A lot of people already have 12GB+ GPUs lying around for playing games, doing video editing, etc. I would not get a GPU or mac just to run LLMs personally, but if one wants to get such a device for other tasks too, it may make sense to eg choose a slightly higher (v)RAM variant if they want to run some bigger models. Then what you pay for the local llms is just the difference.