Apple Inc. (AAPL.US) unveiled its progress in artificial intelligence at its annual Worldwide Developers Conference in Cupertino, California.
The event featured a significant update to Siri, showcasing a redesigned version capable of multi-turn conversations and handling complex tasks like querying event dates, setting reminders, and planning routes.
Core Strategic Distinction
The announcements highlighted a strategic divergence from many of its Silicon Valley peers. Instead of committing billions to massive infrastructure and advanced models, Apple is emphasizing privacy and user convenience as its primary advantages.
Apple's senior vice president of Software Engineering, Craig Federighi, commented on this approach, suggesting some companies are racing ahead with AI for its own sake without focusing enough on the end users.
Infrastructure Collaboration
Company executives confirmed that Apple is collaborating with Alphabet Inc. (GOOGL.US) and NVIDIA Corporation (NVDA.US) to build its most advanced model, the Apple Foundation Model Cloud Pro.
While a partnership with Google was announced earlier this year, this marks the first official confirmation that some Apple Intelligence features will run on NVIDIA hardware.
Apple AI executive Amar Subramanya stated the AFM Cloud Pro model is comparable to Google's frontier Gemini models and will operate in the cloud on NVIDIA GPUs as part of Apple's private cloud compute infrastructure.
Privacy as a Foundational Principle
Subramanya emphasized that the collaboration with Google and NVIDIA extends this private infrastructure to NVIDIA GPUs in Google Cloud while maintaining Apple's stringent privacy standards.
Apple software vice president Sebastian Marineau-Mes noted the company wanted to use NVIDIA's latest chips but required them to be configured with a privacy-first approach, ensuring NVIDIA could not access data on the servers.
He cited recent technological improvements like confidential computing that enabled Apple and Google to build a system meeting their requirements.
Differentiation Through On-Device Processing
Apple is differentiating itself by leveraging on-device data for personalization, contrasting with web-based AI services that collect vast amounts of user data.
Technical Architecture Details
During a technical briefing, Federighi and his team explained the architecture of Apple Intelligence, centered around a "system orchestrator." This component intelligently routes AI requests to either on-device or cloud-based models based on the required processing power and data sensitivity.
Federighi described this orchestrator as central to the entire system's privacy architecture.
Clarifying the Google Partnership
The briefing provided further clarity on the Google partnership. Federighi clarified that Apple Intelligence uses Apple's own proprietary models, not the publicly available version of Google's Gemini, and does not directly use Google's off-the-shelf cloud infrastructure.
Google's technology is being used to assist in building Apple's in-house models, including the third-generation AFM models announced for the cloud, which are specifically designed and optimized to run on Apple silicon.
Subramanya concluded by stating that the suite of models—AFM Core, Core Advanced Cloud, and Cloud Image—are custom-built for Apple chips, trained on proprietary data using reinforcement learning, and optimized based on outputs from Google's frontier Gemini models.
Comments