I think it's because there are a lot of people writing articles about the benefits of running local models. I think it's fair to say that there are daily threads on HN singing the praises or local inference. I also see people buying new hardware where the main trigger is ability to run local models.