You can already run a large LLM (like sonnet 3.5) locally on CPU with 128GB of ram which is <300 USD, but can be offset by swap space. Obviously, response speed is going to be slower, but I can't imagine people will pay much more than 20 USD for waiting 30-60 seconds longer for a response.
And obviously consumer hardware is already being more optimized for running models locally.
But most of the time, we as engineers don't pick the winners. Some C-Suite executive or middle manager, who isn't very technically inclined, picks the winners, and we as engineers are forced to make it work.
As I don't think a engineer has ever had the chance to choose a company's CRM, the CRM with better marketing would always win over the CRM with better engineering.
And obviously consumer hardware is already being more optimized for running models locally.