Until everything starts running local inference. A real Siri that can operate your phone for you, and actually do things like process cross-app conditions ("Hey Siri, if I get an email from my wife today, notify me, then block out my calendar for the afternoon.") would use those increased compute and memory resources easily.
Apple has been shipping "neural" processors for a while now, and when software with local inference starts landing, Apple hardware will be a natural place for it. They'll get to say "Your data, on your device, working for you; no subscription or API key needed."
That's a very big maybe. The LLM experience locally is currently very very different from the hosted models most people play with. The future is still very uncertain.
Apple has been shipping "neural" processors for a while now, and when software with local inference starts landing, Apple hardware will be a natural place for it. They'll get to say "Your data, on your device, working for you; no subscription or API key needed."