Apple probably asked and paid for it. Isn't there potential in accelerating LLMs and transformers on smartphones? Most of what they are good at fits the smartphone use case well.