We got custom inference per model per hardware now. How long before we get custom hardware for model? · AI Intel