“Migrating DeepL's custom model architectures off of NVIDIA's platform to alternative hardware is considered difficult due to engineering overhead and the strategic importance of development speed.”