Beyond the GPU: Navigating Silicon's Next Wave

[00:00:00] Hi, Rob. How are you? Good to see you. You've been busy. You had the Paris ETHE conference. I think you were in Copenhagen, New Hampshire, New York. Yeah I've actually cut back on the conferences a lot, it's very tiring and I think a little bit distracting, you know what, I've been in a reading and researching phase, much like you have, and one of the things I'm really excited about, a thesis that I have been excited about just as an investor for the last five years or so, is energy Computation and connectivity and crypto and what our digital future holds for those three sectors, my background, I'm an oil and gas girl, started on the trading side, went into the upstream exploration production side, midstream, downstream refining, chemicals, I did some mining work materials, physical materials, mining not bitcoin mining, before I got into the crypto side, so it's been really fun to see the re emergence of this narrative around not just the bits and bytes, like the last [00:01:00] decade, has really been about software and venture capital and this idea of Scaling and hyperscaling through digital services and software.

And I feel like the pendulum is shifting a bit. And now we're talking about infrastructure and atoms again, the world of CapEx, which I love. And I've been spending a lot of time just delving into that future. So what's happening on the energy side, obviously a lot of conversation around energy transition.

more efficient utilization of energy resources in different ways, new generation sources, the world of computation. And I think the AI narrative certainly has been a huge catalyst there. But obviously, even just looking at ASICs and Bitcoin mining, the story emerging around specialized hardware, Specialized data centers the economics, both on the CapEx and OpEx side, really fascinating.

And in the world of crypto, I think that's where everyone knows me from. I think [00:02:00] people might have this perception that the majority of my time is spent in the world of crypto, which historically it has been. But again, just understanding that value chain at the end of the day, So many things in our digital world are just so dependent on our ability to secure energy and to secure computation, which is semis and racking data center connectivity to relay that computation.

So I'm excited that we're finally getting to have this conversation because I feel like we're at an inflection point. To that very long response, less traveling, more focusing, more reading, more writing, more just starting to meet with people who are operating in these different verticals because those worlds are starting to converge and you're seeing those headlines emerging in real time.

So it's very fun for me because I feel like this little it's, niche that I've been excited about is finally entering sort of, the awareness the global awareness, if you will, the zeitgeist on [00:03:00] tech Twitter and so it's been lovely because there's so many intelligent conversations happening in so many different forums and it's just, it's been really great to see.

I agree. I agree. You and I had a fantastic lunch a few months ago about semiconductors, energy, and AI. And I realize that we're tracking similar themes, and we're pleased to have Dylan Patel join us, who is the author, and he's got a team, he'll introduce himself in a moment, of Semi Analysis, And I've been reading 500 pages of semi analysis.

I'm just putting it on screen here, just so you all can diligence that. These are the semi analysis newsletters that have been compiled in my sub stack Google Docs. And I've actually been also working with AI to identify the key themes here. But to your point, there are a lot of key trends at work.

One is, Energy. We're excited about energy. We've talked about that before in multiple different ways, including the nuclear renaissance. [00:04:00] You saw Microsoft investing in nuclear technology to generate electricity for their data centers. Their data centers have a growing demand for electricity and also GPU chips and bring us to semiconductors.

And I believe in the next 10 years, we're going to see a crop of opportunities in the semiconductor space. Yeah, and on that, I just want to say nuclear isn't the only generation source. I think more broadly, companies like Crusoe and others have really introduced and helped refine the idea of the economics of co location, particularly for standard energy assets to get to market.

Like at the end of the day And energy is the most relevant form of currency that we have. So the ability to use different types of compute. Okay, there we go. Hey, how are you? My microphone was not connecting properly because I had this up and it was outputting audio through it. So sorry about that.

You're coming through loud and clear. So first off, thanks for [00:05:00] joining us. We're both subscribers and big fans of your newsletter. I was just sharing on screen, I've got a Google Doc that's 500 pages long, and this is what I read before I go to sleep, actually doesn't put me to sleep, but I find it quite interesting and it's been really fantastic content, so if you really want to dig into the semiconductor space, you should subscribe to Semianalysis, it's top flight so there's a lot of content we want to dig in with you Dylan, and, we'll So Meltman and I approached this from an investor perspective.

We see a long term enduring trend around semiconductors and then trends within that. There are a variety of trends depending on where you are in the value chain from round trip investments. We're seeing Nvidia do with inflection to reshoring and semiconductors to the lithography layer. And we're looking for emerging dominant.

monopolists. So there's a lot to discuss. Let's get right into it. Do you want to introduce yourself, Dylan? Sure. I'm Dylan Patel chief analyst of semi analysis. So we're a [00:06:00] semiconductor supply chain and research firm. I'm here in the U. S., got another person in the U. S., and we've got folks in Japan, Singapore, and Taiwan.

And so we mainly analyze the supply chain. I came from a data science background before so still very in tune with the algorithms and such, but there's a lot of various other areas, that we research and discuss around AI, but mostly around semiconductor supply chain.

So that's a quick intro. Great. Let's get into it. So if we look at the value chain for semiconductors, here's my broad summary. You've got the chip designers, the NVIDIA, AMDs of the world, Marvels, perhaps Broadcom. You've got IP providers like SoftBank's ARM, Cadence, Rambus. You've got the EDA software companies that do electronic design automation.

I think of a CAD software. for chip design, the foundry layer, Taiwan Semiconductor, GFS, and obviously there's a lot of CapEx subsidies from government and perhaps Chipsact and other, there's a Eurozone version of that, Japan announced some [00:07:00] subsidy for foundries as well, We'll get into there.

There's a lithography layer with ASML, packaging and testing. Finally, the OEM, cloud and AI providers, the hyperscalers. And we're gonna throw a core weave into that too. So if you look at that value chain, where do you find the most benefits? Opportunity. What layers of them that have the most interest and opportunity?

It's hard to just say one specific spot in the supply chain has the most interest in an opportunity. There's certainly some areas that have, more versus less, right? Like obviously client devices are in a lull right now. With smartphone sales slowing down, PC sales slowing down, but on the flip side of the coin, there's also obviously a lot of tremendous interest for what's going on in the AR VR space but, of course, all of the the attention nowadays is on the supply chain around, whether it be AI or automotive or those sorts of areas that are dynamically changing more.

So there's obviously more interest there, and then there's changes up and downstream, right? From manufacturing to packaging, to IP, to [00:08:00] chips themselves. To the deployment of them for, to support those supply chains. So that's, yeah, there's a lot of opportunities up and down the supply chain based on what's changing in the end market and with society at the, the end goal, end states.

Right on. Product cycles are off, but we have new product cycles emerging to your point AR, VR. Let's double click into chip designers. NVIDIA obviously has captured a lot of focus And they've got emerging competition from AMD and Intel is seeking to do an N around the CUDA ecosystem, which has created a lot of engineer lock in.

And then you've got Broadcom, who's done a number of M& A transactions, cut costs, and they're also on the upswing. You've got Qualcomm as well, that's looking to apply a similar model of licensing their various chip designs. Where do you see NVIDIA's competitive advantage and potentially in decline?

How enduring is that trend [00:09:00] versus new players that are seeking to take share of NVIDIA's dominant market position? Sure. There's quite a few different spots where they are, going to maintain their strength, and then there's other aspects where they won't maintain, when you think about chip design, you might think, oh, Intel, AMD, those are the areas that are going to be more competitive. But in reality the areas where that are where there's going to be more, Competition is actually going to be how most people access AI compute, right?

Which is going to be from the cloud. And so companies like Google's in house chips are just as important, if not more important to watch because they're producing far more, right? And then there's companies like Broadcom that help them design those and folks like that. So those are just as important, if anything, relative to the the main The main markets of, hey, AMD into how are they going to compete now in the area outside of, the cloud silicon, right?

And of course, Amazon has their own silicon and, that's what Alchip and Marvell and, Microsoft has their own silicon that they're working on. And Meta has their own silicon that they're [00:10:00] working on with Broadcom again for Meta. Everyone has their own silicon that they're working on, but outside of these major players that are working on their own silicon, Nvidia will mostly remain the dominant player.

Now that doesn't mean that, hey, this pie is growing so big that AMD or Intel can't grab a slice. Mostly AMD. But that slice is not quite as big as what others the cloud share is, right? So cloud is really where you have to focus on. Yeah. So on that topic, I think one interesting sort of thing to talk about here Dylan disappeared.

Let's see if he comes back. I'm still here, sorry. No worries. So I think one interesting question that Ram and I have talked about a lot, what you're talking about, I think, with the cloud providers moving upstream is this idea of vertical integration, right? And I think we are starting to see the larger players go vertically integrated, where they're producing the silicon, then deploying it in their own data centers, and then offering that to the end client.

I guess, do you think that vertical integration is going to become a more dominant A trend as time goes on, or is it going to be limited to just specific software sectors that are driving that [00:11:00] desire to vertically integrate and control costs at every step of the value chain and really control and off X margin, right?

Yeah, so I think obviously vertical integration is only possible by the biggest tech companies and their share of total purchases is growing. And it has been growing for the decade Amazon versus,

sorry, Amazon versus Google versus the others their share of total compute resources is growing. But that doesn't mean that the clouds are going to own everything, right? Plenty of people don't want to go in the cloud or they want to go in the cloud, but they don't want to do a managed service.

They just simply want to use them as an infrastructure provider. And so the stratification of who gets what compute is very large and wide. And I think that brings up another interesting trend, One of the things I've seen emerging a lot more with enterprises, like historically, the move has been to shift towards cloud.

‍