A Microsoft patent published on November 5, 2026, describes a system that could use spare GPU capacity on users’ devices to help train AI models. The patent is not an announcement of a launched service, and Microsoft has not established whether users will be paid.
According to IT Home’s November 10 report, the patent, titled Opportunistic Micro-training on Edge Devices Using Spare GPU Utilization, proposes splitting training work into smaller tasks and assigning them to devices when computing resources are available. The system would consider model dependencies, communication costs, memory overhead and data location when dividing the work.
The patent describes monitoring GPU and CPU usage, memory use, temperature and power consumption to determine whether a device can take on a task. It also proposes adjusting how often the system checks a device: less frequently under heavy GPU load or at higher temperatures, and more frequently when the GPU is idle or the device is cooler. Microsoft says the approach is intended to avoid disrupting normal use, though the report notes that the effects of sustained training work on a GPU remain a question.
Participation would require users to opt in. Under the proposed system, completed work could earn points redeemable for software, model access or cash. Microsoft also envisions connecting GPU owners with third parties that need training capacity.
The patent does not resolve questions about privacy, how much capacity a device would contribute or what control users might give up. A published patent does not mean the system will become a product, and any eventual reward amount is unknown.
Source: ithome.com (originally published 2026-11-10 14:43 UTC+8)