Microsoft Patent Would Let Gaming PCs Train AI for Pay

Built on PilotFish

The idea has already been tested on Microsoft's own machines. In July 2022, researchers from Microsoft and Shanghai Jiao Tong University presented PilotFish, a system that slips AI computations into the dead time of cloud gaming servers. By detecting idle periods at millisecond scale, it recovered up to 85.1% of unused GPU time without degrading the picture for players. A job only started if it could finish before the game needed the GPU back; otherwise it was interrupted.

The patent carries that logic over to consumer hardware. The system monitors temperature, memory and power draw, and slows or pauses work when the machine is busy or running hot. Each task is sized to the device's power and its availability history, and results are cross-checked between machines and weighted by each one's past reliability.

The electricity question

The filing contains no figures on earnings, but power has a known price. In France, the regulated tariff has been €0.2001 per kWh since 1 August. A card drawing 250 watts for four hours a day uses roughly 30 kWh a month, about €6 to cover before any gain. That scenario is illustrative, since the application specifies neither pace nor duration. The real cost is the extra consumption the machine would not otherwise have used.

The text remains a pending application, and neither Windows nor Xbox has announced a related product. A similar scheme for Macs this summer promised up to $200 a month, but its operator later warned participants that earnings would stay low with no increase in sight.