there wasn’t enough demand for AI to justify the scale of buildout that was proposed. People building the datacenters didn’t want to pay the absurd costs to get them built quickly if they were just going to loose money operating them. So they warehoused chips they were already getting, and cut back on future orders.
NVIDIA and the other suppliers for the build out wanted to keep their share prices high and also didn’t want to get caught holding the bag of inventory, so they found more buyers where ever they could and poured money in to startups to absorb demand.
What really pisses me off is that this whole thing is a giant waste of resources, they’re just pissing away capital, fab capacity, and resources on stuff that has no use outside of a datacenter. Maybe we’ll see a cottage industry of folks removing these chips and putting them on boards consumers can actually use?
Potentially the largest miss allocation of resources so far this century.
Like, it goes beyond just “the fab time could have been used on something useful”, the money spent on scaling up production across the industry could have also gone to rebuilding crumbling infrastructure, or building out public transit in under served areas, or replacing gas turbines with battery solar.
That is not what the Alice Huang case is about though. She definitely illegally imported Nvidia chips to China (where at least some of them ended up in Russia on drones used in Ukraine).
The false demand created by Nvidia investing in AI-startups and then forcing them to buy chips they didn’t need, is a separate case of crime.
It’s part of the same dynamic, “we have too much supply, how the hell do we get rid of it”
Selling it to people you’re not legally supposed to is just as much a problem as paying people to buy them from you, at least in the context of the health of a business. Both are also illegal, albeit in different ways.
They exported the chips to China despite it being illegal for the same reason they funded all the startups. To create demand, maintain the bubble, and get rid of excess inventory.
there wasn’t enough demand for AI to justify the scale of buildout that was proposed. People building the datacenters didn’t want to pay the absurd costs to get them built quickly if they were just going to loose money operating them. So they warehoused chips they were already getting, and cut back on future orders.
NVIDIA and the other suppliers for the build out wanted to keep their share prices high and also didn’t want to get caught holding the bag of inventory, so they found more buyers where ever they could and poured money in to startups to absorb demand.
What really pisses me off is that this whole thing is a giant waste of resources, they’re just pissing away capital, fab capacity, and resources on stuff that has no use outside of a datacenter. Maybe we’ll see a cottage industry of folks removing these chips and putting them on boards consumers can actually use?
Potentially the largest miss allocation of resources so far this century.
Like, it goes beyond just “the fab time could have been used on something useful”, the money spent on scaling up production across the industry could have also gone to rebuilding crumbling infrastructure, or building out public transit in under served areas, or replacing gas turbines with battery solar.
That is not what the Alice Huang case is about though. She definitely illegally imported Nvidia chips to China (where at least some of them ended up in Russia on drones used in Ukraine).
The false demand created by Nvidia investing in AI-startups and then forcing them to buy chips they didn’t need, is a separate case of crime.
It’s part of the same dynamic, “we have too much supply, how the hell do we get rid of it”
Selling it to people you’re not legally supposed to is just as much a problem as paying people to buy them from you, at least in the context of the health of a business. Both are also illegal, albeit in different ways.
They exported the chips to China despite it being illegal for the same reason they funded all the startups. To create demand, maintain the bubble, and get rid of excess inventory.