross-checked | medium |
| Aurora Serverless v2 (PostgreSQL) | aws.amazon.com | Oct 18, 2026 | published price list | high |
| Azure Database for PostgreSQL (Flexible Server) | azure.microsoft.com | Oct 18, 2026 | vendor rates, cross-checked | medium |
| Crunchy Bridge | crunchydata.com | Oct 18, 2026 | our estimate | medium |
| Databricks Lakebase | docs.databricks.com | Oct 18, 2026 | published price list | medium |
| DigitalOcean Managed Postgres | docs.digitalocean.com | Oct 18, 2026 | published price list | medium |
| Fly.io Managed Postgres | docs.fly.io | Oct 18, 2026 | published price list | high |
| Google Cloud SQL (PostgreSQL) | cloud.google.com | Oct 18, 2026 | vendor rates, cross-checked | high |
| Heroku Postgres | elements.heroku.com | Oct 18, 2026 | published price list | high |
| Neon | neon.com | Oct 18, 2026 | published price list | high |
| PlanetScale Postgres | planetscale.com | Oct 18, 2026 | published price list | medium |
| Render Postgres | render.com | Oct 18, 2026 | published price list | high |
| Supabase | supabase.com | Oct 18, 2026 | published price list | high |
| TigerData (ex-Timescale) | tigerdata.com | Oct 18, 2026 | our estimate | low |
| Ubicloud PostgreSQL | ubicloud.com | Oct 18, 2026 | published price list | medium |
| Xata | xata.io | Oct 18, 2026 | published price list | medium |
Not included yet
- Aiven for PostgreSQL: its pricing page lists starting prices for one default cloud; we couldn't get per-plan prices for AWS US East.
- Google AlloyDB: we couldn't confirm which of Google's published rate tables applies to US regions.
- Vultr and Akamai (Linode) Managed Databases: we couldn't read plan sizes and prices from their pages in a form we could check.
- Railway: it runs Postgres as a container you operate, without managed failover or point-in-time recovery.
- Snowflake Postgres: it's billed in Snowflake credits, which don't convert cleanly to a monthly price.
- Prisma Postgres: it bills per database operation, so it can't be compared by machine size.
Corrections
Oct 18, 2026
- Aurora Serverless v2 was overstated by about half: we had counted 4 ACUs per 4 GB of memory, but AWS defines an ACU as about 2 GB.
- Aurora (provisioned) was understated for the larger workloads: we had priced a 2-vCPU instance where the workload needs 4. Production HA rose from $436 to $819, ending what had looked like a tie with Cloud SQL.
- Neon's high availability was described as a standby that scales to zero. Neon keeps no standby; its compute restarts in another zone, which takes 1–10 minutes. Each row now states how failover works.
- Databricks Lakebase's standby is now billed: its HA secondaries run continuously.
- Heroku, Lakebase and Crunchy Bridge no longer count as qualifying for Production HA: Heroku's HIPAA plans are limited to Teams and Enterprise customers, Lakebase's need an enterprise security add-on, and Crunchy doesn't publish its point-in-time recovery window.
- Price history from before this date was removed for the rows these corrections changed (both Aurora products, Lakebase's Production HA and every side-project row), so the history shows vendor price changes only.
- The side-project workload was resized from 1 vCPU / 4 GB to 0.25 vCPU / 1 GB, which is what a side project uses. Earlier history for that workload isn't comparable.
- Committed-use discounts we couldn't source were removed. Crunchy Data's acquisition date was corrected to June 2025.
- Six services were added: Amazon RDS, Render, Heroku, Fly.io, Ubicloud and Xata.
Terms
- vCPU
- One virtual processor. Most services pair each vCPU with 4 to 8 GB of memory.
- Serverless / scale to zero
- Compute that shuts off when nobody is connected and starts again, usually within a second or two, on the next connection.
- Burstable
- A cheap instance with a shared or capped CPU that can run fast for short periods and slows down under sustained load.
- High availability (HA)
- Keeping the database running if a data-center zone fails. Most services keep a standby copy running in another zone; Neon instead restarts its compute elsewhere.
- Point-in-time recovery
- Restoring the database to any moment within a window, such as the past seven days.
- HIPAA / BAA
- US health-data rules. A provider signs a business associate agreement (BAA) before you may store patient data with it.
The full dataset, including the rows behind this page, is described in the provenance record. Every API response names its license in an X-License header.
Compare & guides
Reach the buyer's-question pages
Head-to-heads, "cheapest for…" decisions, and per-provider profiles — each one built on this live map, itemized and sourced.
Use this data
Take the numbers with you
The four-workload table is free to reuse under CC-BY-4.0: credit datafloe and link back. A free API key returns the same rows as JSON. The $49/month license adds the price history, re-checks as vendors change their prices, and support.
Free API key
We'll email you a key. It allows 100 calls a day across every dataset.
Commercial license · $49/mo
For teams that build on these numbers: the price history, re-checks when vendors reprice, the keyed API and support. Tell us you're interested and we'll set it up.
Every API response names its license in the X-License header. The open snapshot is CC-BY-4.0; neither license permits AI training (ai_training: false).
Embed this
Put the live digest on your page.
The open (CC BY 4.0) snapshot, glanceable — as an interactive iframe, a paste-anywhere image, or a one-line script. Attribution links back here and is required by the license.
preview
snippet