The Database in the Room — an LLM metaphor, not a forecast

query · a comparison<br>‹ return to log<br>read-only

> run query

SELECT future<br>FROM llm;

1 row returned · 0.00s

NULL

future IS NULL ↓ so here is a metaphor instead

a metaphor, not a forecast

Maybe large language models don't stay a revolution. Maybe they become a database — essential everywhere, invisible the moment they work.

1994

The Database Wars

the most exciting technology in the building

Oracle, Sybase, Informix and IBM traded benchmarks like blows. Pour every record into one engine, ask it anything, watch intelligence fall out. The company that queried best would win.

vendor keynotes drew crowds

"query your whole business"

careers built on the platform

the future, clearly

the loud ones lost · the quiet ones won

2026

Load-Bearing Furniture

underneath everything, exciting to no one

The database won so completely it disappeared. It runs every app you touched today and no one wrote a thread about it. Fundamental — and thoroughly, permanently boring.

a config value in a .env file

chosen once, forgotten

only noticed when it breaks

infrastructure, not news

Swap one word and the same silhouette shows through. This isn't a timeline of what happens next — it's a mirror. Every marvel gets re-read, in the end, as plumbing.

but re-run the standings

The 1994 headliners didn't win. The trophy went to two engines that weren't even on stage.

-- SELECT winner FROM database_wars ORDER BY reality;

engine<br>1994 rank<br>2026 status<br>cost

Oracle<br>#1<br>still here. bloated, load-bearing, quietly resented<br>$$$$ / core

Sybase<br>contender<br>absorbed into SAP → went quiet

Informix<br>contender<br>absorbed into IBM → a footnote

PostgreSQL<br>not present<br>the serious default nobody argues about<br>$0 · open source

SQLite<br>not present<br>inside every phone, browser and app · silent<br>$0 · ~1T deployments

5 rows · winners flagged where deployment ≫ noise

The point isn't that the database won. It's who won. Postgres, by being an open project instead of a vendor. SQLite most completely of all — a library you've never configured, running in the device in your hand. It won by vanishing. Oracle just persists, mostly as an invoice.

Read a model through the same lens and the advice inverts: don't watch the frontier launch. Watch for the SQLite — the one that wins without ever being chosen.

SELECT hype FROM llm<br>WHERE year = 2036;

1 row returned

NULL

Not a prediction — a reminder that fundamental and thrilling are different states, and most technologies pass from the second into the first. The database did. Electricity did. Maybe models are mid-trip. Maybe the metaphor is wrong. No one knows.

end of result set

‹ return to log