Apple GPT
The announcement
Apple recently announced a major update at its keynote: the Apple Watch Series 9 can now process Siri requests directly on the device, without an internet connection.
Historically, Siri operated exclusively online, sending audio to Apple servers for transcription, natural language processing and response generation. This new approach differs fundamentally by processing everything locally.
Why this announcement matters
Apple’s strategic pivot reflects several advantages:
- Hardware integration: The S9 chip features a 4-core neural engine enabling on-device processing
- Proprietary language model: Apple deployed its own large language model, initially supporting English and Mandarin
- Infrastructure savings: Eliminates dependency on massive data centres like those required by OpenAI
Apple Silicon explained
Introduced in 2020, Apple Silicon processors replace Intel chips in Mac computers. Built on ARM architecture, they prioritise performance efficiency. The “Neural Engine”, a specialised processing unit for machine learning, delivers approximately 11 trillion operations per second while consuming just 39 watts.
Practical applications
These neural processors enable features like Touch ID, Face ID and photo editing without internet dependency. Apple benefits through hardware sales incentives and reduced server maintenance costs.
Vision for the future
Apple’s Vision Pro will leverage similar on-device processing for spatial computing, positioning virtual objects and tracking user movements for enhanced experiences.
This direction from Apple confirms a broader trend: local inference is no longer confined to laboratory research. It is becoming a product reality, with direct implications for data sovereignty and user privacy.