It won’t matter the AI or not at this point. This has baked and changed production so much that even if AI fell out tomorrow, it’d take years to retool and undo the amount of tooling chip makers have done to commit HBM DRAM.
Like a lot of the tooling that was done was done on the back of speculation. Now that it’s done, undoing it would cost a fortune and that price is going to be amortized by the customer. And since we’ve baked this new higher price for so long, it’s going to take forever to cool it back off.
AI or not, these prices are the new normal price. Things like ARM desktops, those are going to be the average consumer computer. Things with discrete GPUs and RAM that is user serviceable, that’s becoming niche. There’s already been way too much investment in that direction to go back at this point. I mean you even look at DDR6 desktops, CAMM2 is the compromise normal there too. You don’t buy 4x8GB, you buy a 32GB block, that’s it. You have the one slot, that’s it.
The AI thing, it’s just pushed a lot of other things past the point of return. While there might be some reduction after the fall of AI, even if chip makers swapped back to desktop product as fast as they can, they’ve invested so deeply into AI, that fast as they can is years before consumers will ever see relief, if ever.
At this point, I see PCs adopting the Apple way of putting it all in a single chip approach. Just because, that’s where a lot of investment has already gone and if RAM tanks, they’ll want to hold onto whatever investments they’ve already made rather try and blaze a new path.
We don’t technically need to retool to go back, HBM could be used by a normal system by having it directly built into the CPU package. Intel Xeon Max CPUs already did this. Of course there are some significant limitations to doing so, and because they were enterprise chips they were horrifically expensive, but it is possible to run them without DRAM.
Especially for something like laptops where there isn’t any significant need for replaceable CPU/ram it could be a cheaper option than retooling. This is especially true if people want to run a local LLM for inference on that laptop, since that kind of setup would be great for it.
I think prices will have to come down a fair bit once the demand from AI disappears. Yes the manufacturers are going to need to pay for the new capacity they built, but they’re also going to need demand to be high enough for the new capacity to not look like a terrible investment, and that will probably mean lower prices when they don’t have data centers filling up the next few years of their production.
Making consumer hardware again will mean retooling and I’d be willing to bet that data centers will be able to swapl their software to cloud based desktops/gaming a lot faster than hardware manufactures.
It won’t matter the AI or not at this point. This has baked and changed production so much that even if AI fell out tomorrow, it’d take years to retool and undo the amount of tooling chip makers have done to commit HBM DRAM.
Like a lot of the tooling that was done was done on the back of speculation. Now that it’s done, undoing it would cost a fortune and that price is going to be amortized by the customer. And since we’ve baked this new higher price for so long, it’s going to take forever to cool it back off.
AI or not, these prices are the new normal price. Things like ARM desktops, those are going to be the average consumer computer. Things with discrete GPUs and RAM that is user serviceable, that’s becoming niche. There’s already been way too much investment in that direction to go back at this point. I mean you even look at DDR6 desktops, CAMM2 is the compromise normal there too. You don’t buy 4x8GB, you buy a 32GB block, that’s it. You have the one slot, that’s it.
The AI thing, it’s just pushed a lot of other things past the point of return. While there might be some reduction after the fall of AI, even if chip makers swapped back to desktop product as fast as they can, they’ve invested so deeply into AI, that fast as they can is years before consumers will ever see relief, if ever.
At this point, I see PCs adopting the Apple way of putting it all in a single chip approach. Just because, that’s where a lot of investment has already gone and if RAM tanks, they’ll want to hold onto whatever investments they’ve already made rather try and blaze a new path.
It was done on purpose, to keep us from building and tinkering with our own technology.
walled garden baby
e: same reason they’re fucking with 3d printing
We don’t technically need to retool to go back, HBM could be used by a normal system by having it directly built into the CPU package. Intel Xeon Max CPUs already did this. Of course there are some significant limitations to doing so, and because they were enterprise chips they were horrifically expensive, but it is possible to run them without DRAM.
Especially for something like laptops where there isn’t any significant need for replaceable CPU/ram it could be a cheaper option than retooling. This is especially true if people want to run a local LLM for inference on that laptop, since that kind of setup would be great for it.
I think prices will have to come down a fair bit once the demand from AI disappears. Yes the manufacturers are going to need to pay for the new capacity they built, but they’re also going to need demand to be high enough for the new capacity to not look like a terrible investment, and that will probably mean lower prices when they don’t have data centers filling up the next few years of their production.
Making consumer hardware again will mean retooling and I’d be willing to bet that data centers will be able to swapl their software to cloud based desktops/gaming a lot faster than hardware manufactures.