Microsoft has filed a patent application describing a system that would tap the unused graphics cards of gamers to train AI models, with device owners paid for the computing power they contribute.
November 5 · Microsoft patent application
Microsoft Patent Explores Using Gamers' Idle Graphics Cards to Train AI
The filing describes splitting AI training work into small tasks for individual consumer devices, with owners paid for the computing power they contribute. It is a patent application, not a product announcement.
How the proposed system would work
A conceptual flow described in the filing. Each step is a design idea, not a confirmed deployment.
01
Split the job
One AI training job is broken into small tasks, each sized to fit one device.
→
02
Monitor each card
Load and temperature are tracked continuously.
→
03
Assign when spare
Work goes only to devices with spare capacity and no overheating risk.
→
04
Pay contributors
Owners are compensated for the computing power they provide.
The core question
Filing stage is not a launch. Several key details remain unknown.
Whether or how the system would be built or deployed
The compensation model for device owners
Security safeguards for participating machines
How data privacy would be handled
The case for it
Pooling many consumer machines could add capacity beyond large data centers, as AI demand grows faster than advanced chip and data center supply.
The open challenges
AI training usually needs tightly coordinated clusters. Scattered devices differ in network speed, uptime and thermal limits, which makes coordination technically demanding.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...