Why are fine-tuned models on proprietary data the real moat, not another foundation model?

Question

Accepted Answer

_The same transformer everyone has becomes a defensible asset only when trained on data nobody else holds._

> Cyriac uses the same transformer that is the basis for AI generally, but he trains it on data from his own wet lab that nobody else has, to get a fine-tuned model that can answer questions nobody else can. The future right now is really fine-tuned models trained on specific data nobody else has. There is a race on data and a race on fine-tuned models, and the same underlying idea is happening in every industry across the globe.

Source: [Tomorrow's Medicine, with Cyriac Roeding (eCornell Keynote)](https://ecornell.cornell.edu/keynotes/view/K052126/)

Q&A, data moat

Why are fine-tuned models on proprietary data the real moat, not another foundation model?