Marketplace Tech · Thursday, September 10, 2026
Nate Sory explains that training advanced AI models requires massive, visible infrastructure, including tens of thousands of specialized chips and significant electricity consumption. He suggests that policing AI development could focus on tracking this large-scale infrastructure, similar to how airplane crashes are investigated, to prevent the creation of dangerous AIs.
“Training one of these advanced AI models isn't something you can do in your basement with consumer hardware. Training a new frontier AI takes like 100,000 of these highly specialized AI chips, which are sort of the peak output of the global supply chain.”
“You know, the other nice thing about this is that to stop, before we build the really dangerous sorts of AIs, we don't need to put the genie back in the bottle. Uh, we just need to not create the next generation of AI swarms.”
“all we would need to do is say, "Hey, we're going to track whenever you have, you know, tens of thousands of these chips aggregated together into a data center that sucks down electricity comparable to a city, where you can see this infrastructure from space."”