SPEAKER_01 0:00

You are listening to the AI Investor Podcast from 24-7 Wall Street. On today's episode, AI Volatility is back again. What else is new? There's a next AI major growth opportunity, and this time it's in healthcare, and it looks like we might be curing cancer. Marvell has a huge win with a massive new Google contract. The memory trade is on the rebound, so anybody who's looking at the optics memory argument, we know who the winner is, at least as of right now. And then we're going to close out with some listener questions. All that and more is next. Eric, it's good to see you. We are in the uh we're in the last aughts of summer here. What are what are you guys doing to celebrate this this home stretcher? Are your kids back in school or is this like a next week thing?

SPEAKER_00 0:37

Yeah, they're back next Monday. Um, you know, Austin, next week, I I've talked about on the show, I'm gonna be at Investicon in Ireland. So, you know, one final plug for this. If if anyone, especially if you're in Europe and it's a short flight, I do have a promo code. We'll put in the notes. I think there's still a few seats left. I'm planning on giving a presentation that's just kind of what I'm most excited about in AI. I'm gonna firm up all my biggest themes, talk about them. I'm planning on giving away a special presentation to people at the event. So anyone, you know, I know I know it is last minute, but if if you are in Europe and you do want to hear a really good presentation, you're gonna get your value back. There's gonna be a ton of stock picks as part of it. I would talk about that. But Austin, you know, this week we've seen stocks declining, as you said, but also some really cool stuff. I know you talked about the cure and cancer thing. Moderna yesterday was up 175%, the largest one-day gain ever for an SP 500 stock, but a day, uh gain that I should say followed some pretty astounding news as well. So I think um we've got a lot of exciting stuff to talk about. It's it's nice sometimes to talk about the aspirational side and and what the end outputs from AI are actually going to do to make the world better. Uh, because we spend so much time focusing on fear around AI replacing jobs or concerns about data centers. It is sometimes nice just to have uh, you know, the brighter side of news, Austin.

SPEAKER_01 2:03

Yeah, let's do a little, let's do a little optimism episode here. You know, certainly uh probably a good time because I I know our portfolios have been, you know, pretty volatile the last few weeks. The AI trade got you know bloody in some pockets. Now there's been some nice rebounds as well, but it's nice to remember what we are you know investing towards and why the technology has so much potential. That's one of the reasons robotics have been so exciting for me, because I feel like that's when a lot of people will actually truly start to benefit from AI in their daily lives, whether it's you know, farming or caring for the elderly or transporting people around who can't drive, whatever the case may be. And and healthcare, which I know we're also going to talk about, is another end state, which we've not spent a lot of time on, but has that same sort of real-world um atoms potential where we're going from bits and stuff on a screen to really the things that affect your everyday life. So I'm very excited to talk about some of those topics. But look, let's let's get into it first. This would not be an AI investor podcast episode if we didn't talk about NVIDIA. So talk to me about some of the new NVIDIA systems and some of the economics that we're seeing on inference, because this really has been one of the major tide shifts in the AI landscape the last year, as a lot of um, let's say investment attention went from model training initially to inference and largely, you know, the just as models advanced so much, um, agentic workflows became so much more possible and plausible. Right now, while we were doing this episode, my ClaudeBot is off doing various tasks on our behalf. So, you know, the last six months we've seen this tsunami of attention shift towards agentic um uh workflows, which favors inference. And that's one of the reasons the memory trade has been so exciting. But but talk to me about the economics of inference, particularly in the context of NVIDIA's new systems.

SPEAKER_00 3:49

Yeah, so when we're looking at what's going on this week, it requires just a little bit of a kind of review of last week. I had said at the time it was probably the most important episode, just because we're gonna establish so much foundation for what's happening in the AI trade. Um, and and right away this week, we really see the entire market driven by kind of the second order effects of what we talked about last week. And and what we talked about, you started this with the NVIDIA systems. What we have coming forward is we have new NVIDIA systems coming out. Uh, we have uh maturation of, or I should say we have um at scale deployment of their more advanced systems from Blackwell like GB300. And what they're doing, Austin, is they're just outputting a lot more tokens than the old system. You know, GB300 produces 50 times more tokens than an H100. You combine that with spot prices, and it creates this incentive for a lot of companies to build and build very fast because they're getting great economics due to these new systems and due to the spot pricing on um tokens themselves. So, Austin, you know, the question we asked last year was how much capacity can come online before this dynamic reverses, right? And that's that's you know, there's gonna be an upside to the situation that most current estimates for what's being built in AI remain too low, especially for 2027. And there's a risk from this situation that the build-out is so fast and so fierce, and you have so much incentives for each individual player that we may reach an overbuild because it is so in everyone's interest to build as much as possible. And what we had this week was on Tuesday, another, you know, you just call it a momentum blow up. Every defensive sector in the market was up, software was up, AI infrastructure was down across the board. Most what you call momentum companies, especially the ones with strong gains across this year, were down 10% a single day. So we were basically at all-time highs again, especially for AI stocks. Most of them had come back. The portfolio was back near its peak, and just in a single day, you're gonna have, you know, I don't know what the portfolio will have been down on Tuesday, but it's probably six or seven percent. It's a relatively big number. And what really caused this is two different things. Number one, we got new data about the run rates for these leading labs, anthropic open AI. Um, the number is for anthropic because, you know, they're getting ready for their IPOs. So information's coming out, they're testing the water, seeing what people think. Anthropic is that they have a run rate of 65 billion. For open AI, it's 40 billion. Now, Austin, these growth rates are incredible. The speed they're growing at is incredible. But again, people expected a little bit more. I had seen a lot of whisper numbers about Anthropic being closer to 80 billion and scaling OpenAI higher. And when those come in below expectations, these two companies are on the hook for so much of the build-out that it concerns people that they're not making as much as some people expected. And the second thing was there's a large Wall Street Journal profile on how much off-balance sheet obligations these companies have: purchase commitments, uh, future leases, areas like that. And Austin, it just spooked the market, right? Because it's a big number. I think the number that they had was something like 3 trillion. So they're saying, hey, if the market spooked about CapEx growth, which how many sell-offs have we seen this year from CapEx numbers rising, right? We've we've probably seen a half dozen. And if you're saying actually there's something bigger than CapEx and you know, lurking in the background, you know, it's like some kind of tiger ready to pounce out. Um, or I guess a bear, I should say. You know, there you go. Come on, it's Mr. Wall Street. A little better. But, you know, Austin, I would say for the Wall Street Journal thing, this is they try and make it sound like some kind of nefarious thing. These are disclosed. There's a reason they had the number, right? These are disclosed, and it's only surprising if you haven't done the math where capital expenditures are going in the coming years, which we've discussed ad nauseum on this show. So a couple, a couple things that basically cause a big market reversion. Now, here's why I'm actually seeing this week a few things. Number one, token consumption is doubling about every 11 weeks. So we continue to see this growing. Second, these numbers from open AI and anthropic. Well, Austin, the thing is we are we continue to see an acceleration currently because what open AI is moving from, they're moving from a revenue base that was largely consumers to a revenue base that's largely enterprise. And as they do that, they are getting massive acceleration. So they they ended July at 40 billion. Well, their run rate already by the end of this month is 48 billion, right? So the numbers are moving at such an exponential pace. It's hard to get upset about one point in time because the growth rate remains at such a high level. And the the driver of this acceleration, like I said, is is agentic users creating massive token growth. I think tokens at OpenAI in this just started quarter are up 2x quarter to date. Now, on the macro side, I think there are some real concerns this week. Um, and that's largely interest rates. Austin, interest rates keep rising. They're at I think the 10% is at 4.7%. The 30-year just hit a 19-year high. And as we've talked about, as we're in this financing phase, interest rates continue to become more and more important, something that you have to monitor. The big question that we're looking at in the background is is this, you know, are interest rates right now a function of, I mean, they're a function of largely inflation and elevate inflation and the fact that uh we don't the large market consensus is that the US isn't going to do things like raise rates as high as it needs to to tamp down on inflation. The question is, and again, we're not a political show, but is is this kind of the result of some bad policy? You know, we had things like tariffs led to a lot of price rises across products, and now those tariffs are being refunded, and we have large structural deficits in the US. Um, and then we also have things like Iran, you know, something that led to higher prices across the board, especially in energy. So, you know, Austin, if if the main reason that we have persistent high inflation at this moment is some of these kind of what you could probably call more temporary mistakes, if you want, things like, you know, tariffs leading to more consumer price rises, things like the Iran war. Well, that's going to be more of a transient concern. If the longer-term concern is just that the market is starting to put its finger on the scale of these persistent US debts, that the US um debt burden just moved over 40 trillion, well, that's going to be a bigger challenge. And that's going to be more persistently higher rates. And that's going to be more of a headwind for the future phases of AI buildouts and a headwind for the growth stocks that are often in our portfolio because most AI stocks are grown tremendously. And when you have rising rates, it's going to punish those in outsized ways. So I think as far as my concerns about some of the news, like the Wall Street Journal report or uh Anthropics growth rates, I'm not as concerned for the reasons I said. I'm a little bit more concerned about where the macro situation continues to head. And, you know, that's going to be something we monitor and it will continue to kind of, as the situation progresses, lead us to do more diversifying in the portfolio, recommending picks like we did for ServiceNow that will probably uh be a little countercyclical to our portfolio and uh, you know, win our stocks in AI infrastructure drop due to interest rates. So Austin, I think I think that's the review. I don't know if you've got anything that you've kind of noticed from the mark in recent weeks.

SPEAKER_01 12:06

Uh I actually enjoyed the WSJ article. Like you, I was not surprised by the numbers since we've talked about them, but I actually think it, I think it's a good article that our listeners should go read just to understand the full scope of what's being said and reported on this landscape. One thing that actually did take me by surprise, though, even as much as we've covered it, is the cumulative off-balance sheet commitments of Alphabet in particular, which are approaching a trillion dollars. I mean, that's a remarkable number for one company. It doesn't mean they're bad investments, to be clear, but it does mean that at this, like these are hitting such a scale that um you have to start worrying more about things like geopolitical disruptions, inflation, GDP interest rates. Like, like this is this is like way beyond what these hyperscalers can withstand on their own balance sheet, right? Like with like like this is the the industry ambitions here are so big, and I believe in them that you have to start paying attention to those other larger, like global uh factors because this industry has to start standing on its own, right? It cannot be funded as um a growth driver from these hyperscalers. And the WSJ article puts the scope of the ambition here, in my opinion, into a really good visual. Now, they're not it they're they're talking about it from the reason to be concerned per se, but I thought that I thought they did a good job rounding up everything industry-wide to sort of show the true scale of it. Whether or not it's a reason to be concerned is for our you know investors and listeners to to determine. I was not scared off by it, but I I thought it actually was, you know, reasonably good reporting for what it's worth.

SPEAKER_00 13:41

Yeah, I think it was a good article, and it's more of just the market reaction, right? That you know, when you have not necessarily new news introduced and it leads to a panic, um, you know, but this is we've been skittish all year, and this trade will continue to be extremely skittish. So this is just the world we're living in.

SPEAKER_01 13:58

Right. And I get I guess to maybe dovetoe those things together, it's reasonable to expect that this trade gets affected by things like inflation and the war in Iran more as the scope of the build-out increases because it starts to hit the size that those things impacted more. Like we're we're I guess what I was trying to say is that this moves beyond like Meta had a good or a bad quarter, and how does that affect their capex ambitions for the next 12 months? And now it's starting to hit the scale of like you know, federal debt levels, interest rates, and and wars. And so the scale of the build out is is now at the level where those things affect it more so than any individual company news, um, frankly. But but let's now now, after we do we had that big macro view, of course, we're we're not macro investors, but it is important to always remember, right, when industries get this big, you know, those are the new trends that affect them. But after talking about the macro, let's go down to the micro. I want to get exciting. I want to talk about AI and healthcare. We have not talked a lot about AI and healthcare on this podcast. We've mostly been focusing on the build out and the growth. We've had a little bit of discussion on robotics, but but healthcare is really, I mean, this is this is one of the original breakthroughs in AI was uh Demis at uh Alphabet uh doing the protein fold demonstration and sort of showing how amazing it was that AI could, you know, you know, uh release all this protein fold data to scientists and something that would have taken, I don't know what the estimates were at the time, decades to do manually. Um, so in many ways, one of the early most exciting moments of AI was related to healthcare, but I feel like we actually haven't seen that much um in terms of real breakthroughs yet. But now we've got Moderna uh, you know, setting records on Wednesday. So what was the news? What happened?

SPEAKER_00 15:42

Yeah, I mean, the news in short was uh late-stage melanoma, which is a particularly nasty form of skin cancer, um, trial results from Merck and Moderna. Uh, the kind of in short view of it is there's an mRNA vaccine, which I trust we all remember mRNA vaccines from COVID. It succeeded in preventing cancer from coming back or spreading in a study of high-risk patients with melanoma. And um, you know, people might recall last year too, there is a mRNA vaccine linked to stopping the recurrence of pancreatic cancer. So, Austin, right, this isn't isolated, this is part of a larger picture uh around this kind of war on cancer and some of the progress that we're making on it. Um, but the big picture is for the first time, you've got AI-assisted personalized mRNA cancer treatments succeeding in a phase three trial. So this is this is pretty, pretty big news. And how it works is you've got Moderna and Merck, and they sequence a patient's tumors and they compare it to healthy DNA. AI then identifies which mutations are likely to trigger an immune response. And from there, you create these individual treatments. And and the bottom line, Austin, is this was a cool headline. There is real tangible progress building uh in this battle with cancer. Um, but I think we're just gonna start seeing a lot more around healthcare. Anthropic, we mentioned earlier, they are preparing for their IPO. They're gonna do their roadshow, they're gonna go kiss some babies, they're gonna do everything that they wanted. And their CEO, he's said that AI could make it possible to solve most human diseases within five to ten years. You know, Austin, which is stuff we've always heard, right? You know, you always hear these kind of ideas, but I think they're going to really push on this. Um, there's a Claude post on Tuesday about protein binding. Um, many drugs work by binding to a target. You had talked about proteins earlier, and then they change or block what it does. So a first step in drug development is creating molecules that can bind tightly to their target. And um, they put Claude on this specific problem and work with uh uh companies like Twist Bioscience, which we'll talk about more in a minute. And they're successful in autonomous design in 14 out of 15 uh targets. So, Austin, the key idea here is this concept of kind of these autonomous labs that you're going to have. Twist was a company I was familiar with them because I we had done um a product at Molly Fool that I was involved with, and we had made Twist really a featured company on. So it was a stock I actually thought about investing in at the beginning of the year and making it a first healthcare um recommendation. Now, I don't deserve credit for that because I didn't do it, and the stock is now up 330%. So the secret's out. But I I think this is going to be that question of how do we approach healthcare on the show, right? That we are not healthcare experts. You can't be, you can only do so much, but you can also know what from your areas of specialty are going to tie in there. And you can know enough to potentially make some bets in the space. And for Twist, they just had a really clean story around their AI tie-in. Um, what they're famous for is they built a better system for synthesizing DNA. Um and if there's a breakthrough in AI protein design, if if if if you're making these designs autonomous, well, then you're going to actually need to physically build and test, which is what leads into their business, because their business is largely reliant on consumables. So, you know, Austin, Moderna's results, it points to growth in many areas of the tools they sell and it creates tailwinds. So you're thinking about this. Okay, if this idea of these autonomous labs actually has some real traction behind it, what are the downstream companies that are benefiting? Another stock that has caught a lot of attention because it was a recent IPO, I think it was maybe last year, is Tempest AI. Um, its pitch is intelligent diagnostics, uh, using AI to make lab tests more accurate and personal. They've got a data licensing business, which I know always perks people's ears up. And they're both a clinical lab and a data provider. So you have clients not just licensing data, but using their products to upload data and build models. You know, that's a great business model if you can get working. You know, the question for them is I think data and applications, it's still somewhere around a quarter of their revenue. Um, it's it's still a diagnostic company, and data and applications is their growth market. Um, and and they had purchased uh personalis, which Moderna uses as a sequencing platform. So again, very a very clean line to this news we saw with Moderna. And then there's other companies like Schrdinger. This is a physics-based simulation um company for healthcare. They're launching their Bunsen co-scientists, and another area that we could look to is things like software. I know you've been interested in the past in Viva, for example. Then there's things like Doximity, and these are companies that sold off based on kind of the factor with software and SaaS. And now they both have potential callous behind them that there could be both a software rebound and there could be this outsized interest in the healthcare space. And you need to think about maybe how some of these advances in healthcare flow to these companies. But I think this does create some really interesting opportunities. We can look at the actual consumables, we can look at some of the data companies, we can look at some of the companies building simulation, we can look at software. There's there's a relatively rich collection of companies to look at in this space. And I think we'll look at them more, even though, as I noted, I'm not super comfortable looking through healthcare companies. We've we've been adjacent to it, we've we've worked with people who are healthcare specialists. I'm I'm probably going to need to lean more on some experts for the specific side, but trying to really think how AI flows into healthcare and the downstream consequences. Where I'll spend my time. But I think it's exciting, Austin. Um, what can you say? I the only the only thing you worry about is you have this area of exponential growth um in AI and the capability of models, and a lot of what's going on healthcare is verifiable, right? We we saw the takeoff in coding because it's a verifiable domain. There's aspects of that to healthcare, and it is just where maybe regulation and other matters in healthcare maybe means you don't get the gains you'd expect. Um, and that's where you need to just understand the industry as well. But really incredible news this week. I think sometimes we we let news like this pass by and we're we're so used to seeing headlines about big medical breakthroughs. But this is this is truly one that isn't just a clickbait headline, it is something that truly is going to change the world and and and save thousands upon thousands of lives every single year.

SPEAKER_01 22:59

Well, one of the most encouraging things about this for me is that this is a technology that basically would have been impossible to imagine or realize at any sort of you know reasonable economic level without AI. Uh medicine is so personalized, right? We all have unique bodies and unique biologies. And I I think uh Siddhartha Mukherjee, I don't know if I'm misattributing this, I apologize. But I think he said something to the effect of, you know, we refer to cancer as one disease, but really it's a it's a set of genetic mutations that are individual to the person, and therefore all oncology is inherently personalized. That is really, really hard to scale, right, without something like AI. Any sort of personalization, and personalization is the holy grail of medicine, I guess is my point, not just cancer, but cancer really illustrates the potential of it because so many, so many people are negatively affected by it, and oncology treatments have to be made personalized, and AI really is the only technology I can think of that would make that feasible at scale, which is important to make it affordable, right? Like personalized medicine is possible to a very, very elite, very, very wealthy um number of people, but not to everyday individuals. So I'm very, I'm very encouraged by this. Um again, we we are not healthcare investors. I don't think any of us or either of us have had a lot of success in the healthcare industry, but but I certainly can appreciate the potential and the impact on people's lives. So I'm I'm I'm hugely encouraged by that.

SPEAKER_00 24:20

Well, and Austin, you like to make the point too of how many things need to come together, that robotics isn't a breakthrough in robotics. It's a breakthrough of a lot of fundamental technologies that essentially gets packaged up. And I think you do have something a little bit similar right now happening in healthcare. So um, it's exciting time that's yeah, the result of a lot of innovations in the background that are finally having a headline um that, you know, is kind of the tip of the spear of everything behind it. So um yeah, it's just exciting.

SPEAKER_01 24:52

Well, you know, back to an area we can maybe speak with with a little bit more authority on. Only a little, only a little though. Uh Marvell and Google. So Marvell, you know, the chronic underperformer, perfectly positioned for the AI industry for a long, you know, for the entirety of this boom. But it will, you know, they didn't get their sea legs until later in the cycle. And and boy, what a rebound it's been. And now they're just you know racking up win after win. So they've got a big deal with Google. Talk to me about this.

SPEAKER_00 25:18

Yeah, and you mentioned earlier Google's $900 billion in commitments. Well, partially it's because people don't understand the true ambition of what they're going for with uh TPUs. I mean, Google is incredibly ambitious, I think, in what they are going to attempt to do in the coming years. And it's gonna make probably their $200 billion in spend this year look um a little quaint when we get down the road. And what they have essentially is a key partner for their TPU program in Broadcom. With Google, TPUs have become so important, they don't like paying Broadcom the margins that Broadcom gets. And also, this project is just too important to have it under one key player. So they've been trying to differentiate or uh diversify, I should say. So they've been increasing their usage of media tech, which is going to build some lower cost TPUs used mostly for inference. And this week they announced a major deal with Marvell that's going to sit around the TPU ecosystem. Now, Austin, I do think this is an example where we're getting to a point. A lot of headlines, like the reaction from the market is just weird. It's almost like everyone just sees a headline and does an action and no one thinks about it. So Broadcom sold off pretty significantly. I think the the two days around this news, Broadcom was off like 10%. And Marvell, after the news, it was it was up a good amount. But it Austin, the bottom line here is that Broadcom is going to be incredibly ambitious in the coming years themselves, as we've talked about. They right now are taking the crown. You know, we talked about what stock is the market the most negative on. If it used to be Marvell, it is now Broadcom. Broadcom is up there amongst most hated and potentially most misunderstood stocks in the market, which is incredible considering its size. But basically, why this is going to be good for um Marvell itself, it is for basically a chip that's going to be kind of similar to what NVIDIA is doing with their Grok by, I believe. It's gonna be a it's gonna be a custom chip meant for some fast inference, I think is what I'm reading from the headlines. But it's also going to be buying all of their networking stuff: storage controllers, memory interface controllers, near memory compute. And the total is huge, Austin. So Marvell's revenue this year, um, they had initially started kind of the rise again because they had forecast 11 billion in revenue this year versus the streets 10 billion. Well, again, that's gonna look quaint really soon because Google is getting warrants to purchase Marvell in this, and they only vest if they spend 120 billion on Marvell's products across the next six years. So the estimates I'm seeing is by 2028, they're already going to be spending maybe 20 to 25 billion on Marvell. That's more than twice the size of the current company, right? So it gives you some context about how big these buys are gonna be. But if I back up as well, Austin, Bank of America this week, they put out some estimates for future TPU market share. And they see Broadcom's TAM for TPUs at 250 to 350 billion, Media Tech at 100 to 150 billion, and Marvell at 50 to 100 billion, which this is just to say Broadcom, they're still getting the vast majority of this business. They're still getting more than 60% of this business. And Austin, the other thing is we're moving to a world right now where there's a few important, you know, for lack of a better word, bottlenecks. But one of them is going to be this interconnect layer. These companies that have a specialty in it, Marvell's great at this, so is Broadcom as well. So they're both going to see benefits from this. It and it's just a little odd to me that the market has such a negative look right now, um, uh, just specifically on a company like Broadcom, because the other side of it is why can't Google just move completely away from Broadcom? Well, Broadcom has incredible allocations towards the amount of chips it can buy at Taiwan semiconductor. Um, they have an incredible amount of HBM locked up. They have an incredible amount of the packaging capacity from Taiwan Semiconductor. And a big reason that they are probably losing some market share as well is because they're having to make choices around TPUs versus their other customers whose projects are also ramping. So I guess if I'm summarizing this up, Austin, three things. Number one, it's great for Marvell. We've already talked about the numbers from this. They are going to see a huge growth in their business just from Google alone, but then they have the acceleration with Microsoft and their custom chip project next year. And they have a similar deal, as we note earlier, with Amazon. Amazon also has warrants with this company. You should probably expect that their growth with Amazon is going to be at a very high level. It's probably a low discount from the market right now. Second, I just think that the market continues to misread Broadcom on this. A big reason that they're probably losing some share as well with Google is that they are having to make decisions around where to put their allocation for high bandwidth memory, for advanced packaging. And they have a lot of other customers they're scaling with. I talked about how they're essentially building their own hyperscaler, which is their XPV project. I still think the market does understand that. I think Broadcom just remains tremendously underestimated. And it kind of is a company that every single headline about it right now is getting read negatively. So it's not a stock that's gonna, it's too big. It's not gonna probably quadruple from these prices, but it is a stock that if you're looking at today, no one is going to want to touch it because it's down. This is exactly the time to probably be adding to that as an outsized position. And third, Austin, this really just does put a lens into the growth of AI networking. When I when I started this podcast, I talked about one of the key themes I loved in the years ahead was AI networking. And the reason is just because the size of these clusters is getting so large, the demands for networking are getting so intense, and every kind of bottleneck like memory that exists, the the the surest way to create a countermeasure to it is going to be to spec up networking itself. The AI networking industry, it's expected to grow from about 10 billion in 2023 to 245 billion by the end of the decade. And that's across optics, that's across most of these components that companies like Broadcom and Marvell are selling into. So the picture just continues looking very rosy for networking itself. Across this past week, a lot of good news across the supply chain. Um, if you're invested in the companies in this networking sleeve, uh, things like optics and demand for it continue to rise. So, Austin, I just think, you know, great news from Marvell. I'm a little befuddled how negative the market's gotten on Broadcom. And overall, this news shows just how much revenue is going to be flowing into the companies in this broader networking space.

SPEAKER_01 32:36

Um, I I think it was a friend of the pod, Gavin Baker, recently was saying, you know, he was doing a Silicon Valley tour looking for chinks in the armor, looking for negative news to understand this sentiment, and basically couldn't find any. I'm hearing it, I'm hearing you echo a similar thing, right? Like uh AI networking, what, yeah, more than what like 25 Xing across a decade as an opportunity. It's just one subset of the AI investment trade. But you have made this point many times on this podcast. You know, when the market is getting nervous, when it's puking over whatever the case may be, deep seek, interest rates, tariffs. Uh if you go look at the fundamental story behind the AI trade, it continues to strengthen at every turn. CapEx continues to increase, commitments continue to increase, ROI remains high. So I'm hearing you echo that similar thing.

SPEAKER_00 33:26

I think there's a pretty good chance for what's worth Broadcom, it's fiscal 2028, it could be a getting close to 10 times that. And it's going to go into that year, growing at astronomical rates. So you the biggest question for a lot of this networking stuff is yeah, we know it's a lot of growth, but uh, what does the multiple look like? And right, if if you can get a very reasonable multiple for these companies, they like you said, there's there's there's not a lot of those chinks in the armor that you talked about. Um, there's risk in the overall trade and and growth of AI, but if we see the levels we do believe are relatively locked in into 2028, this should be a company with incredible growth trading at a very low multiple of it today.

SPEAKER_01 34:10

I I I would love to I want to stay on this point actually, because I would love to get your thoughts on this investment warrant option movement we're seeing. Nvidia is by far the company that's doing this the most, where you know they're making investments in the in the entire industry and options to buy and warrants and companies XYZ, but some they're they're the trendsetter. But this now appears to be the norm, right? We're seeing you know, Google getting warrants for anthropic. You said, I'm sorry, for Marvell, same thing with Amazon. How do you consider that in your investment thesis? Or do you not? Because the scale of what we're talking about here is so big that, like, were these warrants to execute, the bus, the core business value has multiplied so much that the warrants that they're getting in the positions that they're getting in their portfolio it doesn't matter? Or are these things like like like call options that nobody's baking into their models now because it's so hard and so multifaceted? Like, how much would NVIDIA stand to gain if all of their investments in the companies that they've made, you know, pay off? Is that a factor in when you're looking at these companies, or is it just a this icing thing out there that is extremely difficult to model? You don't know what the probability is, and if it happens, it's even better.

SPEAKER_00 35:22

Yeah, I would say the upside is probably higher for the companies providing the warrants because yes, you will have companies like NVIDIA or Google. I mean, Google, it's it's funny. I was looking US corporate earnings this quarter or yeah, I think it's this quarter up like 50%. But so much of that's because Google reports a $100 billion gain on SpaceX. So, you know, something like a material uh an investment like that may become pre-material.

SPEAKER_01 35:51

That's kind of what I mean, right? You SpaceX anthropic. I mean, Google's Google has this sleeper tech portfolio, which we've talked about a little bit, but in the context of how big Alphabet already is, as as b as amazing as that portfolio is and as amazing as those returns are, the comp the core company value has grown so much that maybe they don't matter, right? They seemed like they were this this huge call option seven, eight years ago. Um, but today the core company is so big that they're just a nice dab. Do you think about that when you make investments in these companies or no?

SPEAKER_00 36:19

Well, less on the larger companies, but on the secondary companies, it is interesting because it's allowed them to catch up often when they maybe would not have been able to otherwise. You think of the amount of deals that AMD has done that had some kind of ownership stake? Are they winning those deals?

SPEAKER_01 36:34

Um AMD is a really good one, yeah.

SPEAKER_00 36:36

And likewise, Marvell are, you know, they are behind Broadcom. They are catching up in a lot of elements of this custom market. They compete with them across a lot of networking markets, and it's probably allowing them to win some deals that otherwise may have gone to Broadcom. So I think who has benefited the most from it? I think it's allowed a lot of the second and third place companies in markets to win deals they otherwise might have not have. And yeah, that's the most impactful framework from it so far.

SPEAKER_01 37:04

Interesting. Um, let's get back to a topic our listeners have heard a lot about, and that's memory. We saw the memory trade rebound recently. Sandisk had an investor event that I'd love to get your take on. And some of the things that we have talked about are the margins in this industry. So the memory margins are north of 80%, you know, 80% here, 85 there, whatever the case may be, um, which long term is just not sustainable. And that is one of the criticisms of the memory trade that these margins will at some point, who knows when, normalize back to sort of normal economic levels. But one of the arguments that we have made is like that actually is healthier, right? Like you'd almost rather see these companies at a 70% or even a 60% gross margin because you can bring all this additional demand online. Um, SK Heinex also threw a little bit of kerosene on the fire in a good way with their $29 billion share buyback. So I saw I know Sandisk and Mike Ron definitely got a tailwind there on SK Heinex's vote of confidence in their own shares. So talk to me about the memory rebound broadly, but in particular, what did you see out of the Sandisk investor event and what does it imply about the memory trade right now?

SPEAKER_00 38:09

Yeah, and I I think the biggest thing is maybe give ourselves a little pat on the back for last week. We had discussed this situation right before I think Sandisk held their investor event relatively shortly after we had recorded. But what the news that they gave was that they gave a forecast 2028 to 2030, 80% adjusted gross margins, 75% operating margins, and that these long-term agreements now cover two-thirds of their output. So Austin, the point that we had made last week was so many people are waiting for this first drop in margins to get out of the ship because historically a drop in margins always meant the peak of the memory cycle. And we had said with the dynamics that are happening right now, people need to consider that they had previously peaked at 60%. We're getting into the mid 80s, up to 90%, I think micron, sorry, was looking to hit something like 75 or 80%, if you could maintain that, is going to be a lot healthier to the industry, especially if they're doing that through LTAs and and um you know mechanisms that smooth out growth. And that's exactly what we saw this week. So we've talked about how much it matters to not let how you're investing be impacted by the price of things, right? We've talked about this narrative follows price so much on this show. And we saw everyone losing their minds about memory in recent weeks, mostly because the share price was falling. And it was falling 40% off a gain of a thousand percent in a year, right? Um, the reality is, as we said, the market is much healthier than it was getting credit for. And we do believe that there's a situation where margins can be lower than their peak, but still elevated for longer. And that's exactly what Sand disk guided to. So, Austin, this is just one data point and it is a forecast. This isn't anything that's happened, but I think it really confirms the discussion that we had had last week about memory. And we do see a lot of the memory names now having bounced a fairly large amount off their bottoms, despite that sell-off that we saw on Tuesday that really sold out the um kind of momentum sleeve. So overall, it's it just if you want to go back and listen to it, we've we filmed that memory segment last week. And and I think what we said holds up tremendously well to what Sandis forecast.

SPEAKER_01 40:36

Uh well, good on us, good on us. American flag boys here calling it. Let's go, let's go. We're calling it. Uh, we're patriots and investors here. Um, Erica, I I want to transition to something uh we haven't done in a while. I'd love to get some listener questions out of the way. We've we've always said this is a community. We love hearing from our community. I noticed you've been doing a really good job staying active and engaging with the community on YouTube. So just a shout out to any of our listeners there. Um, please leave us comments, feedback, questions on YouTube, Spotify, wherever you get this. We do try and engage with as many as we can. And let's bubble some up because we know that a lot of people ask questions that other listeners have. So, Eric, without further ado, let's jump into it. Um, we've got one here from Mean Dean, who said, Great pod, as always, fellas. Thank you. Uh, I've been listening since April and never miss an episode. I want to show some love in the comments. We appreciate that. And specifically, Mean Dean said, I'd love to get your take on iTron, ticker symbol ITRI, as a potential asymmetric behind-the-meter play. They sell utility meters, communication networks, software analytics, and distributed energy management. All right, that's an interesting one. That's a name that has not come across my radar, but I am a big believer in the behind the meter play. We also had a segment last episode talking about power, um, power solutions, um, which has been a rough um pick in the portfolio. But as you pointed out and is a thesis, I believe as well, the trend matters more than the individual company. And that was a company that maybe was not the foremost operator, but they were correctly positioned. You also made a recommendation um tied to the $10 billion, I'm sorry, the 10 gigawatt build out of SpaceX, ticker symbol SEI. So if people want, you know, so this is very, this is a really timely question, ties well with our last episode where we commented on behind the meter power opportunities. Um, but what um what can you say to Mean Dean about iTron, ticker symbol ITRI, and how does it play in with your uh recommendation last week?

SPEAKER_00 42:27

Yeah, and let's throw one more in. We've got one from uh Mo, Solomon, Chris, and he asked about H Y L N, and he said they make nice little turbines that can use any fuel source. So that's that's a similar kind of behind the meter um concept.

SPEAKER_01 42:41

And follow and follow Solaris, right? Which was your last recommendation here.

SPEAKER_00 42:45

So very timely. Let's go. Itron, this is a relatively conservative company. Um I would need to spend a little bit more time. My first glance at them, I saw EPS is falling from 713 last year, 642 this year. Um, they did see rapid growth in 2022 to 2025. Um, but again, I'll I would need to understand why the pause this year. Uh looking at its history, it's it's a little more cyclical than I expected. Uh sometimes you probably want a little bit more upside if if you have that cyclicality and the potential downside. Also, revenue today is essentially what was in 2010. Um, so you know that that is a profile that you have 15 years of stagnation and Wall Street expects it to be continue baking in consistent growth started next year. So I I would have some questions on that one. I I I like looking through that theme, but this is a company that um, you know, is having some stagnation in 2026 and and does have that track record, which are kind of the things I would poke at the most. And the other stock, which is Hylion, it looks like it's a broken spec. I'm I'm actually not familiar with this one. They have 9 million in trailing revenue, I think, against a market cap of around 600 billion. Expected to grow 38 million next year. Um, I would need to understand the specific technology um behind their systems more. But what I would say too, I worry about some companies that are trying to ramp new solutions today from zero. Because we're going to see such an aggressive ramp, you want those relationships built. You want that experience with your products being used at scale. You know, if you were buying a chip company during the smartphone boom, you wanted a company that was starting to get into things like RF and in 2008, because it's going to be tremendously hard to break in in 2017 once the market's relatively maturing. And I fear that that might be part of the situation with a company like Hylion that seems um very speculative, which we're we're not afraid to do some speculative positions. You mentioned uh power solutions, that very speculative. It is just always, if you're doing that, keep it to a very small position. You know, we had talked about that one being 1% of the portfolio when it was $500,000. I would imagine if you are someone buying a stock like Hylian, you you don't want to make that any larger in your portfolio than a very small position.

SPEAKER_01 45:14

Uh okay, great feedback there. Thank you so much. Let's move on to the next one. This is from Will Grimes, too. Uh Will Grimes says, I'd be curious to hear your analysis of ASM International, ticker symbol ASMI. I don't think you've covered it before. It could uh complement the portfolio's LAM research and TSM C exposure by adding focused atomic layer disposition exposure. Atomic layer deposition. Oh my god, advanced packaging. I'm gonna set a timer. Eric, you do your bit. I'm gonna go get a coffee. I'll come back.

SPEAKER_00 45:45

Yeah, yeah. This is this is where Austin takes his nap in the middle of the episode. But um ALD, it it is kind of like a lot of things at the bleeding edge of building chips, it's it's kind of sci-fi. It's a thin, thin film technique uh that builds materials. Uh, you know, you have one atom at a time in the layer. So it's it's seeing a lot of growth because some of the newer uh growth areas in semiconductors like 3D transition, the growth of patterning, these are long-term demand drivers for it. Austin, the big picture would be semiconductor equipment. It's been a fabulous, it's been a fabulous industry. I was looking at a chart the other day of the top 15 performing stocks across the past 15 years, and semi-equipment was three of the top 15 stocks. Uh, that's LAM research, KLA, applied materials. We've talked about how this industry, it tends to kind of it's built this big five of companies each with their own unique niche. So when you're looking at semiconductor equipment, the technologies are often technical enough. It's it's hard to necessarily understand advantages. What you want to look for is some end markets where some of these companies have traction in the spaces between where these kind of big five companies are are extremely dominant. And ASM fits that. They they've they've grown to right below the tier. Um, they're they're that tier below the Lamb Research, the applied materials, the Tokyo Electron. They've got five billion in revenue that's uh 10 times larger than Camtech, which is in the portfolio, four times larger than uh Anto. Um, so anytime you find a company that's found a niche, not dominated by that group, and they are defending some pretty good market share, you're interested in them. Now, the second side of that's gonna be it's at about 27 times Ford earnings. That's elevated relative to history. Um and its expectation is gonna be to double earnings across the next four years, which is fantastic, but that's pretty similar to Lamb Research, which is a more diversified company. Um, so Austin, I I think the bottom line is this is a stock I've looked at in the past. It's it's an extremely solid company. I think if semiconductor equipment does outperform in the coming years, because the growth rates in the industry are so good, which we've talked about, it's gonna be successful. My my biggest concern across not just it, but the broader industry remains um valuation. So this is this is definitely one stock, you know, I I keep on the radar. I would I would consider it, but I've probably targeted a little bit more of smaller niches with semiconductor equipment, going for some of the stocks like BE semi uh with hybrid bonding and and Anto and um also Camtech with their kind of attach to things like um high bandwidth memory and uh and advanced packaging. So I I think we may add some more semiconductor equipment stocks because it's a fabulous industry and uh this will definitely be one we'll keep on the radar.

SPEAKER_01 48:50

Uh I love that. And then we've got two related questions here. One from Warden26 who said, Who's buying all this compute? I don't get it. Right to the point. I love that. And then we've got Forest Fire One who says, My question is, how much visibility do we have into who is actually consuming all the compute? So basically an identical question. How concentrated it is with the Frontier Labs, AI startups dependent on VCs versus broad enterprise adoption, which feels potentially more sustainable. This is a very good question. And this also, this also goes to our spot price question or spot price commentary that we had in our last episode, where we're saying, look, you know, the one indication of where how valuable this compute is is going to be that spot price rate. And if that collapses, that's or starts to show weakness, that's something a lot of people are going to look to as a sign of weakness in the market. But spot prices have actually been increasing in strength. I'd love to get your thoughts on this, but I would also remind our listeners there's a lot more consumption out there that we might not see that's coming not just from the Frontier Labs, although they're certainly the dominant consumers today, but there's also all the open source models. And as those models become, you know, take shape, we're gonna, whether it's Kimmy or GLM or Deep Seek or Quinn, whatever, there's a lot of consumption coming from open source that we have far, far less visibility into. And they're also demand drivers on a lot of this compute. Um, so Eric, I'd love to hear your response to this question, but it also occurs to me that it might not be obvious that there's these, these whole sort of we we talk a lot about um Anthropic and GPT as and Gemini as the front runners here, but there's this whole world of additional models consuming that are just not making the headlines in the same way. And I do agree with one of the points that I think it was Forest Fire made here, which is we need to see broad enterprise adoption. I would argue in many ways we actually are. It's just starting with tech companies, you know, that that which makes sense. It's that first derivative away from the AI companies themselves. So it's tech companies that were not primarily AI companies, whether it's Snowflake or you had made an investment in service now, um, largely under this, you know, under a part of this thesis as well. So I would argue we are starting to see broad enterprise adoption. And I would also argue there's a large amount of consumption going on that we have we don't have visibility into that's also pushing this consumption higher. But what do you say to our two listeners here?

SPEAKER_00 51:02

Yeah, just to echo, I think it's a fantastic question. Part of the challenge is a lot of usage is is going to be into companies like OpenA Anthropic that um don't necessarily uh want to always share information like that. They they probably, as part of their roadshows, will have more incentive to open up the books and and give some details around who their end customers are. Uh the other thing is too, there is an incredible incentive for companies to say that their revenue is diversified. I think Microsoft had mentioned something like 90% of their cloud revenue was non-frontier companies. But I think that's kind of a loaded theory that presents it as as good as possible because everything, every mark incentive is for them to say they have less reliance on open AI. Um, now the the primary driver, and I responded to Forest Fire directly, but it is agenc usage, as we've talked about. It was 0% a year ago to 64% today. And this is why, since the beginning of the show, we've been so excited about agentic computing, is partially because if you're in a chat interface, which was the previous breakthrough in 2022, um that is an interface that is rate limited. Something that's autonomous and could be part of a process, has just exponentially higher usage. And if you think about most companies, how you're probably still going to see the vast majority of AI usage is someone going into a Chat GPT or going into a co-pilot and asking questions or giving, you know, um, you know, a specific spreadsheet and asking it to perform some commands. That's going to be a very low percent of overall AI usage as we continue on this path. You know, in years, agentic compute usage is probably going to be something closer to 99% of total usage. And we do see this in there are companies that do release data. Ramp is a company that releases a lot of data around AI usage. And they released something this week saying that the top 1% of businesses, they spend $7,400 per employee per month. Right. So who are these 1% businesses? It's it's probably businesses that are very developer heavy. Um, it's it's smaller startups that have built their operations. Austin, I think we're spending right now about 5,000 per employee per month.

SPEAKER_01 53:26

I I it it's it's yeah, I was I was actually just gonna share some numbers on our own consumption. We are not that that 1% edge case. We're actually not a majority developer-oriented company, although we do have some world-class, very talented developers. But just by way of context, our own consumption has gone up 70x in the last year. That you know, that that's our monthly spend. So that's our run rate. But our run rate every month has increased 70x. And I don't think, I mean, you and I have talked about this offline. Um, the returns that are coming from it are pretty obvious, actually. And it's not just a pure expense growth, right? We are seeing revenue growth associated with it. But more importantly, we're also able to remove expenses that we previously thought would have been immovable. So there, there, we're we're we're we're sort of in that phase of our own um, you know, work with AI now where people are saying, you know, well, hey, the spend can't go on forever. Well, one of the things to remember, I mean, we are an example here, we're probably, you know, AI forward, AI first, but we're not frontier, frontier development-oriented company. Our consumptions increase 70x. That would seem like it can't increase forever, but it's off of a small base. And it can increase a lot longer than people realize when you realize that it's replacing some previously immovable, very, very expensive. So we've reduced licenses like major, you know, five and six-figure licenses that we had with companies with AI tools that we've built internally. There's a lot of other expenses that we thought were just a part of running a business that we're now able to reduce with this consumption. So that spending is not pure cost, it's it's driving revenue and it allows us to cut expenses elsewhere. So that's where a lot of that money is going to come from. As other businesses start to realize the same thing, it um you can you can displace costs pretty easily.

SPEAKER_00 55:11

Yeah, and I I think we've it did very much as an operating for our business. It kind of flows through everything. So, and you know, the other thing too, it's it's an operating system for our business, but we still have people in the right places to add that kind of final judgment layer in other areas. So it's it's very complimentary, I would say. The the thing is, Austin, um, also someone this week did mention the comments. Can can you not say, Austin, so much so it you can slap me if you hear me, if you hear me saying that too much. But the median firm is only paying, I think, $11.95 per month. So we just talked about a top firm at $7,500 or $7,400 per month. We're looking at more $5,000. The median is still very low, which shows how top heavy this is. And I think it is a lot of things, a lot of agenc use comes from the growth of its use in software development. Um, so that's that's still companies that have large developer teams are going to be on the forefront of spend. And if you're someone who works in something like an HR, um, well, you're only going to generally be experiencing AI through those chat interfaces. So I think that's that's the best way to put it right now. And I do think there is probably a universe where in terms of percent spend, yes, venture companies are going to be mini multiples of enterprise companies. And um we're going to need to see a larger distribution of AI across these. If we are going to see, okay, you know, you can you can justify spending at the levels it was last year with this smaller cohort. If you want to justify your $2 trillion of expensive use case is going to need to grow dramatically. Um, and that will be in operations of companies, and that's going to be in also things like new fields like healthcare that we talked about earlier. Jamestown had asked, I said I would respond to it, but he had asked me about AXTI. This is a stock that there's a reason they have so much interest. They were $3 last summer. They peaked at $140 in May. They're down to $37 in late July. They're back up to $100 per share on Monday. And as we film, they're down to 73. So I bring all that up, Austin, just to say this is an incredibly volatile stock that's that's seen a lot of gains. It's one of the main suppliers of NP or indium phosphide. And that's one of several materials in the optics space, but it's essential for lasers. Um, and it's a relatively small industry. I think there's three primary suppliers, and they're all growing at just an incredible rate in terms of what kind of capacity they're going to be bringing online. Now, I will say when it was at $150 per share or that peak, it's estimates for three um 2020 are $3.29 in earnings. That's kind of 50 times that zone we've talked about a lot. What are you paying for 2028 earnings? Because that's where our visibility goes out to. That's very expensive. On that lower end, closer to $50, you're more $15. That's that's going to be a lot more reasonable. But the big thing I wanted to tee up here, I have been putting together a longer explainer on the optics industry to go on our YouTube channel. And we think we're going to release that in the next few weeks. Um, I continue to work on it, I'm trying to finish up. And one of the reasons I wanted to do that is because I think this space is um so important for investors. And also every single time I introduce a lot of these optics stocks, I feel like I need to spend like five minutes teeing up what exactly they are. So if we had that definitive explainer video walking through not only the background of the industry, um how it works, going through all kind of the terminology of it and showing people the stocks and the key themes in it, then we could tell people, hey, just go here, watch this. If you do that, we think we've built the best explainer on this industry um in the world. And that's what our goal is, to build the best explainer for listeners of this show. But I don't think we're gonna probably put it on the podcast. I think it'll probably be YouTube only. There'll probably be a lot of visuals. So the only way to get that's going to be subscribing to our YouTube channel. So I would just say to everyone out there, if you are not subscribed to our YouTube channel, we'll put that at the top of the show notes and we'll have that Optics Explainer coming pretty soon. So if you set alerts for the channel, you're going to get a notification. The moment it goes out, you'll also get notifications for new show episodes and any other content we put there. So um, so Jamestown, I'm gonna dive more into the supply um side in that video. If you're really interested in more on AXT, check that out. But for everyone else, I think this mark is just too important for everyone to not go back and give everyone that really great explainer of the companies behind it.

SPEAKER_01 1:00:03

Uh Eric, thank you for that breakdown. Anybody who saw your ETF explainer on our YouTube channel knows they're going to be in for a treat with this one. Um, that was a fantastic video. You put a lot of effort into it, really great visuals. So I encourage anybody who's listening here to go subscribe to our YouTube channel, check out that prior explainer as a uh uh a preamble to what they can expect on that next one. Eric, that's uh that's a full episode. We cover a lot of terrain, a lot of great companies and tickers, great industry analysis, um, some uplifting news in the medical space. Looking forward to all of the the ways that medicine is going to be made better and personalized for people. And let's call that a wrap. Eric, that's a full episode. Thank you for your time. Thank you for your expertise as always. And to our listeners, thank you for tuning in. Please leave us any comments or questions. If you have time, we really appreciate it. We try and engage with each and every one. And uh until next week, have a good one. The AI Investor Podcast is for educational purposes only and should not be considered investment advice.