It feels like it's too late. Microsoft, Meta and Google all beat Apple to the punch with highly usable and competitive local inferencing frameworks. Apple could

It feels like it's too late. Microsoft, Meta and Google all beat Apple to the punch with highly usable and competitive local inferencing frameworks. Apple could release a Tensorflow-style library with full-fat CUDA and Linux support tomorrow, but they would just be Another Competing Solution next to ONNX and Pytorch. The hardware side doesn't look much better. Apple's biggest userbase is located on iPhone, which poses hardware constraints on what kinds of model you can realistically deploy. The Mac has a deeply-ingrained audience, but it's unlikely that the AI will be a selling point to commercial customers (who have Nvidia) or PC users (that have other models). Honestly, I believe AI research would be a wasted investment relative to supporting third-party libraries upstream and welcoming Nvidia back onto their platform.