Nick Frasse: Well, thank you everyone for joining us again for our quarterly semiconductor call. My name is Nick Frasse. I am the product manager for our thematic ETF suite, including SMH and SMHX, our two semiconductor ETFs.

I have with me as always Angus Shillington, is the deputy PM of our emerging market equity mutual fund, who also covers a smattering of different semiconductor. And he's my partner in crime going through these, these different semiconductor names and trying to navigate the semiconductor landscape for our clients. So this will be posted tomorrow. So yesterday we had our Nvidia earnings call or Nvidia had their earnings call rather Angus was blowing me up.

last night and this morning to go over all the different numbers and the excitement that he had. So first and foremost, let's just turn it to Angus, know, why the incredible new conviction for the space and Nvidia.

Angus Shillington: Thanks, Nick, and welcome, everyone. So you and I briefly talked a couple of weeks ago that I had seen earnings estimates tighten up and had a feeling that we were going to say what we're going to say today, which, as you say, is pretty exciting. So if I think about our mind map, where we've come from, we put Nvidia on the bench maybe a year ago as it worked through some issues, preferring Broadcom and really our overall feeling was the group in SMH and SMHX were the right way to do this. You want to be diversified. Today, I'm officially going to pull Nvidia off the bench if we're going to put this in basketball terms, put it back on the court. Broadcom has played fantastically well. It stays on the court and the sum of those two kind of positions means that the rest everything under the stack is set up in a in a really I Mean the stars are aligned. So I'm trying to stay away from explicit recommendations You know broad comes out performed in video by about 50% For over the past year, I think they now line up with similar kind of organization So that's point number one in video goes back on the court Broadcom stays on the court the remainder of the semi cap stack stays on the court. I will move SMHX, the Fabulous Semiconductor, to point guard, so shooting range. And I'm pulling off, in this analogy, just a bunch of consumer internet names, tired legs, getting overrun by OpenAI. And I don't see this as explicit recommendations. I see this as a shift in conviction around the space where Nvidia moves back up to playing position. And the pain to, where there is always pain within a sector is moving downstream. I think that sort of sums up where I feel right now. I see it as a very big move difficult to overemphasize.

Nick Frasse: Cool. Yeah, I mean, I probably would have gone with a hockey reference being a hockey player, but that's fine. We'll do basketball.

No, for sure. I think as we look at these, we've made the argument for diversification over the last few calls. We're not making individual calls for SMH and SMHX. Obviously, these are passively managed. But as we look at the space and just determine where the opportunities lie, Nvidia definitely has had renewed vigor and definitely is intriguing again as

A quick recap of where we've been over the last couple of calls and what we were thinking. Maybe what moved a little bit faster, monetization of AI, that happened super quick. ChatGPT, driving revenue, is there enterprise? I think one of the big thing was the deep seek moment. Obviously that's still in the lexicon for semiconductors today. Looking at maybe, know, optimization was the path forward and not compute laws and scaling. that

seemed to be not the case, right? They're moving together. Grok was a great example of that. If you keep throwing money at compute, these models will continue to get better. And that's what we're talking about with scaling laws, right? So I think that was a good, maybe a little bit we were thrown off by or just missed the timing of that a little bit.

Divergence of customer strategy. That happened very quick, right? Hyperscalers developing their own TPUs quickly and ASICs chips, whereas Nvidia was more of the generalized option. Geopolitical incentive structures, now China, TSMC, they're all kind of central, they're not peripheral, that's right in the frame of what we're looking at. That happened pretty quickly. And then OpenAI adoption versus the broader peer group of LLMs just took off.

So those are kind of the things that we went over. Looking at our current framework and what we think of the semi-space moving forward, you know, let's look at our thesis update. The foundational mental model, about fabless diversification, how do we see that changing? Has it changed at all, do you think?

Angus Shillington: In a word, it has changed. It's changed a lot quicker than I think we'd anticipated. And I feel like all of our past calls are out there, so we'll attach them to the show notes. And we're accountable to that track record. I think the simple answer is everything just moved a lot faster. We were concerned that this sort of ASICs move from the hyperscalers would sort of disrupt what Nvidia were doing and that was sort of part of our problem. We also felt like the China thing sooner or later was going to roll off in terms of US businesses having access to that massive market and that's by and large happened. But I think, yeah, I mean, I think there's a lot to unpack in what you said, so probably better just to sort of go through it maybe point by point.

Nick Frasse: Okay, so for Nvidia. I thought you'd maybe expand on this. Is Nvidia AI? Because that seems to be the narrative that's put in place.

Angus Shillington: Look, I mean, think that's a really good point. think believing that does not help investors. I think it was originally the key player. And I think if you think about it like a rail network, the original track that was going down was built by Nvidia. Since then, there is different options for track.

And some people would prefer to run their own trains on their own track. And that's effectively what's happened. They've been building their own ships through Broadcom. And that was really the sort of dispersion between the Broadcom and VideoShare price over the last year. The other really important point is like SMHX, and you've mentioned it a couple of times, is the fabless space. It's sort of off the radar names that design, invent, innovate.

I think that's sort of the way to think about it. So if you sort of step back and say, okay, we're feeling good about the top of the stack, right? The kind of monopoly guys. And then as you go down the stack, if they're doing well, then the semi-cap equipment guys will do well. The software guys will do well. And as the new applications roll out, and I think one of the best ways to think about this is power management.

So that's sort of the way to think about it. There are stages, stage one was the original infrastructure, stage two was building models, stage three was implementation. And now you're finding the applications are just atomizing from that level.

Nick Frasse: Sure, it's a whole ecosystem, right? And the rails are down. The entire ecosystem is moving in that direction. Okay, so we go back to ASICs chips and the spending from these hyperscalers, I guess. Is AI spending maintaining? Is it shifting? What is the thought there from like a customer perspective for these semi names?

Angus Shillington: It's going up and it's not stopping, right? So the way I would think about it, and just to really, an arresting point. So if you think that we're, I'm using projections, I'm coming in higher, 3 to $400 billion, three to $400 billion of CapEx that we can see, right? So if you think that sort of the Microsoft and the Met has just reported, they increased their CapEx projections, the sum of the number that we can see is in the three to $400 billion for this year alone. Now, if you think about what the increase of US GDP will be this year at a sort of 2% to 3% growth, that 300 to 400 billion represents about half of GDP growth this year in the United States. These are vast numbers. And as I say, it is not slowing down. In fact, it's increasing. And there are a bunch of reasons for that. And maybe we can step through that as we go along.

Nick Frasse: So I mean, I think the natural question would be, it a bubble?

Angus Shillington: I mean, let's unpack what Jensen said in the Nvidia call. And the reason why I sort of feel like in the mind map I have is it's better position. So one, we've got that incredible news about sovereign AI in the Middle East, right? So what that ends up doing is taking out demand a year out. So if you take out demand a year out, anyone who is thinking they would smooth it who needs it now, is going to pull in their demand. So that's point number one. We're introducing more and more new demand at the bottom, which is putting pressure on the guys who have existing capex plans.

Two, by its very nature, it's a compounding phenomena. So if you already have it in the ground, you want to continue to upgrade and you want to continue to upgrade and scale. You have built use cases. Now, if you look at Microsoft, they told you that the token usage year on year was 5x, like 5x. So think that through. Now that's shorthand for OpenAI, but broadly speaking, token usage is doubling every six to eight months, right? That's a phenomenal number. Now, compute demand is not a one-to-one correlation, but in reality, it sort of ends up being targeted that way because the more complex applications begin to be rolled out because of discovery and just how effective these models are at anticipating and predicting, then that token usage will, and the sort of need by more complex models will mean that the demand is there.

These guys have proven they are anti-fragile, right? So when things get tough, they gain market share. However, we haven't tested them. When things get tough, what their response will be, I would anticipate there is no question, right? They will have to spend to defend. So as the open AI ramps, and we've seen them just hire a monetization team to build an ad layer, really super competent people. They're going to start to run into the competition and the competition feels this, they're going to have to up their game in terms of innovation. That will mean capex. The whole thing. No, it is not a bubble. Every time there's a shakeout in the market and multiples get compressed, you know, the Cisco chart reappears. I do not believe that that is in any way helps investors.

Nick Frasse: Yep. You took the words out of my mouth. I was saying as a recap, this isn't Cisco, right? I think the, just the technology that is AI is lending itself to what seems like infinite capex for compute, right? The more compute the better because either A, your models get better or B, like you said, the competition has to keep up with, you know, to stay ahead. So that's one. And then if you look at, and maybe some people aren't as familiar with token usage, but that's essentially the input words that you put into these LLMs, right? At a very, that's a very dumbed down version of it, but that's essentially what it is.

The adoption rate of AI is still relatively low in regards to just entire user base of maybe internet users. think about how much token usage there is now, what compute is needed just for what we have currently, and then start layering on applications, start layering on to increase the adoption. The token usage is probably going to even though it seems like it's hockey sticked already, it's going to continue to hockey stick and J curve. just something.

Angus Shillington: Yeah, I mean, think just on that, I mean, this thing is growing in multiple dimensions. And I think, first of all, you need to unshackle yourself from this belief that AI is Nvidia, because that's what sort of gets you into a hole when you see the Cisco chart. AI is very, very broad, right? That's point number one. So the point number two is the cumulative nature of adoption, right? So. More adoption equals more tokens, but more adoptions equals more data. More data equals better models. Better models mean more applications. So it's spreading in many directions.

Nick Frasse: Well, let's pivot a little bit. Geopolitics, that's a big question for everybody, especially in the semiconductor space tariffs and all of those different things. They just announced Sovereign AI, Saudi Arabia investment, et cetera. What does that mean for the semi-space as a whole? I know Nvidia was kind of at the front of that conversation, but just maybe if you can hum a few bars on the Sovereign AI.

Angus Shillington: That's always sort of been on the back burner. I think the lead into that was the sort of end of Biden diffusion laws. To me, that was super dangerous, right? Because if you choose who is allowed to build AI and who is not about to build AI, and if you think about how the first internet iteration rolled out, it was basically the world, and then there was sort of walled off China. I had felt that the sort of danger was we were going to build a walled off US, and then the rest of the world was probably going to cluster around China. Now we've moved away from that.

And this idea of building out with friends in the Middle East has created, I think, the pivot I've felt in Nvidia. So you've got that huge piece. Now at the same time, Nvidia's been shut out of China. There's a lot of talk about maybe they'll get back in, maybe they'll not. I think that's of CNBC type drama. And the reason I think that, which is not sort of kind of agreed with, is I don't think Mr. Xi wants Chinese AI built on any US architecture or software. So at the end of the day, DeepSeek was an output of constraining China. are building, developing very quickly. I think it's unlikely that there'll be a material reentry in any form. Which I think actually at the end of the day removes or brings down the risk that was embedded in Nvidia. But I think geopolitics, tariffs not sure, I don't think there's material issue, but definitely sovereign AI is huge and taking China out as a sort of question mark probably reduces risk overall.

Nick Frasse: Slight tangent. And we didn't go over this in the call prep, but it came to me as we were talking. With the rate at which AI is growing and getting better. They've obviously been able to optimize with things like DeepSeek. Does that put them on the back foot if they don't have leading edge technology like we have? Because we have ASML, we have TSMC. Even if they say they want to develop their own chips, do they have that? Can they optimize for that?

Angus Shillington: I mean, it's difficult to say what a head means, am easy answer is I don't think it constrains them.

Nick Frasse: So let's move on architecture talking about, we talked about ASICs chips a little bit. If you just want to talk a little bit about Nvidia, their parallel processing of GPUs was huge, right? It was, it was perfect for what AI was doing. Now we've moved on to kind of this next phase.

How do you kind of parse that out from being a general player or a general option to the specialized chips?

Angus Shillington: Yeah, mean, there's a couple of issues here.

Nvidia had originally said, we're going to do general purpose GPUs. You're going to take them and that's it. They're sort of pulled back from that and they are building inference overlays that are slightly cheaper. I don't think that changes the trend because if you're Amazon and your objective, like it has been in any part of your business is to get costs down as low as possible. Think delivery infrastructure, that kind of stuff. They've got training them to output costs to... cost per user has down about 40% in the last year and with 50% less power. most of the hyperscale customers want their own solutions because they're operating their own business models and they want to be able to customize to scale and they want to do that at a lower cost and they want to extract the CUDA software architecture out and sort of build on top of their own architecture. So I think, bringing in the big sovereign piece does backfill in Nvidia, but it does not change this trend, which is sort of driving Broadcom, which is like designing your own custom silicon if you're, know, Meta or Amazon or Microsoft is inevitable. However, you know, at a foundational level, you build a new data center and your Microsoft, you will use Nvidia GPU. So one of the things that the thing is just the demand on silicon is just exploding.

And I think that's really the point. I was sort of thinking about this on my way to work on the car today. We cannot help being very promotional in this. This was never the intention of these videos. We set out to be informational, right?

But I think, broadly speaking, we built these ETFs over several years, but they are designed for purpose for exactly where we are right now. SMHX is, you know, the top of the stack is Nvidia Broadcom. And this thing just cascades down from there to, you know, semi-cap equipment software. They are the sort of, they're the monopolies on monopolies on monopolies. Now, as we sort of roll into this next phase, know, SMHX sort of broadens, but so it's very difficult to actually express how we feel or I feel about the outlook without sounding promotional. So I do apologize for that, but it really is not intentional. We're trying to be informational. So let's sort of leave it there.

Nick Frasse: No, we appreciate it. And I think we'll touch on that and the winners, the winners and the ecosystem and why SMHX.

Quickly touch on, I mean, obviously the big topic of conversation right now is, you know, the potential of a recession here in the United States. Does that derail AI? What does that mean for capex? All of those different things. I guess if you could just tie a bow on that quickly, because I think I know your opinion on it, but maybe, know,

Angus Shillington: You know, I think if you get market volatility, multiples will be all over the place, right? And the Cisco chart will show up and people will panic. we're trying, you know, part of this is to actually partner up with our sort of advisor partners to say, look, you should not be panicking during these situations. These businesses, this entire space theme is growing. So the intrinsic value of the businesses are delivering as we expect, the intrinsic value of the companies in the ETFs or the companies in the

That you pick a single stock should increase over time. Now, here's the question now, you know, what if things get tight? What if if if you know we go into a recession? You know one rates will probably come down meaning you can leverage for cheaper and debt stacks get cheaper and For the guys who are borrowing to build but most importantly like especially with open AI you can't stop building you just can't because

One, anti-fragile point I made earlier on, they've tended to get stronger in times of economic turbulence. There's no reason that would change. But then you throw into the mix that there's this massive competitor. mean, OpenAI is no longer tinkering with chatbots. I think that's a point to really, really throw out. They are building a revenue juggernaut. Now, at consumer internet level, people sort of think like Amazon makes its money from e-commerce or, that's not the case. These are lost leaders, right? In order to bring in a large community, right? That you can then sell advertising to. So advertising becomes the front line of competition. Now, the more they compete, the more they're going to have to invest. And this filters back up to this whole stack of GPUs and semis and semi-cap and innovation around power management and delivery at cost and delivery at scale.

Nick Frasse: So, semis, if you see a recession, it's a share price problem. It's not a structural foundational problem to the sector as a whole. I think we've been kind of shouting that from the mountaintop set. We're less and less cyclical as time goes on.

So let's go back to the ecosystem of winners, the monopolies on monopolies on monopolies.

We’ll start with SMH if you have one of those or SMHX, but what are some names inside of there that we could really look at that maybe people don't have as good of a lens on that really exemplify that diversified exposure and futuristic outlook.

Angus Shillington: Power is the key primitive to pretty much everything here. Power plus cooling is probably 15% of cost of goods sold of a data center, right? So if you are UAE and you can put in power for 50% lower, you immediately end up with a 5 to 6 to 7% better operating margin. So you can then deliver your tokens radically cheaper than anybody else. here's your problem is, one, if you're short of power, power will spike, and that puts you in a competitive disadvantage. If, on the other hand, you have a friend who has a fabless semiconductor business who is able to give you a solution to get your power usage down 40%, you are immediately in a far better position. The same case is true for, I'm just trying to randomly pick a name, know, Asterra, Asterra Labs inside of SMHX. These guys do specialized connectivity trips. So basically they eliminate data bottleneck. So if you don't have enough GPUs or you don't have enough switches or they're not working perfectly, these guys package and compress data so it flows quicker. So what we're moving onto is this stage now where the demand for applications within your technology stack can be optimized by using other semiconductors and other AI solutions based on silicon that will make the whole thing move quicker. So there are always going to be constraints. There are constraints at the top. Now that you've loaded all the stuff that's coming in the Middle East, you probably don't have the fab capability. Now the fab guys are going to start to invest more. Let's say, ASML moves down the stack into Synopsys, those kind of guys. And then within the stack of technology you've also already put out in your data center, you're having to optimize for data bottlenecks, for power bottlenecks. There's just a constant demand for efficiency. And efficiency, like we said, you touch anything with silicon it becomes more efficient. And I think that's really the way to think about it.

Nick Frasse: OK. I think it's important to note that it's like, Asterra is one example of that. Ther are 22 names inside of that portfolio that are doing something similar in a different part of this, of the stack, right. And in a different part of, that, that working group. it's important to understand that this isn't just, you know, SMHX is not just looking for the next Nvidia, right. They're very well might be the next Nvidia in there, which might be Broadcom or whoever might be, but there, there are an entire, group of holdings inside of that that are doing different things within the semiconductor space, designing different chips that are making things more efficient, that are making things faster, that are making things better, and that are driving AI forward. And that's why it's, you know, I think we've been doing a pretty good job, but I think this is the time to really hammer home like the differentiation there because where SMH is a literally a team of winners of the entire ecosystem. These are the guys that are making the ecosystem efficient in every way possible across every part of the data center. And not just data center, right across the board for every different industry that uses Silicon. So I think I fully agree.

Okay. Close your eyes. I need you to imagine what the next 12 to 18 months of the meat conductors look like. And I would imagine it's probably faster than whatever you could come up with, but we'll see. We'll see what you think.

Angus Shillington: Yeah, I mean, I think we've sort of hit it. You've already answered. Look, we had that, I don't know, everybody should watch that Johnny Ive video that OpenAI posted. Now, people think it was idiotic or it was annoying or I saw it as a very confident declaration of war. It may have been a recruitment drive. It may have been like a call for capital. The reality is the competition has arrived and at that level it is extremely broad. I think that's the thing to really watch out for here. Elon is activated. This guy is tilted. I don't think he's tilted because he's sort of upset with the way that OpenAI has worked. I think he sees the runway, right? It's like a boulevard of green lights right now. So I think that's the big surprise. As we said, the competition ramps. You are forced if you're Google or really anyone in that consumer facing stack, you're forced to invest more, right? To innovate more. So I think the jump, this isn't a conversation about consumer internet, but I think you've got to go to the customers to find out what's going to happen upstream to the guys who are designing and then the guys who are building and then the guys who are building the equipment to build and then the software to build, the equipment to build. upstream of the consumer guys, the next year I think is more positive than I would dare to believe. And maybe it's less positive for the downstream. I don't know. What we've tended to do with this thing is we wanted to throw out some crazy ideas that we believe in.

Nick Frasse: They look conservative.

Angus Shillington: And the thing is, where we've been most wrong is we've got the timeline where they've happened, but they've happened almost immediately. I don't think you can underestimate the sort of, yeah, the build. I don't think can underestimate the build.

Nick Frasse: Yeah. would say, as the kids say today, the video is a little cringe, but I get what they were doing and I understand. I do think, you know, we talked about this moving into the consumer space with something like this. It absolutely is a declaration of war, right? They are coming for everybody and that's what they're saying.

All right. guess any last thoughts before I go over our key takeaways, anything that I didn't touch on that you wanted to go over as far as outlook, thoughts?

Angus Shillington: No, would. No, is the answer. I think we covered everything. My feeling is like everybody's out there trying to pick winners. And I know it feels good to pick a winner and you can tell your friends about it. My feeling is always in the way throughout my career, I've always operated. It's like look for losers, get them out of your portfolio and the winners will find you. Right. And I don't know the winners will be in terms of, you know, share price performance over time. But I think given the reality that both the ETFs optimized to quasi monopolies, the winners will find you and it will be pretty rewarding. It's difficult to say exactly where that's going to be. And I think that's just a function of the fact that this thing is moving so fast. It's very, very difficult to get ahead of it in terms of mind mapping at a micro level.

Nick Frasse: Yeah. Own the basket, own the basket. Okay, key takeaways. Nvidia continues to be in focus. They're back in the starting lineup. Not that we picked them, we're passive, but as far as our mental model goes, you know, they're not on the bench anymore. Token growth scaling 2X every six months, something like that. That's probably a slow estimation, but token going up, compute needs to keep up. CapEx needs to be spent to get it there. OpenAI is a disruptor. And I apologize, I'm going through my list quick here, but CapEx forecast still too conservative. Streets.

behind the curve. We were behind the curve. didn't know how fast it was going to go. Maybe we're staying slightly ahead because we're overly optimistic because we don't have anything tied to us and we can do that. Definitely probably too conservative for CapEx spin. not in a bubble. Bullish on SMH. CapEx cascades throughout the entire semi-stack. SMHX just getting started. Probably the rookie, maybe the MVP 12 to 18 months, we don't know.

If you're looking at the, if we're sticking with the analogy of basketball teams or hockey teams, and the story of AI and semis continues to accelerate, right? They are inextricably linked. If you believe in AI as a technology, the most pure way to play it is through semiconductors. And I think that's full stop. SMH, SMHX are the leaders in the space. And I think the way that the indexes are constructed, has proven that that that to be the case. So, and we don't think that the AI train is stopping the rails are down and everybody's going to hop on. So

Angus Shillington: And I think the last piece, that was a great summary. You can expect these share prices to be volatile. This is a high-bait effector. But I think more and more, you should not expect the business shareholder return generation to be volatile. I think that becomes less volatile. So you can have more anticipation on the ability to buy into dislocations.

Nick Frasse: Absolutely. All right. Well, everyone, thank you again for joining us.. We would love to get comments, questions. We will try to put as many links in the notes as possible for everyone. But again, reach out if you have any questions, feel free to give us comments and feedback. We always are open to it. And if we don't talk to you, we'll see you next quarter.

Angus Shillington: Thanks, everyone.

In this quarterly semiconductor update, Angus Shillington and Nick Frasse discuss Nvidia’s earnings and the company’s return to the forefront of AI infrastructure. They explore the CapEx surge, sovereign AI investments, and the growing importance of fabless designers. Other topics include token usage growth, hyperscaler strategy shifts, and why diversification is key in the evolving semiconductor landscape. Read the blog recap here.