The "Semi-Apocalypse": Market Panic, Enterprise AI, and the 2028 GPU Glut Theory

0:05Hello, everyone. Welcome to another episode of The Circuit. I am Ben Beheron. Greetings, programs.

0:11Jay Goldberg: I'm Jay Goldberg. Okay, Jay, have you heard about the latest shortage? No. What is the latest shortage? It is Apple's newest billion-dollar business, polishing cloths. There is a shortage of polishing cloths. Okay, all right, now hold on. I have to say something about this because there's two products that have entered my life to where the polishing cloth has become essential and has made me believe that everybody is gonna have these polishing cloths. Okay, so the first is I'm now fully a iPhone duo user. There's no going back. I was super skeptical about this. I publicly stated my dislike for the passport form factor.

1:01I'm still getting used to this front screen. I show it to a lot of people. They kind of say the same thing, but then you fold it and you're like, holy crap, this is the greatest thing ever. However, no cloth will polish the screen folded open, but the polishing cloth. Like, I know there's cheaper versions of this. It's not a normal microfiber cloth. Like, you have to find... But legitimately, you cannot polish this thing without this dense... Without that thing right there that you're holding in your hand. This is microfiber. Okay. The other is now ProLED. So I now have a touchscreen Mac in my life that my grubby little fingers trying to put on there.

1:48And guess what you have to clean it with? The polishing cloth. So there's a shortage on my theory, my thesis in Apple's greatest – this is part of the strategy. Give you things to touch that only their polishing cloths will clean and you got another billion dollar business.

2:13Jay Goldberg: Quick, everybody run out and find out who Apple's cloth supplier is. But the greatest thing about this was like, generally, I was like, I don't need that cloth. And then I cannot get these smudges off the nanotexture of iPhone Duo. And I'm texting Max on my team. I'm like, I can't clean these. He's like, you need the polishing cloth. I'm like, no way. I got to be able to clean this with something. Another microfiber cloth, a little bit of water and some t-shirt. Literally nothing would clean it. Only that dense microfiber polishing cloth would actually clean it. And so I just walked away being like, this is unbelievable.

2:51I need this new type of cloth to actually clean because you can't clean it with anything else. Okay. There's both my, there's a shortage of polishing cloths. People are ordering these things like it's nobody's business. There's Apple's conspiracy to sell you more polishing cloths because they're giving you big screens to touch. But I do love iPhone Duo. It is, there's no going back. I know you're waiting for yours. I'll be curious to your take once you get it. But yes, it's a great product despite the mixed reviews. Okay.

3:27Jay Goldberg: So I just did a clod search to see who makes it, and Apple hasn't said. But the suspicion is it's a company called Torre, which makes ultra suede, a polyurethane polyester microfiber chemistry. They've branded it as Alcantara. and it's milled in China. So there's a shortage today, but I'll bet you we can get some on Timu in a month. Well, and that's the other thing. I bet you that people, this is the next like YouTuber sensation is going to be comparing polishing cloths that actually work to clean off your goods. Does the$20 Amazon polishing cloth work as good as the$30 or whatever dollar polishing cloth of uh actually i don't yeah i heard prices went up anyway max on my team was like i bought i ordered more because there's a shortage of polishing cloths okay well anyway there you go there's our there's our anecdote before we go into the crazy news of the sem semocalypse semi apocalypse sem apocalypse all right well okay two two things let's let's get into the integrity of it but i I want you to give your – Jay has always been a bear, as we know, or skeptic is.

4:51However, he even thinks this is being overdone. So we need the curmudgeons. Like people are being maybe not as reasonable even from the far side. So, yes, market-wise, give us the take.

5:10Jay Goldberg: So we're recording this Friday, November 6th. The market is up a little bit today, but yesterday was one of the worst days in SOX index performance. It was a bad day for tech all around. And it was just incredible. Everything was down. Good companies, bad companies, good earnings, bad earnings, didn't matter. Everything was down and down by a lot. Lots of names were down by double digits across the board. And I just looked at all of that and it just struck me as like maybe this is a little overdone. And the other thing too is that nobody quite knows why it's down. There's no obvious news. It was two days after the election, right?

5:54Jay Goldberg: And so there are a lot of people saying, oh, there's going to be more data center moratoriums, which I don't think is going to happen. I really, really, really don't think we're going to ban data centers in the US. That just – it's not going to happen. But okay, so maybe that's it There's a few other things Mostly in the category of animal spirits People just being spooked On Monday, I think Anthropic released an updated S1 Which had a couple sort of concerning numbers in it Maybe we can get into those Maybe that spooked people But it was just, I think it was just People, the market got spooked and it became self-reinforcing at some point.

6:39Jay Goldberg: But I really, I like, as sort of cautious as I have, you know, as you described me on a lot of things in AI spend, I just, I don't see it. The companies are all reporting really good numbers. A lot of companies are reporting good numbers. The companies I talk to all have very healthy order books. I guess those can go away, but, you know, I don't know why. And then the hyperscalers reported last week and they all either affirmed or raised their CapEx numbers for next year. And so there's no one thing that caused this meltdown as far as I can tell. And in those kinds of situations, I have to wonder how much of it is just like sentiment working in reverse.

7:27Jay Goldberg: It's bad sentiment rather than the good sentiment we had last year. Sure. Yeah. But yeah, I mean like it was a weird day. I had to like restart my stock application on my phone a couple times because like this can't be right. Like that one's down too? Like what is going on here? Yeah, I mean I think there's always been the herd mentality in the market. I think it gets exasperated in today's both Twitter and the bot trading. You know, I mean, I took this picture of AMD stock, the second, like literally the second earnings dropped and it was already up like 7%. I'm like, humans can't do that. Like this is just action being praised on quants and others who can initiate those trades at bot level speeds.

8:18And again, I feel like there's a lot of that. That's like, hey, follow the, follow the cell, follow the, um, the, the volume. and automate it. And to some degree, right, there's a lot of that that comes into it, plus the fear, right? It's like, hey, if X, Y, and Z happens, do this, right? So that part is new of the volatility chain, the automation of this, which is hard to predict. But anyway, I agree with you. There's a lot of stocks that, and I'll just say this too, I feel like the sentiment then was exactly what you said. We agree next year and maybe the next year after is somewhat okay, barring some concerns of misses that just get hit.

9:05But it's like after that is now the question. It's just all going to bottom out after that. Yeah.

9:11Jay Goldberg: I think there are also two other sort of contributing factors that occurred to me. One is – I mean high-frequency trading has been around for a long time. Automated trading has been around for a long time. I think there is some AI effect taking place though. I think there are a lot of funds out there who are deeply AI-pilled and have started using AI to design trading strategies. Not just automating the trades, but designing the trade strategies. And so I think there's a lot of elaborate trades in place, very complicated structured trades, which are basically just levered bets, which amplify returns on the way up and then amplify them on the way down too.

9:56Jay Goldberg: I think there's a lot of that taking place. So there's a little bit of us all getting a little too smart for ourselves. That's one thing. The other thing, too, and maybe this is just a function of me, but this is the time of year when we start adding years to our models. You start putting out 2028 numbers. And I think that's what's going on is a lot of people are putting out 28 numbers and going, oh, I don't know what 2028 is. What assumptions did I use? And then you start – like people are literally thinking about 2028 in a more – in a very, very tangible, concrete way. Six months ago, you weren't really thinking about it.

10:38Jay Goldberg: You didn't have a model. I didn't have a model that thought about it. Now I'm thinking about 2028 a lot more. And that's sort of focusing on that year. And I think – I actually think 2028 is going to be a big change because there's lots of things that will converge at that time. and so I think there are a lot of people in the market who want to get ahead of that. I think that's partly what's taking place here. Because 2028, you think about it, we're going to have a whole bunch of fabs come online, memory fabs. We're going to have a whole bunch of data centers coming online. There's a lot of construction taking place and that's all coming up in 2028.

11:20Jay Goldberg: It's also the year when it looks like a lot of the energy bottlenecks will get resolved. So for the first time in the last 30 days, I've had a lot of people come to me talking about gluts of GPUs. And I don't think we're there yet, but maybe some people are looking at 2028 thinking, oh, we're going to have too many GPUs at some point. Yeah. There was also a thesis that I saw a number of the big cell-siders kind of align on. I don't know if they all talk to each other. I know most don't necessarily read each other's notes, but there was a digestion year kind of baked into some things that kind of wavered and went back and forth.

12:04But your point is exactly, you know, the point of one of the reports I did called, you know, next year being peak constraint, which is that, yeah, in 2028, a lot's going to come online. We're going to be able to make more memory. We're going to be able to make more accelerators and CPUs and all those things. And so the question is then, is there enough demand for what is a pretty step increase in capacity across the board to go fill data centers and have the backend compute to go and do those two things? And we'll see. You know, I think there's a lot of dynamics that still kind of go around my brain around this cycle.

12:45I get everybody wants to know that there's ROI and that's a fine, that's a fine concern. I think you can see it if you look hard enough and talk to enterprises. but but you know i was at this event with the big oem who provides a lot of services and infrastructure to enterprises this week and like you keep coming back to every conversation with enterprise decision makers is we're early we're still figuring out our business process for this there are lots of holes we have to plug to be ready for agents in our enterprise both security, both data, storage, all of these things, software services, governance, you name it, right?

13:34All of this, like businesses have to way more so than any digital transformation period we've ever gone through, really go through a structural rethink of their processes. And that's what the top 20 % of these early adopting enterprises are going through right now. You're hearing from them. They're trying to push the limits. They're learning. they're realizing what they need to shore up and they're figuring this out and then you have this other 80 that's going to lag to this that's going to be slower to adopt they're still going to do their pilots they're still going to do this in phases they're going to learn from these earlier companies that that do it but but even that top 20 that's doing this this is such a big structural change on their infrastructure on their software on their processes their paradigm that like once we get through all that like the enterprise is just going to be ripe in terms of dollars that come here it's just we've got to get over this they have to go through such a difficult transition that they've never had to do before to get ready for agents in their enterprise i yeah so i i think

14:47Jay Goldberg: if we build a counter thesis to what's taking place, I think there's a few data points here that would point very much in the direction you're talking about and very much against what the market seems to be thinking this week. And the first was Anthropics S1, right? They amended their S1 early in the week and they gave Q3 numbers. And the big problem with the Q3 numbers was that their growth slowed pretty dramatically. They went from like 150 % sequential growth to 40 % sequential growth, which is a big decline and a big issue for a company that's going to go public in theory soon. And lots of people said, oh, that's it.

15:30Jay Goldberg: Everybody has coded all the things they're going to code and they're not getting the growth because everybody wants to code. Now has it, right? And I certainly understand that. I've made similar arguments in the past. But if you read deeper into the S1, you realize that earlier in the year, they had an outage, like March, April. They had an outage for like a day where you just didn't work. And in the S1, they disclose that that was because they ran out of compute. So now imagine you're a company, you're on the verge of an IPO, and you're seeing very strong usage. You really, really, really don't want to have an outage.

16:09That looks a really bad look two weeks before your IPO, a month before your IPO.

16:15Jay Goldberg: Second data point is in Q3, they spent$10 billion on compute, but only five and a half of that was contracted. That meant they spent$4.5 billion on spot purchases of compute. I could probably even pick out the day that that started if you look at the pricing data. So here's a company that is very worried about surging usage and running out of compute at a really bad moment. And they spent a lot of money on very expensive compute. So they didn't have that. Which is a long way of saying that Anthropic is short of compute still. It's not that they're running out of users. They're short of compute.

17:02Jay Goldberg: And they just couldn't serve everybody that they wanted to serve. and they had to throttle people and delay signups and things like that. And that hurt their sales. That hurt their growth. I don't know that that's true, but it's certainly – there are a lot of things in the S1 that point in that direction. So that's piece one. Piece two is Cloudflare, right? Now, not a company we talk about much, but it's a company I've paid attention to for a long time. I don't think people really understand what Cloudflare does. It's a little bit nebulous, but they're basically, they provide a connecting layer for pulling all your remote container Kubernetes workloads under sort of one networked umbrella.

17:52Jay Goldberg: So that if you're a large enterprise and you have servers all over the world and different offices, they can all sort of operate and interconnect in a much more efficient manner. Right? It's like the networking layer that AWS provides. Right? But it's more than a CDN. It's a little more. Right? And there's a lot of networking that takes place and there's a lot of security that takes place there. Right? So they're a networking layer for distributed compute. and it turns out that that is very important for running agentic workloads and they reported this week and they had phenomenal numbers right i mean the stock is up a lot um they were very very bullish on their call they were very excited on their call and if you look at some of their numbers like the i have one stat where the the number of paying customers, they had been doing about 20, 30 ,000 customers a quarter additions.

18:55Jay Goldberg: Last year, it was about 30 ,000. Every quarter, they added about 20 ,000, 30 ,000 customers. In Q3 alone, they added 158 ,000 customers. It's just like parabolic, straight up exponential curve in their customer additions. And yeah, hobbyists will use Cloudflare, but really like most people who are using Cloudflare are doing it for work. And so this is the enterprise. And to me, that's a really important telling indicator that enterprises are working hard. And this couples with what you're talking about last week with your survey data and what you're just saying now, which is it looks a lot To me, enterprises are really starting to ramp hard on deploying AI, probably agentic AI.

19:44Jay Goldberg: And so I've asked for a long time, when is this going to be in the enterprise? When are people going to really start adopting it? And it really feels like there's something happening. There's real momentum on the enterprise side for adopting AI. Yeah. Yeah, and there's two other parts to this, right? I mean, you know, we, the first half of our survey was CIO, CTO with a bit of security. And then we did one with CSOs as well that I'm going to publish this week. And like, you can't come away from all of this data without realizing that security is the gate that has to be cleared first to agentic scale.

20:20And, and, and that's just, that's a very different muscle for a lot of companies. You know, you and I've talked about this. I wrote about this even back in June, the opportunity for both Anthropic and OpenAI to make dedicated cybersecurity models. GPT-6 is now a cyber model. They launched that yesterday on the back of their cybersecurity efforts. So you can already see them saying, you know, we know that enterprises can use AI, use the frontier, build these new processes into place. And CloudFAR is, I mean, fascinating in this part, but only solves a portion of this problem. Like you have to have, like when I say, you know, people talk about like enterprise AI or AI ops in the same way that they talk about FinOps and realizing that like from a financial standpoint or just in terms of where FinOps was, like that's a governance thing.

21:09Like you as an enterprise have to have processes in place for these very boring terms called permission and observability and all of these things. And that now has to come to agents. And like that's like such a different thing. Like I just don't think everybody fully grasps like how structural of a change this is and at the same time has to go through this like – like again, I say adoption cycle. It is the right way to think about it. But the way enterprises have to adopt these is so much more, you know, dotting your I's, crossing your T's. It's just they're not – like they're not built for what's happening.

21:51and they're trying to figure out how to solve these solutions that, you know, again, like I said, we, this vendor I was at, right, we were having these conversations about what are you guys hearing from customers? And a lot of times they're like, everybody understands this problem. Nobody's sure about what to do yet because a lot of these solutions don't exist to help them scale governed agentic production across, again, permission, observability, data, like just the data. I like problems. I don't think enterprise security is super, I mean, enterprise software is super fascinating, but I like like hard problems.

22:30The data problem just keeps coming up to me as like, I don't know how this gets solved, right? So just basic stat, right? 80 % of enterprise data created at the edge is unstructured data. And so what we had heard was a couple, you know, of these more leading edge companies are using agents to go structure that data. Okay, so now you all of a sudden as a business bring about a whole slew of information you never had context for, you didn't have any semantic data on, you're gonna index it, run context on it, have it be structured, and now all of a sudden you've got another light visibility into your business you never had, both at a productivity level, as an individual, as a BU level.

23:11Like, this is gonna create such a fascinating dynamic for enterprises as agents go and kind of help them solve some business problems that were just unsolvable before, but it has to be depleted. Like that's what I keep coming back to is just this, this is a hairy challenge and technology is going to help, but we're so early in this. It's like there's more answers. There's more questions than there are answers. How we get from A to point B in the adoption of enterprise.

23:44Jay Goldberg: I think oddly I'm more bullish on this than you are right I actually think this is happening sooner rather than later I I think you're 100 % right I all these complex issues have to get sorted out I just my my intuition right now is that a lot that a surprising number of people have done a surprising amount of work and we're further ahead of that curve um it's it feels to me like it's it may be i hope so dude because like we've talked about enterprise either funds this cycle or it doesn't like like this is literally like everybody is like how are we going to pay for ai and the answer is enterprise like they will be the first customers that fund the next few years of this because we don't know what happens in in consumer but literally enterprise is what makes this is go no go and so we we need them to not take their dare sweet time We need technological solutions to this that they can do, you know, deploy in the enterprise that chocks all their boxes from security to governance and all of those things.

24:46So I hope so. I just worry. I am a little bit worried that, like, the longer tail of the enterprise is just going to move a little bit slower because they're a little bit more cautious. They're not your typical leading-edge technology-driven early adopting of technology. But they're going to be hungry for this. Because the reality is if you don't do this, you face a competitive risk by someone who will do it, one of your competitors, faster than you. Right.

25:17Right.

25:18Jay Goldberg: I certainly – I know that in all my company conversations, I hear this repeatedly again and again. There are companies that are just rushing to adopt AI because they just see they can get their employees to be more productive using it. I've had that conversation a dozen times this week with companies. Anyway, but that's part of it. And I agree with that. I think, you know, my hope is they move. But again, right, this is the things that just make me come back to we're just still really early. And, you know, we don't know how fast this moves. But that demand is there, right? Enterprises have built-in demand for this.

25:59They'll pay for this. As they scale, the compute infrastructure demands change. the demands on power change, all of those things. And we're just early, right? So, okay, let's jump to earnings. Let's start with AMD. Lots of stuff to talk about in what is a fascinating contrast with everything we talked about with Intel. So let's get into that.

26:32Jay Goldberg: AMD reported this week, I guess. It feels like a million years ago. They had very strong numbers, both quarter and guidance, and talked up their GPU opportunity and their data center estimates for next year. All those numbers were incrementally higher. This was Lisa at her best. This is a company that's executing very well, I think. And she even sounded optimistic and upbeat, which is, you know, you should take seriously because she's not someone who, say, exaggerates much. So I was pretty encouraged by the results. A lot of it seems to be they're picking up share from Intel, who is obviously stumbling with capacity.

27:19Jay Goldberg: But also I think they're in a good position next year. They seem to be doing really well with getting their Instinct GPUs deployed at the hyperscalers. And so we're seeing the first deployments of MI455 coming on stream now, and then that will ramp up next year. And just for reference, 455 is, I think, on paper competitive on par with Rubin. um so very competitive with with what nvidia has like you know on paper you know the specs are competitive i would argue amd has has the lead in a couple categories um that's not all there is to this so it's more complicated but certainly they have a very competitive technical technical platform and they seem to be getting fans in the hyperscalers so So I want to echo your point that Lisa Sue is not the type that would signal this if there was not something there.

28:25I think, again, we've talked about this for, I want to say, like two years where we're always like a couple years ago, right? Lisa comes to either on an earnings call or at Advancing AI and we're like, get more excited, Lisa. Like, give us your conviction. And she just, that's not what she does. And so I think a lot of people gave her their feedback. Like, if you believe this, like, let's lean into it. And now you see her, like, making these kind of big, big proclamations. You know, the two big things were obviously, you know, they're signaling a bigger year for next year, which I think is noteworthy, right?

29:01So based on purely what they had for 2027, which I think was 70%, they're now saying 80 % growth. which if you if you take with some first that that's a floor which is i think a safe bet barring again apocalypse of the economy and whatever things like trends continue but there's a good chance if you just do some of their numbers right that there's hints of upside amd could conceivably do 100 billion dollars next year in revenue like that's insane like we know they'll get there someday, but the fact just that, you know, and I always hearken back to like my days when, you know, AMD was a$6 billion a year company, pretty flat.

29:46I think it was flat for, I want to say three years and that like, it wasn't growing. And the company was just like, yeah, this, this is what we are. We're a$6 billion a year company. And now you're talking this company could do a hundred billion dollars in a year. We'll inevitably at some point in time. So I thought that was good. I also thought it was interesting that they went out of their way. Lisa went out of her way and others on the call to talk about the supply chain and that they have all the things they need, both capacity, substrates, memory, all the things they need to hit their numbers.

30:20Which I think is, you know, become a little bit of a underlying concern of a lot of folks who are like, look, your numbers look great. But we're now seeing how troubling the supply chain is. Can you get enough supply to do this? And a couple other names that we'll talk about in a bit did this as well. But I thought them going out of their way to talk about they've secured what they need to hit these numbers was important because that actually was the nuance that hurt Intel. They didn't secure those things. And that became a problem to their upside. So it's kind of just the counter of execution, I guess.

30:53Obviously, they're working with TSMC who helps with all of this, gets substrates, gets memory, has the wafers, probably gave them allocation from a competitor, which we won't name, but from a competitor, probably took those wafers from that person, gave to AMD to that FSA, which again, I think is one of those variables that could swing in their direction as more upside for next year than even their guiding. But I thought those points were interesting. And then again, just broadly, they're so confident on where this is going. And they're executing on all cylinders. So despite what happened with their stock yesterday, even though it was up a little bit after hours and to start the day.

31:36So AMD – I mean AMD also for the first time crossed – their server revenue was higher than Intel's in Q4. And there's a good chance that they will have more share, like sell more server CPUs than Intel next year. It's plausible that that plays out that way, which again is – I'm not going to say it's a full shift of tide. It's just a difference in capacity, execution, supply chain that AMD has been gearing up for for some time.

32:13Jay Goldberg: Yeah, I think Prior to yesterday's meltdown I was coming to the conclusion That the theme this earning season Was If you have capacity, you win And if you don't have capacity, you get punished And AMD and Intel Were the perfect examples for that Where it really comes down to Just having Supply and lots of companies don't and they're struggling against that is absolutely a topic that has come up on every earnings call I've listened to. Management talks about it ahead of time, but they get asked many questions about it. How's your capacity? What have you signed in terms of long-term agreements? All of that stuff has come up again and again.

32:57Jay Goldberg: I mean, we had Apple report this week and even they have finally, finally signed, you know, they made$33 billion of purchase commitments in the quarter. Everybody's talking about it. Companies that have supply are doing well and those that don't are really struggling. And then sort of the next order of conversation you want to have about that is like, who's really being taken by surprise here and who is, who is, could have done things better, right? I think there's certainly suspicion that Intel could have planned this out better. But there's lots of other, there's lots of things that are beyond their control and beyond their ability to forecast and it applies to all companies it's the real so i think i think that's going to be a growing topic of conversation is who who's actually executing well who has the best operations and i think we're gonna we're gonna see a lot of that in the next quarter or two as everyone works through all the supply because again some of those stuff like substrates as we've talked about are really hard you know you either got this stuff in advance or you didn't and you're not going to all of a sudden just get the supply so it is a fascinating dynamic and you're right but but i think that's why i'll be curious if more companies in this chain do do what you know lisa did and really talk up right the the brow the um the supply chain right that they've got what they need to hit their guidance just just because i think that had become a question that lingered and they need to just put that to rest as they talk about right to go forward yeah and it it was interesting arm like i who also reported this week um they had mentioned this last quarter and the quarter before because they've they're that they're now entering the chip business so they've started to get a lot of supply chain questions and they have said earlier in the year they've said a few times that their their chief constraint is their chief worry.

34:53Jay Goldberg: The place that the CEO and CFL are spending the most time is securing supply for their pretty aggressive rollout of their own chips. This quarter, they all sounded very different about it. They sounded very confident. They clearly signed a bunch of agreements during the quarter. You could hear the relief, the weight off their shoulders that they'd gotten supply locked down at least for the near future, next year. right um didn't didn't help their stock price much but that wasn't really the issue it was but it was it was a major topic of conversation and they got asked a bunch about it even though they'd addressed yeah i mean i think there's there's there's there's one theme i want to hit broadly on a couple of these names that we were talking about so obviously we're also going to talk about qualcomm in this but but amd and arm you know amd's commentary was also i think super interesting but this is also true of the commentary that i believe both arm and qualcomm want to push more so than these other two than qualcomm although we can discuss this um you know for amd their commentary was you know we're wholesale a data center company now like like their theme their narrative is we have made this shift we are not a client computing company anymore even though they'll continue to sell into that.

36:16They are a data center company. And that's newish. Like, I know that's been coming. Those signs have been on the wall, but you just kind of come out and say, you know, this is who we are now. Like that's, that's telling, right? Arm also, right? No, like nobody's talking up client devices anymore. They're all talking about what they're doing in other areas. In Arm's case, it's, it's obviously data center, but they've got, you know, automotive and eventually robotics will come into that. And same with Qualcomm, right? They're talking about they're kind of saying like, don't, don't pay attention to the mobile stuff over there.

36:47Like just pay attention to our other, our other businesses that are, that are growing really well and becoming a larger chunk of the company in this kind of, you know, again, these devices don't go away. Like, and I'm not here saying, I'm not saying like nobody cares about consumer devices or client devices. Those don't go away. That's just not the growth factor. And, and that's just the reality of this side of the market. So it's just interesting to hear them emphasize the things that, again, people once wanted to emphasize, right? Mobile, client, all these companies had that in common, right?

37:23Their growth was dependent on, AMD was dependent on client PCs doing well. ARM was dependent on smartphones doing really well. Same with Qualcomm. And so now they're all like, look, these are other opportunities we're playing into. And to some degree, these are company-wide shifts that we're taking to capitalize on these new opportunities.

37:44Jay Goldberg: Yeah, I think it's true across all of them. I don't think we're done talking about those things, though. Certainly, they're not anybody's focus right now. But I have a feeling these subjects will come up again next year for all these companies. Because despite all the talk about data center, it's still 60 % of ARM's revenue, 60%, 70 % of Qualcomm's revenue. CPUs are still a much more important thing than GPUs for AMD. So these are still big businesses. It's just everybody is shifting. The growth is all coming on the AI data center side. And so everyone's going to spend a lot of time talking about it, but the other things aren't going to go away.

38:27No, they're not going to go away, but I think they're the growth mechanisms. And like next year, I think Akash was signaling Qualcomm's CFO that data center or the adjacency, so the non-mobile part of the business, will be bigger than the mobile part of the business. I think they were saying it would be 60 % next year. And then once ARM starts selling actual data center CPUs, that imbalance flips almost immediately as soon as that revenue starts being recognized because that's just a lot of money. That's way more than licensing. So then they are in a vast majority of the revenue is selling data center chips and to some degree right their royalties on that so um yeah anyway it's just interesting times right that that where we're at in this in this transition um but yeah so um okay anything else on those three earnings

39:28Jay Goldberg: No, I think we got it all. ARM seems to be doing very well with their chip sales. They, again, raised their estimate of how much they can sell in the sort of medium term. They're seeing a lot of interest for their chip. Qualcomm, a little bit more tepid, but they are starting to – they did mention that they have – It sounds like they have more customers getting more interested. I'm very interested in their, what do they call it, high bandwidth compute. They put out some more technical data on that. Sampling, yeah. It seems to be performing pretty well. So I think that sampling, I think that's going to get them a fair amount of interest.

40:08Jay Goldberg: Their automotive business did well. And mobile phones didn't do particularly well, not surprising.

40:20Jay Goldberg: and I think, yeah, I think that's it. I think we got it all. Yeah. Good stuff. Good point on the HBC stuff. I think that's going to be a fascinating narrative for this. Could be a real advantage for them. So, all right, let's, let's jump to side time company. We have talked about before a company. I just wrote a giant report on yesterday, kind of deep diving on the opportunity that we think we've attempted to size this market through a whole range of things via our accelerator model, networking model, because they're attached to both of those things. We've talked about this before. This company is just so interesting.

41:00And I framed it too. I know we've kind of talked about what they do. I framed it too. I've always been a fan of kind of what happens when something goes from analog to digital, right? You could look at TVs, you could look at automotive, I mean, like all these things, right? You could look at phones. And anytime we went from some analog technology to a digital one, the market exploded, got really, really, really, really big. Tons of different opportunities arise. And there are just early companies in that that take advantage of those shifts. And so in this case, right, the shift from analog to digital is the shift from quartz crystals in timing to MEMS.

41:43And this is what PsyTime does. And PsyTime is now that they've bought Renaissance and Division and that's closed, which does have some courts. So they're actually selling into that. But collectively, I think they have the vast majority of the market for timing, like as whole, right? And where this gets really interesting, and they talked about this on the call, they've talked about this a few prior and we have too, is like, as compute density increases, as racks need to be synchronized or kept in sync so that they can work together to not have a slowdown or a bottleneck in either the networking, in the CPU or the GPU or any of these things, their timing content goes up, right?

42:24And their timing content goes up for higher frequencies, parts of their roadmap where ASPs are higher. And so it's almost like almost everything we've gotten excited about technologically from networking to accelerators to density to scale up compute pods requires a whole lot more of this timing content. And that's side times bread and butter. And they're like right in the middle of this super interesting thing that's technically, technologically very interesting, technologically critical to all of the infrastructure we talk about, but also a small part of the market today growing into a much larger part of the market over the next few years.

43:04And so they're also small growing into a much larger TAM and providing a very critical solution to the wider infrastructure buildup. So all those things are just why they keep doing really well with earnings. And I just think are super well positioned. They check a lot of these boxes for, you know, again, having a large share of the timing market that's, like I said, small but growing because it's absolutely essential to infrastructure. Yeah. So I'm, I'm very fascinated by Sightime.

43:44Jay Goldberg: I've been fascinated with them for a long time. They are an interesting, just an interesting phenomenon. And I've, it's hard to explain to people what they, why they're doing so well, because to really understand what their products do you have to get so deep into like how computers function at a very very low level right and but the the basic premise is like there is time every computers depend on having reliable timing and since before we even had computers we've used quartz clocks that are not even really, you know, they're not semiconductors. They're just resonant materials. And it's a very, very basic technology that we've known how to use for a long time.

44:38And it

44:38Jay Goldberg: worked. It's very reliable. It doesn't do anything fancy, but it was very reliable. And then Sightime came along and said, why don't we do this as an active semiconductor component using MEMS? and we lose some of the technical precision, but we can make up for that in other ways. And you get all kinds of other advantages because it's programmable. And that means you can use it in a lot more places. Yes, yes, yes. I'm oversimplifying this greatly, but that's the basic premise is it's programmable, which means you can use it in a lot of other places. It gives you a lot, you know, someone's designing system a lot more design flexibility.

45:19Jay Goldberg: and so you can start throwing these little clocks everywhere and suddenly people are putting them everywhere, right? I think there's like, I want to say there's like 10 or 20 on every server board, maybe more, maybe 100 on every server. There's probably 10 or 20 in every phone. They're tiny little things and really as of now, SciTime is the only one who is doing programmable mems-based timing solutions and they just bought Rynossus' timing business, which was pretty sizable and is probably their largest competitor. And there's still going to be some things that we'll need to use quartz.

45:58Jay Goldberg: Satellites will need to use quartz. Fine. Here you go. They have a small quartz business. That's the very high end of the market,$1 ,000 parts. Most of these things are$0.10,$0.50, a buck or two. No, actually more than that. They gave pricing on the call. A lot of these things are selling for more than you'd think. And I don't know. It's kind of fun to see because it was such an obscure, purely technical company for such a long time. And they're really able to capture a lot of – they're just growing their addressable market very, very rapidly and winning a lot of it. It's an interesting stock.

46:40Jay Goldberg: It's an interesting – excuse me. It's an interesting company. It's an interesting technology. It's worth spending some time on. Agreed. I have a whole bunch of visuals in case anybody cares. And in our report, I drew kind of a board diagram of how they fit and what the relationship to increased compute density. But you're right. We don't talk about what goes on in networking. We don't go deep in the weeds on retimers, for example. And why does a Broadcom switch need a bunch of retimers next to it? Because it's interesting. It solves a problem. But it's kind of like it's an in the green, nitty-gritty thing just like this.

47:15Like, why is timing? But again, I think you come back to, which is also true, why are retimers right around Broadcom switches, which is within a cluster, within a rack, north to south, east to west, timing plays a role in those racks' synchronization together. Their ability to function as a mesh is dependent on quality timing. And that's just, it's a very big problem. The technology's, like I said, in the weeds, nitty gritty. but I love this, right? It was once analog and now they're doing it in digital and that just creates a bunch of different opportunities. So I know, you know, I've talked to investors about this, right?

47:54Who've been like, well, I mean, are they overvalued? Like, look at the price, like that's all fine. Right. The point is, is that they're growing into a really big market. They're growing fast and they're in a super interesting part of the component supply chain. Um, that's critical, growing in its critical, and giant as a market. So, yeah, Jay's right. They're worth spending time on because they're very, very interesting with a lot of upside.

48:24Jay Goldberg: It's an interesting company. I'm not going to comment on valuation. The market is – yeah, I'm not going to comment on valuation. I have no idea. But I will say that it's – I think it's interesting to read both. Yeah, dig into it. all right everybody uh that is it for today let's hope that next week is i mean it's still gonna be busy and in chaos but let's hope there's not another semi semi apocalypse semi apocalypse uh maybe rationale will continue or we'll find out what the heck happened in this week if it was something other so we will talk to you then and we look forward to it thanks again.

49:05Jay Goldberg: Thank you for listening, everybody. Tell your agents to subscribe.

49:32Thank you.

50:02Thank you.

50:32Thank you.