> Apple Neural Engine is a thing already, with support for multiply-accumulate on INT8 and FP16. AI inference frameworks need to add support for it.

Or, Apple could pay for the engineers to add it.