I've heard a lot recently about how the moat / defensibility for a particular startup building in AI largely centers around data, with the thinking being that as they grow, they'll leverage their expanding dataset to train their model better, hence reinforcing their superiority.