Aiven vs Vercel Postgres: a decision guide
Short version: pick Aiven if the database has to sit inside your own network perimeter, in a cloud and region you name, with an audit trail your buyer will ask for. Pick Postgres through the Vercel Marketplace if your app already lives on Vercel and you want the database wired to the project with no procurement step. The thing to know before either conversation gets going is that "Vercel Postgres" stopped being a Vercel product in December 2024, so what you are comparing is Aiven against a Neon project that Vercel resells and bills.
I run Layerbase, which hosts managed Postgres, so I have a stake in this comparison. It stands without me: most of this post is about two other companies, and I have tried to be exact about where each one beats what I sell.
Aiven or Vercel Postgres: which should your team pick?
Pick Aiven if you need VPC peering into your own cloud account, a specific region outside the US, point-in-time recovery with a contractual retention window, a standby node, or a vendor whose compliance paperwork is already written. Aiven holds ISO/IEC 27001:2022, 27017, 27018 and 27701, an ISAE 3000 Type 2 report aligned to the SOC 2 criteria, plus PCI DSS, HIPAA, GDPR and CCPA compliance, all listed on their security and compliance page as of September 2026. That is the answer to a security questionnaire, and it is the single strongest thing on Aiven's side of the table.
Pick the Vercel Marketplace route if your team is small, the app is a Next.js deployment on Vercel, and the highest-value thing a database can do is appear in your environment variables without anyone opening a purchase order. The integration provisions a Neon project, injects the connection variables per environment, and can give each preview deployment its own database branch. It is the shortest distance between a git push and a working database, and the bill lands on your existing Vercel invoice.
Consider neither if your objection to both is the meter. Aiven charges an hourly rate per service, and Neon charges per compute-unit-hour plus per GB-month of storage, so both bills move with usage and neither is predictable in advance. If you want a fixed number, or you need more than Postgres on one account, that is a third shape and I describe it near the end.
What "Vercel Postgres" actually is in September 2026
This is the part teams get wrong, and it derails the debate, because half the room is arguing about a product that no longer exists.
Vercel's own Postgres docs say it directly: "Vercel Postgres is no longer available. If you had an existing Vercel Postgres database, we automatically moved it to Neon in December 2024. For new projects, install a Postgres integration from the Marketplace." Neon's Vercel Postgres transition guide dates the full move across Q4 2024 and Q1 2025.
So when someone on your team says "Vercel Postgres," they mean one of two things:
- An old store that was migrated. It is a Neon project now, reachable by an ordinary
postgresql://string. ThePOSTGRES_*environment variables still work, and the@vercel/postgresclient still works but is no longer maintained by Vercel. Vercel Postgres alternatives walks through what to change. - A new database bought from the Marketplace. Almost always Neon, in one of two modes. Neon's docs call them the Vercel-managed integration, where Vercel creates the Neon organization and the charges go on the Vercel invoice, and the Neon-managed integration, where you keep your own Neon account and Neon bills you.
That distinction matters more than it looks. Under the Vercel-managed integration, Neon runs the database and Vercel is the merchant of record. Neon's own docs warn that deleting the database from the Vercel side "permanently removes the Neon project and all data," so the destructive button lives in the dashboard where your teammates are clicking around for deploy settings.
The prices are Neon's, published on Neon's pricing page, as of September 2026. Free is $0 with 100 CU-hours of compute and 0.5 GB of storage per project, scaling to zero after five minutes of idle. Launch has no monthly minimum and charges $0.106 per CU-hour plus $0.35 per GB-month of storage, with 500 GB of egress included and $0.10 per GB after. Scale is $0.222 per CU-hour on the same storage rate and adds IP allow rules, private networking and compliance options.
A database that never idles bills for every hour in the month. That is the whole character of the model: cheap when your traffic is spiky, unbounded when it is not.
What Aiven is, and where the comparison is unfair to it
Aiven is an open source data platform, and databases are not really the headline. The pricing page sells PostgreSQL, MySQL, Valkey, ClickHouse and OpenSearch next to Apache Kafka, Kafka Connect, MirrorMaker 2, DataHub, Grafana, Metrics and Aiven Runtime, all on one control plane. If a streaming pipeline is why you are looking at Aiven, comparing it to a Vercel Marketplace database is comparing a platform to a checkbox.
The half that does compare is the PostgreSQL ladder. Verified September 2026 from aiven.io/pricing:
| Aiven PostgreSQL plan | Price | What you get |
|---|---|---|
| Free | $0 | 1 GB RAM, 1 GB storage, single node |
| Developer | $5/mo | single node, limited cloud choice |
| Hobbyist | from $12/mo | single node, limited cloud choice |
| Startup | from $75/mo | single node, full cloud and region choice, VPC peering, 2-day PITR |
| Business | from $180/mo | adds a standby node, 14-day PITR |
| Premium | from $270/mo | 30-day PITR |
Read two things off that. High availability starts at Business, so everything under $180 a month is one node. And the features most teams are actually shopping for on Aiven, meaning VPC peering in AWS, GCP or Azure, point-in-time recovery, and the choice of AWS, DigitalOcean, Google Cloud, Azure, OVH or UpCloud across regions like Frankfurt, Singapore and Sydney, all begin at Startup. The $5 Developer plan is not the plan your security review is going to accept.
Aiven's free tier is real, not a trial. Their free tier page says there is no expiry and no credit card, backups included, one free service per service type: a PostgreSQL with 1 GB of RAM and 1 GB of storage, a MySQL the same, a Valkey with 1 GB of RAM, plus a small Kafka and an OpenSearch. Aiven says plainly it is not aimed at production workloads. I wrote more about Aiven's model in Aiven alternatives.
Side by side
| Aiven for PostgreSQL | Postgres via Vercel Marketplace | |
|---|---|---|
| Who operates it | Aiven | Neon |
| Who bills you | Aiven | Vercel, or Neon if you use the Neon-managed mode |
| Entry price | $0 free tier, $5/mo Developer | $0 Neon Free, then usage |
| Pricing shape | Hourly per service, per plan rung | Per CU-hour plus per GB-month |
| Scale to zero | No, plans run continuously | Yes, after 5 minutes on Free and Launch |
| Region choice | Six clouds, many regions, from Startup | Neon's regions, chosen at install |
| VPC peering | AWS, GCP, Azure, from Startup | Private networking on Neon Scale |
| Standby node | From Business, $180/mo | Neon's storage architecture, not a plan toggle |
| PITR | 2 / 14 / 30 days by plan | Neon restore windows by plan |
| Other engines on the same account | MySQL, Valkey, ClickHouse, OpenSearch, Kafka | No, separate Marketplace products |
| Preview-branch automation | No | Yes, a branch per preview deployment |
The row that decides it for most teams is the last two. Aiven gives you a platform and makes you do the wiring. The Marketplace gives you wiring and makes you buy each further piece as a separate product.
Questions to ask in a proof of concept
A PoC that only proves "we can connect and run a query" proves nothing. Both of these will pass that. Here is what I would actually put on the list, and it works just as well if the shortlist is PlanetScale or Convex instead.
Connection limits and pooling from serverless functions. This is the one that breaks in production and not in the PoC, because a PoC has one developer and production has a hundred concurrent function invocations each opening a socket. Ask for the connection ceiling on the exact plan you would buy, not the top plan. Ask whether the pooler is transaction mode or session mode, because transaction pooling quietly breaks prepared statements, LISTEN/NOTIFY and session-level SET. Then run a load test that opens more connections than the ceiling and watch what the error looks like. If you are querying over HTTP from the edge instead, test that path specifically: it has different limits.
Cold start and scale-to-zero behaviour. Neon scales to zero after five minutes on Free and Launch, which is how the bill stays small, and it is also a latency cliff for the first request afterwards. Measure it. Measure it on your slowest endpoint, not a health check. Aiven does not scale to zero on paid plans, so you pay for idle and never pay the wake cost. Decide which of those two you would rather explain to a user, then confirm whether the behaviour is configurable and at which price.
Region co-location with your compute. Put the database in the same region as the functions, then measure round-trip latency from inside the deployment rather than from your laptop. A chatty ORM turning one page render into twelve queries will make a 60 ms cross-region hop into three quarters of a second. Ask what happens if you later need a second region, and get a real answer rather than "it's on the roadmap."
Backups, PITR, and an actual restore drill. Do not accept a retention number as an answer. Restore a backup into a scratch database during the PoC and time it end to end. That tells you the recovery point objective and the recovery time objective, which are the only two numbers your incident review will care about. On Aiven, note that PITR starts at the $75 Startup rung, so check what the cheap plans actually retain.
Egress and storage pricing. Find the per-GB egress rate and the included allowance, then multiply by a realistic month. Neon's Launch plan includes 500 GB per project and charges $0.10 per GB after. Analytics dashboards and anything that pulls large result sets into the app tier are how teams find this line item after the fact. Ask the same about storage growth: does the number cover logical size, or WAL and backups too?
The exit path. Ask for a pg_dump from the vendor's own tooling and restore it somewhere else during the PoC, while you still have negotiating room. Both of these are real Postgres, so the dump should be portable. What to check is whether you have the privileges the dump needs. Aiven's avnadmin user is not a superuser, which is why their own recipe passes --no-owner --no-comments. Also confirm who owns the delete button: under the Vercel-managed integration, removing the database from the Vercel dashboard takes the Neon project and the data with it.
Extensions. Write down the extensions you depend on before the PoC, including the ones your ORM installs quietly, and check each one against the vendor's supported list. pgvector, postgis, pg_cron, timescaledb and pg_partman are the usual sticking points, and support varies by vendor and by version. Then check version support: which major Postgres versions are offered, how long old ones are supported, and whether the upgrade is in-place or a migration you run.
Two more worth adding. What is the maintenance window policy, and can you move it? And what does support cost, because on both platforms the plan that answers your ticket in an hour is not the plan you priced.
Where Layerbase fits, and where it does not
Here is the third shape, and the reason I could write a comparison of two other vendors without being neutral about it.
Aiven's appeal is many engines under one account. The Vercel route's appeal is zero friction from a Vercel project. Layerbase Cloud tries to be both without a meter: 18 database engines on one account under one flat plan, with a Vercel integration that writes your database environment variables per environment and keeps them in sync, including a pooled and a direct variant.
The numbers are the whole pitch, so here they are. Free is $0 with no credit card: 2 databases, 5 GB between them, drawn from 8 engines including PostgreSQL. Solo is $5 a month for 2 databases with a 768 MB always-on pool, so the database that has to stay awake stays awake. Pro is $15 a month for up to 10 databases, 25 GB, 30-day rolling backups and point-in-time restore on always-on Postgres. If you need more always-on capacity, that is $10/month pool blocks. No CU-hours, no per-GB egress line, no second purchase when the app grows a cache.
Two features from the questionnaire above: 16 of the 18 Cloud engines branch copy-on-write in seconds, which covers the preview-database use case the Vercel integration is prized for, and Postgres over HTTP works from edge runtimes with the same serverless driver a Neon project uses.
Now the part that matters more, because you should not start a migration on an incomplete picture.
Aiven beats us outright on four things. VPC peering and bring-your-own-cloud: Aiven will run inside your account and peer into your network, and we will not. Region and cloud choice: Aiven sells six clouds and regions across three continents, while we run a single United States region and a shared-plan database gets no region choice at all, so if you need Frankfurt or Singapore this is over before it starts. A standby node: Aiven Business has one at $180 a month, and we do not sell a standby on any plan. And compliance: Aiven is a large vendor with the attestations listed above, and we are a small company, so if a certificate has to be on file before signing, ask us first rather than after.
The Vercel route beats us on one thing that is hard to argue with. If you want exactly one database, you want it on the Vercel invoice, and you never want to think about it again, the Marketplace is fewer steps than anything I can offer. Our integration connects a project; it does not put the charge on Vercel's bill.
Where we genuinely fit is the team in between: past the free tier, not near a procurement review, running more than one data store, and tired of a number that changes every month. Neon alternatives covers the Postgres-only version of that argument, and keeping Neon's branching without the compute meter covers the case where branching is the only thing you would miss.
FAQ
Is Vercel Postgres still a thing?
No. Vercel's Postgres docs state that "Vercel Postgres is no longer available," and that existing databases were automatically moved to Neon in December 2024. New projects install a Postgres integration from the Vercel Marketplace instead, which in practice means Neon. If your team is debating "Aiven vs Vercel Postgres," the real comparison is Aiven against a Neon project that Vercel resells.
Who actually bills me for Postgres on Vercel?
It depends which integration mode you install. Under the Vercel-managed integration, Vercel creates the Neon organization for you and the charges appear on your Vercel invoice. Under the Neon-managed integration, you keep your own Neon account and Neon bills you directly. Neon runs the database either way.
Does Aiven still have a free tier?
Yes, as of September 2026. Aiven's free tier page says it is always free with no expiry and no credit card, and includes automated backups. You get one free service per service type, so one PostgreSQL with 1 GB of RAM and 1 GB of storage, one MySQL, one Valkey, plus a small Kafka and an OpenSearch. Aiven says it is not aimed at production workloads.
What is the cheapest Aiven PostgreSQL plan that is production-shaped?
Verified September 2026, the ladder is Free $0, Developer $5/mo, Hobbyist from $12/mo, Startup from $75/mo, Business from $180/mo, Premium from $270/mo. Startup is the first rung with full cloud and region choice, VPC peering and point-in-time recovery, and Business is the first with a standby node. The $5 and $12 rungs are single-node plans with limited cloud choice.
Aiven vs Neon on price: which is cheaper?
Neither, reliably, because they meter different things. Aiven charges an hourly rate per service whether or not anyone connects, so a quiet database costs the same as a busy one. Neon charges per CU-hour of compute plus $0.35 per GB-month of storage and scales to zero after five minutes, so a spiky workload can be much cheaper and a continuously warm one can be much more expensive. Model your own traffic against both rather than comparing headline numbers.
What should I test in a Postgres proof of concept?
Connection ceilings and pooler mode on the plan you would actually buy, cold start after idle on your slowest endpoint, round-trip latency from inside your deployment in the region you would use, a real restore drill timed end to end, egress and storage rates multiplied by a realistic month, a pg_dump you restore somewhere else, and every extension and Postgres version you depend on. The same list works for a PlanetScale or Convex evaluation.
Can I move from either one to somewhere else later?
Both are real Postgres, so yes, and the mechanism is the same: a pg_dump from the source into your new database, or a migration wizard that reads the connection string once. From Aiven, copy the Service URI off the service Overview tab and paste it into the Aiven migration page. From a Vercel Marketplace database, take the non-pooling connection string out of your project's environment variables and use the Vercel Postgres migration page. Neither source is written to.
Wrapping up
The debate your team is having is probably not the one on the agenda. Aiven and the Vercel Marketplace are not competing for the same job: one is a data platform you buy when a security review is in your future, the other is a wiring convenience you install when shipping this week matters more than anything else. Decide which of those two problems you have, and the answer follows.
If the meter is the thing you actually object to, and you want Postgres plus whatever the app grows next under one flat number, create a PostgreSQL database on the free tier and see how far it gets you. No card, and the connection string is the same one your driver already expects.
Keep reading
- Preview environment platforms in 2026: what is in the database when the preview comes upEvery platform on this list will give a pull request its own URL. The question that sorts them is what is in the database behind that URL: nothing, a restore of last night's backup, or the live data as of right now. Here is where each of eleven platforms lands, quoted from their own docs, and what a preview costs while the PR sits open.
- Northflank alternatives: a preview environment is not a database branchNorthflank gives every pull request its own stack, and the database in that stack starts empty unless you seed it or restore it from a backup you already had. Here is exactly how their forks work, what a copy-on-write branch does differently, what each one costs while a PR sits open, and the cases where Northflank is the right answer.
- Layerbase vs Crunchy Bridge: two ways to buy managed PostgresCrunchy Bridge is Postgres specialists selling instance-hours across AWS, Azure and GCP. Layerbase is a flat monthly price, 18 engines and branching. A head-to-head on pricing, versions, extensions, HA, backups and placement, with worked examples.
- Prisma Postgres alternatives: the plan with neither meterPrisma Postgres bills per operation, Neon bills per CU-hour, and Prisma frames the whole decision as a choice between those two meters. There is a third option: a flat plan with no meter at all. Here are the current Prisma numbers, the arithmetic for a chatty app, where operations pricing genuinely wins, and how to move with the direct TCP string.