Maybe, maybe not. They’d decidedly need to make it modular, whether externally or just internally, in a manner similar to how ARA plugins currently, or VST3 plugins, depending on context, work in Cubase now. This could allow having different teams of designers and programmers involved similar to how most truly complex software projects work. Kind of like SpectraLayers, WaveLab, and Dorico now, really – in the first two cases, ARA provides the middleware that allows the connection (not only for SB products, but also for other developers’ products). And whoever the teams are working on these types of add-ons could focus on the specific utility of the add-ons, albeit with some back and forth between the team providing the hooks and the team providing the specific functionality.
The sort of deep AI-based functionality (or at least LLM training) being talked about here would be pretty specialized to get the depth (e.g. specific instrument knowledge) and breadth (e.g. genre knowledge) needed to truly be useful to a broad range of users. But this is a consideration with loop-based virtual instruments, too. For example, if I’m working on a country rock song (probably about half of my most recent album), LLMs that specialize in EDM or heavy metal aren’t likely to be terribly useful to me. Ditto if I’m wanting a fiddle part and the (lack of) utility of an LLM specializing in strummed acoustic guitars.
IMHO, that is really where third-party specialists come into play, but where the right sorts of hooks are needed on the DAW side. The question on Cubase (for me anyway) is really if additional hooks are needed beyond what can already be done. I know some people are using Dreamtonics Synthesizer V (an AI-based vocal synthesizer that has a selection of vocalist models) with current Cubase – I think via a combination of MIDI and ARA). I don’t have a feel for its limitations within Cubase (I’ve read various posts in these forums on the topic, but I haven’t even attempted using it myself), though, no less whether some additional hooks would be needed to provide smoother integration and/or more functionality. I believe there are also some products out there that do something on the instrumental front, though I haven’t checked those out to date, even to have a sense of what specific instruments they can do and what their capabilities and limitations “on paper” are likely to be.
The other consideration on this scope front, though, is that I think it is important for Cubase itself to focus on the core DAW and extensibility hook(s) functionality to serve the widest group of potential users. The more solid those elements are, the easier it is for others, be it at Steinberg or elsewhere, to make the platform even more powerful for those who have additional needs.
I’d use the analogy that, though Cubase undoubtedly provides a wide range of, both generally useful and somewhat specialist, plugins there are very few that I’ve even tried. Why? Because I only started to come to Cubase in late 2018, making it my main DAW in mid-2020, so I already had my favorite third-party plugins I’d come to in the over decade and a half I was using Cakewalk SONAR. For example, I had no real interest in Groove Agent because I’d already been using Superior Drummer and EZ Drummer, with a wide variety of add-ons to support the genres I work in, and the sequencing and programming potential within those has a lot more capability for the specific need of creating believable drum parts than I’d be able to easily do with generic MIDI and Groove Agent. I don’t see any limitations in this, even today in the face of whatever potential AI might add on this front. For example, I can already specify a general rhythm I’m looking for and have SD3 find potentially matching grooves, which I can use as a starting point and tailor further without having to describe to AI what I want (e.g. “push the snare on 2 and 4” or “add some ghost notes in the third verse”).
The old model that comes to mind on just generating tracks is Band-in-a-Box with its RealTracks. I used to use that to experiment with genres for song demos at one point, but, when deciding what software to migrate to my newly built Windows 11 computer, I’m not even sure if I’ll bring BiaB over since I haven’t actually used it in years (my version of BiaB is 2019, and I don’t think I’ve done much with it since upgrading to that version). That is an area where I could see something like SUNO, or DAW plug-in-based functionality, potentially being useful for some. Even there, though, unless someone is just writing in one style, a wide breadth of genres will be needed to be useful to a wide audience. And, of course, not everyone even has a need for this.
Also, there’s plenty of advantages to using samples over humans. One just needs to know where to look.