First, this is one of the conditions where capitalism actually works. Many players in the field dropped out because of the razor thin margins of the past. Fabs take years to ramp up, and are insanely expensive to set up, so getting in to take advantage of a temporary shortage was an unacceptable risk.
Now there is a decade(s) long projected shortage, making the investment attractive again. Plenty of players have experience in Fabs, even though they are not up to date.
Also, the market is going to accept that slightly slower ram is quite fine in many applications, and they are easier to make. DDR5/6 is really not that important. I have an AM4 Ryzen 9 with 64Gb DDR4 that flies, I mean the thing cooks!
This environment is going to make Chinese Fabs competitive in the mid-term, and give them the opportunity to catch up, especially since the Chinese government subsidizes whole sectors to catch up, and often surpass the west (see EVs, solar, airliners, etc.) maybe they’ll take years to get there, and maybe they won’t match the very top end, but they’ll take over.
We are going to have a shortage and obscene prices, but not as long or hard as doomsayers scream.
Another factor is that the AI bubble is going to pop. LLMs are a dead end, and are already at an extreme diminishing returns point. There is no way the major players are going to recoup investment, and the market will eventually wake up. Open source models are at single digit distance of the most powerful commercial models, so much of the resources are going to shift to in-house.
JEPA is one of the next steps in AI, and is way less hardware intensive. There are several new approaches to AI that are way less hardware intensive. LLMs are plain brute force approaches, and evolution makes efficiency a major goal.
Depends on government policy also I guess. What would most likely happen is for US gov to continue sanction Chinese companies like what they did to cxmt and ymtc to stifle competition and keep prices artificially high
Yes. Capitalism is not bad, it’s actually a sane approach. The problem is when left unchecked, which is the state’s duty, but fucked up by politicians, in the pockets of oligarchs. Have you read Adam Smith? Both Smith’s and Marxs’ thesis fail beacuse they depend and assume inherent good in people, while in reality greed is the driving force in economics, and fuck all else.
AI will remain a massively parallel numerics affair with enormous data sets and monstrous memory bandwidth and network crossection. And accrding energy consumption. Jevon’s paradox will eat any efficiency improvements.
Only if LLMs are the only option. A paradigm change is coming. It’s like what happened when European and Japanese performance cars started to take on American muscle cars or SpaceX (yeah I hate the Nazitard too) started recovering rockets and reusing them, or PCs started replacing mainframe workstations…
The article ignores several points.
First, this is one of the conditions where capitalism actually works. Many players in the field dropped out because of the razor thin margins of the past. Fabs take years to ramp up, and are insanely expensive to set up, so getting in to take advantage of a temporary shortage was an unacceptable risk. Now there is a decade(s) long projected shortage, making the investment attractive again. Plenty of players have experience in Fabs, even though they are not up to date.
Also, the market is going to accept that slightly slower ram is quite fine in many applications, and they are easier to make. DDR5/6 is really not that important. I have an AM4 Ryzen 9 with 64Gb DDR4 that flies, I mean the thing cooks! This environment is going to make Chinese Fabs competitive in the mid-term, and give them the opportunity to catch up, especially since the Chinese government subsidizes whole sectors to catch up, and often surpass the west (see EVs, solar, airliners, etc.) maybe they’ll take years to get there, and maybe they won’t match the very top end, but they’ll take over.
We are going to have a shortage and obscene prices, but not as long or hard as doomsayers scream.
Another factor is that the AI bubble is going to pop. LLMs are a dead end, and are already at an extreme diminishing returns point. There is no way the major players are going to recoup investment, and the market will eventually wake up. Open source models are at single digit distance of the most powerful commercial models, so much of the resources are going to shift to in-house.
JEPA is one of the next steps in AI, and is way less hardware intensive. There are several new approaches to AI that are way less hardware intensive. LLMs are plain brute force approaches, and evolution makes efficiency a major goal.
Depends on government policy also I guess. What would most likely happen is for US gov to continue sanction Chinese companies like what they did to cxmt and ymtc to stifle competition and keep prices artificially high
This is your example of capitalism working? Then I don’t want to see it failing.
Yes. Capitalism is not bad, it’s actually a sane approach. The problem is when left unchecked, which is the state’s duty, but fucked up by politicians, in the pockets of oligarchs. Have you read Adam Smith? Both Smith’s and Marxs’ thesis fail beacuse they depend and assume inherent good in people, while in reality greed is the driving force in economics, and fuck all else.
Yeah any extra fabs able to produce ‘last gen’ DDR4 would have a big impact.
AI will remain a massively parallel numerics affair with enormous data sets and monstrous memory bandwidth and network crossection. And accrding energy consumption. Jevon’s paradox will eat any efficiency improvements.
Only if LLMs are the only option. A paradigm change is coming. It’s like what happened when European and Japanese performance cars started to take on American muscle cars or SpaceX (yeah I hate the Nazitard too) started recovering rockets and reusing them, or PCs started replacing mainframe workstations…