Apple knows that Siri has been behind for many years. It's built on 2 generations old methods. If you read the AI papers that Apple is publishing, you'd better understand their approach. Their research is leading edge.
Rather than releasing another "me too" LLM tool, where they would be behind the others, they are working on a whole new compute paradigm. Highly risky but if it works, will redefine personal computing.
Take a look at their new APIs. VisionOS and Spatial functions are designed to build a 3D model of your favourite locations, Home, work, cottage, car etc. This is the setting that will generate location context for your meta data. Have you seen spatial images on Vision Pro? - stunning. Lidar is included in all of their high end devices for a reason.
Apple is pushing the boundaries for edge computing models that will run on local devices - specifically phones. Combined with meta data (private) derived from every interaction within your domain, Apple has the potential to deliver magic.
When the edge devices don't have the compute power, the private servers will process your data and send the results back to your model. They can't do everything locally.
Their vision is bold and they are inventing solutions to bring about this new paradigm. They'll keep inventing until they deliver a worth while product. I'd rather they wait to get it right before release.