On Monday, Apple Inc. (AAPL) unveiled its research and development progress in artificial intelligence at its annual Worldwide Developers Conference (WWDC) held in Cupertino, California.
At this year’s WWDC, Apple demonstrated a redesigned version of Siri. The new Siri can engage in multi-turn conversations with users, marking a significant upgrade over previous versions.
The announcement highlights a strategic approach that sets Apple apart from most of its Silicon Valley competitors: rather than investing billions of dollars in infrastructure and ultra-large-scale advanced models, the company is focusing on privacy advantages and user convenience as key selling points to potential users.
Apple executives told the media that two traditional AI giants, Google (GOOGL) and Nvidia (NVDA), are helping Apple build its most advanced model—the Apple Foundation Model Cloud Pro. This marks the first time the company has officially confirmed that some Apple Intelligence features will run on Nvidia chips. Amar Subramanya, an Apple AI business executive, stated that the AFM Cloud Pro’s capabilities are comparable to those of Google’s cutting-edge Gemini models. The model will run in the cloud on Nvidia GPUs, which are part of Apple’s private cloud computing infrastructure.
Apple is differentiating itself from other companies heavily investing in AI by emphasizing the privacy advantages of its software: unlike ChatGPT or Claude, Apple will not collect large amounts of data. Instead, it uses locally stored user information—such as calendars and text messages—to personalize AI functionalities. Apple executives introduced the software’s architecture: the operating system and software include a module called the “system orchestrator,” which can route AI requests to either on-device or cloud-based models based on the computing power and amount of personal data required for the request. Federighi noted that the system orchestrator is “the core of our entire system’s privacy architecture.”
Apple Intelligence, which runs on Apple devices, uses Apple’s own in-house models rather than a public version of Google’s Gemini; Google’s technology is used to help build Apple’s own models, particularly the third-generation AFM model built for the cloud. These models are specially designed to run on Apple chips. Subramanya stated: “These four models—AFM Core, Core Advanced Cloud, and Cloud Image—are all custom-built for Apple chips, trained with reinforcement learning on proprietary data, and optimized based on the output of the Gemini frontier models.”
