Apple Turns Idle Macs Into AI GPU Cloud, Cutting 10-Hour Inference Queues
Updated
Updated · Computerworld · Oct 15
Apple Turns Idle Macs Into AI GPU Cloud, Cutting 10-Hour Inference Queues
3 articles · Updated · Computerworld · Oct 15
Summary
Apple disclosed an internal system called Minions that links idle Apple Silicon Macs into a virtual GPU cloud for AI inference, letting jobs start within minutes instead of waiting more than 10 hours.
The setup tackles scarce centralized GPU capacity by splitting work across connected Macs: a server queues tasks, an executor assigns them, and a watchdog requeues failed subtasks if a device sleeps or goes offline.
Minions currently supports two inference workloads—embedding generation and LLM completions—using open-source models via Hugging Face and Ollama, with local execution avoiding remote API calls.
Apple said the approach can unlock substantial compute capacity at zero hardware cost while keeping sensitive data inside team- or organization-scoped Mac pools backed by isolated PostgreSQL databases.
The white paper says Apple may extend Minions to distributed model training, but the company has not announced any plan to release the technology outside Apple.