It appears to offer a lot more flexibility in terms of external referencing and custom scene definitions than that of Alembic. If it were to become a truly universal format, or FBX replacement, it seems it would need a better level of animation support rather than model/vertex transform bakes. This doesn't directly suit their needs of course as they have custom methods to pair with their Presto rigs it sounds. I applaud them for continuing to push universal formats. It won't suit my needs but I'm not a direct user by any means at this point in time.
Seems really interesting to me (and our niche ;) ). I don't see that as an FBX contender. I'd rather see it as a scene-version of alembic (while considering alembic as a geometry format). FBX is trying to get original procedural data across as much as possible, and i have yet to see an exchange format that even remotely manages to support a majority of features, let alone plugins or in-house tools.
I really dig the idea of having USD + Alembic at the core of a pipeline.
The only real issue i currently see is that windows support for alembic is REALLY bad. I assume this will be similar for USD. Application support is also "not ideal".
I have not read and viewed everything yet. But from what i saw it seemed that there is an API including everything you need to include support. Also it is not plain ascii, the geometry is binary but still offers an ascii editor which seemed kinda neat. Also it definitely fills a gap for me (Vizualisation, Animation, VFX managing highly complex data sets (mostly automotive)). I can see how it might not really appeal to gamedev tho.