Apple is making edge AI a centerpiece of its latest iPhone lineup, redesigning its flagship chip to run more powerful AI models directly on the device while introducing an overhauled Siri that can act across apps and use personal information stored on the phone.
At its product event today, the first for new CEO John Ternus, Apple unveiled the iPhone 18 Pro and Pro Max and its first foldable phone, the iPhone Duo. The devices are built around the new A20 Pro chip, which Apple said was designed in part to handle the growing computing demands of on-device AI.
The A20 Pro doubles the number of neural engine cores to 32 and provides twice the AI computing power of its predecessor, according to Apple. The chip also incorporates neural accelerators into its CPU cores, improves GPU AI performance and increases memory bandwidth by 50%.
Significantly, Apple said those neural accelerators can run third-party large language models directly on the iPhone. That potentially expands the role of the smartphone from an interface for accessing AI models running in data centers to an inference device capable of executing some models locally.
Apple demonstrated that approach on the iPhone Duo with Detail, a content-creation app that uses Apple’s Core AI framework to run its AI models on the device. Apple also introduced new APIs intended to help developers adapt their applications to the Duo.
The approach reflects Apple’s hybrid AI strategy. Its latest Apple Foundation models include what the company calls its most advanced on-device model, while more computationally demanding requests can draw on models running through Apple’s Private Cloud Compute infrastructure.
That computing architecture underpins Siri AI, a new version of Apple’s assistant that can draw information from a user’s mail, messages and other apps, understand what is displayed on the screen and take actions across Apple and third-party applications. Apple said Siri already handles more than 2.5 billion requests a day and works with more than 300,000 apps.
Apple is extending on-device AI beyond the iPhone. The new Apple Watch Series 12 and Ultra 4 use on-device AI to recognize sounds such as alarms, sirens and a baby crying without requiring the user’s phone to be present. New Audio Intelligence features can also transcribe recent speech and generate summaries and key points from conversations. Apple said the features do not create or store audio recordings and that raw audio used for processing is inaccessible even to Apple.
The company described the new iPhones as an “intelligent personal hub,” positioning its combination of custom silicon, on-device models, cloud AI and access to personal context as the foundation for its AI strategy.