Apple's Minions Turns Idle Mac GPUs into a Virtual GPU Cloud

Apple's research team has unveiled a system called Minions, which pools idle Mac GPUs into a dynamically schedulable virtual GPU cloud for running large-scale machine learning inference. Traditional AI inference relies on centralized GPU clusters, which are costly, slow to scale up, and often come with long waits, while Macs usually sit idle during everyday office work.

The system has four parts: Minion Server handles task intake and scheduling, PostgreSQL stores task and device status, Minion Executor runs inference locally on Apple Silicon GPUs, and Minion Watchdog monitors offline devices and reassigns tasks to avoid duplicate work.

Tests show that inference tasks that once waited over 10 hours in queue could start within minutes under Minions, with no new GPU hardware needed, cutting infrastructure costs. For now, it is an internal research project, and data privacy remains a key concern when sharing devices across teams.