"Local LLMs" sounds expensive compared to a 20$ subscription. You'd have to pay for years of usage upfront by purchasing those GPUs.