MediaTek and the Evolving Custom ASIC Business, More Memory Pain, and More

0:00 This is not your father's media tech anymore. They are taking a brand new approach to all of these markets that they're in. Hello everyone. Welcome to another episode of the circuit. I am Ben Barren. >> Greeting agents. I'm Jay Goldberg. So, there was uh a lot that happened this week, both news and you know, we've been stacking events, at least for those in the analyst circuit. Um I attended Mediate Tech's Analyst Day this week. Um which is an interesting, you know, company. I followed them for actually a a really long time because back in >> the 2000s we were consulting with a company named Zoran who was the who was the DVD player media uh engine and their largest competitor was MediaTek who had 70% share of the DVD uh codec compression media player APs back in the day. So, >> I remember when I back in like 2007 six, I was like I I remember I was I would bring I was at the bank and I would bring MediaTek. I say, "Hey, I just got back from China and there's this company called MediaTek that's doing all these crazy things in mobile phones and like nobody had heard of them. Like investors hadn't heard of them."

1:28 >> Yeah. It's tough also because they're on the Taiwan exchange. So, like I get it. But but you know if if you're savvy investors like the ones that we talked to especially to the subject we're going to talk about about custom AS6 um you follow them because you're just curious do they heard you know Broadcom or Marll or companies you do cover right competitively? Well, I I think that's the interesting thing is for a long time nobody in the US took them seriously in probably going back to DVDs and then nobody took them seriously in mobile. Uh and now everybody knows them as for a long time as a mobile company. And now it's the same thing. They're just being seen as sort of second source lowcost competition to ASIC vendors like Marll and Broadcom.

2:13 >> And if you look I mean they they crush the DVD market. And then if you look at where they are in mobile today, they are fully competitive with Qualcomm. >> Absolutely. >> Technically, technically and margin wise, they're fully competitive. And I don't know what they're going to do in AI, but they, you know, they have a legitimate business here. >> Also, for what it's worth, they've been doing custom AS6 before we called them custom AS6. I mean, they were doing this >> 10 years ago. They were doing custom AS6 for a bunch of people.

2:39 >> They've been in that business a long time. >> Yes, they have. So, so there's two things just to mention. So, so again, if you're not terribly familiar, you know, with MediaTek, as sort of Jay said, right, they have a lot of history uh back into uh silicon in consumer electronics that parlayed into mobile where again they were a huge threat um kind of across the board. I think correct me if I'm wrong I think in like India or somewhere they dabbled with a vendor who co-branded with them in smartphones uh for like a period of time like a very short cycle um because I remember work we were doing with Nokia back then they were freaking out about some joint thing MediaTek was doing with phones and anyway lots of back back history in consumer electronics. Uh their CEO also Rick Sai was the CEO of TSMC from 2005 to I believe 2009ish um and chairman. So deep roots in the Taiwanese ecosystem and uh and supply chain. But what's always intrigued me about this company is they've always been extremely humble when it comes to their technology. And I say that because there was a lot of external belief that I had about this company when I started, you know, coming to a lot of their events seven years ago and they never talked about like their customers or who they were in, right?

4:12 And so we were at this event and all of a sudden we find out that they're in all the Amazon Echo devices like at the time that Amazon Echo was like starting to sell like hotcakes and we're like why didn't why didn't anybody tell tell us like you guys are the SOC in this and actually doing some some some interesting co-work with Amazon to optimize silicon to uh to Alexa and those uh and those Echo and they're like well you know we just don't like to talk about our customers and so for the longest time like the view was you're just an ARM shop. You know, you're not necessarily an ARM lency, which they aren't. They are a lency of CSS and standard IP. They're not doing custom micro architectures like Nvidia's doing and um and Apple, etc. Uh again, they could become one of those licenses, but but but they're not. So, the so the the theory was, oh, you're just you're just doing custom ARM stuff. You don't have any of your own IP. And for the longest time, like they didn't refute that message, even though it's not true. And I was always like, other than the PUMIC, I always knew they had a really good PUMIC just because like analysis I had seen at the board level from some friendlies was always like their power management's really good and that was unique IP and I kind of didn't want to talk about that. But then, you know, anyway, fast forward to now and I'm sitting in this presentation that they're giving on their launch in the data center. And if anybody who knows this space, you know that they've been talking about this for a while. They they have had rumblings of customers. We do know they have customers. It's sort of like a secret, but not really a secret who those customers are. And um and all of a sudden they're like, "Hey, we're going to tell you about our 30s and our optical IP and our packaging IP and all these things." I was like, "Wait a minute. Where the crap was this like two years ago when you first started saying like why we're going to get into AS6? Why was there not a broader technology story of all this IP that you guys actually have when again the world just assumed you were an arm shop? And so I I I thought that was interesting in that culturally they're willing to start be a little bit more aggressive and talking up their IP because they do have um a lot. I'm not going to say they have all of it um but they've got a lot and that's that's a interesting change. And anyway, we'll talk about some of the things they walked us through, but but for the first time, they're starting to tell these more relevant technical stories and show their technical chops.

6:44 And that's a very new approach for media tech. I I think there is something different about the way that US semiconductor companies handle marketing versus pretty much every other semiconductor company geography in the world. Um, it's it's a recurring theme. There are a few countries I can think of that have pretty deep semiconductor engineering talent and marketing is not one of the skills that they teach in semiconductor schools. And it's it's interesting to me like and you know certainly US semiconductor companies can all do a better job marketing but Taiwanese companies in particular I would say Israeli companies too really struggle if with >> Yeah.

7:27 uh with that and and it's it's funny. I've recently been working with a couple uh foreignb born engineers and on a project and getting them to be sort of more confident in what they're offering and sort of talk it up. They're they it's it's it's like a weird humility muscle that they just like they feel uncomfortable promoting themselves. very very you know and it's like it's again it's one of those things where you're totally right like US companies like this is the thing they're like look how good my engine like let our engineers talk and even like at these media tech events like I would ask like these really hard like questions even around their NPU which is is actually pretty good and and some of their own IP and they never wanted to talk about it >> and I was like like just what like and I had sort of told them I said listen guys I get it. But like what weirds me out was I used to ask these same questions to like Huawei and others and their non-answer to me was more of a red flag than it was uh you know that they were just shy about their IP. And so I was like so I don't so I feel like you're hiding something when you don't want to tell us about these things. Like that's just that's just bad stake. But but anyway, that was sort of like my concern. like why don't you want to say or why don't you let some of the technical engineers the designers out of the out of the cages so they can start talking about the things that they're doing even in R&D and it just took a long long time so luckily they are at that stage now where they're willing to tell these stories um our event was kind of like the broader coming out party for ASIC even though Rick has talked about this at a handful of events uh they've led investors to the kind of dollars that they're expecting this year and next year. So, it's been the cat's been out of the bag, but a lot more details about what they're doing in um advanced nodes, advanced packaging, optical, their 30s, all of that stuff was was brought more to light at this event.

9:29 There's a bunch of slides the world had never seen that I was tweeting and like people were like, "Holy crap, where are you?" Like, I've never seen any of these type of I follow MediaTek forever. So, it was a good a good event. I think that you know the true launch will be as revenues start kicking in and it's and it's clear who their customers are and which those products are. Um but anyway, high high level point as we were joking with other analysts and media tech folks at this this is not your father's media tech anymore. They are taking a brand new approach to all of these markets that they're in right not just phones and uh and IoT and compute right at the edge. But now AS6 which is the cash cow of cash cows um in in a in an interesting developing competitive market because it isn't really just you know it was Broadcom Marll G and ALC chip with varying degrees of who brought what to what you know Broadcom being the dominant and now I think MediaTek is going to enter that conversation in a relevant way and and people are going to start talking about this more as they think about the as the custom ASIC landscape.

10:37 Yeah, I mean like I was saying before, Media Tech has been doing custom AS6 for a long I think before Broadcom was doing TPU almost certainly. >> Oh, for sure. Yes, correct. They pointed this out too. Yes. >> And and I I think there is I I think this is a topic that that investors are pretty fully aware of. There's a very active debate now uh about around some of some of the big potential MediaTek customers um which has been widely reported. It's not a secret. It's it's everybody it's very widely reported in the press. Uh it's that's it's really you know Google is MediaTek's big customer >> for for AS6. Um, Media Tech's never said that, but it's it's widely reported and there's a lot of very active debate in the investment community about how much share does Mediate Techch get there versus Broadcom.

11:32 >> Yeah. >> And and so I think investors there are very familiar with it. I I do think there is still a perception that, you know, in in my mind I think of the ASIC market as there's there's really two tiers. There's sort of full solution providers and then they're sort of like like sort of Alchip and GU which are just sort of doing a much much lower level of support where the customer the designer is doing a lot more work as opposed to to Marll and Broadcom where they have a much heavier engineering role and and I think there is a misconception that MediaTek is in the the sort of low tier category and they're not. They're competing against Marll and Broadcom for for some of the biggest accounts to do pretty serious amounts of work. Um, and and we can debate and I do this all the time every day. We can debate where MediaTek is competitively against those two, but you have to benchmark it against those two against Marll and Broadcom and not against Alch.

12:29 >> Yes. Agreed. But but I think again right they what's interesting to me is like I said they for the first time started talking about their 30s. You know they've got uh 224 today on track and and 440s coming. Um again not too far off from the timeline of their major competitors Certi. So the the the networking IP is there. They talked a lot about um all of their fabrics from co-ackaging to uh advanced packaging in 2.5D and 3.5D ties to memories they have. And I think, you know, the the point they brought this up when they first started talking to us at our annual events, I think two two years ago, I want to say, um, they kept talking up like, look, we we have a very privileged relationship at TSMC that I don't think many people understand. And again when that is whether or not right they're shipping smartphone wafers early on they're still the of the earliest adopters so they can go through that learning process so they can have shared learnings with TSMC so that they can um refine some of the things that they will be doing so they're they're in the earliest of discussions with TSMC and and I think that is one an underappreciated metric but one that right now related to uh XPUs and whatnot is going to become an even stronger differentiator and I think a couple years ago you know the market was just talking about it's the chip everybody just cared about the chip like what chip are you making you're going to do custom and now that is the full solution so you have to have a story that's not just I've got what you need to go make an ASIC but I've also got what you need to do the advanced packaging to do the integration to do 2 and a half and three and 3.5D to do scale up scale up networking to have optical IP. They were even showing some stuff they were doing in co-ackaged copper and near packaged copper that I actually think are are farther advanced than other people are doing at the chip level for keeping copper up to you know 400 gigs per lane. So go ahead.

14:44 >> I I I was just going to say that, you know, the the sort of more informed knock I hear on criticism I hear on MediaTek is that their selling proposition in the ASIC market is the fact that they get allocation at T they get preferential allocation at TSMC, which for a long time was just like, oh, these guys can't compete technologically, but I'll I'll use them because I get >> allocation. I I won't debate that point. I don't think it's true, but let's not worry about that. My point here is just that like having allocation to TSMC is probably your biggest competitive advantage for any of these companies today. That's that's, you know, that matters more. It doesn't even matter how good how good your IP is to some degree.

15:24 You get me can you get me wafers at TSMC? And TSMC and MediaTek have a very close relationship. They're essentially next door to each other in Shinju. They kind of grew up together. I think is there's a whole there's a whole long story there which it basically boils down to these two companies sort of were very very close from very early days and um there's a lot of overlap and that gives MediaTek I would I would argue gives them a preferred position at TMCC. They have a lot more than that but you know in this today's market like that's enough like I'll take the allocation.

15:57 >> I I agree. I agree and I think again right if you just step back from one they are they are a larger customer in revenue to uh TSMC than any competitor in the custom AS6 space right so I'm just talking about custom AS6 so the bigger than Broadcom bigger than Marll by quite a lot they also do a lot more wafers they're probably top three in wafers at TSMC um and I say that only because GPUs are gigantic right in terms of monolithic they do uh they do at least top three just a little behind or in front of Qualcomm but Apple obviously does does the most um but their ability to have that wafer allocation and again the priority that they have in all the aspects of under one roof at TSMC um yeah I I think again right the these are things that you can't take for granted when you're the Googles the Microsofts the Amazons others of the world who you know are like yeah like working with you becomes an attractive proposition because of all of these things. Now, now again, so going back to the business model change, I had always believed that the thing that would benefit MediaTek the most would be customer own tooling and we've discussed this as a debate for right a long time. Um but their shared position with TSMC has very specific advantages to the design technology services co-optimization and the system services co-optimization which is a leading edge uh program to refine your process and your technology and you're involved in that with TSMC like you're actively working with them on on on on future process and and packaging whatnot and So the fact that more customers may want to do more of their own IP, which is the trend for Google, the one that Broadcom and Hawktown's been like, "Oh, you know, it's okay, but it's not going to be that big of a thing." Um, I do think the market is moving in that direction for a couple customers, maybe not all of them, but a couple customers.

18:10 And and if that continues at that rate, that's where I think the business model that that that MediaTek is bringing plus the technology story is really the best like opportunity for them, right, as it shapes up. That's a that's a pretty contentious statement and I think I think we need to unpack that a little because this idea of customerowned tooling which is basically saying if you're a hyperscaler you're designing your own chip to what degree do you need an ASIC partner like MediaTek or Marll or a orchip or media or Broadcom?

18:47 It's sort of like it's a sliding scale, right? Some customers need Broadcom to do everything for them. Some need like Microsoft or goo need very little from their ASIC provider, so they go with an LLC chip or GU. And then what MediaTek is essentially saying is somewhere in the middle where Google's going to do some of their IP themselves. Um, and that it's, you know, like you said, Hawk mentioned it on the call as sort of dismissing it.

19:17 People people debate this with me every day. And I I think it's unclear. And I think I think rather than think of it as a point, it's either this or that. It's like I said, it's it's a sort of a sliding scale of some customers are going to do it, some aren't. Some are going to try it and go back. Some are going to avoid it and then come back to it. So all the hyperscalers are going to continue doing chips. Uh it is I think it's hard to it's hard to argue that all the hyperscalers are eventually going to move to fully custom own tooling and just doing it all themselves and just having Alchip sort of do the last couple steps. I don't think that's going to happen. I think some people are going to need to use Broadcom in the future and Marvel in the future.

20:02 I I I don't think but I I could be wrong like it's it's a it's a pretty complex picture right now. It's very dynamic. All the hyperscalers are going through their processes and iterating and I don't think anyone's sort of locked down on the right model. We've certainly seen seen some hyperscalers try one path and have that fail and need >> yes >> more support. >> Right. >> Um >> correct. >> So I think I think what will happen >> I would say two things. I don't >> I I think I think what's going to happen in the future is all the hyperscalers are working on their own chips. Some of those projects are going to fail this year. And then the question is, do the hyperscalers move, you know, up a up a level from an alcheip to a MediaTek or from MediaTek to a Broadcom or do they just abandon do they abandon custom altogether and >> you know go with AMD and that that's I think that's the framework for the whole debate.

20:53 >> Yeah. Yeah. I I I agree and I and and I don't think right again it's a it's a onesizefits-all both chip or business model or vendor program. I think again all these players will compete. Um you know if anything I would be a little bit more concerned about like Juicy for example like you might need them less. Alchip I think can still be there but you know again they're largely a Amazon um who again as I pointed out is a large customer of MediaTek in devices. So it's not like they don't already have a relationship with them there to start leveraging. Um I mean really every customer in the ecosystem you're talking about is the customer of MediaTek in some capacity. So they already have a lot of those relationships. But you know but but what I do think happens is I I to my point again right when you were just talking about the chip and the chip has now moved to the rack and the entire you know element of compute the rack scale compute is the pod I think we're going to see variations of chips from the custom uh from the XPU guys. So it's not just it's a you know matmo accelerator or it's a networking chip. I I think you're going to see variations of training trips some big some small variations of inference chips some big some small all the networking right that go into that. So I I think you'll see reasons why people will pick certain vendors over others for certain certain ASIC programs versus the entire ASIC program from a Google and Microsoft and Amazon and OpenAI etc is at one customer. That's essentially like my thesis that this fragments and each person has a reason in which they'll support a program initiative by a hyperscaler or a frontier lab company or or whoever.

22:51 >> Well, and then that just expands my framework to take in Nvidia because that's that's the spectrum here is you can you can go to Nvidia and get everything including your networking and your s your rack design and your whole data center design. There are companies that are going to do that. And then at the other end of the spectrum is you do you mix and match and do everything you want exactly the way you custom customize it all. You know, Google, that's sort of what Google's doing where it's like everything is they're, you know, they're doing their own networking. They're doing their own accelerator.

23:23 >> They're doing their own CPU. Um, right. And and they're increasingly pushing down down the stack to do to, you know, to do more and more in-house. and everyone else is somewhere in between. >> Yeah. Yeah. I mean, I think with two with Nvidia, it's it's it's that is the argument, right? How much do you want to build versus buy? Um, but I do I still think pain points will emerge for these programs that they go try to solve. The the other thing I thought about that was interesting on the back of sort of the ARM stuff was, you know, obviously, you know, MediaTek has CPU IP. They can make those big, they can make those small.

24:05 And what if the hyperscaler who's building let's just say again Google who's building their TPU program with MediaTek also decides they want to build an agentic CPU so not a cloud CPU like we talked about right the whole point of why does Nvidia believe Vera CPU should exist and ARM believes AGI CPU should exist is that there's a derivative of your CPU architecture that's not cloudnative that becomes more AIC native um you know MediaTek would actually be right there to be be able to say we can help you do an agentic CPU to go along with your XPU and all the networking, right? Because they have those IPs. So, you know, I just think it's interesting because all of this is unsettled. Like, we just don't know. Like, it's changing so much. We have no idea where it goes.

24:52 But my broad takeaway, as I said, is if the market moves to more wanting a more flexible business model and a flexible approach to mix and matching IP, I think that's where MediaTek's trying to say we're we're your right we're your right partner. >> I I think they're firmly in the mix and where they end up exactly is anybody's guess at this point, but they're firmly in the mix. >> Yeah. Yeah. And you know, like I said, revenues will hit this year and accelerate into next year. So, the good thing is we'll be able to see how successful this program is by one particular line item that should keep going up and to the right. Um, so anyway, I I I tweeted a lot of these charts in case everybody follows me on Twitter and wanted to see some of the very specific slides that MediaTek had not shown publicly before, even in in Ricky's uh presentations. Um so so check that out and uh this will be one I'm sure we will talk about again either in the end of the year or as new new developments happen. So um anyway that was good. So from media 2 memory I'll let you I did a brief memory update so I can this week in the diligence chat so I can talk about but let's let's talk briefly about memory. I mean it might be that we talk about memory every week. I don't know. But let's so let's talk about this week's take on memory.

26:13 >> Well, so I I I think we at this point we need like little theme music and a little subsegment for the show because >> we're going to talk about just short things that are in shortage and like top of the list of course is memory. >> Memory is still very short. Uh and >> this week confused me to no end because the memory stocks uh had a pretty rough week. I mean everything had a rough week. we are, you know, at war. But like memory has been getting punished the last two weeks around things like we talked about last week. uh Google has a memory tool and there's some startup that came out or some academic thing that came out of I don't know that that is also claiming to do incredible things with memory and there's this whole debate about you know somebody's going to start using mobile memory more for AI stuff and and so the memory stocks got hit this week on a whole bunch of like weird news about memory and at so that that's going on in the market but then meanwhile you and I were both have both had sounds Lots of conversations this week about people actually trying to buy memory and it's impossible.

27:17 like like all it sounds like all the memory makers are sold out into I mean through next year into like 2028 is numbers I've been hearing for when the memory shortages ease now and I I don't know how to reconcile the streets view and practitioner operator's view of memory because it it is like you know when when memory stocks like these big 10 you know hundred billion dollar companies move 10% a week 10% % in a day up or down. It tells me that the market doesn't know how to value anything.

27:51 >> Yes, that's correct. >> Right now, I certainly don't. So, it's very it's very confusing. >> I I don't either, but just, you know, Micron's a good example of like what it got to I believe was uh I think it might have even gotten to the 330s. I could look. But um the the compression that that multiple took was almost unlike a compression even in cyclical cycles and basically believes that none of the guidance or estimates to next year are real. And you're kind of like it's very very hard to find any evidence that that's true.

28:29 Now again, if your belief is that we're peak right now and it holds at peak for another year, that that's fine, right? Because you want to trade before you right get to the downside of of a cyclical cycle. I understand that. Um but just on the the basis of fundamentals, it was hard to understand sort of that reaction. Now, now again, you're right. We don't know where things are going. My supply chain conversations have made this sound like it's going to go to 2030 and and and a lot of that is because wafer allocation is shifting in the majority to HBM of so all existing memory capacity moves to the majority of HPM and you're not going to have any relief come on for two years at a minimum dude DRAM and NAND and DDR and all of the above SOCOM which trying to now get just like good luck. Prices of those things will only go up into the right and you just watch like spot pricing is no longer the good way to look at this.

29:35 For example, with long-term agreements shifting toward capacity commitments and prices being negotiated on a fluctuating cycle, which means the memory vendors are like, "No, I'm not going to lock you into a price right now because in two quarters it's going to be higher than it is right now." And we know that and you all just going to have to accept it. Like this is ne never say the memory guys aren't going to milk a cycle for all it's worth and this is going to be a multi-year if not longer cycle.

30:12 >> Yeah. I I again I don't I don't know. I mean, I'm I'm I am con confused by all this. Whereas my I actually I talked to my my teenage son about memory a lot because he builds his own PCs and he's into gaming and he hates AI because he can't get anything he wants. He can't get new GPUs, >> updated gaming. >> Everything's AI's fault. >> It's and and he can't get memory. And so he he is and so I was talking to him about it this weekend. And so I I am confused. That was the word he used.

30:47 Confuzzled by what is happening in memory. I I I will say every I mean I'm just going to let me play the devil's advocate. Let me play the other side of it because I Right. And the the I think the other the opposing take would be every cycle you hear that same language like oh this is incredible. It's going to go on forever and eventually it won't. Um >> Yeah. Right. >> And I I think I think what's happening I I think what was really critical to watch you touched on this is this idea of spot pricing versus contract pricing because historically contracts will move towards spot, right?

31:27 And so if spot prices keeps going up, contract prices will keep going up. And we we talked about this a couple months ago where during SanDisk or Micron's earnings where they they've all been talking a lot about long-term agreements and the extent to which that sort of locks in long-term pricing. I don't think that's actually taking place. Like I don't think we're we're having these, you know, nobody's signing a 5-year memory contract with locked in pricing today. But what you're seeing is in order just to in order to have a conversation with the with a memory supplier, you need to sign a long-term agreement that agrees to a certain purchase quantity.

32:05 >> Yep. >> And then prices are negotiated every quarter, which is not that different from the past, but there is that sort of underlying basic volume agreement in place. And this is a lot like we what we saw with Trailing Edge Foundry space during the pandemic when everything was so short and Global Foundry signed all those agreements and long-term agreements that really put people in a tough spot for a long time and that didn't like I wouldn't say that that I think that had very mixed results for for Global Foundaries because it certainly helped their numbers for a period probably still continues to be seen in their numbers, but it also definitely put a crimp in their relationship with customers long term.

32:54 People people remember that. And no one's going to say that now about memory guys. Everybody loves the memory suppliers. They're our favorite people. But I I have to wonder if when this cycle turns in in two years or three years or four years, um how how people still feel about memory companies then. But right like I don't know. I mean, I I there's there's a debate we could have about is this structural and are we just has the memory industry now just shifted to something different or are we just like in a super cycle that's going to last for three years and then >> yeah, >> I mean I would say I would say it's unlikely that GPU demand remains up and to the right and memory doesn't. like those two cycles all feel fairly like I I mean the whole thing is correlated right across the board like we're building the the semiconductor industry as a whole is moving from something that it was when it was a $600 billion a year market to something that probably by 2031 or later is actually going to be now two trillion and if you just follow the lines right it took what 40 years, I don't know, to get it. I mean, we're not at a trillion dollars yet, but we've been waiting for that moment. It's probably going to happen this year, and memory is going to be half of that. Um, and then it's probably going to go to two trillion in like four years. Like, that's essentially the path that we're on. So, you're moving from an industry that sustained very little growth to incredible growth.

34:43 And I would struggle to see like that entire thing go back to half like just half. It might shrink some but it's probably not going to just half. Um so within there are these structural dynamics of then what are what are new baselines? Like that's a conversation I have like what like what's a new baseline? This is not your grandfather semiconductor industry. This is a new era of semis. Okay. But but what's our new like structural foundation that can hold steady and we don't know. We have we have no idea. But this whole thing is correlated. It's growing.

35:20 Maybe it peaks but cyclicality is going to be very very different in the future. It might still be there but again it's not going to go all of a sudden semiconductor industry is back at $600 billion a year or $800 billion a year. Um, go ahead. >> Yeah. I uh but I like I don't think it will ha I don't think semiconductor demand will have. The question is when does it stop growing? >> Yes. >> Right. And >> Right.

35:50 >> We're going to be I I Yeah. We're going to be dating this for a long time. >> Yeah. >> Correct. >> But but it's important to remember that you you quoted a statistic now. What is it? 30 40% of that trillion dollar tam this year, the trillion dollar market this year is going to be memory. >> Is memory and probably more than half next year. Yeah. >> Yeah. So more than half, but units are way down, right? It's all pricing. It's all pricing. It's all pricing.

36:17 >> And that that can move much faster than anything else, right? And so I I think for the moment, you know, I I think, you know, you think about most most Wall Street models go out two years at most, eight quarters. And there's a decent chance that the memory cycle lasts beyond the period in which most people are modeling it today. >> Yeah, I think that's reasonable. Yeah. >> Yeah. So, >> yeah. >> Well, again, shortages are shortages.

36:47 this, you know, I >> weird. >> Yeah. But but in memory it's just it's it's interesting to me like you said people aren't people think it's peaked has hard to value it and then the question becomes like when does everybody realize I guess it hasn't like at at some point in time you have to realize a lot has changed and and the overall group sentiment needs to evolve something will convince them is my is my point and some point once people I think recognize this then I think we're in a little bit of a different different market but we're not there yet obviously relative to memory so >> just I just want to add on as long as we're talking about shortages we also have new shortages to add to to the list this week so last week we had helium as it is in shortage the the new one we found out about this week is tungsten I don't even know where tungsten is used I don't I don't actually think it's a semiconductor thing I think it's a somewhere else power chain thing.

37:52 Tungsten is short supply and like if it if the reasons for it weren't so terrible, it would be almost comical because then we're always we're always finding new shortages. But then I think the real the much more serious one more more significant than helium is is water, right? Taiwan is in a drought. And I think that that is uh that's actually a concern to me because Taiwan is has sort of structural water issues that they've been dealing with for years.

38:24 >> Uh they're an island and >> doesn't rain as much. So that's that's a concern. And looking at all these things, right, there's other other stuff that's I was reading this week about Transformers and >> the list goes on. I I think there is a a real a real concern that like probably not this year but next year we just can't build what what the industry wants just cannot cannot be built. Everything's we'll start to see more and more delays.

38:51 >> Yeah. >> For for data centers specifically, right? And and whether it's >> it's the electricity or the transformers or some other random component >> and it will all be constrained by by all those things plus TSMC's capacity. So the transformer one is interesting because I came across this stat. The average age of a transformer is around 35 years and they cannot stay like these things don't go to 4550 years. So a lot of transformers are needing to be fixed.

39:27 And then I put out a note this week on the shift to 800 VDC. That's going to require a whole slew of of new grid innovation and transformer innovation because they're doing all these power bike side cars right now. That's not sustainable in the long haul. Um, so you've got agided aged infrastructure that is going to become but but to your point like where this gets interesting then is if you're Nvidia or if you're you know Broadcom you all these companies that we like to talk about even the hyperscalers and the the the concerning points that people look at which is like oh your revenues aren't going up as much but is it because of a function of the supply chain You then start to have to explain all of the things about the supply chain that is the reason in which your growth has slowed, not demand. And that's probably going to happen in capacity, but that's going to be a weird spot, right, for everybody to be in. Um, and I'm not sure like it's very like we're stacking constraints at this point. Like it's not like it's oh there's one constraint or there's two constraints or there's three like there's like 10 and that list keeps growing and human labor is another part of that constraint. And so you have all these things where there's just so many and that goes to your point like when you've stacked as many as we are like we we just can't solve all these problems and eventually those dominoes start to hit and you know you might want to spend a trillion dollars in capex next year but but you can't.

41:08 >> Yeah. And it'll it'll be very interesting to see how for me it'll be interesting to see how the street responds to this, right? where the the supply chain at this point is pretty highly scrutinized. People are surfacing these shortages well in advance of them actually hitting anybody's numbers. But let's say a year from now, Nvidia misses its numbers because >> yes, right, >> OpenAI and Oracle and Google can't build their their data centers and so they're just pausing deliveries of Nvidia parts.

41:40 Does the street look through that? I I don't I don't know. I I think if it happened today the street would but you know a year from now different. >> Well I mean but isn't isn't part of this like a lot of investors like conversations I've been having is like let's go let's go as many layers deep as we can into the supply chain to find like who's solving those constraints. So there becomes like >> the deepest level of picks and shovels to rock miners to people who make the shoes for those who are going to mine like you know what I mean like it becomes that becomes the interesting exercise and where they might right put some money in the the most attractive size for who are solving the constraints versus the companies who benefit from them in years or quarters as constraints alleviate. Yeah, I mean that's certainly the trade that's been taking place over the last six months is in investors looking for things that are second and third order derivatives.

42:38 >> Yeah. >> People who have companies that have capacity, companies that have some critical components that's short. Um but you know, I I spent a lot of time looking at those names and we're already pretty pretty deep into it. The the market street is pretty deep into that already. Um, and you know, I mean, just the fact that you and I are talking about power, power power generation, like if you like 18 months ago, it was on neither of our road maps. We would have never been talking about it. And here we are. It's, you know, pretty important.

43:07 >> I mean, 18 months ago, I didn't think I'd have to moonlight as a utilities analyst just to get deeper insights into the freaking data center constraints and builds. And then I find myself I found an Equinox data center that's in South San Jose and I driving around for it looking to see what machines are on the back side for you know power utility generation like what what's happening now? I'm like a data center spy trying to figure out you know what's sitting back there in in components or or how the sidec car is working. Um, so yeah, it's it's again it's interesting, but I also wonder like maybe that's why, you know, maybe this is just the bubble that I'm in that but then I am seeing so many people want to move and talk about the optical names because even even if you don't all of a sudden expand uh shipments, right, in doubling, the sheer shift from copper to fiber will benefit people even in a constrained environment.

44:06 And so I do wonder if to some degree maybe that's somewise the I keep seeing a lot of these optical names start to come up and people hunting for it's not just lumenum or or coherent but you know uh testing with a here and um you know even cornings come up more you know in my conversations because you might have structural benefits from just replacement cycles than just pure growth cycles I guess. I >> the idea that I I have to start being an analyst covering replacement cycle of testing equipment is >> I don't want I don't want to contemplate that.

44:49 >> I'm not going that deep. I'm just saying like these are the conversations that keep coming up like they're just looking for those things like structural benefits of a transition again from copper to fiber without a growth cycle. Like that will be a structural benefit of a shift. So all right. Uh this week also our favorite our favorite company Intel has um purchased back their share of the Apollo agreement for the Ireland fab. So give us your thoughts on this because this I think the main point was misinterpreted which I know you'll hit. Um so let's so let's start there.

45:32 So yeah, just some background. Two and a half, three years ago when Intel was in very tight financial situation and it had a very shaky balance sheet, they did a series of deals where they um they did a sale lease back. They sold their the the real estate underneath their fabs to a real estate company, Brookfield, and then leased it back from them. Very common structure. And they did a few of those including they did one with their main sort of production fab in Ireland which they this is like the third the third one they did and they in rather than doing with a a real estate company they did it with a private equity fund Apollo and this week they announced uh they they they originally Intel had sold that plant for I think 11 billion and this week they announced that they're buying that plant back from Apollo.

46:28 They're buying back Apollo's 51% stake for $14 billion. So Apollo gets to profit gets to, you know, make pocket $3 billion off that deal. >> Yeah. >> Plus whatever payments they got in between. And Intel now has complete control of that P. And in the press release, it was it was uh it was a very uh loftily worded press release like about all the great things and how they were such great partners and how they love working together and what you know and Intel said this is you know our strong commitment to the future of CPUs whatever >> and and I think the sort of the the the deal was viewed very positively by the street. Um it's it's essentially created the earnings, >> right?

47:15 >> Um >> and I I saw a lot of takes where like oh this is Intel betting on the future of semiconductors again >> and yes all those things are true. Oh, well it is it is a creative I I believe and uh it is uh you know a strong I would say strong sign that Intel feels it's in better financial shape and I don't think there was any question that they believed in the future of semiconductor manufacturing like we we haven't asked that question in six months >> right >> but I also think it was a that a deal that deal with Apollo was a terrible deal and and I say that because the the first two deals they did with Brookfield were very straightforward real estate deals.

47:58 The one with Apollo had a lot of other things in it around operating metrics. It was a much funkier structure. Um, and we don't know all the details and believe me, I looked the the real details were redacted from the filings, but it there there were performance metrics in in the contract. Intel had to hit certain performance targets. If you all recall, um, you know, a year ago, they were not hitting those targets. uh with yield around 18A specifically. I believe that's actually in the contract.

48:28 >> And so it just didn't make sense to be in that deal. I can only imagine, you know, what the future future terms would have required. The fact that Intel's willing to pay $14 billion to get out of it should tell you that it was pretty ownorous. And so it's a it's a good move for Intel uh to get out of from under that for a lot of reasons, >> but not just >> the future of CPUs.

48:58 >> Correct. Yeah, I think the the right take as you said is that that was a very bad deal. Um that deal would have taken an EPS hit which they get back and are in a better position financially because of it. that that's the like like again like what what was Brookfield going to do with their 40some percent share like they weren't making they weren't taking out like capacity like Intel had that capacity like capacity was still there it's just it was a terrible deal um so but I also find it interesting that kind of the reaction was um oh you know Intel's so confident in the future you know they're willing to make like these investments and and then the streets like reward boarding these investments when like two years ago it was like Intel get your spending under control like like why aren't you spending like stop spending so much money and now they're like Intel what the crap are you doing spend more money there's capacity there's there's there's share to be had there's there's chips to sell man and you're you're at a constraint so I just the irony of it is just that that's what's got me um but then I want you to comment on I saw some people big names from sellside bank say, "Hey, we'd love to see them get out of this deal with Brookfield for Fab 52." And I didn't have a thought on it, but you did. So, I want you to share that because that might not be a right the right take.

50:25 >> I So, I Yeah, my my take is that the the Brookfield deal is much more benign. Um, I haven't gone quite as deep into the filings in for that one, but it certainly looked much much more straightforward. It's a real estate deal. It does not I do not believe it has any operating metrics in it. Um and they're just like renting a piece of land from Brookfield. And yeah, they could they could get out of it and >> maybe they will someday, right? Who knows?

50:55 >> Maybe they will someday. But like you like the the the metrics there, the the motivation to get out of there, I think, is much smaller. It's much more plain vanilla for whatever reason. They they got a little they they got a little creative, let's call it, with Apollo. And my my guess is that they don't have the same urgency to get out of the the Brookfield deal. Maybe maybe you know again I there's maybe there's something in there I missed but my take is it's it's not you know they like just to be clear they still you know don't have a great financial outlook in in the sense of they still need to spend a lot of money on capex and I would rather have them buying tools than doing you know paying buying back real estate at inflated or not inflated prices but higher prices. So my guess is they don't get out of the Brookfield one. I also don't think it, you know, matters much one way or the other, but you know, at this point, I'd rather have them writing big checks to ASML than big checks to >> Yeah, keep expanding capacity. That is that would absolutely be right. You you you need to keep making CPUs as the fastest rate possible and CPU and stuff for other people. Um, you know, one point on the MediaTek thing, not that they confirmed this or others, but let's just say there was a strong uh suggestion that as we were talking about that they were an early um design technology partner with uh TSMC, they also pointed out they are in that position as well with Intel and customers like that. Uh so not saying you know oh you know we're going to be anybody's confirmed to Intel there was a strong suggestion that there is there are people who want to do stuff on Intel and I believe EIB was was a big part of that which has been the going theory that I think is true. It's been mine forever but um it it sounds like MediaTek's doing stuff there.

52:57 >> Yeah. I mean MediaTek has been talking about Intel Foundry from the get-go practically. They're they were the first real sizable company to sign up as as far as I can remember at least publicly for Intel Foundry and I again email pack advanced packaging is very hard to come by today and makes a lot of sense right and I think again it's one of these weird things where people just sort of discount MediaTek and the fact that Intel has been able to talk about MediaTek for so long and nobody sort of gave them much credit for that speaks as much to what we think about, you know, is as much to how we sort of think about MediaTek as it does about Intel because it is I mean Media Tech's big customer. They do they ship a lot of units.

53:42 >> Yep. Yep. Well, something's brewing there is >> good but good for Intel. Good for Intel. Like I said, the street rewarded it. >> I mean, again, your I think your main point is everybody's feeling like they're in a better capital position and I agree. I think that's the right take. Um, so anyway, that is it for this week. Thanks for listening everybody and we will talk to you next time. >> Thank you everybody. Tell your friends, tell your agents to subscribe to us.

54:15 Leave us a review. >> Bring us all your claws. >> Bring us your claws.