best 12 managed postgres providers

crunchy publishes a worked example matching our anchor almost exactly, which nobody else does: standard-8 at $140 plus 100gb of storage at $10, for $150 flat. pgbouncer is bundled free, backup storage is included in the base price with no separate line, and there is explicitly no charge for networking or egress. against supabase's $100-a-month pitr add-on and aiven's tier-gated recovery, that inclusiveness is worth real money.

billing is by the minute and prorated to the second, with 42 regions across aws, azure and gcp — the broadest multi-cloud footprint here after aiven.

the caveat is ownership. crunchy data was acquired by snowflake in june 2025 for a reported $250 million and is being folded into 'snowflake postgres'. the product runs unchanged today, and an enterprise data-cloud parent is a different long-term proposition for an indie developer than an independent postgres company was. there's also no free tier and no scale-to-zero, so intermittent workloads pay full freight.

aiven runs the same managed postgres across aws, google cloud, azure, digitalocean, ovh and upcloud, which no other vendor here matches. the startup-8 plan bundles exactly our anchor spec — 2 cpu, 8gb, 100gb ssd — at a flat $150 with all networking included, so the bill is the plan price and nothing else.

there's a real free tier too: a dedicated 1 cpu, 1gb vm with 1gb storage, though aiven reserves the right to shut down unused free services.

the recovery model is the frustration. the startup tier gets a single backup and no point-in-time recovery at all; getting pitr means stepping up to business at $180 or more, and it cannot be bought as an add-on. connection pooling is likewise unavailable on the free and developer tiers and costs extra above them, where crunchy bridge simply includes it. for a $150 production database, having to change plan to get pitr is an odd gate.

$127.57 all-in makes rds the cheapest genuine 2 vcpu / 8gb instance here, and reserved instances take it lower still. it also has by far the broadest region coverage — 30-plus globally — which for data-residency requirements is often the deciding factor rather than price.

two costs sit outside that number and both are ones production systems tend to need. connection pooling isn't included: rds proxy is a separate service billed per vcpu-hour of the underlying instance. and multi-az, the standby that makes it highly available, roughly doubles the compute line to around $232.

the egress allowance has a subtlety worth catching: the first 100gb a month is free at the aws account level, shared across every service you run, not per database. a busy s3 bucket can consume it before your database sends a byte. backups are included up to your provisioned storage size, with pitr inside the retention window at no extra charge — better than supabase's paid add-on.

supabase is postgres plus auth, object storage and edge functions in one platform, which for a small team replaces several vendors. at the anchor it comes to $146.50, competitive with crunchy and aiven, and the 250gb of included egress means our 100gb costs nothing where neon and aws would meter it.

the free tier is the most generous in this ranking — 500mb database, 1gb file storage, 5gb egress, two active projects — with the caveat that free projects auto-pause after a week idle.

two structural points about the price. compute is an add-on stacked on a mandatory $25 subscription rather than a standalone rate, so the real cost is easy to undercount from the pricing page. and point-in-time recovery costs $100 a month for each seven days of retention, never bundled — a database needing 28 days of pitr pays $400 a month for that alone, more than doubling the total. crunchy bridge and aws include it.

our anchor is deliberately hostile to neon: a database running 24 hours a day is the one case where scale-to-zero earns nothing, and $189.76 is the most expensive real instance here except planetscale's oversized tier. suspended instead, the same database costs a little over its $35 of storage.

for development and preview environments that's transformative, and neon's branching is the cheapest here — ten branches free on launch, twenty-five on scale, then $1.50 a branch-month. monthly minimums were removed in december 2025, so it's pure usage-based with no floor. pitr is included via storage history rather than sold as an add-on.

two things to plan around. neon was acquired by databricks in 2025 and the roadmap direction under a data-and-ai company isn't yet clear. and azure regions are deprecated for new projects with existing azure projects required to migrate by 5 october 2026 — a hard, dated deadline that anyone currently on azure needs in their calendar.

at $98.08 xata is the cheapest verified figure here, roughly a third below crunchy and aiven, with storage billed by the minute rather than the hour and idle branches suspending compute entirely on every tier. copy-on-write branching shares deltas to keep branch storage cheap, which suits the ephemeral-environment pattern well.

it also rebuilt itself: the original product was a spreadsheet-style database with search, and it is now positioned as vanilla postgres with branching, with its core open-sourced in april 2026 — a self-hostable escape hatch none of the commercial rivals here offer.

the gap is disclosure, and it's material. egress pricing, backup charges and connection pooling are all absent from the pricing page, so the $98 figure is the floor rather than the bill, and three of the levers that move a real postgres invoice are simply unstated. there's no persistent free tier either — a 14-day trial for the cloud product, with the open-source version free forever.

railway bills cpu, memory and volume storage per second at published rates, so you pay for exactly what you provision with no instance-size rounding. at our anchor that's $142.49, mid-field, and egress is a flat $0.05 a gigabyte — the cheapest metered rate in this ranking.

the structural point is that railway has no postgres-specific tier at all. the database is a container running at railway's generic compute rates, which is why the metering is so clean.

it's also why there's no postgres story to evaluate. connection pooling, point-in-time recovery and backup pricing are all undocumented for databases specifically, and autosuspend behaviour for postgres isn't confirmed. compared with crunchy bridge including pooling and backups in $150, you're getting compute and providing the rest yourself. the free plan is $1 a month of credit after a 30-day $5 trial, which is not a free database.

the $297 figure needs the asterisk stated clearly: planetscale's skus track ram, and there is no 2 vcpu / 8gb option, so the anchor forces ps-160 at 16gb. you're paying for double the memory rather than being overcharged for the same thing, and a workload that actually wants 16gb sees a very different comparison.

the tooling is the draw. branching and backup behaviour carry over from planetscale's vitess heritage and are more mature than most here, each branch getting backup storage equal to twice its disk size at no charge. billing is prorated to the millisecond, the finest granularity in this ranking, across 20-plus regions on aws and gcp.

the free tier is gone — confirmed removed, with the smallest paid database now $5 a month. and storage and egress rates vary by cloud and region, from $0.06 to $0.137 a gigabyte, which makes like-for-like comparison across providers harder than it needs to be. a local pgbouncer is free; dedicated pooler instances run $18 to $551 a month.

prisma postgres charges per database operation rather than per provisioned hour, so there is no instance running up a bill while nothing happens. for a low-traffic application that inverts the economics of everything else here, and the free tier — 100,000 operations, 500mb, fifty databases, no card — is a real starting point.

the business plan at $129 includes fifty million operations and exactly 100gb of storage, so our anchor's storage sits inside the allowance. what it can't tell you is what the compute costs, because the anchor specifies hours and prisma bills queries. depending on your query volume the same workload could cost a fraction of aws's $128 or a multiple of it.

connection pooling is included with limits scaling by plan, backups are included with retention from seven to thirty days, and egress gives the first kibibyte or two per query free before charging $0.08 to $0.09 a gibibyte. it ranks here not on weakness but because the comparison this page performs doesn't apply to it.

what tiger does publish is good: no charges for data transfer, ingest or egress at all, and backups included with no separate fee. point-in-time recovery is tier-gated at three days on performance and fourteen on scale, but it isn't an add-on purchase.

the differentiator is the postgres extension work it's built on — hypertables, continuous aggregates and the time-series tooling that made timescale worth using — layered on vanilla postgres rather than replacing it.

the rate card doesn't exist publicly. tier minimums of $30 and $36 a month are stated with no per-instance pricing behind them, and the actual figures live behind account creation and an in-product calculator. we tried the pricing page, the docs and third-party trackers and could not price our anchor. the company also rebranded from timescale to tiger data, so older references and comparisons use a name the site no longer uses.

the free postgres instance is hard-deleted thirty days after creation rather than suspended. supabase pauses free projects after a week of inactivity and neon suspends compute while keeping your data; render removes the database. anyone treating it as a persistent free tier the way the others work will lose whatever is in it.

the paid product looks reasonable — tiers named by ram from $6 to $6,200 a month, billing prorated to the second, storage overage at $0.30 a gigabyte, and tight integration with render's application hosting.

we could not price it. the pricing table renders through javascript and static fetches returned only navigation, so the exact cost of the 2 vcpu / 8gb tier is unverified, along with egress rates, backup charges, pooling and regions. high availability requires the pro or accelerated tier and doubles the cost.

nile's premise is tenant isolation built into postgres itself, with per-tenant billing granularity, which for a b2b saas is genuinely different from bolting row-level security onto a normal database. the free tier gives fifty million query tokens a month, 1gb of storage and no card requirement, and there are no pauses because there's no instance to suspend.

pricing is in query tokens, so our anchor doesn't convert. the scale plan at $350 a month includes five hundred million tokens and 50gb of storage, and reaching 100gb would add roughly $75 to $150 more — but there is no honest way to translate 730 hours of provisioned compute into a token count, so we haven't invented one.

it's also the least finished product here. provisioned compute up to 8 vcpu is explicitly listed as coming soon, branching is coming soon, and multi-region placement is coming soon. the company is small — around thirty-one people with no funding announced since a january 2024 seed round — which for a database you're building a business on is a consideration in itself.