Choosing a database in 2026: a consultant's cheat sheet
Supabase, Postgres, ClickHouse - we get asked which one constantly. The honest answer is that they mostly are not competitors, and the decision is easier than it looks once you ask the right question first.
A version of this question comes up on nearly every project: which database should we use? People expect a nuanced, it-depends answer. It mostly is not nuanced. The reason it feels hard is that the options get presented as competitors when they are usually answers to different questions.
Here is the cheat sheet we actually use.
First, the question behind the question
Before naming any database, answer this: is your workload transactional, analytical, or are you just trying to ship fast?
- Transactional — lots of small reads and writes, fetch-this-record, update-that-row, keep it all consistent. This is what an application does all day.
- Analytical — scan huge amounts of data, aggregate it, report on it. Dashboards, metrics, historical analysis.
- Ship-fast — you do not know your scale yet, you have no users yet, and the priority is being live, not being optimal.
Almost every confused database decision is really a failure to answer this question first. Answer it, and the database mostly picks itself.
The cheat sheet
Shipping fast, early stage → Supabase. It gives you Postgres plus auth, storage and APIs, immediately. For an MVP, the value is not the database, it is everything bundled around it that you would otherwise build by hand in the week you do not have. And because it is real Postgres underneath, choosing it now does not corner you later.
Transactional, long-term, you want control → Postgres. Self-managed or managed, on whatever cloud. This is our default for serious long-term builds, and it is the right default. It is reliable, deeply understood, does far more than people remember, and scales further than most projects will ever test. The component holding all your data should be the one you worry about least, and Postgres earns that.
Analytical at scale → ClickHouse, alongside Postgres. Not instead of. When analytical queries — big aggregate scans over lots of history — start straining the same instance serving your application, that contention is the signal. Move the analytical workload to ClickHouse, which is built for exactly that shape, and let Postgres keep doing the transactional job it is excellent at.
That is the whole sheet. Notice that for most projects the answer is "Postgres", possibly wearing a Supabase wrapper, and ClickHouse joins the picture only when a specific, recognisable problem shows up.
The mistakes we see most
Choosing analytical for a transactional workload. Someone reads that a columnar database is "fast" and reaches for it as a primary application database. Then they try to do many small updates against it and have a bad time, because that is close to the worst thing you can ask a columnar store to do. Fast at what is the only version of "fast" that means anything.
Choosing the second database too early. ClickHouse is a real operational commitment — another system to run, another place data lives, a pipeline to keep them in sync. Below the scale where contention is actually biting, a well-indexed Postgres plus a read replica covers analytical queries fine. Adding the second database before you have the problem is just buying overhead.
Choosing exotic over boring. New, specialised databases are interesting and occasionally exactly right. But "interesting" is a cost — less documentation, smaller community, fewer people who have hit your bug before, more unknowns. The bar for picking the exotic option over Postgres should be a specific need Postgres genuinely cannot meet, not a feeling that Postgres is unexciting. Unexciting is a feature here.
Forgetting that more databases means more pipelines. Every database past the first one needs data flowing into it, and that flow is a real component that lags, breaks and drifts. A two-database architecture is not twice the database work — it is the database work plus a pipeline you now have to operate.
What we would tell you
- Ask the workload question first. Transactional, analytical, or ship-fast. It resolves most of the decision.
- Default to Postgres. Wrap it in Supabase when speed is the priority. It will carry you far further than you expect.
- Add ClickHouse when contention forces it, not before, and run it next to Postgres, not instead.
- Make the exotic option earn its place. Boring and well-understood is worth a lot.
The database decision feels weighty because it is hard to reverse. But it gets much lighter once you accept that these tools are mostly not competing — they are each the clear answer to a different question. Figure out which question you are asking. The rest is short.
Working on something similar?
We have probably hit the exact scaling or architectural bottlenecks you are navigating. Book a call and we'll walk through the technical trade-offs.