[LoganDark]

Apple is definitely using custom silicon designed for PCC, as none of their existing chips have the memory capacity or bandwidth to serve LLMs to even a single client performantly, let alone many clients. The speed with which Apple Intelligence worked back when it actually used Apple's own PCC and not Google's cloud was still much higher than could be achieved with anything they sold to customers. Now of course, Siri AI is too big and expensive for even that, but hopefully Google's cloud is just a stopgap...