Not sure if you already know, but - sophisticated large language models can be run locally on basic consumer-grade hardware right now. LM Studio, which I’ve been playing with a bit, is a host for insanely powerful, free, and unrestricted LLMs.

Cloud AI services will always be more capable. You can do more with near-limitless RAM and processing power than you’ll ever be able to do locally on your mobile/desktop system.

But locally-run AI’s usefulness was never in question either.

An example of this is Google who made the Coral TPU and then the Pixel TPU chips specifically in a bid to bring AI to edge and mobile devices. And now they have been developing a version of their Bard AI that is built to run on mobile systems.

There is a huge market for it but there are a lot of challenges to running AI models on lower power hardware that have to be addressed. In its current state most AI on these platforms performed only get specific operations using very condensed models.

I’m sure lots of big companies do, but there are lots of big companies and they all have their own goals. The article goes into it a bit, but lots of companies aren’t exactly leaping to send off their most sensitive data into a third party cloud. They want AI to work with their data, but they want it locally and on-prem.

Plus you also got Intel, AMD, Nvidia, Qualcomm etc who want to sell as much as they can regardless of if it’s going to be a cloud customer or a customer looking to train AI locally.

Being good for the consumer is just a side effect of the blending of money, business paranoia, competition, sensitive data and rapid expansion of an industry