What Investors Are Overlooking in AI & Semis | Val Zlatev
01:16:51 Share

Before we get started, I want to do a quick disclaimer that nothing we say here is investment advice. As well, nothing we say is marketing or advertising for Analog Centrally Capital Management or any of their funds. Everything is meant to be informative about the fund management industry.

Welcome to another edition of Other People's Money. I am joined today by Val Zlatov, Portfolio Manager and Senior Partner at Analog Century Capital Management, a hard tech focused long short hedge fund manager based in New York City. Val, thank you so much for joining me today. Thanks for having me, Max. Appreciate it. All right. Well, we're going to be talking a lot about hard tech, semiconductors, AI. Why don't we start with what is hard tech and how did you become a hard tech investor?

So, hard tech is, think of tech, technology, but take out software and internet. So it's really all the hardware that goes inside any electronic device. So this will be semiconductors, communication equipment, the hardware itself, the equipment being used to making the semiconductors, and so on. Within that space, the majority of it is semiconductors, is by far the most prolific and most important space.

Many of the other things like hardware, I think servers or storage boxes is more on the commoditized side and it's also very concentrated over time. But there's a ton of innovation and proliferation on the semiconductor side. So that's the way to think about the space. It's a large space, by the way. It's about half, a little bit, probably more than half of the overall TMT space.

The market cap is big. So by now it's probably over $14-15 trillion worth of market cap in that space. So in that range, it is about 500 to 600 companies which have over a billion dollars of market cap. And there's a ton more, which is sub-billion dollars, but that becomes a long tail.

So that's the way to think about hard tech. Obviously, with the emergence of the AI megatrend, more focus has been going on hard tech.

But for a while, software was taking sort of all of the TMT talent and focus in the industry. And then prior to that, obviously, the first sort of tech bubble, you had the Internet names, but there was a lot of hard tech involved in that. Can you talk about the evolution of the hard tech sector and how finance has shifted its focus towards it and away from it over time?

Yeah, that's actually a very good point that you made. Hard tech was the it thing at some level in the 90s. And I think that back in the 90s, many of the sort of famous hedge fund managers are still very large and very famous. They made a lot of their personal wealth, overall hedge fund wealth in the hard tech space. And then in the 2000s, the whole thing shifted. And the reason for that, so we moved away from hard tech more towards internet and software.

And there was a very good reason for that, which is ultimately software and internet were the secular growth drivers in the 2000s, between, say, 2000 and 2014-15 timeframe, in that range. That's 15 years is a long period of time, right? And meanwhile, semiconductors became sort of slower growth and more cyclical. So instead of being secular drivers of growth like in the 90s, the way they were,

they transitioned from secular to more cyclical. And cyclical is really even cyclical from an EPS perspective, which is to say you go from these up cycles and down cycles, and the EPS goes up and down, but it just wasn't growing through cycles. So which basically made the hard tech space more of a trading vehicle as opposed to a secularly investing vehicle, which is hard. Being a trading vehicle, it's not fun.

And it's very hard to compound alpha. It's very hard to compound returns. And that's why it was very natural for all these managers to move much more to software and internet. And that's why as a result of that now, you don't have much of knowledge built into that space, into the hard tech space. There's a ton of knowledge into internet and software.

But when you look at the hard tech space, the knowledge base is very shallow on average across tech managers or TNT managers, pods, you name it. If you talk to most people and you ask them about a hard tech name, like they'll know Intel, they'll know NVIDIA, ASML. Like I start running short, you know, once I get to about five or six hard tech names and

And I think investors themselves are largely underweight hard tech, despite the performance of semiconductors over the last few years. You're absolutely right about that. They're actually underweight despite the performance, not just over the last few years, which is sort of very recent and noticeable. But the outperformance of semis has been noticeable since 2015, even more. So it's been like a decade of that outperformance.

And most investors have actually missed that development. And I think, by the way, they continue to miss it. So maybe part of sort of the hidden question here is, or the question, the question is, why is that, right? Why do people keep on missing it? I think the, so there are a couple of factors for that. One of them is what I just spoke about, which is the knowledge base just wasn't built there.

because everybody was focused on internet and software. It's kind of easy to continue doing what you know how to do. And by the way, it's a good practice to continue doing what you know how to do. The second one is, and by the way, building knowledge in Heretic is hard because the Heretic space is highly fragmented. There are many layers of the supply chain. Understanding all these

different companies with very different business models takes a large investment. It is not easy to understand that. So it's much more complicated than taking the internet subscriber model and effectively replicating that across a hundred names across the world, or taking a SaaS business model and replicating it across a hundred SaaS names in the world. They're kind of the same business model. The metrics are slightly different, but it's sort of the same metrics.

When you look at hard tech, oh my gosh, you have like 15 layers of the supply chain, semi-caps have a whole other one set of metrics, which is super different from what an analog semiconductor metric could be versus what a foundry metric could be. And basically the business models are very different, right? The spreadsheets are very different. So it just takes time to build that knowledge and people just haven't spent time to do it. So that's one part of it. The other one is, it is just fear of cyclicality.

At the end of the day, hard take is a physical product. These physical products go through distributions, they sit in inventories somewhere, they sit in boats in the oceans between Asia and the United States, whatever it is. So inventories would naturally build up and they'll get depleted over time. So understanding these cycles is also important.

At this point of time, over the last 10 years, there's a lot more secure growth drivers than cyclic car. But cyclic car is still there. It hasn't gone. And people are afraid of that, right? People are much more comfortable with investing into something that doesn't have a supply chain element to it. That just you push a button and it doesn't build an inventory like a software piece. So I think that fear factor is also there.

I think also when you have strong performance, you said the outperformance is not just recent. It's been since 2015. There's more commentary on when to sell NVIDIA than whether it's a buy, I feel like, than there is, you know, is it cheap? Is the AI trend going to continue? I mean, people are scared of buying things that are up hundreds, thousands of percent. It's just mentally, it's very difficult to do.

You're absolutely right. Literally, I've been facing these questions of, isn't that the peak of the AI trade? I have faced these questions since June of 2023. So it was about eight, nine months after ChatGPT came out. Remember, ChatGPT came out in November of 2022, right? And people started talking about it and it became a thing. And then in May of 2022,

In video had this huge print. That was the April quarter and the stock just flew, right? And the whole space flew in that one day. And then really since then,

well, a month or two later, almost every question was, well, oh my God, isn't that the peak? Isn't that like 2000? Isn't that the bubble of 2000? It's sort of coming back to what you mentioned at the very beginning, which is people made money in that space in the 90s. And then there was an internet bubble. And cause the bubble burst, the space didn't do much for 15 years. So right now, everybody's trying to sort of hit that inflection point.

to the inflection negative trade that says, oh, the secular piece may be gone and now we could be going back to cyclicality and everybody somehow is trying to do this pattern match or pattern recognition to the internet bubble and to match it to the current situation. And to be, basically, that's what people are trying to do. And that's why, by the way, most people have been missing it also because people would build a position...

I don't think many people even had Nvidia before 2022, before chat GPT. It wasn't really a thing back then. And then everybody started to latch onto Nvidia and then everybody started to want to trade it. They were like, oh my God, I just made 40% on it. That's so cool, right? And they will cut it in half in the position sizes. So they'll cut like, they'll sell 75% of it.

And they would try to justify that by saying, oh, maybe that's the next Zoom, right? I mean, look at Zoom. Zoom had one big print back during COVID and hasn't had anything else since then. And I remember these discussions, people saying, if it is kind of like the next Zoom, had one big print in May 2022, it's game over. And I'm like, seriously? I, like most people, certainly fell into that as a generalist and...

And you just have to, though, when you look back on things, you have to say, what did I get wrong? And is there a potential I'm still getting it wrong looking forward? But to the cyclicality, I think there's a unique aspect with the quote unquote end of Moore's law. I think everybody's familiar with the idea of like overcapacity and traditional industrials where, you know, there's more steel being made. All the steel plants are up and running and then suddenly you're oversupplied. But, you know, with chips, you

doubling their power every few years, you were getting like two steel beams for the amount of iron ore you were putting in. And I think that really made the cyclicality sort of exaggerated. And now that is coming to an end. Can you talk to me about the end of Moore's law and how that is changing the cyclicality somewhat? That's a great topic. So

The end of Morse law, it's actually, it's a big thing. So Morse law, just for you listeners, is a thing that every two years, it was an empirical law that said every two years, you pack more transistors into a semiconductor, which is like building logical blocks of a semiconductor. And

You pack twice as many, and the result of that is that you get twice as much performance at half the power. It's a beautiful thing, right? And that's an exponential law because it compounds every two years. So that's the reason why now our simple smartphones, our iPhones, are more powerful than the, probably they're more powerful than the Oak Ridge National Lab supercomputer that I used to work on when I was in undergrad back 30 years ago.

That's the compounding effect of Morse law. Now, after packing and packing and packing these transistors onto silicon areas, which is tiny, which is basically the size of my fingernail, at this point of time, we're putting two, three, five billion transistors on my fingernail. Billion, not million. The spacing between these transistors is, by now, it's about five atoms apart.

So there's nothing more complicated from a manufacturing perspective than doing atomic scale manufacturing when you're really tweaking things about five atoms apart. That's like an amazing thing to me. And then the quantum mechanical effects. So if we're trying to control electrons between the walls of these transistors,

Well, if you have only five atoms spacing, well, they're quantum tunnel. They don't just sit there. And quantum tunneling means power leakage, which basically means if you don't do amazing amount of tricks, you're going to have a phone that drains its power in 10 minutes, which is unworkable. Nobody's going to buy a phone that drains its power in 10 minutes. So that drove a ton of tricks to be done. It became very expensive, very complicated.

To be able to control quantum mechanics, basically to be able to fight quantum mechanics, that's the price for the new fabs. A new fab, now fab is a facility to build semiconductors in. These facilities, I remember back in 2000, when I was in my prior life, I was in McKinsey, I was actually working in these fabs as a consultant. They used to cost, like, state of the art was like a billion dollars. You know, a billion dollars is like a big deal, right? It's like, oh my God, that's so expensive.

Well, now it's 50. The same size fab now is a $50 billion investment. It's 50x. Nothing went up in price 50x. It wasn't inflation. It wasn't the brick and mortar. It wasn't like the workers or the salaries. It was really the machines that go inside these fabs. And the machines became 50 times more expensive.

because they had to be able to deal with quantum mechanics and control it and create these super complicated structures to stop quantum mechanics. I never thought, by the way, that I would talk about quantum mechanics in my whole life. I mean, I have a PhD in physics, but I felt like, you know, I did my PhD, I left, done, checkmark, never talk about in my life about quantum mechanics. I used to do quantum gravity kind of stuff way back in time.

But here we are. For the last seven, eight years, I've been talking about quantum mechanics in almost every allocator meeting that I have had. So that's the reason why, by the way, it is very slow to build these facilities. It takes like five, six years to build them.

It's very expensive. Like I said, it's about $50 billion for a state-of-the-art leading-edge technology facility. And when you do that, you don't do that willy-nilly. So you don't just go ahead and throw a ton of capacity into the ground just to capture the profits.

And the high cyclic clarity in that sector was driven by profit pool goes up, you add capacity very quickly to capture the profit pool. And of course, before you know, the demand goes down and the profits vaporize because now you have extra capacity, which is sitting there idle. That kind of doesn't happen at that level anymore. I don't want to say, I don't want to create the impression that cyclic clarity is gone. It's not true. Cyclic clarity is still there, but it's much more subdued than what it used to be. So it's more moderate. It's palatable.

And that's what makes the space actually much more investable also. Not just that. So one aspect of the space being much more investable now than being a trading vehicle is that piece, the more slow.

The other aspect is that you do have now secular growth drivers like AI, for example, that's exploding, which I believe we're in a very early stage. We can discuss that if you care. But these are meaningful demand drivers on top of constrained capacity. So you have constrained capacity, subdued security, hard-to-earn capacity, and at the same time, you have growth that actually wants more units of this stuff.

So that makes it much more investable than what it used to be. Is there any capacity being added? Is anybody dropping $50 billion? Yeah, definitely. TSMC is dropping $50 billion more or less on a yearly basis as we speak. They're building fabs in the U.S. as well as in Taiwan. So that's an ongoing process. There's plenty. I mean, like if you think about the overall capacity investments,

They're large on a yearly basis. The equipment that's themselves, just the equipment, is in the, right now, this year, it's about $100 to $110 billion across the world, just for equipment, for capacity. And then the bricks and mortar and stuff will put the overall CapEx product at $150 billion or so. So it's in that range. The biggest vendors will be like TSMC. They'll be in the $40 billion plus, $40, $50. Samsung will be $25, $30 billion plus.

Intel is trying to do something to save their life in the $15 to $20 billion age. So this is kind of... And then, of course, you have Texas Instruments in the $6 to $8 billion and so on and so on. So there's a long tail of investments. This may be a good place to transition into the AI trend and where you see it. I mean, how does that compare to the growth in demand that we're seeing? Is that enough to keep up? So let's talk about AI because that's...

It's a very interesting topic. And like you said, most people are trying to compare AI to the internet boom and bust. And hence, everybody was trying to compare NVIDIA over the last two and a half years to Cisco back in late 90s. Max, I bet you can see that chart of overlapping NVIDIA to Cisco on Twitter probably showing up every four weeks, shows up from somebody there.

So it's cute, right? It's very cute to do these overlaps and to try to make a parallel, but the reality is actually very different. So the reality, I like to make a different parallel for AI. I think AI right now is much closer to the cloud's development. So the cloud started in the 2010, 2012 timeframe.

And it grew over time. And it has been mostly driven by a set of like three or four American companies and a handful of Chinese companies, right? So we're talking about AWS from Amazon, Azure from Microsoft, Google, GCP, and a couple of others. So that's the cloud. If you think about the cloud, they have these huge companies. They invest from their free cash flows. They're not borrowing money. Like back in the 90s, it was all borrowed money or venture capital money.

zero borrowing, it's all free cash flow. They have a set of very large customers, Fortune 500, Fortune 1000 customers, different from the 90s where the customers were a bunch of startups. And they have, the cloud has had this strong business case that basically says cloud compute is more efficient or more secure, whatever it is. And more and more of the compute should be moving to the cloud, which has been exactly the case for the last 12 years.

And so if you think about it, CapEx for cloud from these four or five large hyperscalers plus a couple of Chinese has been on the upward trajectory since 2012. We actually tried that every single quarter, the CapEx of the cloud companies. It has basically never gone down for the last 15 years. It just hasn't. Every year it's up. In a horrible macro environment, it could be flattish. And then it pops again the next year.

From my perspective, the AI CapEx investment is on a similar trajectory to the cloud CapEx investments. The reason is it's the same companies that are investing into AI. It's the big hyperscalers. They're by far the biggest spenders on AI. They are building that not...

because somebody is throwing venture capital money at them. They're doing it because they have the big pipelines of like 50, 100 plus, sorry, the Fortune 500, the Fortune 1000 customers. They have large number of proof of concept cases for these customers. These customers want the AEI.

This is not a push. It's actually a pull from their customers. Their customers, like the big financial institutions or the big industrials or the exploration chambers or whatever it is, they're the ones who are pushing them for more and more capacity because they need these use cases to go live. And at this point of time, they have enough capacity for like 25% of the use cases to go live, even less than that. Even on that, they're very supply constrained. So based on our research,

The return on investment is there. ROI is there. They're building because the ROI is there. And the largest companies in the world are seeing this ROI, and that's why they want it. So from my perspective, we're in a very, very early stage. Like we're year two and a half in what I think it will be a multi-decade investment cycle. That's back into the cloud. And you talk about those use cases that are already there, but as the models improve, the use cases expand too.

Yeah, exactly. So the use case expands, which is kind of the power opposite of what the reaction was during deep seek back in January, right? So you're absolutely right. The models improve. The cost goes down. So if you think about the cost for inference, which is the actual use of AI, this cost goes down dramatically, like 5% to 10x a year. That's a lot. 10x, not 5% to 10%. I'm talking about actually 10x drop. And the use case is proliferate. You're absolutely right. And the model proliferates.

As long as you have return on investment capital on this, the more and more you spend on it. It's as simple as that. If you make money on it, you're going to spend more on it. So what do you think people got wrong about DeepSeek? What was the common perception of that event when that came out? Obviously, people thought, well, they're not going to have to spend as much. Yeah, exactly. That was the gut reaction at the very beginning, which was... So there were a couple of mistakes that people made there.

One of them was to believe that the deep seek, whatever deep seek introduced, was a quantum leap compared to from a cost structure perspective. So people felt like, oh my God, these people build this G-Wig gizmo for like 6 million bucks. That was their white paper compared to the city Americans that spend like billions of dollars.

So the smart genie, 6 million, the silly Americans at billions of dollars investment. Oh, look at this. It does a good job compared to the American AI. Well, even if it's all that wasn't true, just it was like apples to oranges. So that took a little while to pass through and realize that, you know, DeepSeek actually did spend many, many hundreds of billions of dollars to achieve what they were achieving. Number two, people realized over time also is that they were just the next iteration of

on the cost structure of reducing cost for inference. So inference, like I said, goes down 5% to 10%. Every three to four months is the step down, step down, step down in the cost. They just happen to be the next step down. And the reality was that three to four to six weeks later,

a bunch of the other models like from OpenAI, Mistral, or Anthropical, whatever it is, they went at the DeepSeq level. And after that, actually many of them went below the DeepSeq level. I mean, DeepSeq is not that cheap at this point of time. There's a number of international models which are cheaper than DeepSeq over the last six months. That's because it wasn't revolutionary. It just wasn't. It was like a myth. But it wasn't.

So that was one thing that people got wrong. The other one was, yes, oh my God, now we can do a lot more with a lot less, right? That was a reaction. Well,

you have something cheaper, you do more of it. If it's useful, right? If the return is there, it's more useful and you're going to use more of it. And that's exactly what happened. Basically, usage skyrocketed across the board because not only DeepSeq was cheaper for inference, but everybody matched it and went below it. So for everybody, inference became cheaper. And of course, the usage just skyrocketed from there. Now everybody's short. Yeah.

So instead of being, you know, having an abundance of chips that you don't have any idea how to sell because nobody wants to use them. Now everybody's begging for chips again, like within two months. Now you talked about, you know, the size of the companies and it's the biggest companies in the world that are in on this trend. There is one company that is bigger than all of them, and that's the U.S. government. What about the U.S. government's involvement in chips and AI? And then maybe we could talk a little bit about China as well. That's a great question.

Yeah, I mean, it's not just the US government. The Chinese government is also deeply involved. It's semiconductor for, God knows how many years already, like five, ten years. So let's talk about that. It's a great question. So ideally, I personally kind of hate government involvement in any kind of private sectors or any private, public companies.

Partially that comes from my personal background, clearly from my accent. I was born in Eastern Europe. I grew up in Bulgaria. I grew up during socialist times. That was a miserable, miserable experience. Socialism just destroys economies. And of course the government was deeply, deeply involved in any kind of economy over there. And it was a mess. It was a total, total disaster from that involvement. So I don't think that government involvement here is at that level. It's not even remotely close to that. But, um,

It has been government, the U.S. government, let's talk about the U.S. government first and then the Chinese government. The U.S. government has been involved over time in a number of ways in the semiconductor sector, mostly. And a lot of that has been an involvement to

A, constrain the amount of exports of American semiconductor equipment to China or the export of a semiconductor chips to China also over time. And that has become this existential sort of

Cold War around AI or military use of semiconductors or spying with semiconductors, whatever you want to call it. All of these things in that whole umbrella where the U.S. government is trying to restrict what the Chinese government needs. The reality is that probably most of that's excuses, but it's coming down to AI. The U.S. government ultimately would like to presumably win the AI war, if you wish. The Chinese want to win the AI war also, and it's a race to who wins it.

I don't think, by the way, from my perspective, I don't think there's winning of that war. I don't think that's a war. I think it's a race. There is no set goal that if you reach it, you win it. This is not like the space race where, oh, you put a person on the moon, you won it.

I mean, how do you define winning in AI? I mean, there's absolutely nothing that you can use to define that. Maybe one way to define winning in AI would be if you have the whole, most of the world using your AI stack, then you can probably claim that you have won because the world uses what you have. You sell more to the world in terms of AI software, hardware, chips, whatever it is,

And, you know, there is some control there over AI will not be useful in horrible ways. And AI will not be used to track what you do and ship it all to CCP in Beijing. Maybe that's a way to define that winning. But that's sort of been the target. And the U.S. government has been both sticks and carrots. So they have been the sticks that say you cannot ship this, you cannot ship that. And every 12 months, they ratchet it up and ratchet it up gets harder and harder.

The worst by far was that diffusion rule that the Biden administration left a week out before they left office. And the diffusion rule was basically restricting many, many nations in the world, including allies that we have in Europe and potential allies in the Middle East, from buying as many AI accelerators as they need for the AI developments.

That, from my perspective, was a horrible rule because if you restrict your allies, they need it, they'll buy from somebody else. And the somebody else would be China. So effectively throwing partial allies and to a big extent important geographies like the Middle East into the hands of the Chinese as opposed to drawing them towards yourself and having them use your AI stack. You're throwing them into the Chinese AI stack.

So that rule got rescinded, thank God, like a couple of weeks ago. So at least that involvement of the US government is gone from that perspective. But you know, many restrictions are still there, right? Nvidia got restricted from shipping all kinds of AI chips to China, including the lowest level, H20s, which is called like very subpar,

chips, but they're still constrained. They cannot ship them to China. So there's a lot of that involvement. There was also an involvement from the U.S. government from a carrier perspective through the CHIPS Act. About four years ago, there was something called the CHIPS Act that said, we're going to allocate $50 billion for U.S. facilities and companies should go ahead and apply to get X amount of money to build more capacity in the United States. And there was additional grants for tax reductions, less taxation when you do that.

So, that was a good carrot. In principle, we need more facilities in the United States. All that this act did was to really bring the cost of building a facility in the U.S. on par with building a facility in Asia.

The reason these facilities are getting built in Asia is because it's just cheaper to build it there. It's regulation, labor cost is lower, it's not unionized. The machines are the same, by the way. So 70% of cost is the same because you buy them, they're the same price everywhere, from Japan or like Netherlands or from the US. They'll be the same price no matter where you build the facility. But yet at 30%, 40%, which is all the other stuff,

is way more expensive in the US than just about anywhere else in Asia. And that chip stack was supposed to bring that on par, bring it in balance so that people can build companies with this game. Good thing. I think the logic was there. Very good thing. Now, the tricky part, of course, is

How did it really get implemented? And who became the champions? And who really got the most money? And guess what? It was Intel got a bunch of money from that. And Intel is still in a horrifying state, struggling. Who knows it will survive over the next quarter of five years? I have no idea. Another remarkable example was like there was allocation of money to Wolf Semiconductor, which makes the silicon carbide. And that company is about to go bankrupt. They cannot even pay their debt.

So again, the amazing wisdom of government bureaucracy would give money to some of the biggest losers in the space just because. Do you think the U.S. government has the technical knowledge to be meddling in this way? I don't think that they know. I don't think they have the knowledge. I think they hired people to be part of this conversation.

efforts, who had some knowledge. And of course, they were consulting with most of the companies. The people they had, they had some knowledge. They were talking to all the companies. So I don't think they were blind about it. But at the end of the day, I would actually much rather defer to the market to decide who is going to be the winner or the loser as opposed to somebody working at the government anywhere in the world, by the way, for that matter.

By the way, the same story is true for the Chinese government. So the Chinese government has been deeply involved in trying to stimulate the Chinese semiconductor developments. And that has been full subsidies, heavy subsidies over the last 10 years, more than 10 years.

Because China, over 10 years ago, they decided that semiconductors are essential, right? They're absolutely essential. Oh, by the way, they already sold all the software. There's no fighting over software. Think about it. Which software company actually makes any money from China? It's not like Microsoft or CRM or ServiceNow where they have any meaningful revenue streams from China. It's all gone. It's all copied, replicated.

Facebook has zero, sorry, Meta has zero revenues from there because they're not even allowed to be there because they replicated the Meta business model. They replicated the Amazon business model. So the Chinese found a way to replicate everything but semiconductors.

Why? Because that stuff is extremely complicated. High barriers to entry, deep knowledge, takes decades to understand it, decades to try to replicate something, and it takes them forever to do it. That's why Semiconductor is a battleground, because that's the one place where it is the most complicated, most involved battleground, and everything else has been replicated anyways. They have dedicated a ton of money. They have done a ton of progress on a number of dimensions, not everywhere.

You know, they're getting pretty close on memory, more so on NAND than DRAM. They're getting pretty close on some other types of semiconductors. You know, they have some good digital semiconductors like Huawei has been quite good at developing semiconductors for digital circuits. So without going into details, they have definitely had progress.

Are they at the bleeding edge now? Are they definitely dead behind? It would take years and years. But again, even in that case, there have been countless, countless failures in China. I mean, you can probably write this in the hundreds.

failed Chinese companies trying to make semiconductors. And I don't think the Chinese government cares if that fails. They just see the bunch and if something comes out of it, so be it. But they continue to see a ton of companies over there. Is there a dominant player in the Chinese market? For digital semiconductors, yeah, I guess Huawei is by far the biggest. Even though they're restricted, they

There's many things they cannot buy, but they still find ways to be about, say, three to five years behind everybody else in the US. There's a dominant player on the manufacturing side called SMIC, which is the semiconductor manufacturing company of China. Again, it's far, far behind TSMC. They have a couple of dominant players in memory like CXMT and YMTC.

that are becoming pretty real. They're getting close to like 20% of the memory capacity in the world. That's nothing to sneer about. That's pretty real stuff. And that's kind of about it. And they have a couple of semi-cap companies which are very, very small, which are trying to compete with applied materials or LAM research. And they do well on some

sort of not very complicated equipment, like cleaning, etch kind of stuff. If the chips aren't as good, are they only selling to Chinese companies or are they also selling to the rest of the world who are being limited by that U.S. policy? If the diffusion by the rocket continues to be in existence, many countries will be forced to buy from Huawei because Huawei is really kind of the one other player that has some accelerators.

for AI. Now, these accelerators are nothing compared to the NVIDIA GPUs. I mean, I shouldn't say nothing. They will target that their latest is comparable to some of the hopper series of NVIDIA. Maybe true, maybe not true. That's highly questionable. Everybody can tune up benchmarks to achieve something, but the reality could be very different. Also, it has been hard for Huawei to actually scale up, scale up facilities, which means

you probably have enough capacity to build 50,000, 100,000 chips, which is enough for one data center. But can you actually, do you have enough capacity at SMIC or anything else in their facilities? Can you build 5 million? Can you build 8 million of these chips? Which, by the way, 8 million chips is not a ton to talk about because there's billions of chips being built per year from the smaller chips. But 8 million would be

big from an AI perspective. Like NVIDIA, for example, this year will probably sell 5 to 6 million accelerators. That's about it. They're expensive. Each one of them are like $30,000 or $40,000. So basically, think about every accelerator from NVIDIA, which is not just an accelerator. It comes with memory and a bunch of other stuff with it.

That's like a Tesla, right? Same price as a Tesla. For something that is 1,100 the size of a Tesla. That's because it's much more complicated. A car is a simple thing. Chips are very, very complicated compared to cars. So I don't think they have right now enough capacity to supply the world. They don't have enough capacity to flood the world with...

with these chips. But I don't know, give them a couple of years. I would not in any way take for granted that the Chinese are way behind, they cannot catch up. I would actually give them a lot of the benefit of the doubt. I think they're smart people. They're highly educated. By the way, when I was at Penn in Princeton, most of the students in the physics department around me at Penn in Princeton were Chinese. So

These are smart people, highly educated. They have invested a ton in R&D. I think that Jameson Huang, the CEO of NVIDIA, was publicly talking about more than half of the patents in China. So don't take it. You should take them serious. They're very gung-ho about it. They're very good at it.

It's all a question of capacity for them. And who knows? They may find a way. So I want to ask about other areas beyond semiconductors. Obviously, that sort of steals the show in the AI supply chain conversation. But hard tech, although dominated by semis, is not a monolith. What else in the AI supply chain do people sometimes overlook? We can talk about the AI supply chain. Let's talk briefly about that. And there's other parts of the supply chain, which is not AI related, which is also interesting stuff. Mm-hmm.

So on the AI supply chain, you're absolutely right. When you build a data center, and let me step back for a second, when people look at these big data centers, right, and people talk about them, they talk about electricity, that electricity may get constrained and so on. The reality is that these data centers, when you look at the cost structure, 85% of the cost structure is hard tech equipment, is a sub that goes inside the data center.

The other 15% is the brick and mortar shells, is the land, is the electricity, is the transformers, the power equipment, the HVACs, the current equipment, all this stuff. Like all the obsession with electricity and power, that's about 15% of the equation. So let's talk about the 85%. The equation is on the 85%, which is the hard tech piece. This 85%, you're absolutely right. NVIDIA is a decent chunk of that. I mean, they...

They're fairly dominant. I think NVIDIA is probably 80% of the 80%. It's about that. Because NVIDIA is not a chip maker. NVIDIA at this point in time is a system maker. Because when NVIDIA sells stuff, they don't just sell you a little chip from a wafer. They sell you, at the very least, a board which has several of these chips with a bunch of memory, with a bunch of power management, with a bunch of other networking stuff. At the very least, they sell that. And a lot of what they sell to the hyper scalers is...

full racks with the networking. So they actually have a very heavy networking company at this point in time. About 20% of their data center revenues, about 15% of their data center revenues is networking-related stuff because they actually hook up their chips in their racks with their own networking to make it work seamlessly. By the way, it's a very important point. It's related to the more slow dev that you were talking about, that we were talking about. The only way to really now extend more slow

It's not for making the chips bigger to be more powerful. This doesn't work anymore because they have reached what's called the radical limit. Radical limit is a technological term that says the equipment doesn't work anymore. Like the lithography equipment from ASML cannot make a bigger chip anymore. You're done. You maximize the size of the chip. The only way to... Now the trick here is to make...

Like 70, 80, 100, 1,000 of these chips work as one. You want to treat them into being one, to think in the same way, to think that they have the same memory. They're basically instantaneously clicking together. That is done with networking. So that's why we're actually excited about networking companies also in the data center, in addition to NVIDIA, in addition to the other accelerator companies out there, like the TPUs from Google or the Tramiums from Amazon. There's like...

There's a set of companies that do all kinds of other. So NVIDIA is not the only one that does networking. They do it for themselves, for their systems. But many companies like Amazon or Google, if they're buying boards from NVIDIA or something from AMD for that matter, or they have their own boards for their own chips, then they hook it up with their own networking. And there's a bunch of companies that you can buy or own or short for that matter, if you want, which are involved in hooking up these chips also.

or connecting chips from like CPUs to GPUs, GPUs to ASICs. So there's a lot of this crosstalk going on and that's not working. So these are the companies which people are not talking about much usually, but exist there. And they're part of this data center build out. It's also like, you know, companies that assemble these, right? Who build them. It's like, these are like assemblers. Yeah, the margins profiles are puny compared to NVIDIA or many of the similar companies. These assemblers are working for like

I don't know, 8%, 9% gross margins, right? Or 5% operating margins. Fine. But the EPS, if the EPS grows fast, SOC will do well and vice versa. So there's a whole slew of these companies. Now, that's the AI supply chain, which is exciting because there's an exponential growth. And I was talking about, I don't think it's going to be exponential for decades, but I think it will be sustainable for decades, like meaningful growth for decades. There's the other side, right?

Like one other side of the semiconductor supply chain would be, we can briefly talk about power management. So power management, or like analog semiconductors in general, analog semiconductors are semiconductors being used in any kind of electronic system in the world. Anything that hooks up to an outlet to get power or hooks up to a battery to get power has to power up the voltages and then power them down many, many, many, many times in a circuitry of some sort.

So this is done by companies like ADI, Onsemi, Texas Instruments, Microchip, STMicro, Infineon, you name it. A bunch of these guys do these chips. And these chips are a lot. They're not like the six, seven million of Nvidia. They're like in the billions. Now, the end markets tend to be mostly automotive and industrial. These names that I mentioned now, they're like more than 50% of the revenues come from automotive and industrial. So it's a very different end market, right? Very different application sets.

So these companies tend to be more cyclical because...

There's countless customers, and they have purchasing managers behaving weird ways for their own incentives just to make sure they have enough, and they get scared out of their minds when they have way too much, which builds up inventories and depletes inventories quickly, but not so quickly. And this is the kind of companies that were in shortage back during COVID. I don't know if you remember, Max, but during COVID, there were countless discussions in the newspapers about shortages, and people couldn't buy cars, right, because analog chips were short.

And there was a lot of bellyaching about the shortages. Well, guess what happened? Human psychology never changes. So the purchasing managers were overhauling like 3, 4, 5, 10x overhauled these chips. And before you know, by 2022, the revenues of these companies were artificially high. And the customers could build a ton of inventory. So since then, they have been depleting their inventories.

for like three years. That's like an extremely long period of time of depletion of inventors, three years. Their revenue is now cut down in half. They're literally cut down in half. Can you imagine how it feels if your revenues are cut in half? I mean, just nuts, right? Your facilities are half empty. Your margin profile is squeezed. I'm sure it feels like hell for them. And just now, literally like two weeks ago, during all the last month during learning season, because we just finished learning season as we speak,

Over the last month, everybody was going into earnings and everybody was expecting that the guides from these companies will be wholly firmly bad. They'll be like, they'll be guiding June down. And the analog was, guess what? There's always an analog, right? The analog was COVID because COVID was a shock event, right? Everything was shut down. Nobody could buy stuff. And so the revenues went down back in June of 2020. These companies went down 15%, 20% sequentially just for one quarter.

And people now are saying, hey, look at the Trump tariffs. Trump showed up in the Rose Garden with his whiteboards and they were like 50% tariffs, 60% tariffs. And everybody's like, holy cow, that's a total freeze, right? I mean, who will buy anything if you don't know how to build stuff? Why would you be buying anything? So the expectation was Trump tariffs, same as COVID. All these companies should guide down 70%.

7% to 15% to 20% down sequentially. Well, guess what? They got it up. They showed up and said, no, no, no, no, no, no. We're actually going to grow. We're going up 7%, 8%, 9% sequentially in June quarter. They were like,

The analog is not correct. People haven't panicked. People actually continue to live their lives for the time being, at least. We'll see what happens. For the time being, they continue to live their lives. And the inventories have been so much depleted that now we're starting a new upcycle. So the June quarter they got into is the very, very first quarter of a multi-year new upcycle in analog semiconductor practice. So I thought that was kind of interesting also over the last month.

And you said it's automotive and industrial. Has that always been the case? Were other markets like phones, computers? Yeah, it wasn't always the case. It was traditionally back in the 2000s, back to the very beginning of our discussion, when semiconductors became sort of much more cyclical and more popular.

still in growth, that was much more consumer driven. It was really PCs first and then smartphones. There were no smartphones until 2008, right? But it was PCs for early 2000s and then smartphones after that. Then the whole thing, I mean, but there has been no growth, right? In PCs and smartphones. But little known fact. Well, the last time, what was the peak of iPhone shipments? 2015.

Apple has not really grown their iPhones since 2015. They have been sort of stuck in a range between $200 million a year to $250 million. But that's kind of it. There's no growth in phones. There's no growth in PCs. So there has been much more growth in cars and industrial. And not necessarily because of the unit growth of cars, because units of cars are 1% a year, 2% a year tops. It's much more about content growth there.

These cars are becoming much more like smartphones on wheels over time. And what used to be 10 years ago, like I call it like a hundred bucks of semiconductors per car. Now it's at least seven, eight hundred for non-electric vehicle. If you go to electric vehicle, it's like three thousand bucks. So you went from a hundred to a thousand to three thousand. And that's content growth, which is fantastic, right? We love that stuff.

That's why these companies are much more supposed to be automotive now. Industrial is a similar situation. There's robotics, there's automation. You're trying to move factories now to the Western world and North America to re-domicile manufacturing. Well, I mean, you do it, but you do a ton of automation, right? You don't want to be paying 50 bucks per hour for a person. You're paying a ton for like little...

I want to ask one question before we move on to talking about investment strategy a little bit more about Analog Century, which is we've talked a little bit about, I think, the winners, who you think are going to be the winners. Hinted at maybe a few losers that are struggling to keep up.

But also, what about the fakers? Obviously, the mentions of AI in earnings call transcripts has absolutely exploded. Certainly not all of those companies are actually doing anything meaningful in AI. So when you are trying to identify AI winners, just general semiconductor winners outside of the AI trend, what are you looking for? And then, you know, same question for losers and then the fakers.

Well, that's a great question. Yeah, so fakers. So everybody's trying to be an AI company, right? Right? There we go. Sort of like, but becomes a marketing gimmick. So there's a number of, there was this whole thing called edge AI, like AI at the edge. So like device-level AI. So there's a number of like all the smartphone companies, including Apple, were trying to pretend to be AI companies, right? And PC companies decided to talk about AI PCs, right?

So AI PCs, AI smartphones, AI whatever at the edge. But let's get real, right? The market sees for that. Market realizes very quickly it does just a marketing BS versus reality. And the reality is that there is no real reality

AI PCs yet at this point of time. I mean, what it really means is there's no real AI PCs. It's not like you can go ahead and buy yourself a PC that has an AI PC sticker on it. But what does it do really? I mean, is it like 10 times better at doing some AI stuff than my normal PC does it? Can it be running chat GPT or whatever the hell it is that I run at like 10x speed? No, that's not the case. So the functionality just wasn't there to justify the marketing level.

And as a result of that, you don't have a situation where people are buying a ton of PCs just because they're AI PCs. No, PCs are still stagnant. They're just not growing. They're like a couple percent a year, if lucky, you see. Same thing with smartphones. I mean, you can talk about Apple intelligence, but it's not intelligent at all. I mean, Siri AI, I mean, I hope they fix it because Siri, I think it's like the dumbest AI out there.

So that's why there's no growth in iPhone units. You can call it all you want. The units are not growing. People see through it, and it doesn't work well. So this is sort of the posers at this point of time. That may change. Who knows, right? They may find a way to really integrate AI in a very, very smart way into these end devices for the consumer, but that hasn't happened yet. So these are kind of the fakers and the posers which are not winning at this point of time. There's also other companies which are pretending to be really smart

benefiting from the data center AI. People are always trying to spin a tale of, well, maybe storage, right? Maybe these things need a lot more NAND memory, flash memory, because it needs to sit next to whatever, next to the accelerators and needs to be able to draw all this information quickly. Well, it wasn't the case, or at least it's not as much as what people thought. So the storage cameras were not benefiting as much from that perspective. So there's a lot of these things

companies out there. And also, you know, you have like the slew of NVIDIA killers. Okay. So you have the AMDs on one side, which is a pure play, right? GPU manufacturer that competes directly, which hasn't made a dent at this point of time. I mean, it's growing. It's growing very well, but it's not a big dent. Then you have the hyperscalers that make their own chips, like the TPUs from Google or the trainers from Amazon and others that come in

Again, that has been true for like seven, eight years. They have coexisted. They haven't killed NVIDIA. They haven't killed anybody else. They continue to buy a ton of merchant chips, as we speak. And then you have the venture capital investor startups. And I swear to God, ever since we launched the fund back in 2018,

From the very first conference I went to pitch the fund down in Florida, I kept on hearing, "Oh, I just put... I'm a venture capital investor. Sorry, I'm an investor in both venture and hedge funds. And I just put a bunch of money into a venture capital fund, which is investing in the Nvidia Cure." And I've been hearing that for the last seven years and nothing has really come out of it at this point of time. Intel bought a couple of these, like Nirvana and a couple of other things.

nothing real has come out of it. And there's many, many reasons for that. At the end of the day, it is from a venture capital perspective, it is very hard to fight the barriers to entry. I mean, we just spent like $12 billion a year on R&D. Writing a $12 billion check a year, it's not easy.

That's why the venture capital community is much better positioned to invest in the next SaaS, well, probably SaaS is Dynos with speed, but whatever, AI software of some sort, AI application of some sort, because how you do it, right? You get six graduates from Carnegie Mellon, you give them Red Bull for the next nine months, and maybe something great comes out of it. But there's no investment, right? Because it's just humans writing codes.

Here it's like, oh my God, just a mask set for a chip right now for a very advanced technology set is $500 million check. And you don't even know if it's going to work. You may have to respin the chip. It's another $500 million check.

This is not exactly a Silicon Valley kind of game, to be honest. And going back to the talent, it's where they've made their money over the last however many years. The knowledge base isn't there. VCs used to fund hard tech. There are certainly funds that specialize in doing hard, deep tech sort of stuff, but

They're not the big name funds that even have the ability to raise the money to write the big checks for the most part. Exactly. And by the way, Peter Thiel, I was listening to Peter Thiel was making a presentation at some conference somewhere with all the other luminaries. And he was very blunt about it. He said, think about it. Since the late 90s, how many people actually spend time on semiconductors? Not many.

The vast majority of the know-how or like the venture capital community brainpower, 99% of that since the late 90s went into internet and software. And right now it doesn't exist. The talent doesn't exist. So that's awesome from an investment perspective over there.

Yeah. And to the AI comment, there's a joke that I've been making that's, you know, they can't even get autocorrect right. Like if you think Apple is going to win on AI, if you think Apple is going to win on AI, you're ducking crazy. It's yes. I would totally agree with that. Yes. It's I hope Apple actually figures it out. We'll see. We'll see how it all goes. But I mean, from all the huge ones, like the trillion dollar market cap companies, Apple is the one that really hasn't invested meaningfully in it.

And it's way behind. Now, I want to shift a little bit to Analog Century and how this all flows through into your investment process.

So what is the research looking like that you are doing to keep track of all of these trends, both the cyclicality of analog semiconductors and these sorts of secular trends? It's a good question, right? Because I was talking about the universe size and say it's like over 500 companies of over a billion market cap, right? So how do you go from 500 companies down to a portfolio, say 60? Because our portfolio is usually about 60 companies, like 25 for the long, 35 for the short.

So it's not extremely concentrated, but it's not an index, right? So how do you get that? So we have a couple of tools for that. First, we have a framework that we use to categorize what we feel like is a good short versus a good long. So we have a four-point framework that we use. And here are the four points. And I'll elaborate on the long side and then the shortest, the reverse of that.

So, point number one of this framework is we want to identify companies which are exposed to inflicting growth.

AI, obviously. There are other things. I was talking about electric vehicle counting graph. So it's not just AI. There's more than that. But there's certain areas that are inflecting in growth. So we want to be exposed to companies on the long side which are exposed to that inflection. On the short side is the reverse of that, right? We want to be more exposed to mature stuff like PCs, smartphones, panels, whatever. Stuff like that. That's one. Second is we want to identify the winners into this inflecting growth. And the

easy to say, right? Tricky to do, to figure out what's the real winner. But we want to see the companies with large market share, which actually have a chance of keeping it. Because if you have large market share and you can keep it, then you have high pricing power. You can have high gross margins or increasing gross margins and increasing operating margins because of that. Number three is we look for strong balance sheets on the long side. In our case,

The vast majority of our loans have huge balance. They're sitting on cash. They're spilling out free cash flow. They have dividends, buybacks, you name it. So there's beautiful companies from that perspective. And number four is on multiples. So we're not slaves to cheap stocks, but we just want the multiples to look acceptable, palatable. If we look two or three years out and the earnings increase has been such that the three-year-old multiple looks pretty darn low.

Now, during the post-DeepSeq thing, everything was looking very low, right? I mean, video was showing it like 18 times for multiples because everybody felt that, oh my God, the EPS is going to vaporize, which of course it didn't. So that's how we think about it. So that's our framework for longs versus shorts. Then you mentioned cyclicality, right? And how do we deal with that? And I mentioned also it's important to be cognizant at the very least and be aware of what's going on from a cyclical perspective.

So we actually have created, we have a database of over 500 companies, sorry, over 400 companies over 20 years, where we have accumulated all kinds of metrics like inventories, inventories of distributors, inventories in the supply chain above them, below them, bookings, backlogs, lead times, utilizations. A lot of that stuff is very anecdotal.

And it has come from, you know, some of that's Bloomberg, a bit of it is Bloomberg. A lot of that's coming from us listening to earnings calls for like 20 years. I mean, I've been doing that for a number of years. And I started developing a concept almost in McKinsey because that's where the sort of understanding of supply chain came through. And then I built it out and it was built a lot more, of course, through the existence of the fund to become a robust framework of actually understanding where we're in these cycles. So that's how we do it.

Are you letting AI listen to the earnings calls now? Yeah. We are testing out some things. We're not quite there at this point of time. The AI is still us so far. Okay. All right. That's funny. I mean, Bloomberg is actually embedding AI in their transcripts. So at this point of time, when you look at Bloomberg transcripts, there's an AI component to them on the site, which actually started to do a pretty decent job in terms of narrowing down quickly on topics you want to care about.

Now, what about the dispersion of outcomes within the sector? Obviously, when you have a huge secular trend, they're going to be big winners, but it might be hard to find shorts that are going to be producing negative returns. You might be able to get outperformance and alpha within the sector, but what about finding companies that are actually going down that are exposed to these positive tailwinds?

In terms of companies going down, there's a set of companies which are just not exposed to these trends, right? Which are just missing it out. They have nothing to do with exploring trends. They're stuck in PC land and smartphone land. There's a lot of them, right? That's one set of companies. Now, I think you're asking the question more like within the AI thing, right? So even within the AI trends, we did talk about that there's companies that pretend to be AI, like the edge AI companies, or like

Some system makers or storage companies that pretend to be AI companies, but they're just not because they actually get hurt. Because if the budget of the data center goes very much to true AI, then the general purpose compute gets hurt because these budgets are not infinite, right? So they're relatively short from that perspective also.

And then also, even within the accelerators or CPUs, which should be benefiting, not everybody's winning, right? I mean, not everybody can be pacing up at the pace at which NVIDIA is pacing up, for example. I mean, Intel has been known to be the

the third horse in the two-horse race at the end of the day because they just never got there. Yeah, then they're years behind in that situation. So that's how we do it. And there's always, by the way, there's always technology transitions where somebody can just fall behind on technology. Like that was happening in power management for instance. There was a power management company that was very much levered to NVIDIA data center chips and

And then they screwed up in the technology transition and a couple of their competitors picked up their share very quickly. So that happens also. Now, what about portfolio construction? I know you have the two core strategies, the long short, sort of like a 40, 60 net, 50 net, and then the market neutral. Are all the same names making their way into the book? Yeah, basically all the same. I mean, it's like 97% of the same names. So it's the same names. The difference between the strategies is

So the strategies are similar in most ways. They're similar from a pipeline perspective, research perspective, team, PM, everybody's the same. The vast majority of the stocks are the same. The difference is more in the portfolio construction. So yes, the longs are the same. So we don't have longs in one strategy, shorts in another, which are the same names. That's not the case. Longs are longs.

But the sizing will be different. And especially, so the longs actually, even on the longs, the bigger longs on one strategy are the same as the bigger longs in the other strategy. In the smaller ones, small and involves strategies. But again, the relative sizing can be different. And also in the shorts, the sizing of the shorts can be very different. I mean, in one strategy, we can keep a 3% short. The other one is a 1% short.

To a big extent, that's driven by our desire to actually be factor constrained also within the market neutral strategy. So in the market neutral strategy, we want to be at least 70% EDO, exposed. And so all the style factors like momentum, growth, whatever you want to call it, we don't want that to exceed combined 30% of the risk we're taking in our strategy. So that necessitates tweaks from a sizing perspective.

Now, what about the trading aspect of it? Market neutral and specifically the large multi-manager pod shops have been well known for their sort of extreme focus on the print and what is going to happen from quarter to quarter. Is your market neutral approach as focused quarterly? Are you able to take longer term, more secular views or even longer term cyclical views?

We have been able to take longer-term views. So our turnover is for sure lower than the vast majority of the pots into these platforms. And I know that from experience because some of the platforms happen to be our investors. And often enough, we have discussions around turnover levels. So I know it's lower, but that's by design. It's by design because our strategy, especially on the long side of the book,

is to really identify the winners into these big trends and compound alpha. And the reality is that if you look at our 25 longs, I think about 15 of them have been there for a long period of time. So we're not always right. We make mistakes. We churn stuff out. But, you know, if two-thirds or like more than 50% have been there for many years, that tells you that there's a lot of staying power and alpha compounding over a period of time.

This doesn't necessarily mean, by the way, that the size is the same. The sizes vary, and we tweak stuff around earnings. This doesn't mean that we always cut size in half before an earnings call just because we're scared. That's not the case. Sometimes we actually increase it. It depends on our conviction level. But there's tweaks around earnings, but this is not our primary objective. The primary objective is not to make a 10% on an earnings call. The primary objective is to

to really be able to compound and get that staying power. Which, by the way, I think it's probably the biggest differentiator between us and many of the pods out there. It's this ability, this knowledge and understanding that there's a lot of staying power into these companies. And that knowledge or conviction that you're better off holding to them as opposed to using them as training vehicles. The concept of

These being secular investments as opposed to training vehicles, it's a relatively unknown concept in hard tech. And we have had that concept since we actually started the fund. We know what the percentages of appreciation of some of these names have been. I'm sure you have to sell down some of the names too, or else they just would have been too large in the book. Oh, for sure. Yeah, absolutely. For sure. Yes. Definitely. Yes. There were specific names, like you mentioned, which went up 8, 9x.

We have had to sell a lot of that over time just to keep within the position sizes that we have because we have risk parameters. For example, in long-short strategy, we don't exceed more than 7% at market. So some positions can be sold numerous times.

And every single one has been a bad sale, but it's risk parameters. Risk management is important in our business. I know the long short is more long lived than the market neutral. That came later. I think a lot of people generally start out with a little bit more market exposure when they're coming up in the sort of single manager, long short world. But

a lot of money is coming from those big multi-manager platforms. You successfully raised money from a large multi-manager platform. And I think just about everybody would love to do that. It's hugely important for the sustainability of the business. Um,

How did you garner interest without running the market neutral strategy that they're so well known for? How did that work? The market neutral strategy, you're right. We started that about two and a half years ago versus the long short has been around for almost seven years, as we speak. But the market neutral was actually came also from an interest of a

of an investor we have known for a while. So it was an investor we have known for like a decade, basically, that was with us in the prior funds. And that investor, they had changed their business model to really move to the market neutral side of the world. So all the investments were market neutral. They kept on tracking us. They kept on tracking our long-short portfolio. They liked the offer we were generating. So

There were ways for people to actually take our long-short portfolio, do a number of calculations of how much is alpha versus how much is better in the long-short. And they realized that about half of our returns was alpha and the other half is better on the long-short. And the alpha was very meaningful. So they basically wanted to extract that. And they were like, okay, so let's see if we can extract that and we can create a market-neutral portfolio from that. And we actually were running paper portfolios for a while, for over a year.

We'd upload stuff on Bloomberg, see how it behaves. We tweaked the approach many times. At the very beginning, there was some index shorting or just scaling up shorts. It wasn't quite working out as much as we wanted it to work out well. So we went for like a year-plus evolution before we went live. But we did go live with an investor that trusted us, that knew us.

We went live then with another investor that actually had done a ton of cutplaces on our alpha. They had high conviction that our generation is very, very meaningful and much higher than many others they have seen. And that's how it all built up. And the sort of the more the platform types, they saw that interest. They saw our data also and they saw the alpha is there. And for them, it was okay to jump into it.

Now, it sounds like you were pretty free in sharing information on how you guys traded. A lot of people are very close to the vest with their cards, don't want to let their secret sauce out. Do you think that that's the wrong approach and that...

You know, people can't recreate and copy what you do. It might probably be easier for you with such a specific sector focus. People don't need to get down to the individual level position on the visual level traits to get conviction they suffer.

For many allocators, it's enough to get daily data for the long book versus the short book, the exposures of the long and the short side of the market neutral book, the performance of the longs and the shorts in the market neutral book, and then get like...

position level data on a monthly basis to just to see how that thing changes. That's often enough for them to derive the Sharpe ratios and alpha generations to be able to see what's going on. Obviously, you have a tremendous amount of conviction in this AI trend. I'm sure personally, you probably want to take on some beta. Do you lean more in your own allocation towards the long short strategy as compared to the market neutral?

I definitely want to take more better from that trade. The vast, vast majority of my money is in the long shot. I mean, the vast majority of my money was created during the past seven years anyways. Thank God for that existence of this fund.

But I do have a bias towards compounding it more. And I do believe that the better for this stuff will be there. What about demand for the two products? I mean, is there a different investor archetype, a different style of allocator that leans more towards one versus the other? I would say that when I'm thinking of our investors on the market neutral side of the book, which by now has become a much bigger side of our business than the longshore book.

It's really investors that have had long-short portfolios or continue to have long-short portfolios. It's just that their long-short product is not growing. So for them to allocate to a new long-short account, they would effectively have to take money from somebody else and give it to me. It's because there just has been no growth in the long-short side of their businesses. But they have actually been growing their market-neutral businesses. So I guess they as streamers, I guess they...

pension funds or whatever it is, or government funds, sovereign funds. I think these Uber allocators, they're the ones who are actually looking for much more on the market neutral sides as opposed to the traditional long short. So do you think other long short managers need to seriously consider whether, could my strategy be turned into a market neutral strategy if they want to garner greater assets?

Oh, it's up to them. I don't know. I guess I would say for us, it's been definitely a lot more assets have come on the market neutral side. I think there's a much bigger market from a market neutral perspective. It does come in SMAs. It's just a different product per se. I have found personally over time that SMAs

have been just as sticky as Commingled investors. I know there's that perception that SMAs are fickle and Commingled is less fickle. I don't know about that. From personally, I think it's much more individual as opposed to a product-based thing. So I like our SMA investors. I have no problem with SMAs. I'm very happy with them. I think they're fantastic investors. And we have seen that there's a lot more money on the market-neutral side, ultimately.

And I think it's natural to some extent. It's on the market neutral side, you're just paying for alpha, right? I suppose on the long short, you're paying for alpha and for beta. And for many people, it's easier to just pay for alpha. I would love to get paid for both, but...

It seems like you're getting paid for both right now. But I want to close with a question about the team and the history of the fund. You and your partners worked together at a prior fund, launched this together, found incredible continuity of team. How has that helped you both operationally, but then also with allocators who do really value team continuity a lot?

So it's hugely helpful from both perspectives. So many allocators care about team continuity, right? So that's very helpful from that perspective for getting new accounts or new investors. That's true. But also operationally, trusting people to do what they're supposed to do as opposed to micromanaging them, it is very important.

I would have been, especially when the business scales and more and more investors come in, if I didn't trust my chief of operations and my CFO, if I didn't trust this person to handle the increasing number of

accounts, making sure that these increased number of accounts, all the numbers are correct. And if I had to micromanage, check everything, I would go nuts. I mean, there'll be no time for me left to actually do research, right? And ultimately the money is made if I'm focused on the research mostly and also on portfolio construction. That's where my money, that's where my time should be. My time shouldn't be spent on

looking at spreadsheets that goes to the TV guys or to the accountants or any of this stuff. Also helpful that, you know, the analyst that I've been together with, that he's been together with me for like well over a decade, he's grown up into a head of research. He's super helpful to me.

Not just from a research perspective, but we discuss factors and combination of factors in the portfolio. So having the sounding boards on somebody that you've been together for a long period of time and you can trust is super important also.

Val, we'll leave it right there. Thank you so much for coming on. Where can people find you? Are you at all on social media? Are you one of the fund managers who stays away? I have a LinkedIn account. So sometimes I post stuff on my LinkedIn account, but I'm not like some of these guys who have a podcast and every week something new comes out. So a little bit of that. All right. Sounds good. Well, thank you so much, Val. Thank you, Max. Appreciate it.