Last week, Sam Altman, founder and CEO of OpenAI, remarked that memory components in consumer computers are often underutilized—essentially "wasted"—because they are not being fully leveraged, even for tasks like training and developing AI systems, given that this memory sits idle much of the time.
In a recent development, Microsoft has filed a patent that would allow the company to remotely tap into the unused capacity of users' graphics cards to train its AI systems. Through this training process, computer owners could potentially be rewarded with credits or financial compensation, commensurate with the amount of processing work performed by their machines.
According to the patent documentation, this feature would require an explicit opt-in from the user; it would not be enabled by default.
Interestingly, this patent filing is not entirely new. Reports and the patent documents themselves indicate that Microsoft filed it back in April 2023—a period when the adoption of AI technology began to surge dramatically.
The concept of leveraging graphics card capacity for remote data processing is not new. The *Folding@Home* project, introduced by Stanford University in 2000, utilized the processing power of CPUs and graphics cards to simulate protein folding—a method used to better understand and address complex diseases.
For now, Microsoft has merely filed the patent; no official announcement has been made regarding the integration of this AI training capability into its *Copilot* AI assistant.
