Microsoft Could Pay Gamers to Train AI Using Their Idle GPUs, New Patent Reveals - Wccftech

The proposal, titled "Opportunistic Micro-training on Edge Devices Using Spare GPU Utilization," describes a distributed AI training system that would leverage spare GPU resources from consumer devices. Filed on April 30, 2025, and published on November 5, 2026, the system would divide computationally intensive AI training workloads into smaller tasks distributed across participating devices.

How the System Works

Microsoft's system would analyze AI model computational dependencies before splitting workloads, considering communication costs, memory requirements, and data locality. It would continuously monitor GPU utilization, memory consumption, temperatures, and power draw, assigning tasks based on available resources and predicted idle periods.

Training workloads could be reduced or suspended when temperatures rise or users launch demanding applications. The patent also includes mechanisms for validating returned results and identifying unreliable devices.

Compensation Model

Owners would need to explicitly authorize the use of their hardware, with the system tracking completed computations to award credits. These credits could potentially be exchanged for free software, access to AI models, or money. However, compensation is presented as a possible implementation rather than a requirement, as none of the 20 published claims explicitly specifies payment.

Profitability Questions

The concept resembles Folding@home but applies distributed computing to commercial AI training with potential financial incentives. However, additional GPU workloads increase electricity consumption—250W sustained for four hours daily consumes approximately 30 kWh monthly. Any compensation would need to account for electricity costs, higher temperatures, and additional hardware wear.