It's not really just startups - honestly I think 99% of those startups will fail.
The problem with generative AI as a business is basically what that leaked Google memo said, which was something along the lines of: "OpenAI doesn't really have a moat, but neither do we."
There are a few large companies who are creating these successful models (OpenAI, Microsoft, Google, Meta, perhaps Amazon with their Anthropic investment), but there are tons of other startups that are trying to come up with "individual products around AI", and I think it will be incredibly hard for them to differentiate. That is, for a lot of the "dedicated" early-stage products I've seen, I've thought "I can pretty much do this all with ChatGPT, maybe ChatGPT with plugins - why would I want to download your special app?"
I don't think all of these companies are doomed to fail (for example, I think Harvey AI was smart to focus on a very specific, profitable niche: lawyers have a lot of money to spend on this stuff, and they have specific needs e.g. around compliance, auditability, privacy, etc. that you could see a specific legal-focused AI tool addressing), but I think a lot of them are.