You’re right.
That’s economically inefficient, but apparently seems to be the only way to create models that compete with the bigs.
IMHO this repo will die within few years, and that’s both a pity and a waste of public money.
It may or may not die within few years (I'm placing my bets on the optimistic side), if it delivers value today, and the alternative is a cost prohibitive walled garden unsuitable for sensitive data, then it's well worth it cost in public money already.
Do you any idea of how much collaborative suites cost to government every year? 10's if not 100's of millions! They have millions of public servents. The investement to build Docs is a drop of water in the the ocean in comparison.
