I had an idea that I think could be interesting for both OpenAI and users.
What if ChatGPT users could voluntarily share some of their device’s unused computing power in exchange for ChatGPT credits or a discount on their subscription?
I’m not suggesting that OpenAI should simply use everyone’s CPU without limits. I’m thinking of an entirely opt-in system where users could choose to participate and decide how much computing power they’re willing to make available.
For example, a user could choose:
- 10%, 25% or 50% of available CPU/GPU resources
- only when the device is idle
- only while plugged in
- a maximum amount of compute per day
- or simply turn it off whenever they want
In return, OpenAI could give users credits based on the amount of useful compute their device contributed. Those credits could reduce the cost of a subscription or provide additional token/usage allowances.
I realize that a phone or consumer PC isn’t equivalent to a datacenter GPU, and obviously not every part of ChatGPT’s workload could be distributed this way. But there may be background, distributed or latency-tolerant workloads where this could make sense.
I also think privacy would be extremely important. The system would need to be designed so that users aren’t giving OpenAI access to their personal files, conversations, or other private data. Ideally, the workload would run in a sandboxed environment and users would always have control over when their device participates.
The interesting part to me is the potential exchange:
OpenAI gets additional computing capacity and users contribute resources they aren’t currently using and users receive something of real value in return.
It could potentially create a new type of relationship between users and the infrastructure running AI.
I’d personally be interested in participating in something like this if it meant getting additional ChatGPT usage or a meaningful reduction in my subscription cost.
I’m curious whether something like this is technically or economically realistic, and whether OpenAI has considered a user-provided compute model.