How do you figure that? Given the speed that the gpt4 API returns data I feel like it's easily going to cost more than pretty pricy hardware.