Seconding that. Several ‘AI’ startups I worked with, do not require a massive amount of data to create MVPs/first releases.

The do however invest early in sufficient data science functions to have a good liaison between the researchers and data acquisition.