A Microsoft idle GPU patent application describes splitting AI training across spare phones, laptops, and gaming PCs, and paying their owners, TechPowerUp reports. The filing was submitted in April 2025 and made public on 5 November 2026.
The idea turns the graphics card sitting unused in your gaming rig into rented AI compute. Owners would have to opt in, and the system would track each device’s work and apply credits. Users could redeem those credits for free software, access to the trained model, or money.
The application, titled “Opportunistic Micro-training on Edge Devices Using Spare GPU Utilization”, comes from Microsoft Technology Licensing. It describes chopping large training jobs into micro-tasks sized to what each device can handle, then sending them out whenever a machine has spare capacity.
The system would monitor GPU and CPU load, memory, temperature, and power draw, and check in less often when the GPU is busy or hot. Because some devices may be slow, faulty, or dishonest, results would be cross-checked against other machines, with more trust given to those with a good track record.
FourWeekMBA counts 20 claims in the filing, and none cover paying owners; the money appears in a single paragraph of the description. A published application also isn’t a granted patent, let alone a product.
publication makes the filing public, and it does not mean the claims have been examined or allowed.
FourWeekMBA
There’s no published payout rate either, and the electricity bill would land on the owner. Microsoft hasn’t commented, and it hasn’t announced a Windows or Xbox programme built on the idea. Privacy and permission questions also go unanswered in the filing.
It’s the latest in a run of AI filings from the company, after Microsoft’s patent on live speech and lip movement. Nvidia has also been patenting on the gaming side, with its AI game optimisation patent.
The filing does show Microsoft treating phones, laptops, and gaming PCs as a pool of spare AI compute. Patentlyze puts the trade-off plainly: the technology “would add to, not reduce, a participant’s electricity use”.
