410
TSMC is apparently considering getting into the RAM business at the request of Nvidia.
(www.theverge.com)
this post was submitted on 07 Nov 2026
410 points (99.3% liked)
Technology
89639 readers
2604 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 3 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
and the reason manufacturers aren't building many more factories is because they're extremely expensive and there's no guarantee the same demand will still exist in 5 or 10 years. What if the AI bubble bursts, the market is flooded with second-hand supply, and all the new factories aren't needed?
Micron is building new factories in New York, and they expect the first factory there to be online by 2030 at a cost of $50 billion (which is more than their entire 2025 revenue).
We had this issue during the crypto boom and the manufacterers that added capacity then are glad they did.
What else could use up that RAM? How about standalone VR? We could get mainstream adoption by 2030.
The point is building factories sometimes pays off
I agree that more supply would be great and building factories can pay off. It's still very expensive though, and some banks are growing skeptical of the ballooning costs of AI-related projects.
That's why I don't think we should limit the AI boom, give the supply chains a chance to pay for more capacity for the future
There are a few other reasons to stop building data centers.
My ask is that DC e held to adhere to environmental regs, rather than be given free rein like they have now
Like what? The economy is basically not in a recession because of a data center boom
Did you miss that google illegally clearcut 250,000 trees for a data center, or that they use tons of water and power and they're building them in deserts. Or that they're an absolute nuisance for miles around them?
How about they shouldn't be allowed to do that? You can build data centers without breaking the law. They also don't use that much water. They have been farming almonds in arid California areas for decades and nobody cared
There's definitely some negatives about data centers.
However, there's also a lot of misinformation. All the data centers in the entire USA use less than 1/20 of the water consumed by golf courses, and modern designs use closed-loop water cooling rather than the older evaporative cooling. Evaporative cooling is the type that uses a lot of water.
I live in Silicon Valley, and most of the data centers here are in Santa Clara County. 65% of the water usage by data centers is recycled water.
I'm spitballing here, but one way it might be possible for TMSC (or whoever puts up the capital, which doesn't have to necessarily be TMSC) to mitigate risk, given that this is Nvidia allegedly talking to them, is to build some kind of new memory specialized for AI use that would displace existing memory for AI use. Like, for neural nets, you don't have the same constraints that you do on conventional DRAM. If you can have someone who builds neural net systems and a memory maker work together, you might be able to produce a new sort of memory that would be a better buy for neural net users.
A few things I can list off-the-cuff:
Access patterns on neural nets are predictable long in advance. That means latency doesn't matter much (so having far higher latency than current memory may be fine) and requesting large blocks of memory might be valuable. Accesses will be sequentially requesting memory over a large range, so it also may mean that one could change how cooling works in the memory, if one has some guarantees on a single spot not being frequently re-requested a lot. A given byte of memory won't be re-written, once written, for a long time, so you could relax how rapidly the hardware needs to be able to sequentially do writes to a given byte.
You have a lower ratio of writes to reads. I read one article talking about how HBM isn't yet optimized for AI use on write/read bandwidth availability grounds. That's talking about the memory interface, not the memory chip itself; they were just arguing for changing the ratio of bandwidth reserved for writes and reads. However, I would bet that you could also optimize the memory chip around this.
So you'd basically come up with, I dunno, "neural net memory (NNM)" or something like that. It's not that you have to use it for neural nets, but that its characteristics would be optimized for neural net use. That only makes sense if there's enough scale of demand for a single application to put engineering time in, expend the fixed costs, but there is presently a lot of demand, and might be more down the road.
That doesn't make risk of demand dropping go away, but...it'd tend to shift that risk to conventional memory manufacturers, since if demand doesn't meet expectations and you're doing neural nets, a buyer would probably buy the specialized memory chip before the non-specialized memory chip. Still risk, but it'd be shifted to Micron/SK Hynix/Samsung.
At the start of the AI boom, we had hardware that was not optimized for AI use. It was generic random access memory with most of the hardware stack designed for rendering 3D graphics. I'd guess that it's very likely that that hardware stack is substantially different from one optimized for neural net use, and I'd bet that Nvidia would be the most-likely party to be at the heart of attempts to re-engineer the hardware stack to optimize it, as that's somewhere that they're uniquely-well-positioned to make changes.
EDIT: I remember a while back when John Carmack was tossing the idea out there of using a loop of fiber-optic cable as a memory for AI use. That may not be the optimal design, but one does see that there is a lot of room to think about specializing the memory to the very different demands that neural nets place on memory rather than the demands that conventional RAM users have.
Not an expert on AI (or even software development), but geniune question: wouldn't there still be the problem of silicon allocation involved?
Silicon allocation?
I mean, at the end of the day the same fabs that build "normal RAM" are going to be the ones to build NNM, so as NNM has higher margins, they would opt for more NNM production, and the problem still wouldn't be solved, as instead of using HBM, they'd use NNM, while at the same time, consumer RAM wouldn't be as much produced.
I think that the best thing for us consumers is a unified standard, that both consumers and AI-centered compute can utilize, as in that way, there would be just one type of RAM needed to be built for everyone, but even then, it wouldn't be really good, as the demand for this ideological unified RAM would have price through the roof since its demand would come from AI, so I really have no clue what would be a good solution for RAMpocalypse.
Or maybe, if you still want to have NNM, another possibility is to create entirely new companies, that are independent from the established RAM suppliers. Maybe in this way it could work, but I could be huffing too much hopium here...
While the bag of experts approach has allowed segmentation of memory into vram and dram, the access in the particular expert isn't predictable, access is spread across all the memory for that expert, and latency dramatically affects bandwith.
The kind of memory the current manufacturers are focusing on is specialized for data-center-grade GPUs. These GPUs are insanely power-hungry and require data-center-grade cooling. When the bubble pops and data centers go under, you're not going to see consumer-grade RAM flood the market. So if TSMC wants to enter the market conservatively with the assumption the AI surge will not last, going into the consumer RAM side would be a better bet.
Compared to today the PCs of 1999 were extremely power-hungry, and cooling was not so simple too.
I suppose those who want that compute could start making such home farms. That might not be everyone and their dog, but if such a gigantic supply increase occurs, it might make sense for a while.
Which could also be the real business model of big AI companies, to stockpile the resource that's going to be very expensive when the demand raises due to popularization of local AI.
OK.
There's a lot of regular RAM in the servers too though, since the servers can't operate entirely from VRAM.
And yet there's so much talk about this and demand putting pressure upon them. that they are certainly taking a hit when/if that bubble bursts, even expecting that.