339. PS6 and RTX 6090 Delayed by Shortages? RAM Executive Explains | Dave Eggleston

At Capella University, learning online doesn't mean learning alone. You'll get support from people who care about your success, like your enrollment specialist who gets to know you and the goals you'd like to achieve. You'll also get a designated academic coach who's with you throughout your entire program. Plus, career coaches are available to help you navigate your professional goals.

Learn more at capella.edu. Wix is the website builder that just keeps on giving. It's giving... I just built my whole site in under an hour. It's giving... I actually made that? Whoa. It's giving... Our sales are through the roof. So whether you need help expressing yourself or just need something to make your life easier... Wix has you covered. Try it out at Wix.com. Mazda crafts cars for those who choose to do more because every day is a chance to explore, to experience, to feel.

And in the Mazda CX-50 Hybrid, you'll spend less time refueling and more time discovering. And welcome to Broken Silicon, a gaming hardware podcast. I am your host, Tom, and this episode is brought to you by cdkeyoffer.com, where you can get Windows 11 Pro for just $24.

with Offerco Broken Silicon, and then also Silver Knight PCs, where you can get 10% off everything on their website, including things that have RAM in it, but that same Offerco Broken Silicon as well. And speaking of RAM, there's a guest we have today. Honestly, I can't...

Think of a better guest. In fact, I was putting to get for this moment, like I was putting together this big report on RAM shortages a few weeks ago. And one of the contributors from War's Law is Dead, really the editor, a carbon cry said, you know, have you spoke to Dave Eggleston?

Recently. I mean, this is literally like the person, you know, who could probably speak to the cyclical nature of the Ram market and what's going on right now. And it's like, what happened? I reached out to you. You responded. Well, I'm getting a bit ahead of myself, as usual, talking too much. Please introduce yourself. And if we could, because it's been, I think, about three years since you've been on.

everybody your extensive history, because I do think it's worth setting the table for why people should listen to you, possibly more than some person who just puts videos out on YouTube. Oh, no worries. Yeah, thanks so much for having me back. It's always a pleasure. It's always thought-provoking to dive into some of these topics, and memory is certainly an important topic, and as we see shortages right now.

in particular interest across the industry and for your listeners. For myself, I've been in the memory business now. It's 40 years. This year is 40 years. joined AMD working on non-volta memory, and then eventually joined a tiny little startup, which was called Sundisk at the time, later we called it SanDisk. Great to see SanDisk, by the way, back as an independent company again.

And then moved on to Micron. I was at Micron for quite a while and worked in the actual systems engineering group, which was doing controllers for memory. And then moved on to work on emerging memory, which is, in my case, was both resistive RAM and then at Global Foundries working on magnetic RAM. And I'm back. working on IP licensing around memory these days, but always just kind of kept myself right in the memory business and all aspects of it and consider myself a watcher of that.

Titles, yeah, I've had all kinds of titles, CEOs, presidents, VPs, et cetera, et cetera. But those things don't matter as much as just being interested in a topic and pursuing a topic. So glad to talk about that today. You're quite humble. I know quite a few people that would, when you gloss over just saying VP and president, they would go, to some people, it does matter quite a lot.

Honestly, not so much to me. And once you've had those titles and interacted with folks at that level, they're just folks. Yes, I can imagine that must be true. It's funny, you mentioned like people are about to show probably a lot more interest in memory right now than they usually would. And I have to imagine, like, I've just been thinking. I work on all these videos about NVIDIA, AMD, the RAM market, and my brother is often the co-host of episodes, also a nerd like me, so I can talk to him.

about this whenever I want. Probably my best friend too. He comes up all the time on the show, Brock. I can talk to him about this about as much as Dan. Outside of that, though, and the people in the Moore's Laws Dead community online, it's like I can talk to my parents about this loosely. I can talk to my... old friend loosely but it's never as in-depth of course as the people that are headfirst in this industry like you but this is i think going to catch a lot of people off guard

therefore right maybe kind of like the chip shortages that caused automotive companies to halt production a few years ago right that took people off guard not you or me probably but Wouldn't you agree? I think a lot of people are about to have a rude awakening because when I brought up, oh, I just put out this, like to my parents on the phone, like I put out this report about ram shortages and they're like, oh, is that bad? And I'm like, yeah.

I think some companies might just stop making stuff next summer.

yeah you can really it's true it can bring the lines to a halt it's interesting that you bring up the you know kind of the whole automotive industry grinding to a halt and learning that you can't treat chips the way you treat rolled steel You know, it's just not a commodity and definitely saw that up close and personal at COVID time because all the other, you know, people were buying dishwashers and refrigerators and toys and all kinds of things.

that took that same silicon and we have something very similar going on right now in the DRAM business and a kind of big picture if you zoom out there's three legs to the DRAM business there's client There's mobile, and then there's data center. And if you roll back the tape a couple decades, all three of those industries were being served by the exact same product, you know, DDR, DRAM.

And client was the largest at that point. But what we saw is over time, mobile became important enough that you had to have your own version, which was LPDDR. And LPDDR has deviated enough from kind of generic DDR to where, you know, basically it's his own product. And one of the things I saw coming a while back was when does data center get big enough?

that it's going to split off and kind of do its own thing and need its own device. And that device is HBM. There's been a lot of different things that have been tried. But what we're seeing now is the distortions, you know, HBM just a handful of years ago, less than five years ago was, you know, a billion dollar a year business. And then the memory business, that's kind of nothing.

Not that important. And now it's forecast to pretty soon be $100 billion a year of business. So you've seen a change where the data centers are requiring HBM. And that warps everything, that distorts everything. And I think that's the bottom line of what we have going on, is HBM coming to the forefront and being so important.

I dove right in and DDR, LPDDR and HBM and what are those and how are they different? I mean, that may be something that we can dive into, but I think big picture, that's what's causing... some of the DDR4 shortages that we're seeing. Or DDR5, yeah. Or DDR5, right. Because what happens is, a company like Micron, we used to talk about it and say, what's Micron in the business of? And Micron has big memory factories, big fabs. Micron makes wafers.

And at the end of the day, what you make on that wafer could be an NAND product, could be DDR5, or it could be HBM. So where can they make the most money on producing a wafer? And so again, that's the kind of warpage we see, the changes we're seeing in the memory business that are leading to some of these shortages.

Yeah, and I think that modularity and what you can use the same equipment for to make different types of products is something the average consumer often misses. I remember during the COVID days, you know, people were just saying things like, well. Why can't they just make more PS5s? And I'm like, well, because those same chips are actually the chips that are used.

to make all of these ai and data center chips that are that they can nvidia can sell for 10 grand so why you would use the same amount of space to make two ps5s or something i don't know why you would do that And it went on and on and on. Like the automotive manufacturers used not the most advanced silicon, but...

close actually compared to where they probably would have in the past. And so then it's like, well, maybe the same IO dyes are being used by Zen 2, you know, using the same nodes at some of these facilities or whatever other client products. And that's where you realize like, well. How much are they paying for this chip that goes into a car versus a super expensive chip that might go into a laptop? And then they get in. That's why they're not getting parts for their cars. Absolutely.

Yeah, anybody that runs a fab, whether it's a foundry like you're talking about for the automotive shortages or a memory fab, again, it's how much money can I make for each wafer that gets produced. Now, one of the things about HBM. you know, HBM in the way it's configured, it just doesn't pack anywhere near as many memory bits per millimeter squared of silicon. And, you know, we can get into the reasons why that's the case.

but you can think of it as it's less efficient in terms of bits of silicon but it's highly profitable so as you shift your production over to doing hbm you have to crank up more and more wafers just to get the total number of terabytes or whatever somebody needs in their system to crank that out. So the efficiency is not there on HBM. The reason for that, I started to talk about that, but the reason really is because you have to wrap a lot of logic around HBM.

to get the performance that GPUs need. And I think that's kind of the key piece. There was a number of different things that were tried. Some people might remember Micron had something called the hybrid memory cube a while back. And that was another case of putting logic in front of lots of stacks of DRAM chips, and they weren't.

DDR chips, they were custom designed, but that one didn't work out. That was a little too abstracted. HBM was kind of the brute force approach of just giving that huge fire hose of data. and giving that to the GPUs. I mean, in memory, you can go broad or you can go deep. You can't really go both at the same time.

And HBM maximizes going broad, giving the widest possible interface, but you can't get the depth. So that's one of the drawbacks of HBM, and that's why it tends to consume a lot of wafers. to get the capacities that these GPUs need. Let me dive in a little bit more to just kind of get around mental model. If there's folks that don't really understand, hey, how's HPM different? And why is it so important?

Think about, especially when fintech really came to the forefront in trading. Well, what was one of the things was, well, how close can I locate my servers to the Wall Street servers? Yes. You know, can I put them on the other side of the wall? And I actually had someone who's an engineer that specializes in building that sort of stuff on the show before. And it's fascinating. He just he wants the highest clock chip, the closest to Wall Street. You know, that's what they want.

Yeah, and I think the analogy here for HPM is really good because in the case of GPU, a GPU is just a beast in terms of how much data it needs to feed it. And we'll go into that a little bit more later. How is that constructed and why does that matter? But in the case of HBM, what you're doing is think of that wall.

On the other side is the Wall Street server. Here's your server. You're punching as many holes through that wall as you possibly can to put as wide a bus into there as you can. And you're limited just by that area of wall that you have. to go through. So again, HVM is that brute force approach is just putting as many data bus paths into that other side. And then you just want that run to be as short as possible.

So that's why you see such tight integration of HBM with GPUs. They have to really be co-designed. And that's where, you know, SK Hynix has done just a phenomenal job. And I understand that one because if we kind of look at the different players, Micron, SK Hynix, and Samsung when it comes to HBM.

SK Hynix, even going back to DDR generation, was pretty famous for, I'll just do whatever Intel tells me to do. And they would stay very much in sync with the CPU design teams, the memory controller design teams inside Intel. all the way through the validation, characterization, etc. So when it comes to HBM, I don't know personally, but I would imagine they followed exactly the same script.

with NVIDIA, which was, hey, NVIDIA, just tell me what you need and we'll do it. I'm focusing so much on HBM because it's just changed so much as a business. And now that third leg of the stool. which had just been same as client, maybe a little bit different than client, but DDR, which you can use for client, you can use for data center.

Now Data Center is using a unique chip, and that's HBM. And what you need to do to be a successful player in that is very different from doing a great job at client. And I think, yeah, this kind of gets us into the first big subject I want to get into with you, which is... At Capella University, learning online doesn't mean learning alone. You'll get support from people who care about your success, like your enrollment specialist.

who gets to know you and the goals you'd like to achieve. You'll also get a designated academic coach who's with you throughout your entire program. Plus, career coaches are available to help you navigate your professional goals. A different future. is closer than you think with Capella University. Learn more at capella.edu. Ever see an idea so clearly in your head but struggle to make it real? We've all been there.

With Wix, you can build a website for your business just the way you pictured it. Get the best of both worlds with AI and drag-and-drop tools all in one place. Stay in your creative flow and use AI when you want, how you want. Create a website your way. Try it out at Wix.com I had that dream again.

My small business needs IT help, but I don't use LinkedIn. And I hire wrong. So then I'm doing IT. And when I go to plug the servers in, they become sentient. And they won't let us access our network unless we forward their chain emails and memes to more and more and more people. And then...

I wake up. Don't let hiring nightmares ruin your dreams. LinkedIn Jobs' new AI assistant finds and invites best-fit candidates to apply so you can discover talent you may have missed. Get 50% off your first job post at linkedin.com slash Pandora offer. Terms and conditions apply. You know, like what's causing all of this and the misconceptions out there, because there are, you know, understandably.

to a large degree, gamers. And a lot of people, I'm sure, are mad about these RAM shortages. And I've noticed just in the comments and some videos I've put out or like, people are just getting mad if you bring up pretty much anything. But I think a lot of this rage is misdirected.

it really seems like if we're really breaking it down you know like you said you said they used to do whatever intel told them now they who's the new big player nvidia they now do basically whatever nvidia tells them nvidia tells them to make hbm hbm takes up lots of wafers those wafers are not going somewhere else yeah yeah

Those wafers are not being used to make the stuff we would put in any of the things you would use to game anymore. And because of that, that's really what's causing the shortage. And I've done a bunch of reports on like open AI and apparently, you know, buying up 40 percent of the global supply with these surprise deals.

hoarding them in warehouses, everyone freaking out, everyone buying up everything they could get. I've even heard reports of a lot of these AI companies are now literally walking into Best Buys and micro centers and trying to buy up $10,000. I'm like, what? I had to ask like a dozen people about that before I ran with that, too, because I'm like, really? They're really just walking in and buying it. That sounds like desperation. Yeah. What are they buying? DDR4, DDR5?

That's a very good question. One of the contributors at Moore's Law said was like, oh, Tom, is it ECC memory? And it was like any RAM kit over 64 gigabytes, suggesting that it's in case they need it in the future, or maybe they have the drivers for... The RAM sticks of ECC and they could repackage the modules. I don't like you say, I think it's just desperation. They're just doing it to be safe. The only thing I've seen supply wise is.

And I did some work for them years ago when they were getting off the ground with CXMT in China. And a very fun company to work for, by the way. That was definitely helping them build a plane while flying it. Good times. Now as a U.S. citizen, can't work on those things anymore. You know, CXMT has kind of come up. They were years behind. And then most recently, they're considered to be about one year behind the leaders. But they just stopped production of DDR4 because they were, you know.

had a fair amount of supply of DDR4. Now, when I say a fair amount, they were like 8% to 10% of the world's supply. But... So in the DRAM business, commodity business like that, even just a 2%, 1% change in supply can have a lot of ripples. Because it's about the bottleneck, right? Do you have enough or not? If you have 10% less supply than the world needs every month, that can... raise prices 50%.

Yeah, yeah. It becomes a huge problem as time goes by. So that's the only one I've seen as far as supply changes other than this mix change that we spend a lot of time talking about and kind of the mix going to HBM instead of other things. But also on the other side, yeah.

uh supplying in in china itself uh 6mt you know that ddr4 they stopped producing it to climb the ladder and now they're producing more ddr5 or starting to produce more ddr5 and most of their production if not all their production gets consumed inside china but china itself is 30 percent of DRAM consumption both for internal and also for re-export

So there's a lot that can go on there. And that's another thing that I've seen that's unrelated to the rise of AI, but another shift in going from DDR4 to DDR5. That's a little more, let's say... Historically, that's a little more standard that as we go through these transitions, probably about every 10 years that there's a big transition. And DDR4 to DDR5 certainly was a big transition. Obviously, they're not compatible on the same bus.

They're quite different in the bus architecture, et cetera. So you haven't been able to bridge that. And then I saw even in one of the questions, one of the comments was there was this idea that, okay, for the data center, DDR4 to DDR5 will use CXL as an abstraction, which will allow us to supply RAM no matter what configuration was there. That hasn't happened.

I mean, this idea of having kind of a memory server or memory, well, in some cases it's called just a bunch of DRAM, which is after just a bunch of disks or just a bunch of flash JBODs. where you just have a sled full of whatever ram you can get, and then you abstract it enough. And that, like I said, that just hasn't taken off. And I think that also has... meant that people have to lock into.

exactly the version, the type of memory, the type of DRAM. Because I've seen people proposing lots of interesting Frankenstein solutions. And I mean, I'm sure to a degree, like you could take laptop DDR5, remove the chips, put it on a desktop. ddr5 stick that you can do but this idea of like forcing ddr4 into a future product it's unsatisfying but no they really need what they need that's right I think you use well said. And I think that's, you know, that's.

You're kind of locked into a particular solution and there's not a lot of mix and match you can do. But let's go back to folks walking into a store and buying up as much RAM as possible. I mean, that's when you shared that. That reminded me of there was a time in the 90s. I'll give an example. You know, we came to work one day and everybody tries to turn on their workstation and nobody's workstation booted up because somebody had come in and.

broken into the building and stole all the ram that was in the workstations oh my god yep the workstations just took the ram okay just took the ram yeah and there was a thing around silicon valley where people were hijacking the trucks

that were full of modules DRAM modules what year was this this was in like the mid 90s okay early to mid 90s so you know when you We're not quite as bad as hijacking trucks of RAM yet, but there have been times like that in the past where there have been severe shortages where people can't ship.

systems because they can't get enough memory and that that's what you're talking about do you remember how much prices spike like i don't remember i was in the mid 90s i was four years old but you know yeah Not to make you feel old, but I'm saying I do remember, though, when I was that little, the PC my parents would buy, the one PC we all shared was like... you know i don't know like two three four thousand dollars and today they can get something better for 500 well better for

10 bucks but they can get something equivalent to the status of that one now for like a thousand bucks i i'm wondering like how what was the percentage ram went up then did it go from like a hundred dollars that was like within it was a short spike and it was like

doubling and tripling in a matter of weeks, you know, when you just look. That sounds kind of similar to now then. Yeah. Yeah. Real shock. And those kinds of things can come through. And what's funny is you just wait a couple of years and then.

You can't give it away. I mean, there's too much. And, you know, for whatever reason, the keeping the supply in balance. Now, one of the things. And thinking back, not at that time, but a little bit earlier, if we talk about the 1980s into the 90s, at one point, I think when I first joined Micron, or shortly before I joined Micron,

Micron was number nine in the memory business. Think about that for a minute. Now, there were companies beneath them that were producing DRAM. So, you know, there was a time when you had 20, 30 different companies all producing DRAM.

And that created a lot of supply chain dynamics because of overproduction, underproduction, back and forth, companies getting out of the business, etc., etc. So by really narrowing down to three major companies producing DRAM now, we haven't seen for a while, we haven't seen these kind of supply shocks. Because you just have three producers and they know their business extremely well and they know how to plan production, bring on production, etc. But now...

Again, back to that whole point with HBM, we see a warping of this whole ecosystem. And so that's what's unusual to me now. So it's not supply chain from, you know, we talked a little bit about maybe CXMT around the margins causing some trouble, but it's really that shift in mix that's giving a big... making a big difference. I did see one of the questions too, you know, should we expect that this affects NAND and SSD pricing? And the answer is yes.

It will because those same fabs, it's not very easy. And the equipment mix is different between DRAM and NAND. And for the most part, these guys now, the fabs are dedicated to either DRAM or NAND. But as these things kind of ripple through the industry, you could start to see NAND capacity coming offline and being converted to DRAM capacity for some period of time. So, yeah, I mean, that's, it is. likely that we're going to see some ripple effect on NAND and SSD pricing because of this.

Transpose 1195 writes, and it says, I recall reading about the boom and bust nature of the memory market in the glory days of Tom's Harbor and Anantech 20 years ago. What makes the industry so cyclical and have those forces changed over the years? do memory producers try to shield themselves from this volatility yeah so so again back in the day it was it has changed a lot and i i talked to those points of just number of producers

better control of supply chain, et cetera, that's gone on. So that's kind of dampened it out. There's that old joke about the bear and I only need to run faster than you. I don't need to run faster than the bear. That joke used to be told around Micron because I remember when I was there and going through a bus cycle, and it's like, well, how many quarters of cash do we have to stay alive? And it was like, well, we've got about three quarters of cash, literally.

And then we don't know what we're going to do. So the market better turn around before then. So sometimes that's the case that it's these memory producers do have to look at their whole financial situation. These guys are a lot bigger now. Micron itself is a much bigger entity than it was decades ago. Of course, Samsung is part of a very large conglomerate. SK Hynix has such a unique position like we've talked about.

But I don't think it's going to really be the financial pressures that used to be in the past, which is what your reader is talking about a little bit, that there were literal financial pressures.

During one of those down cycles about 20 years ago, and it was when NAND and SSDs were starting to have some impact on the hard disk drive business, Seagate came in and said, we're going to buy Micron. And you look at the market caps today, right? I mean, exactly. They're reverse, right? I mean, Seagate's a much smaller company than Micron these days. So those were the kind of things that went on in the past.

Now these companies are just kind of the behemoth, almost too big to fail type of mindset in terms of their stability. All right. So... If someone were to ask you, though, like just point blank, what's going on with RAM? What caused this shortage? What would be your answer? Yeah. And I hit it right at the top. It's the shift to HBM. and the AI data center demand for HBM. You just have to have it. And then I talked about it a little bit, just the efficiency.

The efficiency gigabits per millimeter squared is just not the same as you're going to get. And so as you're producing wafers, you're producing less bits of DRAM on HBM than you would on some of these other types. So that's the biggest thing that's changing. And then everything else to me is ripple effects. Like everything else, like a company buying up this, this company doing that. You're saying macro, macro level. It's this explosion of HBM.

And the AI boom, if we really want to blame some, like the big picture. Going from a billion dollar to a hundred billion a year business. I mean, that's a big change in the DRAM business. And SK Hynix being well positioned for that and putting a big hurt on Samsung as a result, that's a big shift in leadership as well. So that to me is the biggest trend.

Even at what used to be called Flash Memory Summit, but FMS during this past summer, HPM was all a topic. I mentioned SanDisk a few minutes ago. Great to see them again. Of course, Sandus got up there and talked about doing instead of HBM, but doing HBM with Flash. Of course, you know, you talk about Frankenstein type. This is the Frankenstein type. So when you see these kind of things. That's the big industry trend that everybody's trying to get onto.

Nothing writes in and says, Tom, thanks for your critical and well thought out content recently. I've listened to more slides for years, but it was your reporting on crucial and that being killed by Micron. That was a rational take rather than spouting rage. That made me want to join the Patreon. And I've seen that. They're like...

That was a thing that flew under the radar for me. Sometimes I'll jump on the Steam Machine from Valve. It's like, ooh, this could compete with PlayStation and Xbox. It's a new entrance into the console market. I bet people are interested. I was correct on that. thinking that would be a trend that I should definitely cover but like when Micron ended Crucial I was like oh I didn't like Samsung doesn't make their own memory

direct consumers. No, they didn't. That's an interesting story, too. I mean, Crucial was... So, look... It's Kingston's world, right? I mean, they just dominate in this space. And I saw one of the comments saying, what will happen with Micron pulling back on Crucial? Yeah, it just means Kingston gets more space.

Kingston actually said they were going to stop buying any Micron DRAM to put in their modules because Micron started Crucial. And it's like, really? Really? I mean, it didn't happen, but...

It kind of showed that, you know, again, Hynix and Samsung weren't competing with Kingston in their own module business at the end. I think Crucial was a nice way for Micron to just kind of... see where the customers are at you know see explore that consumer space explore the gamer space and they made some interesting products over time in that but in terms of where that lands and importance for micron it's just not that important

I mean, they've got bigger things to do. Following Jensen Huang around and, you know, what's NVIDIA up to is much more important for Sanjay Marocha to be interested in right now. So, I mean, that's the kind of thing that they have to stay on top of. rather than what's going on in the end markets. So I think Crucial was a nice little, let's say, side business for Micron and allowed them that insight. But it never was going to be that impactful for them financially. Right. And because it's not.

impactful for them financially it's not really like this giant impact on the market either right i mean like g scale and mushkin buy up modules from samsung and and micron and sk hynix and that's who's mostly making the ram that you buy

to put in your gaming desktop. Crucial going away is going to have effectively no effect. It's a channel. Crucial is a channel, effectively. It's just another channel where you could buy pretty much the same thing, like you're pointing out, that the other guys can... provide to you and the net ram chips on the market people can still just buy that from micron probably in the future if that becomes more lucrative it'll be the same net amount of supply to gamers nothing changes i agree with that

Okay. I just, and I think it's important because I just, I don't think people get like, it's Ram is like water. and there's all these that's poured into all of our systems to power it like a steam machine if you like steam look at that steam machine and the water can flow from about a dozen other brands that buy from three big companies it's not the big companies like

How much RAM they make and how much they can supply is what will affect when these shortages end. It's not going to be if one brand does. And I don't know this business very well. The only thing is if they're doing something very unique with the module design itself.

But if it's in the DRAM chip, yeah, like you said. Yeah, and Crucial tended to be very – basic ram for gaming like g skill or sometimes kingston will be the ones with the most overclocked ram or the flashiest stuff right not crucial right um This piece of content is brought to you by cdkeyoffer.com. Believe it or not, what you are looking at is my co-host Dan putting together his current Zen 4 desktop.

almost exactly a year ago of course that code for windows 11 came from cdkeyoffer.com and we've been using cdkeyoffer.com at moore's laws for a very long time i use it for my office software on both my desktop and my laptop

and that's because they're reliable they've always been there for us and they offer reasonable prices for microsoft products that are usually set in my opinion at monopolistic pricing and you know they're actually having a christmas super sale right now so if you're looking to get the best pricing

on microsoft software or games or other products on their website as well please show support for a company that's shown support for the moore's law is dead team and community for years by using offer code broken silicon to get 25 off all microsoft The holidays are here, and that means it's the most wonderful time of the year to save with Rakuten. Use Rakuten to stack cash back from your favorite stores on top of holiday sales. That's savings on savings.

with Rakuten you can get cash back on gifts for everyone on your list from toys for the kids to kitchen gear for the person who loves to cook to electronics for everyone you can even save on something for yourself just shop the stores you Love and cash back is automatically added to your account and you can get paid with gift cards, PayPal or check or eligible American express card members can even choose to earn membership rewards instead of cash back. It's truly.

a no-brainer. Join for free today and get a new member bonus after minimum qualifying purchases. Just go to Rakuten.com, download the app, or install the browser extension. That's R-A-K-U-T-E-N. Terms and conditions apply. Your business is one of a kind, so your website should be too. With Wix, it's easy. Almost too easy to create a website that's perfectly yours.

Just tell AI what kind of site you want to build or choose from thousands of templates. Change whatever you want, whenever you want, and get everything you need to start running your business your way. No matter what you sell or what you aspire to be, you can do it all. yourself on Wix.

All right. Melodic Warrior writes in and says, when I first heard about the wafer deal OpenAI secured with SK Hynix and Samsung, I was literally trying to pick my jaw up off the floor. I'm still baffled at how OpenAI pulled this off, though. How would it have been possible for them to get to. separate deals with both Samsung and SK Hynix without SK and Samsung knowing about each other's massive deal and thus win the ink dry and securing 40% of the RAM market overnight. Color me skeptical.

You know, OpenAI is very good at announcing deals. And I would expect that there's a lot of outs in those deals. would not expect them to be locked down. So, you know, I saw another comment which was talking about, you know, DRAM pricing and deals and, you know, five-year deals, six-year deals versus three to six to nine months. DRAM market can change so quickly that historically nobody wants to be locked into a long-term agreement. You do it.

You do it when things are tight. You say, I need a long-term agreement. I'm going to go sign it. And then the market softens and everybody goes back. Yeah, I want to renegotiate my deal. So I'm pretty skeptical of that's really locking things down. Let me give another analogy, right, was Pat Gelsinger at Intel, you know, rides back in as CEO.

announces all these new fabs he's going to be building around the world for Intel Foundry. Yes, former automotive workers in Ohio will now be chip engineers, apparently. Yeah, how many of those will be built? The answer is zero, right? Because it's a heck of a lot easier to do a press release and build a shell and stuff like that. So I'm kind of skeptical on LTAs, long-term agreements on DRAM. Yes, people sign them. Yes, they exist. They rarely work out the way that...

was initially in the initial document. Let's just put it that way. So now you've like really piqued my interest, actually. You seem to be suggesting I would, or if I may extrapolate what you're saying. It's mostly panic then right now. I mean, yes, they signed these deals for 40 percent of global supply and yada, yada, yada, all the other stuff. But at the end of the day, were these companies. Yeah. When you look in the rearview mirror.

A couple of years from now, it's going to be like, yeah, that isn't really what happened. Maybe we all shouldn't have panicked this much. And if they and if Samsung and SK Hynix want to sell to other people, they will find a way to do it without. The second open AI isn't the most lucrative person to sell to. You know, I'm going to go off on a bit of a tangent here, but you brought up the automotive industry finding out all of a sudden they couldn't buy chips to ship cars.

damn, I can't ship a 50, $60,000 car because I can't get a $2 part. And literally that was what was happening. On the other side of that, the foundries and I've been close to the Global Foundries guys, worked there for a number of years. The Global Foundries guys, let's talk to their SVP that handled all their contracts of partnerships. He's like, this is the greatest time in my life. He said,

I've got folks stacked up and we're signing long-term agreements. And he was even insisting that they buy him fab equipment and put it on his floor that he could run. In order for them to just buy chips, just buy wafers. And a lot of guys like companies are like, no, I'm not going to do that. And it's like, yeah, I've got people behind you in line. So please move to the end of the line.

So these things happen, and that was a panic situation. How much of those long-term agreements, how many of those long-term agreements, which were five-year-plus agreements, are still in force? Probably zero. Right, because that five-year would be ending next year. Two years go by, you get this impulse, it comes back down, and the customers go, you know what? I don't like that long-term agreement anymore. I want out. And then in order to keep the customer...

You strike the, you kill the agreement. So, or renegotiate. In some cases, you pay a little money to get out or something like that. So, yeah, I'm skeptical. It doesn't make sense for a cyclical business. You're exactly right. It's panic. You put something like this in place. Sam Altman is extremely good at capturing interest and putting... You know, press releases out there. No, no. I mean, he's trying to just show flex muscles and show he's dominant and everybody else has to.

chase to keep up with them if they also panic which i doubt they will um you know they're they're i i don't see it playing out that way so And here's another thing that's weird about the deal that at least as far as I'm aware, and granted, I'm very far from an expert on this, but.

It they didn't buy finished sticks or HBM from Samsung or SK hynix. They bought the wafers and I guess they're hoarding them somewhere before they decided to do that. Yeah, I saw that a little bit and you can do that. You can park it. what's called wafer bank. Dye bank, wafer bank. These are big nitrogen purge chambers that hold the wafers. Seems like an expensive way to hoard things.

It's not too bad. I mean, all companies have this capability. It's just, you know, for OpenAI to set up a bunch of nitrogen chambers, I mean. They probably spend more on electricity in a day than it would cost to set those things up. So I don't think that's that big a deal. That makes sense, too, in that you don't always know the module configuration you're going to need.

And so that makes sense to kind of hold it at that point. Like which one of their products, which types of their servers they'll build out, they can save it, get a finishing company to finish it in. That's right. Yeah. Yeah. You can get the module company to build what you need. I think what's also interesting is you can resell. There is a buy and resell that can occur.

You know, I saw one of the questions in here. I don't remember which. Yeah, I was looking for it too. It's like basically I think what it comes down to is when. And I have two ways I would go with this, right? Number one, if this AI bubble pops, if people realize they bought up too much RAM, I mean, obviously then they will just sell off the RAM they've hoarded.

That's number one. Number two, do you think there is I'm just throwing this idea out there. Do you think there's any chance some of this as well is like open AI going? We have the HBM that Nvidia needs to finish their AI chips. So now when Nvidia has so much capacity. Part of a deal could be, well, we can give you the RAM you need. Well, you're short on it right now if you give us that capacity for our servers. Because you have to admit, like just...

That's kind of like a common sense bargaining chip they now have with NVIDIA and AMD. Yeah, but you have this from here that Micron, Samsung, and SK Hynix would be willing to let OpenAI kind of step into their role. and take their leverage away and that would surprise me because that's effectively what you're saying is that the DRAM companies have now handed their negotiating leverage over Jensen Wang to open AI.

And I mean, definitely not long term. Right. But I mean, short term, isn't that just true? I suppose that's possible. But yeah, it seems again, it seems unlikely to me because. they're going to keep such tight relationship with nvidia that you know they're these guys are constantly talking what do you need what what can we do for you um there's from engineering level all the way through executive level, that kind of interchange is going on. So I guess it's possible.

I guess what you're saying too, right, is if they did try this, Samsung and SK Hynix would take notice and probably immediately be renegotiating their contracts then. Or... Suddenly, yeah, yeah, or failure to deliver or et cetera, et cetera. Again, because none of those memory companies want to give their leverage away. Right. They want to retain it, and they're not just going to let some other third party kind of soak that up.

I think there was some attempt at this, but I think back in time, there was a time when Dell and HP would do the same kind of things in server memory. Some of these strategies work. over a short time horizon, months to quarters, but not really longer than that. So again, if OpenAI is gaining some negotiating leverage out of this, it's a short-term play.

mm-hmm isn't that plausible though like just in general um that every like a lot of this going on with this ai build out especially with open ai it seems is all short-term plays though you know this is just my opinion feels like this ai market or bubble if you will i will i will call it a bubble it seems like it's

but increasingly turning into short-term plays and less long-term plays. Like how can we get the most soonest get ahead? I have to be faster than you. I don't need to be faster than you, not the fastest person for the bear to eat the person behind me. But the circular Ponzi is lasting a lot longer than any of us expected. I mean, that seems to be the case. Well, I never claim to be sure of when anything ends because you never really know, you know.

The market can remain irrational longer than you can stay in it. So but it's something real, too, because I think. You know, one of the things that I always thought was going to provide gravity was the CFOs at these companies going, hey, what am I getting for all this investment? And so it's an arms race. But the thing that's different.

with this one is this is a nation state arms race. You know, when you see Nvidia chips, that's the narrative that they've sold successfully so far. No, I think it's, I think it's real. Because when you see NVIDIA chips being used as an export capability or, hey, Saudi Arabia wants this many GPUs for their... their newest data centers are going to be delivering. I mean, that also is something different because now you're outside the normal economics that companies would interact under.

And as a nation state arms race, hey, we've got to beat China. We've got to keep going. That's something different. So I... Don't sell it short. I think this is going to be longer lasting than any of us might have thought, you know, two, three years ago. It still seems to be that there's a lot of momentum and a lot of possibility in it.

Just don't get too hung up on the press releases that, you know, we're going to buy. What was Sam Altman going to do? I mean, he was going to invest a trillion dollars into chip development.

He's looking for somebody in Saudi Arabia to build him fabs, and it's like, yeah, buddy. No, no. Where do the people that know how to do that come from, too? It's not just buying land or... putting in machines or any of that you know look at tsmc struggles just to get up and running in arizona i mean that's been hard because it's where do the people come from and how do you get them lined up behind that certain objective

So, no, back to your key point. OpenAI is playing a short-term game. And, yes, some of this is good headline fodder. Like I said, a while back, color me skeptical. Yeah. And just to, you know, put into perspective, like where I'm coming from, I think the same thing as you, or at least I did. We'll see if I even think that's what will. pop this bubble anymore it's like i thought you know i think people are going to spend a billion dollars at a company on ai they'll fire people and then they'll go

oh, we need to rehire people and we've only made back a million dollars. We haven't made back a billion. Like definitely like AIs, again, everyone, this is where everyone thinks I'm like.

some luddite it's like no i use ai every day it's here to stay i'm not saying it's going away but could it be overvalued at this point i'm in the same place as you it's just but i was especially convinced by this other argument which really kind of dovetails with what you mean you just said to be honest but um ed elson uh from the professor galloway markets podcast was just on and he said that his opinion is it's all narrative at this point

you know and so the thing that changes it is narrative when does the narrative change because like when you look at palantir's earnings multiple i don't think it's just like short term they're making that back and it's

Whenever the public looks at it differently is when they might start scaling back how much spending is going into this. And to think that we know what will really change the narrative is probably naive or egotistical. Yeah, I guess I can understand that and buy some of that. But I think AI is already proving to be a lot more impactful. I think there's a lot of these tech companies have been doing layoffs.

There's also kind of a... They've been rehiring. They've been starting to rehire, though, too. A little bit. A little bit. I was going to contrast it. Let me back up for a second. I was going to contrast it with Zuckerberg's play in the metaverse. Right? I mean, he just... folded that up and how much did he spend on it well the number they came up with is like 77 billion which compared to ai is is still nothing but nation states weren't interested in the metaverse

You know, it just didn't have any traction. And again, Trump traveling around with Jensen doing deals, allegedly doing deals. Press releases at a minute. Press releases for sure, but that's different. I mean, that's something that's not been the case for boom and bust cycles around the memory business or around data center growth, et cetera. So we got new players on the field. Don't you think that that's kind of like what...

A way China's kind of trying to fight this as well is kind of like how they dumped steel on the market for cheap. They're dumping AI models on the market for free constantly from tons of companies.

that don't require all of these chips. Could that be part of two? Again, if the narrative were to change, people take a step back and go, maybe we don't need all of this infrastructure. I think that's, yeah, that's a reasonable point. And I think there's some of that is yes. I mean, that's... a way for them to play it but i think also deep seek is an interesting one because it's it's clever engineering and it's clever engineering that comes from constraints i can't get the number of gpus i want

I can't throw the whole fire hose of data at it. So what can I do in my algorithms? How can I do training different? How can I adapt with these constraints? I'm a little more respectful of it in terms of his clever engineering. Oh, I don't mean to put it down, you know, like. I wasn't saying that, but I was saying, I think, of course, you know, if I'm going to put myself Chinese government side, yeah, I'm going to make the most of this. Right. And this is.

little company over here um interesting too they're high frequency trading guys you know which is move fast break things etc uh they figured out how to do it And I think it was a great wake-up call for Google in particular. It seems like Google's... kind of back on the playing field now you know it seemed like they were out of it for a while yeah it is better their their new gemini definitely works better i can say that and they're starting to make their own chips as well

Yeah. So I think, you know, I think David Saxman at this point a couple of weeks ago, just about, you know, for the U.S., there's kind of this concern. Well, there was kind of this concern that, hey, is the U.S. itself going to backstop OpenAI? and financially because you know sam outland's out there shaking the can again for donations um david sachs is like hey we have access in this country to multiple frontier models i don't think that's really going to be the thing that

cause us to fall short. China, same thing. You know, I think there's multiple teams working on different things. The constraints are just different. And so they're going to come up with solutions. But yeah, I mean, look, that's the... That's the arms race that we're in. And that's what's, you know, go back to our key point. That's what's warping the memory business. The holidays are here, and that means it's the most wonderful time of the year to save with Rakuten.

Use Rakuten to stack cash back from your favorite stores on top of holiday sales. That's savings on savings. With Rakuten, you can get cash back on gifts for everyone on your list. From toys for the kids to kitchen gear for the person who loves to cook to electronics for everyone. You can even save on something for yourself. Just shop the stores you love.

Love and cashback is automatically added to your account and you can get paid with gift cards, PayPal, or check. Or eligible American Express card members can even choose to earn membership rewards instead of cashback. It's truly a no-brainer. Join for free today and get a new member bonus after minimum qualifying purchases. Just go to Rakuten.com, download the app, or install the browser extension. That's R-A-K-U-T-E-N.

Terms and conditions apply. At Capella University, learning online doesn't mean learning alone. You'll get support from people who care about your success, like your enrollment specialist who gets to know you. You'll also get a designated academic coach who's with you throughout your entire program. Plus, career coaches are available to help you navigate your professional goals.

A different future is closer than you think with Capella University. Learn more at capella.edu. Wix. It's where website creation meets AI and where your boldest ideas become real. It just takes one platform to build a site that looks great and does everything you need it to. And it just takes one person, you, to start taking care of business like a 10-person team.

With AI tools for creating an entire website from scratch or testing new ways to make money, Wix is there with you from day one. Try it out now at Wix.com.

uh moist queasy bartlett lake soup 999 this was a person you had quite a few questions about their name which i have questions yes i have questions but um people in the comments can explain to everyone else what this name means because long-term Wars, a lot of dead fans will get all of the jokes, but three years or so into AI and it's left a wake of victims, energy, water, and hardware, which now includes Ram, not just GPUs.

What do you think will be next? Capacitors, PCBs or something or other things no one knows about, but we'll notice when it's gone. Actually, if I may say, I think energy is still not quite hit people yet. I think that's.

Yeah, no, I think that's a good one. But let me stay in my own lane on hardware. I think there's two things. I think one we talked about already, which is I think there's going to be a ripple to NAND and SSDs. And that's just because the memory companies themselves. I mean, where do you invest? These companies like to keep a pretty decent balance between NAND and DRAM, but DRAM's really profitable right now. So that's getting a lot of the focus and a lot of the funding.

um different topic too by the way tom we may come back around to is technology wise i mean what can also move DRAM forward so Maybe placeholder on that. We'll come back and talk about it. The other one that I'm hearing about is TSMC themselves is kicking customers out of like their 40 nanometer fab and fabs. And then multiple of these. Well, that's kind of an older technology. Why? Why are you doing that? In this case, the business I'm pretty aware of is microcontrollers. Why are you doing that?

And the answer is that I can come up with is because they need those 40 nanometer fabs to do some of the silicon substrates, which help match up HBM and GPUs. So again, it's ripple. It's another ripple that comes out. So good. Microcontrollers start getting some shortages because these guys have to find new fabs, et cetera. And TSMC is making a very logical decision there that.

hey, feeding this GPU and HBM beast where I need silicon substrates and I use my older fabs for that, that may cause some other shortages. So those are the two in the hardware space. nand ssds i think we can expect some price increases some changes there and then maybe something like uh SoCs and microcontrollers that are built in some of these older process nodes just because those fabs are being reused to, again, feed this beast.

Dark Side of the Force writes in and says, it seems like direct consumer RAM pricing has gone straight up. When do you think people will see similar increases in more consumer products, phones, tablets, but then eventually maybe smart TVs and even fridges? What do you think is the extent?

of the effect of it would we see complete supply disruption to no products or somehow will manufacturers find a way to keep things going just at a higher price and because what you just talked about 40 nanometer i'm thinking of like budget smart speakers or smart fridges and stuff. What I've said is...

I'd be more afraid that they just don't have a stockpile and they run out of capacity like what the automotive industry had during COVID. But I wonder what you think. Like what's the next thing? I think that automotive one we've touched on a couple of times. I think that was a real shock. And I think the automotive guys were particularly bad at planning their supply chains. I mean, they literally treated chips like rolled steel.

They can just get it anywhere, and you can't. You've got one supplier. Yeah, made to order. We don't need to hold any supply of it. So if we think about white good manufacturers making a washing machine, et cetera, et cetera. these guys have multiple supplies i mean they're better at managing their supply chain they're thin margin businesses they have to do it um unlike

What was the one? Oh, it was the Ford Ranger truck. They were going back to an older model because of, again, like a $2 part. They just, you know, they had to downscale. If I'm buying a washing machine and somebody can't get the certain 8-bit microcontroller they're used to getting, they're going to figure out how to buy a 16-bit or some off-brand. They'll just redesign it.

and just come up with something different. So those guys will be pretty flexible. But back to the core question, price increases after Christmas, right? I mean, we're doing Christmas shopping, et cetera. I keep telling people, get the PS5 now. Get it now. Get these things now. Yeah, I think it's one to two quarters, and then that's when things start to ripple through and affect the overall prices. I mean, it's been...

You know, tariffs have been interesting in not changing some of the pricing where it might have. Yet, though, that takes time. It's coming. Yeah, I think that's the right way to look at it. Well, so CarbonCry writes in and says, what are the usual lead times for deer in contracts? I think it's, what, around three to nine months. The OpenAI deal has a five-year lead time. Did the purchasing time...

I didn't realize the purchasing time horizons expand significantly, or are they always around three to six years? And I also want to follow up on his question by adding, I've had some contacts of mine in retail say that usually, yeah, they'll get like a three month. to six month or something lead time for if they need to buy a bunch of RAM to make PCs for gamers. And recently they're being told 13 months or 24 months. Like how real do you think that is?

is another question I have. And what did you see in previous shortages, like the jump in lead times? Yeah, you will see a push out. And again, I think just like the pricing, you'll see a doubling to tripling is... pretty typical so you know talking about kind of a when things are normal three to six months sounds you know that's about right um yes things can push out

The 13 to 24 months sounds like the old 52-week lead time when you didn't want somebody's business. You would just quote them and you'd say, yeah, 52-week lead time. And that was kind of the way of giving them the finger without turning down their business. But some of that could be real. That could just be, hey, I don't have visibility.

you know 13 months or 24 months i don't have visibility i don't want to lose you as a customer i want to keep things going so you know let's do something with the lead time and then like we talked about you know several minutes ago We can come back and renegotiate later. We can adjust later when both of us have better visibility. The thing that has caused a lot of disruption in the memory business in the past is double and triple ordering. Right.

Which I think is happening right now with some people. Yeah, it is. And you don't want customers doing that because you just, even as the supplier, you just completely lose visibility. And that's when you stop believing what you're hearing in the field and you just go on models that you developed internally.

And you just, there's no other way to put it because you can't trust the data coming in from the field and knowing that folks are placing orders here, there, and the other place. And they're going to cancel later. And that's going to lead to a crash. I mentioned it quite a while back, but 1% to 2% change in the supplier demand has big impacts. A lot of people have asked me like, again, usually very gamer focused here, like.

They'll say, what about the steam machine that launches early next year? And I'll go, yeah, it might be affected. Pretty bad timing, actually, on that one. But who knows? I would assume they would have stockpiled some RAM for the initial manufacturing months ago. So we just don't know the extent to what it's affected. But yes, I think it will be. affected but then they'll bring up like the playstation 6 the next gen xbox there's a few other things as well and it's like

My understanding, like the documents I've seen behind the scenes, the PS6 is supposed to be manufactured mid-2027. And based on that, I go... I don't think we really know if it will be affected yet. I would suppose it probably won't. But then there's also like the RTX 50 Super Series from NVIDIA, which was supposed to come out last year, then early next year. Now, apparently, they're telling their partners quarter three of 2020.

I think quarter three of 26 is just a placeholder, don't you? Don't you think all of this is just placeholders? Yes. Nobody knows. And I think that was one of, if you even look into Carbon Cry, was talking about the fabs.

Yeah, let me ask that now then. Carbon Cry writes in with a rather extensive... question and he says in 2026 sk hynix put two new fabs into full production one is in bring up right now micron opens a fab in 2027 sk hynix does as well it's first in their fab super cluster project then in 28 micron will open

Another fab. Samsung will open another fab after a rebuild in SK Hynix. Probably another fab in their supercluster. Then in 29, Micron opens their new New York fab. First module, but in capacity full fab. SK Edge, probably another one in their supercluster then. 2030 Micron opens a second module in New York. The total expansion is like eight to 10 new fabs over the next five years, like two a year at a minimum. And these are mega fabs too. These are not small little.

little deals you know this is these are big fabs and he says yeah recent articles like pc gamers memory crisis and sky high ram could run past 28 and sk hynix and samsung are now minimizing the risk of oversupply he says this is one example There's many others. Many people are being whipped into a frenzy over the evil DRAM barons refusing to expand. Isn't the industry not already expanding at a break net pace? Is there even enough tool production to sustain this rapid buildup?

Yeah, it's very, very thoughtful. So if all those fabs were to come into being and open, and we used the example earlier of Intel announcing a lot of fabs, but... None of those ever came online. But again, these are serious companies that are talking about everything. And Micron's listed several times there. Micron's definitely on a build out right now.

you know, trying to keep up on capacity. They're in third place in the DRAM business, and it's, you know, a challenge for them to keep up with, certainly Samsung. Samsung is... part of Samsung Electronics, part of Samsung Corporation. So there's deep pockets there that allow them to expand. So, you know, bottom line is, yes, there are a lot of fabs that are on the drawing books. And these are mega fabs that can move a lot of silicon. That's one of the factors. And if, if.

these fabs all do get built, then that'll help relieve some of this supply shortage. But it takes time, as pointed out here. And then even the tools, that was a very good point, the tool production to sustain a more rapid build-out. Yeah. I mean, the tool manufacturers, Applied Materials or Tokyo Electron, et cetera, there's only so many fab tools that they can build and produce at any given time. So there is an entire supply chain into build out of fabs.

And those guys are also competing with the expansion that's going on. We didn't talk too much about nation states getting into the fab business. So that's going on at the same time. Whipping people in a frenzy over evil DRAM barons? Yeah, I think there's a good headline there. Let's just put it that way. But the industry is looking to expand. The thing we haven't talked about is, yeah, but how much more DRAM gigabit per millimeter squared can you produce year over year? Historically...

That was 30% more every year. That was like clockwork. This has nothing to do with pricing, but you can think about cost down or, again, gigabit per millimeter squared. What can you put in a certain area? And it was just 30%, 30%, 30%. It was just a beautiful trend. That started to bottom out about 10 to 15 years ago. And probably.

It got as poor as about 10% a year of thinking of like cost reduction. Let's call it cost reduction. Cost reduction per year. And it just kind of bottomed out. I think they've gotten back on a better trend. Some of that was better embracing of EUV equipment and getting down into some pretty advanced nodes.

But I think that's still another challenge for the DRAM business is how to get that curve. Can they ever get back to 30% per year as a trend? I don't think so. But if they can get back towards 20% or something in that range. That starts to be pretty effective. The one thing, this is the placeholder I told you to set, the one thing I do see, DRAM has not, unlike NAND, adopted the trick of going 3D.

And this has been the opportunity for the longest time because NAND really kicked in and ran away from, you know, what you can do in hard disk drive or SSD. When they could go 3D. And now you're talking about gigabits per volume. Yeah. Millimeter cubed. DRAM is still area. It's just two-dimensional. Now, we talk about stacking chips, but that doesn't change the economics at all. You're just stacking. What we're talking about is how can I pack DRAM bits into the third dimension?

IEDM is coming up this week in San Francisco, and 3D DRAM continues to be a good topic. I think that that's something that within a handful of years here. somewhere towards the end of this decade, we're going to start to see 3D DRAM kick in. That's what gets the DRAM guys back close to maybe that 20% or maybe 30% per year is unlocking that third dimension and going for volume. How many bits can they pack in?

Now, you're going to see that, we talked about at the very beginning, the kind of three legs of the DRM business, client, mobile, and then server. You're going to see that in client first. You know, because that's the kind of business where you could... go a little bit slower speed and get away with it you could have some poor bit error rates and get away with it and kind of surround it with logic etc so you're not as stressed as you are on the power consumption that you are in mobile

or the performance that you are in data center. So I think that that's, in addition to the fab build-out, it's keep an eye on DRAM technology and can 3D DRAM come into being. All the companies are working on it. No question about that. And I've seen they get a little more serious. They're using some new materials to try and get it going. And it's a hard challenge because the transistor right now is built down into the base silicon.

And it's really hard to build good transistors up in space or good capacitors, for that matter, up in space. So that's the technical change that needs to take place. I think to really bust through beyond just fab expansion is also technology scaling and getting into 3D DRAM. And I mean, and I hope I didn't misunderstand what you were talking about here, but like that also allow for massively higher bandwidth, too, is right in like. Not necessarily.

It could be something because you might sacrifice, it might go the other way. You might sacrifice bandwidth because now, let's say, you know, think of I have a dirham bit in a cube instead of area. And then to get there, I have to take multiple paths to activate that cell and read out the data. So again, you'd probably sacrifice performance and you'd sacrifice bit error rate in order to just get volume.

Think of it like the early days of NAND SSDs, too, where, you know, you might jump to the next node. NAND is a pretty crappy memory overall. It's a bitter array. It's pretty bad. but you correct it with logic on the outside. So you might do something similar on a 3D DRAM. Yeah, so like how SATA SSDs got cheaper by just cramming in more bits. You're wondering if we could do that with what...

we would put into ram sticks and it might not be as fast or efficient, but a lot cheaper machinery, right? Could maybe make something they said earlier. It's breadth and depth. Yeah. And so you have a certain bandwidth, but now I want more depth. Well, I'll sacrifice the breadth. I'll sacrifice the bandwidth in order to get more depth and get more, you know.

How do we get out of 24 gigabits being, or 32 gigabits being the highest capacity DRAM chip you can get? You know, we've got to get much, much higher. You know, you think about NAND devices today, I mean, hundreds of gigabits to terabits on a single. single chip you know how can how can DRAM kind of get on that curve it's gotta adopt 3D

That's interesting, too, because I think what you would see, for example, I put out leaks recently about Zen 7. Just hypothetically, this is going to bring 32 cores to the standard desktop. They might even, apparently they're aiming for seven gigahertz. I mean, get ready for some really powerful CPUs.

That would probably need the fastest, best memory. But all the way at the bottom of the lineup, you're saying, where you don't need as much bandwidth, you could probably have a lot of cheaper laptops using this cheaper memory. That's right. That's exactly right. And that's the first place you'd see it. And then as it migrates through, again, that's now you're starting in your fab, you're starting to generate, you know, use the analogy of micron makes wafers. And if each wafer is you.

that much more DRAM bits because now I'm using the full volume, well, I've got more wafers left over to do the high-performance stuff, etc. This piece of content is brought to you by Silver Knight PCs, a disabled veteran owned and operated pre-built PC and parts company headquartered in North Carolina that ships to... fpos and apos now why would you choose silver knight pcs well let me tell you

First of all, they have some of the fairest pricing that I've ever seen. And I know they're trustworthy, too, because they're literally friends of the channel. In fact, you can check out an interview with the owner, Ted, on Broken Cellcon 113. So I stand by them. But it's not just...

about you know me liking them or them having fair pricing they have great quality as well they offer three year system warranties on parts and labor with paid shipping both ways in the continental united states and they also have per component warranties in the event you change

maybe a part in your system, but then another leftover part breaks, they will honor that as well. And, you know, most of their parts are the higher quality offerings as well. Like they don't cheap out on PSUs or cooling. They use the best stuff out there. Most of their power supplies, in fact, have five to ten. year warranties oh and by the way that three-year warranty i mentioned

it's transferable without any paperwork involved as long as someone has the order number someone has they can just send it to silver knight pcs if they're a new owner and they'll say yep this pc was sold this amount of time ago with these parts we will on

your warranty that means you can gift this to someone sell it in a couple years and tell the person you're selling it to hey this still has a year warranty left on all of its parts they also have a trading program as well if you want to upgrade sooner rather than later maybe send in a graphics card and get a better one back they also do that and finally they are doing a phenomenal tiered pre-built promotion right now where you can decide between a $1,000 $1,500 or $2,000 build and

all of those come with the high quality components and amazing warranties i've talked about and also you can get 10 off them right now with offer code broken silicon but you know what they also assured me they did plan ahead and they did stockpile dram months ago because they saw this coming. They are one of the companies that planned ahead a bit. But I suspect these things will still sell out very quickly with how good the deals are. So please...

Support Moore's Law is dead by checking out Silver Knight PCs through the link in the description below. And if you do, make sure you use offer code BROKENCELCON to get 10% off everything on their website. The holidays are here, and that means it's the most wonderful time of the year to save with Rakuten. Use Rakuten to stack cash back from your favorite stores on top of holiday sales. That's savings on savings.

With Rakuten, you can get cash back on gifts for everyone on your list. From toys for the kids to kitchen gear for the person who loves to cook to electronics for everyone. You can even save on something for yourself. Just shop the stores you love. Love and cashback is automatically added to your account and you can get paid with gift cards, PayPal or check or eligible American Express card members can even choose to earn membership rewards instead of cashback. It's truly.

a no-brainer. Join for free today and get a new member bonus after minimum qualifying purchases. Just go to Rakuten.com. Download the app or install the browser extension. That's R-A-K-U-T-E-N. Terms and conditions apply. Ever see an idea so clearly in your head but struggle to find the time to get it all done? Wix helps you go from, eh, I'll get to it, to done.

Build a full site just by describing your idea. Let an AI agent handle daily tasks, plan your next marketing campaign or help out customers so you can grow your business the way you want without it taking over your life. Try it out. at Wix.com. Now, I'm going to throw this one out there. I think I know your answer, but Nicholas Buckner writes and it says, do different countries need a strategic ram reserve similar to strategic oil reserves at the government or joint venture?

Or company level to help smooth out boom and bust cycles? Well, first of all, there usually are backup reserves of RAM. They were just pretty empty this cycle at companies. But, I mean... The thing about what you're describing, like fundamentally new technology to make it or evolutions of technology to make it cheaper and all these build outs happening at a breakneck pace. I think this is a very unsatisfying answer for gamers, really anyone.

who needs RAM right now. But the answer is actually there is more RAM coming. They are building out as quickly as possible. There will be new tech that makes it easier. And it is... Again, an unsatisfying answer, but short term people are actually kind of doing everything they can. And there's just going to be a shortage next year. Yeah. And the evil, evil DRAM barons, as we've turned them.

They're rational economic actors. You know, what they're doing makes sense financially. And... you know i mean i guess that's the nature of capitalism uh do we need nation states to get involved well i mean we did have a national cheese reserve for the longest time okay so the government has done some

crazy things in my mind in the past i think that was just to keep dairy farmers happy but that's another whole nother story um i'm not personally a fan of having the national bitcoin reserve either but you know i mean that might wind up happening um so no i don't i don't think that's an answer i don't think that getting governments involved in in stockpiling ram um and in ram you know we talked about keeping it in nitrogen cabinets etc

But it is perishable. It's something that as time goes by, there's better stuff that's coming out. The industry doesn't sit still. I don't know that oil... for the, for that matter. You know, if I sit on oil for 10 years, I believe that goes bad eventually. Yeah. Yeah. There's probably some sell by date, but you can probably hold on to oil for a pretty long time.

But I think in terms of DRAM, no, it's fairly perishable. And that's why we talked about it earlier. This is a lot of short-term strategies in trying to make sense of the supply shortages. Well, that's the thing, too. The more we talk, the more my opinion on like who to blame and it's like, well, all of these companies before these shortages, you know, they're not dumb. They were like.

I think we'll need a lot of RAM soon. And they were building out. They've already been planning to build out as much as possible. And so really, if you want to get mad at someone, it is the companies, I would say, acting the most irrational or short term. enforcing people to panic like that's really what it is if you want to be mad at somebody right now you know it's not the companies that are hey and not to say they've always been angels and every year but

I want to be clear, too. We're not saying that. It's just I don't think what's going on right now is like price fixing or hoarding the RAM at Samsung. Right now, they want to make as much RAM as possible. Ram executives, what was it, early 2000s? DRAM executives went to jail for collusion. I think it was Dell that brought the U.S. government into it. And, you know, we're talking about SVPs, etc.

that literally were locked up because it was alleged, and I don't know if it was sufficiently proven, but that they colluded on pricing. So they're not doing that now. You know, that was a lesson learned decades ago. And the guys who are running these companies know that history. So, look, Samsung and Micron and SK Hynix, there's no way.

They're running the risk of getting together in a room and saying, let's all raise prices together. Now, are they sitting back watching each other on a daily hourly basis to see what's going on with pricing? Absolutely. Are they going to take advantage of it? That's capitalism. Yes, they will. Surge pricing, right? I mean, you can rail about it, but they're rational economic actors, and this is something that they can go off and do.

Yeah, and just to bring it back to something I brought up before, like future products, you know, when people ask me how long will this shortage last, because again, they're worried about the next gen of NVIDIA, they're worried about the PlayStation 6, my answer is, you know...

Nvidia seems to be like a pretty smart company with pretty smart people working there. And if their placeholder is quarter three, 2026 for their series that was supposed to come out, honestly, quarter four of this year, my guess. is that that's when they think it will be over soonest. Like they're not sure it'll be over, but that's the soonest. That feels right to me. I mean, honestly, by the time...

Memory shortage is the narrative and reaches the zeitgeist. You're probably, again, peak fear. Right now, yeah. Yeah. Yeah, and I think the unwind is going to start to occur, is largely underway. And so you think it's unlikely things are this bad mid-27? Correct, I do. Okay. A lot of people want to hear that, by the way. I mean, obviously we don't have a crystal ball, but. Yes. Yeah. It would surprise me because, again, there's multiple factors in play here.

Something is going to change. It's hard to predict what, but yes, if we're at the evil, evil D-Ram Baron narrative. then we're quarters away from a pretty good unwind. So Colt Marshmallow writes, and it says, just a quick one from me. Obviously for consumer, the DRAM news is pretty devastating, but what is certain other industries?

Think about this like automotive or military. I can't imagine, you know, Ukraine is happy or Japan and Germany are happy if they have cars with all this tech in them now that they can't make. You seem to think a lot of that won't be as much of an issue, but I. I might have a defense engineer on soon to talk about this because it sounds like he does think there could be some issues here. Like, do you think, though, to a degree, the amount of short term surge buy up that's happened?

is this a national defense issue like security issue to a degree possibly so you know let's be honest there's a lot of stuff blowing up around the world and when it blows up you got to replace it um There's a buildup in European defense capability that's been a good trade over the last year that also is going on. So there will be shortages, but not to the same degree.

Now, I mentioned the specific example of a company like TSMC taking older fabs and redeploying them in order to feed the AI beast. That will have an impact. Now, I do see new fabs being proposed that are being built that will come online that are, you know, you need to think of an equivalent to older logic fabs.

So there are adjustments. So there is supply chain scrambling that's going on. So there will be some impact, but it's not anywhere near the level of impact we're talking about at the top line here.

of just memory impact overall because of the distortions from the AI buildup. So do you think like there's any degree to which they're like, Not even just the U.S., any government that touches like the manufacturing of RAM might want to sit down and go, hey, in the future, we might want to have oversight over some of the deals that are signed, because if you spike this, all of a sudden we couldn't make a thing during one month.

Or what do you think? I don't think so. Because honestly, the military, so defense industry, you know, if you think about that just on the logic side. Defense industry shifted years ago to buying a lot of FPGAs. And why? Because, first of all, the volume of what they buy is so small that it's not worth doing an ASIC.

You know, even if it's something you might produce on a missile for 10, 20 years. FPGAs are very flexible. You know, I can reprogram it and I can use it here. I can use it there. So that's on the logic side. On the memory side, there's... A lot of stuff is more embedded memory that's used, and then there's not as much external memory. They don't like using charge-based.

So a lot of aerospace and space and defense doesn't like using DRAM and NAND because they're charge-based memories. And in weapon systems, those can be disrupted with strong energy fields. and corrupted data so they like other types of memories and those tend to be very specialty tend to be produced in some very old fabs on some very old technologies but that are particularly robust. So again, defense, military, no. In terms of if you scale out and it's David Sachs, our AIs are talking about

hey, you know, I talk to Jensen all the time on GPUs, but now Jensen's telling me I've got a shortage of memory. Could you see him start to knock on Micron's door and say, hey, what can we do? What can we do to support you? What can we do so you build out your new fabs in New York faster? What can be done there? I think that's the kind of place where the government weighing in to just...

Just have that conversation, you know, that this is a shortage. It's affecting our national agenda on AI. I think that's much more likely than done in weapons systems. Moist Queasy Barlet Lake Soup999 writes in yet again and says, could TSMC or especially Intel with their unified capacity switch to making memory wafers for direct to consumer? No.

No. See, my gut answer was no, but I don't have enough experience. The equipment mix in a fab for logic and memory is so different. It's just so different. How can I give just one quick example? Tokyo Electron. Tokyo Electron, the oxidation chamber that you use for DRAM capacitors or for NAND.

is never used in, in logic wafers, you know, so the, the, the equipment's become too specialized specifically to memory technology. So no, even between, we talked about it earlier, even between DRAM and NAND, the capacity is not fungible. But between... DRMA Logic. And then historically, Logic guys, Foundry guys sometimes look at the memory business and say, oh, this would be a good fab filler for me when things are lean.

And then once they look into it, they always wind up going, no, I don't want to be in that business. No, thank you. So Jay Buffay writes in and his question, I'm just going to paraphrase, is basically about. being allowed to sell older manufacturing machines in Korea to Chinese markets so they can build out and upgrade because there's this cycling that goes on. But you kind of touched on this already, right? Like the answer is.

There are a bunch of companies scrambling to get this stuff right now, or is there not? Or is there something where they can be done? Yeah, it's an interesting point. China in particular has been blocked from buying most advanced fab equipment. They're really kind of stuck now. Two things are happening there. Well, and obviously the U.S. companies can't sell into China, but there's kind of been that gray market.

of used equipment, older equipment, repurposed equipment, retooled spares, technicians, you know, kind of all that. That has been important for you know, some of the China build out. But again, that's really hard to do advanced notes. I use CXMT as an example. I mean, CXMT, you know, is trying to break past the 18 nanometer barrier on DRAM.

Because that's where the U.S. government held them up for a number of years. Now, they had R&D going on long before that, which was much smaller than that, much farther advanced. But, you know, to get down to 10 nanometer. eight nanometer or something like that they're going to need fully advanced equipment so i don't know how those guys solve that particular problem however where there's that need and you look chinese engineers

they can deploy an army of, of young engineers and they'll figure out, they'll reverse engineer some equipment somewhere and figure out how to make it themselves. So I think the biggest thing in blocking the Chinese companies from buying state-of-the-art fab equipment was to in five to ten years there's going to be a chinese supply of fab equipment that's going to be

Almost as good, if not as good as whatever you can buy in the world. Maybe one example where it's not the case would be EUV equipment. That seems to be just too specialized. The holidays are here, and that means it's the most wonderful time of the year to save with Rakuten. Use Rakuten to stack cash back from your favorite stores on top of holiday sales. That's savings on savings. With Rakuten, you can get cash back on gifts for everyone on your list, from toys for the kids to kitchen gear.

And you can get paid with gift cards, PayPal, or... Check or eligible American express card members can even choose to earn membership rewards instead of cash back. It's truly a no brainer. Join for free today and get a new member bonus after minimum qualifying purchases. Just go. At Capella University, learning online doesn't mean learning alone.

You'll get support from people who care about your success, like your enrollment specialist who gets to know you and the goals you'd like to achieve. You'll also get a designated academic coach who's with you throughout your entire program. Career coaches are available to help you navigate your professional goals. A different future is closer than you think with Capella University. Learn more at capella.edu.

Whether you're gifting, decorating, or treating yourself, Lowe's December Deal Drops are here to help you save more all month long. Get up to 50% off select holiday decor. Plus, get up to 45% off select tools and accessories. Shop new December deal drops every week this month and get so much more out of your holidays. Lowe's. We help. You save. Selection varies by location while supplies last.

Planet9 writes in and says, Hi, increasing deamer manufacturing capacity is going to take a lot of time and money. But once that capacity has transitioned and then the market cools off. What do you think will have changed in the landscape of products available and their relative prices? So I think he is talking like, you know, we think shortages next year, things start to recover 27.

What does 2028 look like in your opinion? Yeah, and that's a good question. I really don't have a good feeling on that other than I think, I do think this AI trend lasts a lot longer than.

than we might imagine. So I think that train is going to continue to run. The question is going to be, though, again, three legs of the DRAM business. If we deemphasize the data center business for some reason, does client or does mobile pick up the slack now mobile's been kind of dead for a while i mean just not a lot going on there we don't see mobile phones taking more you know the ram capacity in mobile phones been pretty

But one thing I can think of is we're not yet doing a lot of inference in the phone itself. And the question is going to be if we start doing a lot more AI inference in the mobile phone, it's going to need a lot more RAM. And so that's a place that it is likely to shift to. I don't know the gaming business well at all, but it's the same sort of thing. If the AI features, AI inference, we want it local in the machine.

then you're going to need a lot more memory. And so I think as we push from data center back towards edge, whether that's mobile or consumer, then that could pick up the slack and that could now tip the balance away from HBM. back to DDR or LPDDR or whatever their success cycle where HBM's going to the things, building the stuff, and then it flips back to DRAM when consumers needed the devices to run it.

That's right. I mean, it doesn't change. Whether you're doing training or inference, you're doing a ton of matrix math. And to do a ton of matrix math, you need a lot of DRAM. And right now, today, we're doing all that training and most of the inference is being done in the data center. It's not being done in your device. And so suddenly, when that...

Apple finally figures out how to do inference in the device and gets on board with that, then suddenly you're going to be using a lot more LPDDR, maybe. Will it be HBM in the phone itself? I don't know. But I think that's what we're going to see is we're going to the data center is not going to be the hotspot anymore, but it's going to move back either to client or to mobile. Yeah, because to give an example.

I put out some leaks on like what the specs of the PlayStation 6 and apparently a handheld version, which is interesting, of the PlayStation 6 will look like. To put things in perspective, right now the PlayStation 5 and the Xbox have 16 gigabytes of RAM. The previous gen, they had eight gigabytes. But I've seen documents where it shows the PlayStation 6 handheld. It's a codenamed Canis.

literally it said up to 48 gigabytes of RAM, and there was a lot of emphasis on AI. So that tells you they're not looking to double RAM Dengen. They're probably looking to go to, it said up to, so maybe at least 24, 36 gigs of RAM. And I think... that yeah sony's trying to time that right you'd think these companies are going to realize that if a playstation has more ram than what our laptops have now that this is going to be the next big rush probably right

Correct. Correct. And that's what we're talking about is that hotspot moves back to one of these other markets, again, likely driven by AI inference local in the machine. Let me see here. Magnetic Manual writes in and says hardware shortages seem more frequent and severe than in past decades. There always seems to be some shortage every couple of years. Does the do it yourself market stand a chance in the long term or will all the expensive factory capacity be reserved for only?

huge customers like Dell. Well, but Dell buys a lot of the same parts. I know it kind of sounds like you're saying it's just going to be cyclical, but... it will swing back towards the consumer it's going to swing it's a pendulum um the higher the high the lower the low you know it's going to swing back um all right we got a few last questions here that are just

for you because i don't think uh anyone else can answer these questions but melodic warrior writes and it says welcome back dave it has been a while ddr5 has impressed me overall as a technology it feels like the timeline between it was first launched to when higher-end kits become available was short. Have you ever seen a generation of DRAM where the technology improved at this rapid of a pace in such a short period of time? And what generation of DRAM excited you the most over the years?

Yeah. Yeah. Good question. Cause I have exactly the opposite take on DDR5. I kind of feel like DDR5 took forever. You know, I've been hearing about it, hearing about it, hearing about it. Part of it was because I was pretty involved also at JEDEC, which is the standard setting. And from the time ideas start...

percolating on what to do for next generation DRAM until the time it actually comes out is about a decade. I mean, it's a long cycle. So, you know, the guys that are working on standardization today are, you know, thinking about, you know, what is a DDR? six or is there such a thing as a ddr6 etc so it does take a long time at least that's my perspective i'm glad to hear somebody else sees hey that came out really fast and worked well i think that there's enough

First of all, the number of players, as we talked about, just on making DRAM chips is pretty small. So that means as somebody who's making... you know, some other component that goes into a DRAM module, et cetera. You only have to talk to a couple of players and kind of figure out your strategy and that allows them to work and work successfully together.

Back to that question, I found that I smiled when I saw that DDR2. DDR2 was kind of the real proof point that, hey, this is really something now. This is really a good technology. and if you look at bandwidth per channel and things like that ddr2 is like a big big improvement over ddr and so it was a a good jump ahead um so that's the one i remember and just saying wow because like what was I remember DDR3 was typically 1600, sometimes 2133.

um you'd have 1067 a lot but like wasn't ddr2 like 800 and ddr1 was 800 and yeah 800 was the one that really kind of put it out there and if you compared it to ddr and that was just a big leap forward And like I was saying, I mean, it didn't have some of the other issues of putting too many modules on the channel, which you later got into in order to get into depth.

And and so, you know, performance wise, it was like, wow, what can we do here? This is this is pretty interesting. So so anyways, that made me smile because. caused me to look back at DDR2. I remember seeing Micron put something together years ago, which was comparing the different generations.

And I can't remember exactly what metric, but DDR2 was still the champion. And I can't quite remember which performance metric it was. But anyways, that was a good one. Do you think DDR6 will be more exciting than DDR4? five like do you think no you think it gets more boring not if it's evolutionary i mean i would like to see so yeah if we dive into it i still think the bus is way too wide

And I think the bus is way too wide. I get so many gamers that write in and go, why can't we have three channel memory, four channel memory? Why can't we get to a wider bus on desktop? You think that's not the issue. I'm thinking, again, I'm probably biased because I spent more time thinking about this for the data center. And it's just how many channels can you get, right? And they're just pin constrained.

And so it's your pin-constrained, and when on a server, you know, a data center processor, you know, two-thirds to three-quarters of your pins are just your DRAM channels. And then... we're straining on number of channels and you look at the routing the routing is a disaster on the pcb the number of layers so i would like to see go to very high performance serial links or or something like that

Again, I'm thinking about it more from data center processors and getting very high performance, narrow links, and then just putting tons of them out there. That's what I would like to see. That's revolutionary, right? I mean, that's not an evolutionary DDR approach. So, again, we talked about it at the very beginning. HVM is a brute force approach.

punching holes through the wall and put as many holes as many data channels as you could i'd rather go to i'd rather go to optical right i mean you just have some kind of link It's a very high-performance link between the memory and the processor. And then, you know, why is a processor so limited in terms of six channels, eight channels? I mean, why don't we have 64 channels?

When we have 256 memory channels, that would be awesome because then we can do a lot of math. And so that's the way I'd like to see it go. But, you know, that takes some... innovation and leadership and figuring out how to do that yeah you almost wonder if like where that's going to happen is just attaching the ram onto the soc because it doesn't need to use all the old interfaces that are compatible but

I mean, and AMD's already experimenting with that. All of them are, actually, but it's all the smoke. just years decades ago i mean ibm came out with a paper and it was like doing wafer on wafer and it was optical interconnect between the logic and the memory itself all produced you know wafer stacking is the way to think about it Pretty wild stuff. MinG3825 writes in and says, hi, Tom and Dave. Will we ever see ECC become mainstream for consumer platforms? IRC, even the DDR5 ready.

Out there has ECC built in, but it only works by itself and won't report the error to the OS or users, unlike the one on server platforms. What's stopping AMD or Intel or the RAM manufacturers from implementing actual ECC on? all products, especially the ones that already have some kind of ECC in the DDR5 already. What pros and cons will it bring other than data integrities if the server-grade ECC is brought to mainstream platforms? Cheers.

I read that question and I was like, I don't think I have a good answer because clearly you can do it. As the person points out, there's a case. cases out there already that implement it and implement it successfully. Uh, I can't imagine that it has much cost to the system. So I I'm stumped on this one. I don't have a good answer. Oh,

Yeah. I mean, it seems like they're almost there already. Right. But I don't know. Yeah. You've got all the pieces and it's not a performance. It's not a latency thing because you can do it in a more high performance system. So and like I said, cost, I can't.

I can't see. So I kind of thought about it from these angles. Is it performance? Is it cost? What is it? And I don't have a good answer on that one. Sorry. Stumped me. Dylan M writes in with the death of Optane. Could you see a future for persistent memory? No. Because we talked about that a lot, I believe, the last time you were on. It hurts me. It hurts me because I invested a fair amount of my own time into it. I think the big takeaway.

well, two, and I did a deconstruct on this at FMS two years ago, was when it came down to it, the customers just wanted cheap memory bits. And the persistence, they not only didn't value it, they didn't want it. It caused problems for them. And I use this simple example. I think even on your show, I use this example of. You know, if it's some tech in Amazon AWS, somebody takes an Optane DIMM out of a server box and sets it on the desk, that has customer data on it.

That's a liability. I see. Even if I unplug it, it's gone, right? I don't have to worry about it. Now, that's just one very simple example. But at the end of the day, customers not only didn't value persistence, they didn't want it. Yeah, they wanted the cheap capacity, but that never really happened, right? They just wanted cheap memory. Right. Now, this whole argument, and we talked about it at the very beginning about CXL, where you abstract away.

the memory type you know that was a later play from intel it was like well what if i hide this all behind an abstraction layer i still think that that has merit We'll see where that goes with data center construction and architecture and design as we go along. The basic memory being persistent is a problem. Second thing is, why did Intel do this? Well, Intel, if you remember at the time, had like 99% of server processor sales.

they couldn't grow any faster than the market so they looked around and said what can i grab oh i'm going to put this memory special memory in with my server processor and if you want to do business with me You've got to buy this as well. So it was their way of grabbing a share of the memory business to help them grow overall. Now, what happened is AMD snuck up behind them and just...

put out a great product and series of products that competed in head to head in the server processor business. So if Intel itself couldn't manage it of kind of bundling these two together. and make a go of it then nobody else is going to throw billions of dollars at this yeah well all right the future of nand is nand the future of dram is dram

And I stole that probably 20 years ago from a presentation. And I still repeat it every so often in a presentation. Yeah, I don't want to dwell on it too much longer, but I just remember. When I heard my perspective was I remember when they announced Optane and they're like, it's going to be a thousand times faster than an SSD and only half the speed of RAM. And it ended up being.

not even a quarter of the speed of RAM and not a thousand times faster than an SSD. I know the latency was very low and it just never felt like it quite got to where they promised it would be. either and so what you're saying about persistency eventually that's i'm sure what intel tried to market it as well it's about the persistency because they weren't getting to the cost effectiveness that they wanted to get to

which is what people wanted. They wanted cheap RAM. They didn't actually care about any of the other stuff. We're seeing that playbook again. We're seeing that playbook again from SanDisk saying, oh. you want hbm but hbm is really expensive well we'll just plug flash in there and give you cheap bits i don't know you know we'll see we'll see it's a little bit different but uh i i think they're going to run into similar problems

All right. I think this will be the last one. And it always comes up because you were the VP of Embedded Memory at Global Foundries. And so undoubtedly someone has to ask something about that. Do you think that Global Foundry's stopping at 7 nanometer in EUV development was the right move long term? Yes. All right. You still do. Yes, I do. I saw an interview from Tom Caulfield, who is now stepped up into the chairman role, but was the CEO.

that stepped in after sanjay jaw and he's the one that ultimately killed the the seven nanometer tom caulfield did he said his first board meeting it was i can't remember the third one but he said Tom, there's some really important things you got to do. You got to get into the seven nanometer business and you got to get your fab up and running in China. And everybody laughs in the audience because Global Foundries has done neither of these things and strategically went away from it.

Global foundries under Sanjay Jha was trying to compete with TSMC head to head. And they didn't have the resources. They didn't have the money. They didn't have the people. They didn't have the technical capability to do it. Global Foundries is very effective at competing with UMC. And so once they kind of figured out that this is where we're good.

and then putting specialty technologies into older process nodes, it's the right decision. It's the only way they could keep up. Seven nanometers, they were late. They didn't have the capacity to produce. They only had two customers at 14 nanometer, which is IBM and AMD. And they lost AMD as a customer. So, you know, it was not the right thing for them to keep trying to go into seven nanometer.

Even if they could get the technology developed, they couldn't afford the equipment to come in and build out the fab. So absolutely right decision. And you don't think it's like a... You know, saving yourself short term, long term, they're doomed to kind of shrink. I don't think so. I think it was more coming to grips with, you know, who are you really? And I think they figured out who they are really in their place in the ecosystem and stopped trying to say, hey, we're going to take on TSMC.

All right. Well, that's everything I wanted to discuss with you. I mean, I think I just try to close this episode one more time, like summarizing. I think the main thing most people who click on this are curious about, and it's that. And you correct me or add on anything you think I get wrong or that you want to say as well. But, you know, the DRAM companies are expanding as quickly as they can right now. They are actually scrambling us.

much as they can they actually do want more capacity it's not that they are hesitating to build more because they'll make more by raising prices they're raising prices because they're out of capacity and they're trying to make as much as they can And as much as it sucks and it's not satisfying that we really have not that much to say that we'll make things better early next year, at the end of the day, it is your opinion that this will probably be over.

by the time there's next gen xbox playstation and by that i mean late 2027 so as much as it sucks now unfortunately there's not much we can do right now but it should be over quicker than previous shortages Like the giant crypto GPU boom will probably not last as long as things like that, right? Yeah, I think you summarized it well. I think it balances out. Okay. And otherwise, I mean...

I want to thank you so much, Dave, for coming on again. Is there anything you want to plug? Anything you want to tell people where they can find you or anything you like, you know? No, not in particular. Not doing any plugs. It's always a pleasure getting my mind into thinking about some of these topics. It's always fun for me, so I really appreciate it. As you know, you reached out and they said, you know, I'm not really watching it.

that part of the market these days and but as always i guess i just have opinions and i'm okay to express them but appreciate the opportunity so thank you so much absolutely and thank you again and then of course everybody as usual like Share, subscribe on YouTube, comment down below. I guess there's now something called hype points on mobile. Give us hype points, whatever that is. Ring the bell button.

And subscribe to Broken Cellcon on your podcast app of choice and give us a review there as well so people find us outside of YouTube. And otherwise, thanks for watching. Thanks for listening, everybody. And have a good rest of your week. This podcast was brought to you by the YouTube channel and website Moore's Law is Dead. Moore's Law is Dead and Broken Silicon are trademarks of their creator.

Tom, that guy is me. And I am indeed the creator, editor, writer, and showrunner of Moore's Law is Dead podcast, videos, articles, and other media. However, it's not just me. Moore's Law is Dead is a team with Broken Silicon co-hosted by my brother. and audio editing by Gerard Cortez, renders being done by the industrial designer Jean-Philippe Clermont, and special assistance is also provided by Karmacra and Kari Nosugat as well. Find all of our information at www.mooreslisdead.com.

on the about slash support page in the event you do want to hire me for consulting work, hire Gerard for audio work, hire Jean-Philippe for industrial design work, or you're interested in working with Carbon Cry or Cary Nosugada as well. You can also find our long-term

sponsors on that page if you want to show them some love for putting food on our tables. Or you can also mail us some love. You can send letters or hardware donations to the following address. Moore's Law is Dead, PO Box 60632, and Nashville Tennis. Although, to be honest, the best way to show Moore's Laws Dead some love is to support us on Patreon. Patrons are what makes Moore's Laws Dead content truly.

possible. Every month and really every day, depending on who you're talking about, me, Gerard, Dan, and John Philippe are working tirelessly to provide a steady stream of content that we could not keep doing unless we knew the work was possible. without being reliant on sponsors dictating every little thing we put out. Don't get us wrong. We love our sponsors, but we love directly working for you, our fans.

much more. If you have any extra money, even a couple free dollars a month, consider supporting us directly on Patreon. Those couple of monthly dollars will get you access to the exclusive podcast, Die Shrink, voting on subjects of future... podcast episodes, the ability to ask guests questions, and of course, access to the Morris Lodzdead discord full of like-minded people who I am sure would love to meet you.

I am one of them. Additionally, higher tiers get access to early ad-free episodes of Broken Silicon, the ability to ask questions in all Broken Silicon episodes and loose ends live streams ahead of the recording and the entire back catalog. of Moore's Law Z podcast. In addition to having thanks in the credits of videos and podcasts, depending on the tier with other perks available as well. And hey.

If you cannot afford to support us directly every month, please do share Moore's Law is Dead videos and podcasts with friends and family and on social media and websites like Reddit. And give Broken Silicon a five-star review on Apple Podcasts or your preferred podcast app of choice.

All of this does really help us so much. But like I said, this podcast would not be possible without it. The patrons directly providing predictable and reliable support every month. And so now it is time to give a personal thanks to the greatest of the fans. The following supporters are at the 10 gigahertz or higher supported levels. We'll be right back.

Chief, Travis Gooding, I Should Win Wing, Zoot Suit Taylor, Mads, Patty Cakes, Craig, Michael McGee, Stephen Coates, Mark Mitchell, James Anderson, Mark Raidmaker, Salty Internet Pirate, Henry Zhang. We'll be right back.

Wright, Slushpot, Jansen, Anguema, John Swinn, The Forbidden Juice, RB Racer, Matthew Cassells, Alex Fega, Solarize 80, Albergun, Kekum, Sprutnik, Brian P, Hudat 42, Kornster 671, Penta Winta, Omega Doge, Dan M 1988, Hugh Guru, Ander Keevy, Mr. Tiger, John Wick Boomstick, Penguins Love AI, Porcelain Oats, Nixu's Mania, Smash Axley, Kareem, Al Masri, Al, AI Soldier, 789, Rye Hill, Cheebol,

Exclamation point. Fred, Fred Berger, Kevin Kuntz, Man Nullis, Ragacousis, Jonathan Harris, Kylan, Reichel, Kainoush, Kagani, Zabir, Ahmed, Truna, Johnny Trans, Josh, Ark, AO, RK0, Penguins. From NKVD, Zephyrin, an angel of cake. And of course, thank you to Sahara for the music. Kelly University, learning online doesn't mean learning alone. You'll get support from people who care about your success, like your enrollment specialist who gets to know you and the goals you'd like to achieve.

Plus, career coaches are available to help you navigate your professional goals. A different future is closer than you think with Capella University. Learn more at capella.edu. Introducing new Sylvania LED Ultra fog and power sport lights. This powerful upgrade delivers cool white clarity and crisp visibility on the road ahead.

so you can reignite the love for your car with lights that make it feel like day one. And built with a compact design for a DIY install, these fog lights are backed by over 100 years of lighting expertise to bring you LED ultra night vision and help you see better at night. it's time to fall for your car all over again sylvania life that lasts i had that dream again

My small business needs IT help, but I don't use LinkedIn. And I hire wrong. So then I'm doing IT. And when I go to plug the servers in, they become sentient. And they won't let us access our network unless we forward their chain emails and memes to more and more and more people. And then...

I wake up. Don't let hiring nightmares ruin your dreams. LinkedIn Jobs' new AI assistant finds and invites best-fit candidates to apply so you can discover talent you may have missed. Get 50% off your first job post at linkedin.com slash Pandora offer. Terms and conditions apply. Hey, guys. Back at the playground again, huh? Yep. You know what this playground could use? A wine country. Heck yeah. And some waves. So we could go surfing.

A redwood forest would be cool. I'm in. Ha ha, ski slopes. Let's do it. Um, can a girl go shopping? Yeah, baby! Wait! Did we just invent California? Discover why California is the ultimate playground at visitcalifornia.com. As the daughter of immigrants, financial struggles were part of my everyday reality. In high school, I became homeless and had to live in a woman's shelter.

Thankfully, being an APIA McDonald Scholar enabled me to attend college and begin a new chapter in my life. And now, my reality is filled with endless possibilities. McDonald's has awarded nearly $4 million through APIA scholars to support students. Learn more at apanext.com.