There are already free versions that are useful or rather ones you could run locally.

Almost all boring jobs can be automated with this technology (will still take human developers to do this and years, if not decades to complete) and indeed, almost all people have boring jobs.

The question is whether that actually will happen, because the elite doesn't want to have a few billion people sitting at home plotting against them. McDonald's restaurants could have been fully automated a decade ago, but it didn't happen.

I don't think it's hard (compared to designing an Airbus A380 from scratch) to build a ChatGPT clone, as evidenced by every Big Tech building their own version of it. There is no moat, as Google engineers explained. There is no secret sauce (because it all leaked).

Everyone that wants to pay me a ten million dollars per year (a tiny amount of money when billions are thrown around) can hire me to design their ChatGPT clone. I doubt anyone is willing to give that to me, which shows the market price for this knowledge is near zero (rounding by a million dollars); the value right now is in the temporary scarcity of the AI-hardware, but that too will pass. I think this has commodity written all over it. Perhaps someone has crucial patents, but I can't think of anything, because all the hardware techniques also apply to other silicon (or other existing hardware architectures). That's also why every Big Tech is designing their own AI-accelerators; it's relatively easy.

I also think many of the people interested in AI hardware do not result into sales immediately, because even if I know a shit ton about technology, I didn't shell out $22,000 for a high-end GPU. The demand for AI hardware is just in cloud companies that are already running certain computations for clients and just want to do more of those cheaper. Perhaps the clients might at some point consider that they can afford to compute more things, and so on, until there is no economical benefit anymore (if Netflix can improve recommendations for movies from 95% accuracy to 95.3% for double the cost, they just won't do it). In our current economy there already are a lot of automated systems; the AI approach is not necessarily better.