AI based human mocap is rapidly growing indeed. It's a quite interesting field to keep an eye on. But I haven't seen any real footage of 3D Ai based monocular mocap behaving correctly yet. What we normally get to see are very controlled results coming from researchers or companies offering AI mocap solutions which look really cool, but as the author of the video said, they are complicated to get working for the final user.
A practical drawback of AI based mocap is that you need high processing power. The cloud based companies seem to deliver better results because they are probably using high end processing machines. This solves the processing power problem for the final user but it's often offered as a subscription based model, taking away the freedom of owning and using a mocap suit. And I would also like to know if they are able to stream the mocap results in real time.
As everything in mocap, there's not a single answer of what technology is better. It's always a balance bettween cost, accuracy, flexibility, and the fit to a particular use case. I'm personally interested in applying mocap to live stage performances, and monocular AI based mocap would probably behave really bad there because of all the lighting changes. There might be other fields where AI based mocap will become the norm sometime in the future.
If you try some of the softwares or services listed in the video please share your results here, I think it would be a good contribution to the community to have concrete results of what this new type of technology has to offer.