General thoughts:
How this all potentially evolves depends on what models are going to be trained off. Stable Audio's been pretty strictly trained off of audio libraries that they're happy to be transparent about. Suno on the other hand (more popular for now due to 'better' results), isn't stating what it's been training it's model off of, so everyone suspects they're scraping copyrighted content without getting permissions from anyone to do so. How that pans out from a legal perspective over time will be interesting to see.
For what I generally do, it's very unlikely AI-Toolsets will be able to match up to the kind of niche sub-genres, detailing quirks and complexity of some of the content I've already done, and the content I'm working on in the background right now across the course of this year. For others: very case-by-case basis, but I'd say that those who have pretty unique design skillsets, voices and instrumentation will remain resillient to ever being replaced by this. However, people trying to churn out House, Techno and Corpo-BG Fillter content are already in trouble.
How it's mainly playing out now and in the future:
In theory: It's solid for content-creators to get some background music in a hurry for their vids to avoid copyright situations, and there's a humungous amount of possibilities in the future for most of us to take advantage of SFX-generation.
In reality: I've seen with my own eyes lurking within AI discussion & sharing groups about plenty of individuals selling AI-gen'd content either through shovelling it out for sale the usual online streaming & marketplace methods, or charging close to full-rates on commisions that 'traditional human-made music' would usually charge close into the range of (on a per-minute basis). Then they congratulate each other on their profits, while also sharing 'rage pics' of musicians/composers losing work because of it and just have hundreds/thousands of individuals laughing at it. Basically a bunch of grifters who take a malicious pleasure in scraping a few quick bucks merely as 'prompt engineers' and getting one over the 'artists'. Coincidentally, alot of those users are names I remember seeing mass-generating 2D-Art through the earlier models and selling them all as NFT's before that area of the market took a massive nosedive in it's value and reputation a couple of years ago. They're the ones who defend their actions by regurgitating the same old tired counter-arguments of "define 'art'" and "Use AI to your advantage instead of complaining!" (despite not being able to clarify a single valid point how to do so, then throw ad-homs when called out on it).
If you're unwilling to search around for those kind of FB and Telegram groups to witness it, then all you have to do is a youtube search for stuff like suno.ai demonstrations (pretty easy to spot: the thumbnails have the typical soy-looking suprised/shocked expression from the uploaders) and trail through the comment sections of those for some examples out in the wild.
On a lighter note: there's been some pretty funny meme-content coming out of this, so I'd imagine that area to get even better over time.
However it's also worth remembering before we bash the tech entirely, that a pretty decent chunk of individuals have been taking advantage of 'machine learning' based toolsets for the last decade such as with iZotope's suite of plugins like Ozone/Neutron/Nectar, while other vendors are opening up 'Online AI Mastering' options for pretty affordable prices. I've never found the results on any of those to be 'great', but merely 'ok' if you need to shovel something out in a hurry, with the quality of the results greatly diminishing when you're running extremly complex tracks through such toolsets. Those types of complex projects still require an experienced human with decent surgical tools is still 100% necessary to get the best results (which I can't see as many people doing going forwards, since it requires alot of study, ear-training & practice to get 'better' than the existing ML-based toolsets).
Want to get ahead of the curve and still do well with your human-made content in spite of the situation with AI-gens? Get on the album and portfolio-grinds now, carve yourselves out as the 'real deal' while you still can with your existing composing skills you've had for years/decades already, and ride atop the avalanche of AI-Music slop which is starting to grow and swallow those who can't push themselves above the waves.
Albeit we've also got to contend with non-AI issues still and shouldn't lose sight of, such as the piss-poor earnings from streaming services and other oppressive copyright minefields thanks to how the big labels and lawmakers have behaved for decades.