Because local AI models run well on a GPU, better than on a CPU