Right now the closest we have to that is running ampere clusters. I'm saying that because it is going to be some years before any phone GPU/CPU is going to be able to effectively run a decent AI model. I don't doubt there will be some sort of marketing for 'boosting' AI via your phone CPU/GPU but it isn't going to do much.

It is far more likely that it will still continue to be offloaded to the cloud. There is going to be much more market motivation to continue to put your data on the cloud instead of off of it.