SPEAKER_00 0:00

On today's AI Investor Podcast, we look at blowout earnings from NVIDIA with the company forecasting a growth rate that is well in excess of Wall Street's targets. We'll also touch on a brand new chip from OpenAI and briefly cover earnings season. All that before we talk with Jorge Aragon, who has been doing research in areas like new memory technologies. Hey everyone, just Eric this week. If you're watching on YouTube, you can tell I am on the road right now. I am about to talk at uh Investicon, which I've brought up several times. Uh, I am building, I've I've spent the past couple days building a presentation for that event. I'm very excited for this. It's it's really cool. I built it using some uh, you know, some of the agentic design services, and it touches on basically where I see the AI market today. And we're going to release that after this event, probably through our YouTube channel. I might also put on the podcast, but I should say this is something that's really going to benefit from the visuals on it. So, again, if you haven't subscribed to our YouTube channel, we're going to have that coming up in the next couple weeks. It's it's something you're not gonna want to miss. Uh, now let's talk about actual news. What I'm gonna do on this episode is it's it's gonna be a little shortened. Uh, I gotta run, I gotta present in uh less than two hours. But what I wanted to do is just talk for 10 or 15 minutes about what's happening this week, and then we're gonna pitch it over to Jorge, where we're gonna talk about some of those areas uh I had teed up at the beginning of the show, such as memory pooling, and he's gonna give some stock ideas that were just a lot of fun. Now, as far as news this week, the main event was NVIDIA reporting earnings yesterday. Uh, that's Wednesday, August 26th. And the headline figure is a 70% growth estimate for next week. Wall Street was at about 44% for comparison. And over and over again in the conference call, NVIDIA took pains to reiterate this would be much higher if they were not supply constrained. Now, my estimates put NVIDIA after yesterday's earnings at about 410 billion in sales uh across the coming year. That is going to accelerate to 695 billion the year after. So the numbers here are NVIDIA's trading for about 13 times its calendar 2027 earnings, which is a 32% discount to the broader market. They're trading for about 10.5 times 2028 earnings, which is a 41% discount to where the broader market uh trades. Once again, coming against a backdrop of growing their sales by 70%. And that number would be much higher if they were not supply constrained. So I've said it over and over again. Broadcom appeared to NVIDIA, and a company riding extremely similar dynamics, is trading for somewhere around 10x 2028 earnings. The market the estimates from Wall Street just haven't caught up to that reality yet. But NVIDIA just gave the market a forecast that makes this broader situation extremely clear. Second, beyond NVIDIA's earnings news, there's also news that the company is buying Hugging Face. The simplest explanation for Hugging Face is they're um some of a GitHub for AI models. And this comes after a $6 billion licensing deal with Poolside for a model factory that's a software system to build open way models. Now, I need a little more time to digest what's going on here. This was just announced uh very, very recently before I started filming. But I do think part of this situation is Intel's uh founder, Andy Grove, had uh published a book, Only the Paranoid Survive. And I think that's a little bit what's happening with NVIDIA CEO today, Jensen Huang, where he is being paranoid and making sure that NVIDIA has a leading capability in models. And I think the next major news story this week says why that is, and that is OpenAI announcing a new chip named Jalapeno, a spicy chip name. And the news on this, the semi-analysis headline, uh put it pretty bluntly. Their headline was OpenAI Jalapeno better than Nvidia Blackwell. And of course, what they are referencing here is that from their initial tests on jalapeno, it's going to have, in many ways, superior total um cost of ownership compared to Blackwell. And this is a chip that looks very competitive relative to other leading chips. Now, some caveats here I gotta talk about. This isn't a chip that's going to be built in enough volume to threaten NVIDIA, at least anytime soon. And it appears to be more specialized. It would be an encroachment on what you'd call Cerebrus' turf at OpenAI. But why is this meaningful? Well, this project moved extremely fast and appears to have created extremely capable chip in its first generation. For perspective, Google's TPUs, which are the standard bearer for custom chip projects, they're talking about their ninth and tenth generations right now. And TPUs took a long time for the project to scale. Um, this news really points to some capable custom projects for accelerators coming out in the next couple years. And the actual chip design was handled by OpenAI. Uh, we've talked about the idea of a recursive self-improvement for models, the idea of crossing over this threshold where models are going to be able to effectively iterate and improve themselves with relatively little uh input from the designers, the actual humans working on the projects. Well, it appears that jalapeno is some form of RSI for chip design, where a very small team from OpenAI had presented this at um a conference named Hot Chips. That's kind of the Super Bowl for you know nerds in the chip space. And the entire team from OpenAI came out on stage, and I didn't do a count of how many there were, but it appeared to be 30 or fewer people, and they were able to design a chip in an extremely narrow time frame that's extremely competitive. And what the experts that I'm reading are saying, they're doing this with maybe 99% of source code being AI generated. Now, a few further notes on this. I I should note too, well, it was designed by OpenAI, both Broadcom and Celestica are industrializing it as part of their manufacturing and engineering part um uh uh partnership. So the those two stocks, Broadcom and Celestica, are two stocks that have been punished recently, despite really positive business momentum across the next couple of years. I've talked about this with Broadcom. I talked about earlier in the episode, and and they continue to have some great news that points towards where their business trajectory is going to be headed in 2027 and 28, and how apparent it will be uh that they are trading at multiples similar to NVIDIA while delivering uh similar growth to NVIDIA. But a couple really big questions I think I'm gonna have to dive into in later episodes, because again, this is just spending a little time catching everyone up on the news before we throw to that segment with Jorge. Um, you know, what does this news say about future moats in the semiconductor space and and and what it's going to mean? What are the broad implications of um this kind of RSI moment in the design of chips? That's that's that's gonna be a big question to bite off and chew. And what does it say about the future of um EDA tools from companies like Cadence and Synopsis? We still have so few details here. I'm I'm gonna be digging into that across the next week as well. But um, yeah, as I said, jalapeno. Spicy news here, more unknowens than knowings at this moment, but but something I'm gonna be spending a lot of time on when we get to next week's episode. And and I'll say too, in broader news, I did want to just quickly touch on earnings beyond NVIDIA, even though that is going to be 99% of the media's attention. Uh, NVIDIA wasn't the only company reporting earnings this year. We have passed kind of the peak of this earnings season, but we had Semtech also reporting this week, and they guided to 45% sequential growth in data center. That that's not 45% growth uh in 2027. That is 45% growth next quarter. And uh a key part of this is uh fast acceleration in active copper cables. They remain a strong partner to MediaTech's cheaper TPUs. They're expected to ship in volumes across 2027. Uh they're gonna see a massive scale-up. The market, I don't think, is really fully absorbed. And and keep in mind, too, still in the background, the industry is racing towards faster networking speeds 1.6 T and 3.2 T. Um, that's really gonna be felt kind of in this 2028 through 2030 era. And SemTech has said that they have a 10x revenue opportunity as uh these newer technologies scale. So again, this is a company that's gonna see massive growth across 2027 and it's gonna have tailwinds to grow significantly in excess of the broader industry once we get to you know what I've called that zone of uncertainty in 2020 through 2030, where it's a little less clear what growth rates we have. So I really like SemTech. I had recently re-recommended it at that market bottom moment. If it's one that you haven't had to your portfolio, uh they continue to fire on all cylinders. And and let's get to the main event today. Once again, this is going to be with Jorge Aragon. He is an embedded uh systems engineer. Uh, he he reached out to Austin and I saying he wanted to do some work with us. He's been publishing articles on 24-7 Wall Street that you can find by searching his name, or we'll also put in the show notes how you can follow him on social media. And uh the key topic I wanted to talk about was some approaches to what you'd call disaggregated memory, essentially creating uh pools of memory that can be accessed by CPUs and servers. And there's lots of implications here. You know, first of all, the obvious one, this is one strategy to combat this memory bottleneck of too little supply for memory relative to demand in coming years and and uh the cost that this is going to build for a lot of companies. Um, but there's also the side of it, too, that there are some key beneficiaries from memory pooling, such as Marvell. So we're gonna get Jorge's take on uh what companies benefit most broadly from this trend. And and then we also do cover just some broader stock ideas. So that's gonna be a lot of fun. But back to a normal schedule next week. I won't be recording again uh with a hotel room in the background. And uh, like I said, I have been across recent days, I have been working almost nonstop trying to build this presentation that uh I'm going to first unveil at InvestiCon and then we will bring out and put on our YouTube and across other channels on 24-7 Wall Street. And on that, I've got some new stock ideas that aren't a part of the portfolio. So we'll have some coming recommendations from that. And in addition, it just really encapsulates what has happened in 2026. I I have some really stunning stats around just what the takeoff of agentic AI has looked like. What does this mean in numbers? I've got stats around what this has meant to the growth of areas like CPUs and memory, and you know, that obviously flows down to recommendations across the portfolio. And then I have some of the key themes that we need to look to as the calendar turns from 2026 to 2027 and and even beyond, and some of the big questions around what the future of AI looks like. So I am incredibly excited to get that presentation into uh the figurative hands of AI investor uh podcast listeners. And and you know, I'm being a little bit biased here, but I I do believe in terms of a broad overview, in terms of being topical, in terms of being something where there's not a lot of fat on the bone. This isn't some 200-page paper, this is something that's very accessible, is is one of, if not the best, presentations on the current AI market. So get excited for that. I I I am very proud of the work that's been done, and uh uh I'm excited that we're gonna be able to get that to podcast listeners very soon. So, with that, let's uh send it over to this segment with Jorge. Hey everyone, I'm here with Jorge Aragon, which is actually a wonderful thing because he had reached out to us on 24-7 Wall Street and said, Hey, I'm a big fan of the podcast, and I would just like to know how I could get involved. Um, it's it's one of the coolest stories we have. And I I've worked with Jorge, I would love to work with him more. Um, you know, we're just always kind of looking for projects, and we have him on a specific research project right now around memory and some of the kind of competing ideas in memory that could be really exciting. And he's also going to talk about everything from Nokia to natural gas stocks today. So, Jorge, I just wanted to say um let's talk a little bit about your background. Let's let's how how did you discover us on the AI Investor Podcast? And and you know, what in your background made you interested in this space?

SPEAKER_01 15:20

Sure thing, Eric. Well, um, my background is related to embed systems. That's um basically computers that are connected to the uh to specific to specific systems, for instance, let's just talk about um a keyboard or a mouse controller, all of those have embedded computers. GPUs are no exception, they have a specific software design for them, firmware in this case. And well, with the with all this um news about AI, I just got very interested in what else can I do. Um maybe the solution is to get more involvement in this, and in my area is just like semiconductors, firmware, and basically we all know that NVIDIA is right now the star, and that's why I got involved with it. Um, December, I think, or maybe January. I discovered the podcast in YouTube, and I uh was really impressed by all of the ideas that were presented there. I think uh they are kind of unique in in some sort, they have um like financial details but also like conceptualization of the future, so that's why I believe the podcast was very interesting. Then I discovered that the podcast was a little bit small. However, their presenters, like Eric and Austin, uh, they have a whole career behind them and have been involved in Wall Street and so on, and that's why I um think that was like a high high hide and jewel hide and jewel in in the mind, right? So there's a lot of knowledge, and I really got interested in in the podcast. Then I saw one of other episodes in which Austin mentioned that we can collaborate, not just like watching, but also like writing. I got an interview with him, and after like five or six months, I've been here writing for 24-7 Wall Street and have a couple of um words, and well, I have been enjoying my time here. It's like a very nice hobby.

SPEAKER_00 17:33

Yeah, and for anyone out there, we'll put in the show notes, but it would be J-O-R-G-E space A-R-A-G-O-N, and you could search 24-7 articles underneath that. You'd be able to reach his author profile. So you'd you'd be able to see what Horror K is writing about. But today, again, we want to talk about this memory wall, this memory bottleneck. As we've talked about a lot on the show, the biggest problem that's happening right now is we have an explosion of agentic AI. We've talked about all year. Um, you need to hold the weights for the model. Mythos, which is the the grand papi from anthropic, reportedly has 10 trillion parameters. Uh to hold the model weights, you need close to an NVL72 server from NVIDIA, which is pretty crazy. And second, we are now running agentic loops, which is putting a lot of strain on the KV cache, which essentially is the notes that you're making as you kind of repeat again and again and again. And with models, it's not that you are needing to redo it for a run. It's you know, each word that you're processing is the fact that you need to go back and and utilize all this memory. So we have this memory bottleneck that is founded by the need for memory in AI, and it's also founded by the idea that memory hasn't advanced at the same pace as logic historically. And what we're seeing right now is because of this memory intensity, especially in the agentic era, memory as a percent of hyperscaler spent has gone from 14% in 2025, 37% in 2026, up to an estimate 63% next year. Um, so Jorge, as we're filming this, which is on Monday night, August 24th, NVIDIA over the weekend announced that they were gonna raise prices by 15%. Normally, that would make a stock pop. Oh my goodness, you have that kind of pricing power, but nope, NVIDIA had its seventh losing day in a row because this wasn't read by the market as NVIDIA has so much pricing power. It was read as NVIDIA is needing to pass through costs because they are basically beholden to memory prices. So let's let's talk about CXL. And this is kind of Austin keeps bringing up, oh, we've got homework. Homework for most listeners is very bad. For you, it's good. Because I have given you homework to talk about CXL because this is maybe a little bit of a technique around this memory bottleneck, and it's also maybe a little upside for some stocks in the portfolio and some stocks we may recommend in the future. So, what I'll just ask for you. I know this is a very in-depth key up here, but just let's talk about CXL, and maybe we could talk about what it is, and then we could start paving the path for some stocks that might benefit for it. Um, so what have you seen from your research about CXL?

SPEAKER_01 20:54

Sure. Well, CXL is basically an open standard that runs over PCI Express. You know, uh PC Express is one of the computer protocols that allows certain peripherals, such as GPUs, CPUs, RAM, in this case, SSSDs, to communicate all over the place. And it's a very fast link. And basically, this new protocol allows for certain operations that normally in PCI Express you cannot do. For instance, memory pooling, which is uh very interesting, and maybe we can discuss about it. But um it allows to um to recycle memory, it allows to virtualize memory, so basically it can help reduce the stress in the memory consumption for data centers.

SPEAKER_00 21:43

Yeah, and so everything that we're seeing across basically all of AI, right? Um it is we we we were talking about this before we filmed, but um essentially there's there's a series of bottlenecks where memory I began with memory. Memory is essentially bringing memory closer to the actual GPU die. It's it's being able to create faster speeds for communication data. Um co-package optics is is a very similar idea. So let's maybe just dig in a little bit on this. Um so you mentioned CXL being a technique to offload memory. What does that mean in terms of we we know that memory and uh kind of bandwidth is so important? How does this work that you'd be able to basically create some hot memory or you'd be able to move memory into other places? Like how are how are you seeing this in terms of what this means for the overall memory market?

SPEAKER_01 22:48

Okay, just let's just think it about like a way to optimize the memory usage. For instance, um let's just think about the case in which a GPU has the amount of DRAM in its system. Let's talk about round numbers, 1000 gigabytes. And the CPU maybe doesn't have that amount of of um of storage or sorry, DRAM. Let's talk about 500 gigabytes. In this case, the CPU is constrained by that kind of memory, that uh amount of memory. And if it requires more, it just needs to wait, uh depopulate the the memory it has, and then wait and then do the next processing event. But with CXL, basically you can optimize the use of that memory. For instance, if that memory from GPU is idle, the CPU can actually use it as part of their like let's call it buffer. So it's one that's an advantage is memory optimization throughout all of the server rack. And um basically that's like the main advantage of it.

SPEAKER_00 23:53

Yeah, and so it is really interesting. I I've got a few follow-throughs on this. Number one, you know, we're going to see this. By the way, this is what probably makes Marvell's acquisition of Celestial AI maybe like one of the best acquisitions in the past 10 years. Is we're going to see memory pooling and we're going to see a broader fabric, which is how you kind of tier memory usage. And Celestial AI is going to make this possibly a home run for Marvell. But you've got some other stocks. So you had learned to me that you're really interested in Sterra Labs, which is a stock I haven't recommended in here. And um, you know, this is a company that has a lot of success in read timers. A part of the reason that I've never recommended them as part of the portfolio is just evaluation when we get into the 2028 period. So I'd be curious what your thoughts are. Another thing you had mentioned to me is Synopsis, which is a battleground stock at this point. And you had said um maybe, maybe has some benefits here. So I would just love if you start maybe um talking about kind of the stock side here. You know, we've got this uh technology that's going to be something that dramatically improves um kind of this memory bottleneck, and there's there's going to be stocks that benefit. So how would you frame the leaders versus the next tier versus the next tier in terms of what CXL could do for stocks?

SPEAKER_01 25:25

Yeah, maybe I forgot to mention like the most important part that I visualize with CXL is that you can virtualize memory. So I was reading uh one document from Marvel, and they say that this technology, at least for them, allows them to recycle all their generation DRAM. Because normally you have like the CPU, they have like a socket in which they physically need to be put in the PCB, which is like the main board of the server rack. And that PCB socket is limited, like physically limited, by a certain amount of RAM sticks you can put or like HPM uh models you can put. But uh they need to be like um they need to match the CPU architecture. So for instance, if it only supports DDR5, uh it's only limited to that. And you need to draw away all of the older models that still work, right? The thing is that with CXL, it requires virtualization. So for the CXL controller, um, either if it's DDR4 or DDR5, it doesn't matter. And in that way, like older generation of DRAM can actually be still used and will still provide value so that it actually lowers the supply constraint of memory because you already have that in the data centers. You're just using older technology with a little bit um a laggier, lagger behavior, but still working, right? And with that in mind, um I believe like maybe Marvel is like the top tier in this case. Um not only because they already have like the structural family, it's compromised of structure X, A, and S, I believe. Um all of them have like different advantages. X is just for like memory pooling. And A is for like also processing or making calculations in the memory without actually going to the main accelerator or the CPU. You can just like, for instance, process um um searching for a certain uh value in the in the memory, right? You don't have to work with the CPU or the GPU, that's that's better. And you also have the switches, which allows for this polling and for memory disaggregation, which basically stands for having uh certain parts of the memory virtualized for a GPU, for a CPU, or for uh for whomever for who needs it, like whichever accelerator accelerators needed. Um like that's like the neat part. They also already have a strategic uh partnership with Samsung and with SK Heinix, which you know they are the biggest uh DRAM producers in the world. Well, DRAM advanced nodes, right? And they um have the alliance and they do the CXL controllers for them, so just imagine in that potential. Um, it's not the case for micron, they still do their own um controllers, but seems to be a little bit laggy. We have different tiers of CXL, CXL1, two, or three, and in this case, micron um controllers, at least to my knowledge, are limited to CXL2. And and in this case, Marvel are limited to CXL3 and beyond.

SPEAKER_00 28:40

Um go ahead. Uh I was just gonna say it's gonna be really interesting. The input output, which again is just interconnects. Yes, you know, you look at you look at the uh Marvell deal with Google, and everyone focused on the basically custom accelerator, which a lot of I talked about in the last episode, a lot of people were using it as an excuse to sell off Broadcom. It it's actually not that, it's it's not the same sock as Broadcom at all. It is a completely different kind of accelerator. And where the majority of the revenue for Marvell is coming from is just this broad data center connections, this interconnect, and and what you're describing here is just Marvell's dominance across this broad layer, which I don't think anyone understands how much money AI networking is going to provide. And Marvell is very good at this, very good at this.

SPEAKER_01 29:38

Yes, and well, you were also asking about um my thoughts on Astera Labs and Synopsis. Um, after a little bit more of research, I'm not really sure about Astera's mode here, mainly because um, at least of what they present in the product lines, it's just limited to memory polling. So they just have their own um, let's call it like PCI Express uh card with uh specific DRAM models that they include and their CXL controller, but they don't have this integration with the memory manufacturers that is like the real mode here for Marvel, right? That's my chat. Um regarding synopsis, well, um basically all of the IP regarding PCI Express depends on synopsis. Uh we uh we have PCI Express 5 right now, but they are already working on PCI 7, so that's like the neat part here. That um maybe it's not like a hardware company, it's more like a software and IP company, but basically everyone depends on them for like improving this kind of um like bandwidth and capacity in the in the link.

SPEAKER_00 30:51

Yeah, Synopsis is really interesting in terms of companies in the portfolio that probably don't matter. I don't know if a lot of people who watch the show own it because I recommend it so early, but it provides me more heartache than anything because I can go back and forth on an extreme bowl case and also a bear case on it. There is so much uncertainty right now on what gets commoditized right now in EDA. Um, but I I I appreciate your perspective on this because it is again a stock I've I've had to think a lot about and remains a relatively outsized holding in my personal portfolio. Also, the Astera Lab thing, yeah, like it's interesting. There's a big conference going on named Hot Chips right now as we record this, and there's been a lot of discussion on custom memory and who's gonna play into that, etc. So it definitely seems like the winds are blowing a little bit more on Marvell's fortune than Acera Labs. But again, this this does get deeply technical. So I would also just bring it kind of to the we wanted to talk about some just kind of in the weeds stuff. Uh we we have you looking at CXL, but you have also been looking at Nokia. I'm very interested in this stock. This is one that we had made an initial investment in Siena. We were incredibly ahead of the curve on that. I don't know what's up today. It's probably four or five hundred percent from our initial cost basis. It has come back from recent highs, but Nokia is a stock that when I look at it, it's down 41% from June. A very common story amongst AI stocks. I would put it in a little bit of the bucket we've talked about with some like a corning, where broadly diversified plays um across scale across, which is basically connecting data centers. Um, they had a certain level of exuberance and their main reality. And now the question is with stocks off the recent highs, are they deals? So when I look at Nokia, um, you know, Jorge, you're a longtime listener of the podcast, the number one thing I look at right now is I believe we have visibility into 2028. And I want to take a multiple from there and I want to evaluate the attractiveness of a stock. I love stocks that got trading 50 times 2028 have re-rated down dramatically because you you don't know what the growth rate from that moment is. So Nokia is at 17 times. So you've been doing some research on them. I would just be curious because I love the scale across theme. I think the In Fenera acquisition for Nokia could be one of the better acquisitions we've seen this past second. What are you seeing? Um, do you think we should be adding this one to the portfolio? Just tell me about what your research is shown.

SPEAKER_01 33:42

Well, yeah, maybe I can think I can talk a little bit about Nokia before and after that we can make conclusions. But well, you know, Nokia, um, they made all the indestructible fonts, but it's yeah, it's not kind of it's not that kind of company anymore. They basically focus on telecom, on patents, on um data centers, and also in military. But the Tailwinds for Nokia are, I think, are also in telecom and in data centers. They can be a strategic partner for robotics and edge computing, basically because in the 6G generation of the network communications, um, they're establishing um like in the in the definition of 6G that you have to be including mapping of like the whole visibility of the run station. That means that phones, people, and objects can be tracked in all the map of the city with these kind of stations, and that's like a necessity for like robotics and humanoid robotics because otherwise maybe um it's not that easy to get around the city. So that's one thing. The other thing is that they are working with NVIDIA for AI AI RAN, which is basically the evolution of like their telecom uh stations. Um instead of providing their own ASICs for computing, they are working with NVIDIA in parallel and are bringing GPUs to the RAN stations. That means that for Nokia, it's an advancement because uh now the RAN is software defined, it's not hardware-defined. You can um improve it as it's needed, but also you can provide a little bit of inference capacity for edge devices. So, in the case of just think about an automotive or um maybe like um a robot, you can have your edge processor, but also you can have fast access to a supercomputer right next to the corner. So that's one of the tailwinds for um for Nokia. And they also have, um, as you mentioned, the Infinera acquisition last year is starting to be present because most of their expected revenue this year and the coming years is from data centers. I heard the CEO mentioned they are expecting to grow like 25% in the coming years. And as you mentioned, Infinera provided DSP knowledge, optical interconnects, also switches technology, optical switches technology. And well, yeah, it's like uh a new player in these old ecosystems of communications, both in data centers and outside of the data centers, like for the commercial use. Uh, they have a drawback, as you mentioned. That's because um, like maybe um there's they're spending too much. That's how the market is seeing it. But yeah, this is due to the acquisitions to um changes in the company, like restructuring. But seems like we are still playing for the long run with them. NVIDIA themselves invested $1 billion in the company, it's not that much, but still um reflects that uh NVIDIA and Jensen Huang believes in them, right? So that's why I I think uh Nokia is like a very good company in this matter.

SPEAKER_00 37:13

Yeah, I I love the thesis. It's funny, I I hadn't discussed this with you. In 2014, I actually flew to the northern end of Finland, a town named Ulu, uh Olu, maybe. I'm so bad pronunciations, I even go places and can't name it. But um, it was a previous RD station for NVIDIA and uh, or sorry, sorry, Nokia. And why I was going there was because they were doing a lot of internet of things, um, ideas. But it was it was interesting at at the peak of Nokia's presence, they were such a percentage of the economy of Finland. But it's what you were talking about. My initial thesis on Sienna was basically uh a thesis built around them being a leader in a market I thought was about to have a broad refactoring, which I thought there was always a limiter in terms of how high you could rate as a company that was traditionally telecom, because I thought they were about to have such explosive growth rates in what you call datacom or scale across, as we've talked about, connecting data centers. And that played up. And that's why that's been such a massive winner. And it was just a matter of getting ahead in timing. Now, Nokia, it's it's gonna be a little bit of a matter of what market expectations are. You talked about growth rate at 25% for next year. I I've recently seen a few things. So, number one, Cisco's recent earnings, Cisco dropped like actually eight or nine percent the day after. People didn't like them. But the reason that they like the earnings was um their telecom segment was actually fantastic. So Nokia is transitioning from telecom to this kind of scale across opportunity, but their telecom opportunity might actually be pretty good this next year. So, what a lot of Wall Street's believing is that their growth is going to be actually closer to above 30% next year, which they're modeling in closer to uh 20% growth rate. So if they do that, they're gonna deliver in a big way. JP Morgan recently put them as a top pick with 100% upside potential. I know Jeffries has them as another top pick. So I really like this. Um, we'll keep talking about this one. I think there's an above average chance I'm I'm going to add no key to the portfolio. And what I've talked about is just to some degree I don't see this as an entirely risky position. I think scale across this growth is gonna be pretty great in the years to come. You've got stock that's off 40%. Um I like its position right now, so I I I really appreciate you bubbling this up, but I think beyond this, what we should talk about. You've got a few other ideas, and these are gonna go into pretty crazy territory. These are gonna go into pretty crazy territory. I've only briefly talked about very briefly, I think in a recent episode, I talked about um um ceramic capacitors, which is MLCC, if you see the acronym. You have an idea here. You also have a high density density battery idea, which I love, which is so crazy. And you also have an energy idea. So maybe we'll take these one by one if you want to talk about them. Um, but we are definitely introducing a poo-poo platter of new ideas to people on the podcast today. So let's um let's talk about your first idea, which I'm not even gonna try and pronounce it. You you actually know people working with this company, so let's just talk about your idea around uh some uh uh power components.

SPEAKER_01 41:00

Sure. Well, um the first company is called Yageo. Uh I don't know if it's pronounced correctly, it's a Taiwanese company, more like a conglomerate of several um semiconductor um companies. In this case, um I personally know one of them. I have visited the plant here in Mexico, and I saw how they made like the ceramic capacitors. The idea here is that there are reports that establish that basically with all these changes in power demands from the racks uh for from the blackwell to better rubbe in transformation, they're expecting like um like a value per rack of uh from $1,500 to $4,300 just in ceramic capacitors. So they specialize in that kind of stuff, and basically it's a bulk um, it's it's like a bulk industry. You need to produce as much as you can so that you have um like profitability. And since like JKO is like a monster. That's one of those.

SPEAKER_00 42:07

You know, and I I can't even remember the names because there's there's a couple other companies, is that Mirada and um Samsung has a smaller division. There's you know, the the thing about these MLCCs, this is when I talked about the other day um in a recent podcast, is they went through a cycle that normally takes three years and three months, and they went all the way up and they went all the way down. And this is largely a commodity component, but it's a commodity component in a really great space. And I think what we'll look at in coming episodes. I've I was talking with Jorge before the show. I'm not gonna reveal which stock I was talking about. You're gonna have to wait for a future podcast for it. But one company I actually think might be an enduring winner in this kind of power bottleneck we have coming. But you know, you have companies that you can buy as commodity producers that you think just the overall market would be better, and then you have companies that you might buy for some differentiation. So I think these are companies that just the overall tailwinds are so good you might want to consider them. So let's let's talk about um you had another company which is a uh high density battery breakthrough, which is really fun. Like, oh my god, the upside if you can actually do something really innovative and high density batteries is incredible. Um, I don't think we've ever talked about this idea in the podcast, even though it's so good. So, what are you looking at in this space?

SPEAKER_01 43:41

Well, basically, you know, uh the next move for AI is the physical AI, like the robotics barrier, and that's it stands for mobile robotics. Um, the only way to move them is basically to use batteries, and you want them to stand a lot of time like up. Right. So the idea is to go for the producers that have the highest capacity or highest energy density for those kind of batteries. One of those is Amprius. They actually are like they specialize in batteries for drones. And I believe they also provide batteries for like the defense department. So they're very good. And well, this idea came to me because I was reviewing this DARPA lift challenge. I don't know if you have heard about it. It's basically one of the DARPA challenges in which a drone was required to lift like I believe it was like 100 pounds for three kilometers. And you have to make the drone as light as possible. The relation between weight that it can carry and the weight it it weights, it should be like three to one. So it should have in this case weights 30 30 pounds and lift 100 pounds. And the only way to do this is to reduce the the energy density the weight density while increasing the energy density. Since like Ampius is um the global leader in that, and maybe we can see them like a big winner in the robotics race.

SPEAKER_00 45:13

Yeah, and that would be uh ticker symbol AMPX. Now I should note for them, you go from a SPAC in 2022 all the way down to I'm just looking at a chart right now at sub $1. They got over $20 in June, and they're trying at $10. So this is a very common, you know. We we we noted many, many times. We felt things were very elevated in June. And uh you can only do so much to talk about that, and that we expected a reversion. We are experiencing a reversion, and the question now is are some of the things that have experienced that reversion, are they deals now or um is there more to fall? But you know, I I haven't looked at this, and I think this will be something that's very interesting to follow, but um it is in that kind of zone right now where you decide if it's if it's an opportunity that it will continue the acceleration or if it will um maybe move back. But let's talk about a stock that has a completely different profile right now. That in recent episodes I've talked about this idea that we could have a bull market for natural gas. And it is interesting if you if you buy uh Bloom Energy or many behind-the-grid companies, if natural gas accelerates to an extreme degree, it it could be bad for your portfolio because um it's gonna make that fuel source a lot less competitive, but you can in a way hedge it, or you can just go directly to buying more things that uh benefit from natural gas increasing in price. This is a very tight market. Um, we talked about the invest like the best episode. I think it was with Matt Smith. Um, it was maybe three or four episodes again, but he had proposed that natural gas prices would grow by three or maybe even four X coming forward. And you have some ideas in the space. So, what are you looking at in the natural gas space?

SPEAKER_01 47:20

Yeah, basically it's just like infrastructure provider for oil and gas companies. One of those is water breach infrastructure. Um they work in the Delaware basin, is like, well, I think it's the Premium Basin, is one of the biggest oil rigs or oil basins in the whole US. It produces like three million barrels per day, uh, which is like if it was a nation, it would be like the 10 biggest producer nation in the world, like just this field. And they provide um strategic uh resources. In this case, it's just like water, but they provide the water that the the oil and gas manufacturer companies require. So that's like the remote, they already positioned, they are in this area, which is one of the um biggest producers in the world. And well, we know that basically most of the data centers are also being established in Texas because they have like energy um energy supply that is um very high, and basically I think is one of the the lowest price in the whole US, right? For like energy.

SPEAKER_00 48:35

Yeah, it is it is really interesting. That that point that you made about like how much energy the US is outputting is relatively crazy. You know, you think about Pemex in Mexico. I mean, just how much incremental supply the US has brought on is just totally crazy. Um yeah, but it is the only thing I would say on natural gas stuff is I I've highlighted some of the recent work there. Um it is such a tight market. Um I think I would love to get some more plays there. I I think we're gonna do some recommendations in the space. Um, if anything, again, it is something of a hedge against maybe some of the other companies we're buying, but it's it's it's hard. And yeah, I appreciate I appreciate getting some new ideas because you never know where following a specific technology is gonna take you. Last episode we talked about healthcare. I think AI is gonna be pre increasingly going to healthcare. I think it's gonna be increasingly going to energy. Um the the thing that's gonna blow people's minds next year is we are going probably to $1.4 trillion in CapEx. What are the next level plays on that? What are the next level plays on that? And uh we we called at the beginning of this year CPUs, we're gonna be big. Um, we've we've spoken many times about the impacts of memory. We've we called that beforehand. We you know, one of the other areas that we continue to talk about is something like optics, right? That if you can build a system of 576 GPUs working together, well, they're all going to share a memory resource. And how you're gonna do that is you're gonna use more optics. So you're gonna get around more memory spent by using more optics. And we've had our bets there. And it is just kind of like what are the next things? What are the next things? I think this is what you've done a great job today talking about. Um is something like battery density and next. It's something like natural gas and next, is something like memory pooling and next. So I really appreciate the research that you brought today because we've won big in the past and we want to win big in the future. And uh we need to set up for that, right? So um, Jorge, uh I I don't know what else to say. Is is there anything else you want to say before you kind of conclude this episode?

SPEAKER_01 51:22

Um, thank you for having me here. I also send in one greetings to my spouse, to my wife. Um thanks, thanks, Eric. And also Austin, he's not here, but I have to say thank you too.

SPEAKER_00 51:38

All right, very good. All right, everyone. Well, we will see you on the next episode uh when I am back from Dublin, where we do not need to record it so far in the future. And uh again, thank you for Jorge. We thank everyone for tuning in today, and we'll see you next week.

SPEAKER_01 51:54

See you. Thanks.

SPEAKER_00 51:56

The AI Investor Podcast is for educational purposes only and should not be considered investment advice.