My last post leaned hard on the DB-Engines ranking, and I admitted that ranking measures popularity, not installed base or revenue. Several of you called me on that, politely and otherwise. Fair enough.

So this time I went looking for evidence that does not come from one ranking site. Usage surveys: where the money is going, what cloud vendors are building, and what AI tools reach for when nobody tells them what to use. My conclusion has not changed, but I think the case is much stronger than I made it out to be.

Beyond DB-Engines

The best usage number I can find comes from developers themselves. In the 2025 Stack Overflow Developer Survey, 55.6% of all respondents said they use PostgreSQL, up from 48.7% in 2024. Among professional developers it was 58.2%, roughly 18 points ahead of MySQL (summary of the survey). It also topped the admired and desired lists again.

That is still a survey, not an install count. But it is people saying "I use this," not people searching for it, and that is a different signal from the one DB-Engines measures.

Three caveats on my own napkin math from last time:

  • Straight lines lie. I projected the MySQL and Oracle crossovers assuming this year's pace continues indefinitely. Adoption usually follows an S-curve, so 2028 and 2029 are best read as "if nothing changes," which is never true.
  • The scores are relative. Some of the decline at the top is the whole relational category losing attention to Snowflake, Databricks, and DuckDB. Not every lost Oracle point is a PostgreSQL gain.
  • Scales matter. Jeremy Schneider pointed out in the comments that the log-scale chart flattens the story. He is right. On a linear scale the gap between PostgreSQL and the others closes visibly faster.

Follow the money

Last time I wrote that big data and AI vendors have been buying PostgreSQL companies, and I did not name them. Here they are.

Date

Buyer

PostgreSQL company

Reported price

Oct 2025

Investors (Accel, Peak XV)

Supabase, $100M Series E

$5B valuation (Orrick)

Jun 2025

Snowflake

Crunchy Data

about $250M, unconfirmed (TechTarget)

May 2025

Databricks

Neon

about $1B (TechTarget)

Supabase was valued at $2B in April 2025 and $5B six months later. Snowflake and Databricks are the two companies most often named as the threat to the relational leaders, and both decided they needed PostgreSQL in house.

The more telling detail is what the incumbents sell. Every one of PostgreSQL's top-four rivals has an owner that also sells PostgreSQL:

  • Oracle runs OCI Database with PostgreSQL, and Oracle's own MySQL blog has published a post on why you should move PostgreSQL to OCI.
  • Microsoft announced Azure HorizonDB, a new cloud-native PostgreSQL engine, at Ignite 2025, alongside Azure Database for PostgreSQL. Microsoft is also one of the larger corporate contributors to PostgreSQL itself.
  • Amazon, which owns no top-four database but runs the biggest cloud, sells RDS, Aurora and the distributed Aurora DSQL, all PostgreSQL-compatible.

When the companies you are supposedly beating are building products on you, you are doing something right.

The protocol may be winning bigger than the database

Here is something DB-Engines cannot see: many databases that are not PostgreSQL now pretend to be PostgreSQL. CockroachDB, YugabyteDB, Google AlloyDB, the PostgreSQL interface on Google Spanner, Aurora DSQL, Materialize and RisingWave all speak the PostgreSQL wire protocol and some flavor of its SQL dialect.

That matters because it means psql, JDBC, and every PostgreSQL driver, ORM, and GUI tool already work with them. (Yes, that includes DBeaver.) A new database vendor in 2026 picks PostgreSQL compatibility for the same reason a new laptop picks USB-C. It is where the users and the tools already are.

In the 1990s, "SQL" became the common interface for relational databases. PostgreSQL's dialect and protocol are becoming the common interface for this generation. That is a form of winning that never shows up as a PostgreSQL score.

AI agents already made their choice

I said last time that pgvector makes PostgreSQL a reasonable default for vector search. That undersells it. The bigger story is who is choosing the database.

When Databricks bought Neon, it said more than 80% of the databases on Neon were being created by AI agents, not people (TechTarget). The "vibe coding" app builders mostly reach for Supabase, which is PostgreSQL underneath. Ask a coding assistant to build you an app with a database and see what it picks.

That is a pipeline. Every one of those prototypes that turns into a real product brings PostgreSQL with it, and nobody had a meeting about it.

And for those who say PostgreSQL is fine until you get big: in January 2026, OpenAI described running ChatGPT's core database on a single-primary Azure PostgreSQL server, with nearly 50 read replicas across regions, serving 800 million users and millions of queries per second (VentureBeat). It took a lot of careful engineering, but "not big enough" is a hard argument to make after that.

Answering my own objections

Last time I listed the obstacles in PostgreSQL's way. Several of them have better answers than I gave.

"The community moves slowly." The core does, and on purpose. But the extension API means a lot of innovation never has to wait for a core release. pgvector, TimescaleDB, Citus, PostGIS, ParadeDB and pg_duckdb all ship on their own schedules. A slow core with a fast edge is a pretty good design.

"The installed base of Oracle and SQL Server code is huge." True, but the bridges keep getting better. Babelfish lets PostgreSQL understand T-SQL and the SQL Server wire protocol. Ora2pg and orafce handle much of the Oracle grunt work, and cloud vendors give away migration tools because they want the workload. None of it is magic, but rewriting PL/SQL is a smaller job than it was five years ago.

"A free Oracle or SQL Server would remove the cost argument." It would remove the easy one. But look at what happened to "free" and "open" elsewhere. MongoDB moved to the SSPL, Elastic and Redis changed licenses, and HashiCorp went to the BSL. Each was the owner's right, and each surprised its users. PostgreSQL cannot do that, because nobody owns it. The license is short, permissive and not going anywhere. That is a feature you only appreciate after you have been burned.

"The PostgreSQL 19 reverts look unpredictable." I said a CIO would see unpredictability. Let me argue the other side. The project shipped on a schedule with fewer features rather than shipping features it did not trust. Most of us have lived through a vendor release that went out on the sales calendar with known bugs. And if AI tools are finding more bugs, the code being shipped is getting more scrutiny, not less.

One more I missed entirely: a lot of PostgreSQL never shows up under its own name. Several Chinese national databases, including openGauss, KingbaseES and PolarDB for PostgreSQL, are built from PostgreSQL code. Data-sovereignty concerns in Europe push the same way. Those users rarely search for "PostgreSQL," so the rankings undercount them.

The other side of the ledger

To be fair to the rest of the field:

  • MySQL is everywhere. Decades of LAMP stacks, WordPress sites and SaaS back ends do not migrate because of a survey. MySQL's installed base will outlast many of us.
  • SQLite is the most deployed database on the planet. It is in your phone, your browser and probably your car. "Winning" depends on what you count.
  • Analytics is moving somewhere else. Lakehouse platforms are taking the big analytical workloads that used to live in Oracle and SQL Server. PostgreSQL is winning the operational database fight, which is not the whole data fight.
  • Oracle and SQL Server still own the most demanding enterprise systems. RAC, Exadata and the deep SQL Server tooling are real, as Ludovico Caldara reminded us in the comments.

So, is it winning?

Yes, and by more than one ranking shows. Developers say they use it more than anything else. The money is buying PostgreSQL companies. The competitors' owners sell it. New databases copy its protocol. And the AI agents writing tomorrow's apps pick it without being asked.

What I said last time still holds: popularity is not dominance, and the hard part is the installed base. But the trend is not one site's search-volume math. It shows up everywhere I looked.

I will check back after PostgreSQL 19 ships on November 19 and the December DB-Engines numbers come out. If you have data I missed, the comments are open.

Sources