Idle Apple Silicon devices were turned into a distributed GPU cloud
Apple researchers described Minions, a system that pools unused Mac GPUs for machine learning inference across macOS devices.
Minions turns idle Apple Silicon Macs into a shared GPU cloud for machine learning inference, according to an Apple Machine Learning Research paper published on 2026-10-12.
The system is built for the kind of AI work that needs fast inference, which is the part where a model makes predictions after it has been trained.
Apple says Minions can turn idle macOS devices into a practical distributed GPU inference platform.
Apple says Minions spreads that work across macOS devices instead of relying only on centralized GPU servers. That can help when those servers are expensive, crowded, or slow to free up.
The paper says the setup uses GPUs already sitting inside Apple Silicon devices that are otherwise idle during normal use. In other words, the hardware is already there, just waiting for a job.
Apple says the system can run large-scale inference workloads, handle failures, and use available resources more efficiently. It is meant to help teams use existing hardware rather than buy more GPU capacity.
A simple way to picture it: imagine several Macs around an office quietly helping one another with AI work when they are not busy doing anything else.
The write-up also points to Apple’s wider machine learning efforts on Apple silicon, including MLX for running and experimenting with LLMs on Mac.
The source does not show benchmark numbers, how many devices were tested, or whether Minions is available outside Apple’s own research work.
Source: Apple Machine Learning Research