Are the LLM Wars the Database Wars?

Are the LLM Wars the Database Wars?

Large language models may not stay a revolution. They may become<br>infrastructure, the way the database did, used everywhere and noticed by<br>nobody.

In 1994 the database was the most exciting technology in the building.<br>Oracle, Sybase, Informix and IBM competed on benchmarks. The pitch was that you<br>would put every record your company had into one engine and ask it anything.<br>Keynotes drew crowds. Careers were built on the choice of platform. The company<br>that queried best would win.

Thirty-two years later the database has won and disappeared. It runs every<br>application you used today, and nobody writes about it. It is a line in a<br>configuration file. You pick it once, forget it, and think about it again when<br>it breaks. The same description could fit a model in 2036.

But look at who won. Sybase was bought by SAP and went quiet. Informix was<br>bought by IBM and became a footnote. Oracle is still here, mostly as an invoice.<br>Neither of the engines that actually won was in the 1994 race.<br>PostgreSQL became the default, and it is an open project rather than a product.<br>SQLite runs inside every phone, browser and app, roughly a trillion copies, and<br>you have never configured it.

Nobody chose them. The engines that ended up everywhere were not the ones<br>being argued about.

If models follow the same path, the launch is the wrong thing to watch. Watch<br>for the SQLite, the one that wins without being chosen.

This is not a prediction. Important and exciting are different things, and<br>most technology moves from the second to the first. The database did.<br>Electricity did. Maybe models are partway. Maybe the comparison is wrong. Nobody<br>knows.