Apple Outlines On-Device AI Capability Matrix Across Its Entire Product Lineup
At the Jamf Nation User Conference (JNUC), Apple presented an official hardware matrix detailing on-device AI inference capacity across its entire product lineup. The breakdown ranges from 14 billion activated parameters on iPhones and iPads up to 1.6 trillion on a 2TB Mac Studio cluster. This official matrix provides developers and enterprise IT teams with clear hardware targets when optimizing and deploying local LLMs across Apple devices. It highlights how Apple's Unified Memory Architecture (UMA) enables running high-parameter AI models directly on local hardware without depending on cloud processing. According to Apple, mobile devices like iPhones and iPads (up to 16 GB RAM, 76 GB/s bandwidth) can host models up to 14 billion activated parameters for tasks like Siri and image editing. Professional workstations such as MacBook Pro (128 GB RAM) and Mac Studio (512 GB RAM) support up to 120 billion and 480 billion activated parameters, respectively.
## BACKGROUND
Jamf Nation User Conference (JNUC) is an annual event focused on enterprise Apple device management and security hosted by Jamf. In modern Mixture-of-Experts (MoE) model architectures, activated parameters refer to the specific portion of network parameters calculated during an inference step, allowing machines to execute large models more efficiently.