to me, ML creates an interesting question concerning monopolies when strictly looking at data-results. If I have the best data possible, you can't build a better-competing model. However, forcibly splitting data up among competitors creates worse models, which is not best for consumers.
As for durability, I personally fail to see how google and facebook are not durable monopolies - the network effects are so strong. In google's case it is even more than that - it is almost impossible to have a product of comparable quality without the users and the information about their behaviour, which is impossible to obtain because everyone uses google. Gone are the days when good search rankings could be inferred from the link graph of the web.