Microsoft has filed a patent for a system that leverages idle gaming PCs to train AI models, offering digital rewards in return. The patent, filed in 2025 and published last week, outlines a distributed network similar to Folding@Home, where powerful gaming PCs could contribute their unused GPU cycles to AI training tasks. The system, titled Opportunistic Micro-training on Edge Devices Using Spare GPU Utilization, operates on an opt-in basis, ensuring users have control over their participation.
The proposed system monitors key components like memory, CPU, and GPU usage, as well as temperatures and power draw, to identify optimal times for AI training. It avoids overloading hardware by ensuring the GPU is cool and idle. Users can manually start or stop jobs and set usage limits to prevent excessive energy or hardware strain. However, the patent does not address whether such usage would be covered under existing warranties.
While the idea of earning rewards for idle GPU cycles could attract users, especially if the payouts outweigh electricity costs, it comes amid debate over the efficiency of using consumer hardware for AI training. OpenAI CEO Sam Altman recently suggested that idle memory in consumer PCs might be better utilized in data centers, highlighting the intense hardware demands of major AI companies.