正文 · AI 翻译
Thanks to IT Home reader 华南吴彦祖 for the tip!
IT Home reported on November 10 that a new Microsoft patent suggests the company may soon allow users to earn money by putting idle GPUs to work training AI.
On November 5, 2026, Microsoft published a patent titled “Opportunistic Micro-training on Edge Devices Using Spare GPU Utilization.” The patent describes an entirely new system for scheduling computing resources: it breaks intensive computing tasks that would normally require large data centers into smaller subtasks and distributes them to devices when those devices are idle.
Put simply, Microsoft may in the future use temporarily idle GPU capacity in users’ computers to train AI models, and users could be paid for it.
The patent describes the system as follows: “This is a system and method for training machine learning models through opportunistic micro-training using idle GPU capacity on edge devices. The system monitors edge devices to detect available computing resources and divides training jobs according to model dependencies into microtasks that can run independently. When dividing the jobs, the system considers communication costs, memory overhead and data location to determine the most suitable points at which to split them.”
The core idea is to divide complex computing tasks that would normally be handled by large data centers into microtasks, dynamically adjust their size according to each device’s hardware capabilities, and assign them to devices with temporarily idle computing capacity.
Microsoft says in the patent that this approach can make effective use of distributed computing resources without affecting normal device use. However, whether sustained AI training workloads would cause wear to GPUs remains a concern. Microsoft does not plan to require participation; users would need to actively authorize their devices to join the system.
To encourage users to contribute their devices’ computing capacity, Microsoft also plans to offer compensation. Once users opt in, the system would record how much computation their devices complete and award points to their accounts based on the completion of each computing task. These points could be redeemed for free software, access to models or even cash.
This patent is not intended solely to provide computing capacity for Microsoft’s own AI business. Microsoft actually hopes to offer the solution to third-party organizations to help companies that want to train AI models but lack sufficient computing resources.
The patent notes that training AI models requires substantial computing resources, that the necessary hardware is expensive, and that running the training also consumes substantial resources. This has already led to a persistent shortage of GPUs.
Under this model, when users lend their GPUs to third parties for AI training, Microsoft would mainly act as an intermediary, connecting those who need computing capacity with GPU owners.
However, this approach also raises questions, including how users’ privacy would be protected, how much GPU capacity devices would need to contribute, and whether users would have to give up some control over their devices after authorizing them to take part in the computing work.
At present, the patent does not directly answer those questions, but it does describe in detail how the system would determine whether a GPU is idle and whether a device is currently suitable for additional computing tasks.
Specifically, the system would monitor a range of indicators, including GPU and CPU utilization, memory use, device temperature and power consumption, to determine whether a device has idle computing capacity available.
The patent explains: “The system continuously monitors devices to identify when they have available computing resources. To avoid affecting the user experience, such as by slowing applications, draining batteries faster or overheating devices, the system makes use, as far as possible, of idle computing capacity that might otherwise go to waste.
In addition, the system dynamically adjusts its sampling frequency: when a GPU is under heavy load or running at a higher temperature, it checks less often; when the GPU is idle or the device’s overall temperature is lower, it checks more often.”
IT Home cautions that a patent being granted or published does not mean Microsoft will necessarily put the technology into practice. Whether this technology can be deployed, and how much users might ultimately be paid, remain to be seen.