This all would be due to optimisations within model inference code and techniques, hardware and packaging of software like the above.
Don't see billion dollar valuations for lots of AI startups out there to materialise into anything.
Why? It's much more efficient to have centralized special purpose hardware to run enormous models and then ship the comparatively small result over the internet.
By analogy, you don't have a search engine running on your phone right?
Will not happen any time soon. Consumer hardware can't even run GPT-4 locally, and won't be able for a looong time. Each GPT-4 instance runs on 8 A100. The cost of such system is ~$81K. Not even in the ballpark of what most consumers can afford.
But in a few years we might be able to have LLMs running on our phones that work just as well if not better. Of couse as you mention the LLMs running on large servers might still be much more powerfull, but the local ones might be powerfull enough.
Privacy, security, latency, offline availability, access to local data and services running on the device, just to name a few.
I think that the models will evolve and grow as more powerful compute/hardware comes out.
You may be able to run scaled down n versions of what state of the art now, but by then the giant models will have grown in size and in required compute.
The 6 year old models will be retro computingish.
Somewhat like how you can play 6 year old games on a new powerful PC but by then the new huge games will no longer play well on your Old mach
That's why I think these private companies will have the best AIs for many decades.