What we know
Microsoft has filed a patent for a system that would use idle GPUs in users’ devices to perform AI training tasks. The system monitors device activity and assigns AI workloads during periods of low or no usage, but only with the user’s consent. Participants in this program could receive compensation, such as payments or credits, in return for contributing their device’s processing power. The patent outlines a method to leverage idle GPUs on consenting users’ devices for AI training. Device usage would be monitored to identify optimal times for assigning AI workloads. Users would be compensated for allowing their devices to be used in this way.
Why it matters
This patent reveals Microsoft’s interest in distributed AI training by tapping into the unused processing capacity of consumer devices. Offering compensation could incentivize more users to participate, potentially creating a large, decentralized AI training network.
What is still unknown
No details are available on when or if this system will be launched commercially. The patent does not specify which AI models or languages would be supported, nor does it provide information on the accuracy or efficiency of training using this method. There are no independent tests or demonstrations of the system’s effectiveness or energy impact.