The massive data required for machine learning presents an opportunity to create a moat around those businesses. Releasing the algo but not the data doesn't help competitors much. There's also often a first-mover advantage in data collection: operating a service can give you a unique opportunity to collect relevant data, making it very hard for competitors to catch up. The canonical example is Google search: a competitor can scrape the web (at great cost), but they can't get Google's result click data until they're competitive with Google. And without the data it's damn hard to be competitive.