But you’re not selling that as a service. A raspberry pi can run a basic Ai model, but you’d never get $5 a month for access from even a single user in all likelihood.
No, but…is your concern the ability to run a commercial service involving AI compute? I mean, there are certainly a ton of startup companies doing that. Heck, there are people selling access to their GPUs for parallel compute on vast.ai.
I’m just saying that I don’t believe that the ability to run neural nets on parallel compute hardware is something that is going to be terribly exclusive over time, and certainly isn’t today something limited to someone with a net worth of a billion dollars.
As I’ve commented before, we’d need a lot more RAM than exists in the world today if everyone’s going to do it. Like, AI companies are buying more 2026 RAM production than the rest of the world combined, on the order of two-thirds of global production. If they get something like 100% capacity utilization of their hardware, and a typical user doing local AI compute would get something like 1% capacity utilization of their hardware, then we’d need about a hundred times that much memory to let everyone run comparable stuff locally. That’s a pretty stupendously large amount of memory. But if, over time, there’s demand for it, I expect that it’ll happen. We’ve scaled up parts of the computer industry by orders of magnitude in the past.
But you’re not selling that as a service. A raspberry pi can run a basic Ai model, but you’d never get $5 a month for access from even a single user in all likelihood.
No, but…is your concern the ability to run a commercial service involving AI compute? I mean, there are certainly a ton of startup companies doing that. Heck, there are people selling access to their GPUs for parallel compute on vast.ai.
I’m just saying that I don’t believe that the ability to run neural nets on parallel compute hardware is something that is going to be terribly exclusive over time, and certainly isn’t today something limited to someone with a net worth of a billion dollars.
As I’ve commented before, we’d need a lot more RAM than exists in the world today if everyone’s going to do it. Like, AI companies are buying more 2026 RAM production than the rest of the world combined, on the order of two-thirds of global production. If they get something like 100% capacity utilization of their hardware, and a typical user doing local AI compute would get something like 1% capacity utilization of their hardware, then we’d need about a hundred times that much memory to let everyone run comparable stuff locally. That’s a pretty stupendously large amount of memory. But if, over time, there’s demand for it, I expect that it’ll happen. We’ve scaled up parts of the computer industry by orders of magnitude in the past.