Data is still a moat. I understand the point, that it's not a big differentiator among foundation or demonstration models. But most nontrivial commercial ML app

Data is still a moat. I understand the point, that it's not a big differentiator among foundation or demonstration models. But most nontrivial commercial ML applications are not going to work with public datasets. What you can scrape off the web is just not representative.