Latest / In the Loop AI / How Gavin Baker Invests in AI, and Where the Bubble is Going Next

Transcript

How Gavin Baker Invests in AI, and Where the Bubble is Going Next · In the Loop AI

0:00
Ejaaz: Gavin Baker is one of the most prolific AI investors that almost no one has heard of.
0:03
Ejaaz: He spent the last 20 years investing in some of the biggest AI companies before
0:08
Ejaaz: they became household names.
0:09
Ejaaz: He was an early backer of NVIDIA, as well as Cerebris, which IPO'd very recently.
0:14
Ejaaz: And he has a very concrete thesis, which is AI isn't in a bubble.
0:19
Ejaaz: It's quite the contrary.
0:20
Ejaaz: It's in a super cycle. He says that by looking at the watts,
0:24
Ejaaz: wafers and tokens, the infrastructure of AI.
0:27
Ejaaz: He's identified some of the key bottlenecks and constraints,
0:29
Ejaaz: and he has one simple thesis.
0:31
Ejaaz: The biggest returns that you can get in AI is in electricity,
0:35
Ejaaz: power, and silicon fabrication.
0:37
Ejaaz: It's got nothing to do with SaaS software as a service.
0:40
Ejaaz: It's got nothing to do with chatbots, such as Anthropic or OpenAI.
0:43
Ejaaz: It all filters downstream from
0:45
Ejaaz: semiconductors, the picks and shovels that build the entire AI industry.
0:50
Ejaaz: And he's been expressing that interest to the tune of $4.1 billion.
0:54
Ejaaz: While most people are calling the AI industry a complete bubble,
0:57
Ejaaz: he thinks that it is a generational buying opportunity specifically for AI infrastructure,
1:01
Ejaaz: and he makes the thesis very clearly in his fund.
1:04
Josh: And if you hear these constraints that he's talking about, this AI infrastructure,
1:08
Josh: you realize that this kind of sounds pretty familiar.
1:11
Josh: We've heard this thesis before, and that's because on the show,
1:14
Josh: many times we've covered an investor by the name of Leopold Aschenbrenner.
1:17
Josh: He has been around for three years. He just published his most recent 13F,
1:21
Josh: and he has a lot of the same core philosophies.
1:24
Josh: Now, here's the difference. Leopold's been around for, what, three years?
1:28
Josh: Gavin's been doing this for 20-plus years. And when we're comparing these two,
1:31
Josh: Leopold actually has almost three times the assets under management.
1:34
Josh: But there's this great quote that I was reminded of from our producer Luke before
1:37
Josh: the show. It's like, okay, you can beat Warren Buffett over a year,
1:39
Josh: but can you beat him over multiple decades?
1:42
Josh: And Gavin Baker has a track record that proves that he might have a slightly
1:45
Josh: different outlook on this investment thesis.
1:48
Josh: For those who don't know, Gavin Baker, he is the basically founder of this company
1:52
Josh: named a Trades Management.
1:53
Josh: They are an investing fund, and he has spent 20 years investing in NVIDIA.
1:57
Josh: Now, if you've invested in NVIDIA for 20 years, it's a miracle he's still working
2:01
Josh: because those are some pretty incredible returns.
2:03
Josh: Some of the early wins that we've recently seen include companies like Cerebris,
2:07
Josh: which he was a very early investor in,
2:09
Josh: Cerebus, if you'll remember, just IPO'd for an ungodly amount of money.
2:13
Josh: Same with this company, Estera Labs. And there's a lot of other companies that
2:16
Josh: I don't think you've probably heard of before that we're going to cover in this episode.
2:20
Josh: As we walk through his portfolio and his guidance on where he sees the AI industry
2:25
Josh: going in terms of investment and where all the opportunity lies.
2:28
Ejaaz: So then the question becomes, what is he investing in and why is he investing in it?
2:32
Ejaaz: Well, if we look at his recent 13F for a tradies management,
2:35
Ejaaz: which is the name of his fund, They have about $4 billion worth of AUM.
2:38
Ejaaz: We look at some of his biggest positions and unpack what some of these companies actually solve.
2:43
Ejaaz: It kind of points towards the bottleneck that Gavin references in a bunch of
2:47
Ejaaz: his different interviews as to where AI is going.
2:50
Ejaaz: So he has some pretty sizable positions in some companies that a lot of people
2:54
Ejaaz: may not have heard of purely because they're quite unsexy.
2:57
Ejaaz: So he has an almost 9% position, so 10% position of the fund in this thing called
3:02
Ejaaz: Astera Labs. Now, Astro Labs can be described as kind of like the connectivity layer between GPUs.
3:09
Ejaaz: So if you imagine a data center, you have GPUs, they're kind of like the engine that can like pre-train
3:14
Ejaaz: post-train your model and also like inference your model. But in order for these
3:18
Ejaaz: GPUs to work, they need to traverse a bunch of different data.
3:22
Ejaaz: They need to send data between themselves. They also need to access data from
3:26
Ejaaz: all these memory chips that they store this data on.
3:29
Ejaaz: Now, in order to access this, you need some kind of a plumbing system.
3:33
Ejaaz: And I'm being very high level on this because I'm not going to pretend that
3:36
Ejaaz: I know the intricacies of all of this.
3:38
Ejaaz: Astero Labs is a company that essentially fixes that.
3:41
Ejaaz: The problem it solves is as AI clusters scale to hundreds of thousands of chips,
3:45
Ejaaz: The bottleneck stops becoming GPUs specifically, but it starts becoming that
3:49
Ejaaz: transfer window and sending the right data at the right time,
3:52
Ejaaz: accessing the right data at the right time is what the plumbing system that Astro Labs builds.
3:57
Ejaaz: Now, I haven't heard of Astro Labs until we started researching for this episode,
4:01
Ejaaz: but I remember another company being the exact same case called Cerebrus Labs,
4:07
Ejaaz: which is what Gavin was talking about almost like, I think like six months ago, which is quite long,
4:12
Ejaaz: given the relative scale or timeline of AI.
4:15
Ejaaz: And then the next thing I know is that they IPO'd for like, what was it,
4:18
Ejaaz: like $60 billion, and it's just been like, it's up 40% since the IPO.
4:22
Ejaaz: So just kind of point towards these different trends. Asher Labs might be something
4:25
Ejaaz: significant on that horizon.
4:27
Josh: Yeah, Cerebrus is one of his earliest investments. I mean, he was in Cerebrus
4:30
Josh: very early in the company's lifetime, which means he's bet on this thesis for years.
4:34
Josh: There's also a few other ones that he's been betting on for a long time.
4:36
Josh: I mean, NVIDIA being that flagship one, being involved in NVIDIA for 20 plus
4:40
Josh: years is pretty incredible and still having conviction all the way through is impressive.
4:44
Josh: Gavin was recently on two podcasts where I was listening to him speak about his NVIDIA position.
4:48
Josh: And it's very clear that he believes that they're going to be able to maintain
4:50
Josh: these profit margins and maintain the demand as well, which means he's putting
4:55
Josh: NVIDIA on a clear path to getting close to $10 trillion in market cap.
4:59
Josh: It's only halfway there right now.
5:01
Josh: A few other noteworthy mentions are Micron, which we discussed on a previous
5:04
Josh: episode, which I would highly recommend going to watch in terms of the AI investment
5:07
Josh: stack and where all these companies lie.
5:10
Josh: Micron is one of the largest memory makers and a crazy statistic about Micron.
5:14
Josh: A year ago, it was sub $100 billion.
5:16
Josh: And as of recording this, it just eclipsed a trillion dollars in market cap.
5:21
Josh: It got a 10X in a single year. And it's a testament to how important that memory problem is.
5:27
Josh: Now, perhaps some of the less noteworthy companies that are interesting,
5:31
Josh: EGES, there's one I want to highlight for you in particular,
5:33
Josh: because I feel like you're going to like this one, is Unity Software.
5:37
Josh: And for those that don't know, Unity, I mean, I know this very well as a gamer, Unity is a game engine.
5:41
Josh: There's a lot of popular games that are built using this 3D rendering software.
5:45
Josh: So why would someone who's investing in AI be investing in Unity software,
5:49
Josh: the thing that makes my video games? And the answer is the 3D game engine.
5:53
Josh: Unity is a world model builder, and it has a really deep understanding of physics
5:58
Josh: and the way the world works and the understanding of textures and lighting.
6:02
Josh: And when these AI companies are trying to build AGI, they're trying to build humanoid robots.
6:06
Josh: A big part of that is simulating virtual environments and virtual datasets that
6:10
Josh: allow these robots to be trained in.
6:12
Josh: And Unity just still happens to be one of the best ones. So I feel like that
6:15
Josh: one I want to highlight specifically for you, EHS.
6:17
Josh: As the world model Maxi, there is a clear path in which a gaming company that
6:22
Josh: is known for a scanning engine becomes a pretty serious player in the world of AI.
6:25
Ejaaz: Yeah, the whole thesis behind world models is pretty simple.
6:29
Ejaaz: It's AI models or LLMs currently understand the world through text, through books.
6:34
Ejaaz: It's kind of like a student sitting in a library, but it doesn't actually have
6:38
Ejaaz: experience of the real world.
6:39
Ejaaz: World models basically unlock that. It's like putting a game character into
6:43
Ejaaz: a simulated environment and understanding the physical reality of how the world works.
6:47
Ejaaz: If it works, if I drop this phone, if I kick a ball, like what happens?
6:50
Ejaaz: What are the next consequential steps? What do you do?
6:53
Ejaaz: World models effectively fix that. And there are very few players that have
6:57
Ejaaz: like built this out at scale. I think currently the leader is probably Google with Genie 3.
7:01
Ejaaz: They actually released a new model called Gemini Omni recently,
7:04
Ejaaz: which kind of like does this at scale, but it's not quite like where it's meant
7:08
Ejaaz: to be. It hasn't quite had its ChatGPT moment.
7:10
Ejaaz: What I like about Gavin in particular is he kind of has this barbell.
7:14
Ejaaz: I don't know if you noticed this, Josh, where he's kind of old school and he's like,
7:18
Ejaaz: people are going to need GPUs. People are going to need memory.
7:21
Ejaaz: I'm going to invest in the biggest players, Micron and NVIDIA.
7:24
Ejaaz: But then he has this kind of forward-looking thing where he's like,
7:27
Ejaaz: I think that's where the puck is going to go.
7:30
Ejaaz: And so I think, let me invest in Cerebrus because I think inference is going to be super important.
7:35
Ejaaz: And then let me invest in Unity because I think world models are going to be
7:37
Ejaaz: the future of how we train robots and future LLMs.
7:41
Ejaaz: So he has this kind of like barbell approach. Now, one thing that I see in his
7:45
Ejaaz: portfolio over here as well is this, well, there's two companies, actually.
7:48
Ejaaz: It's this company called Positron, which kind of creates inference chips.
7:52
Ejaaz: Now, if that sounds similar, it's very similar to Cerebrus. That's exactly what they do.
7:56
Ejaaz: And it's around this entire thesis, which Gavin has spoken about on his recent
8:00
Ejaaz: interviews, which is the infrastructure stack, specifically the training stack
8:05
Ejaaz: for AI models, is moving from pre-training to something more focused on post-training.
8:10
Ejaaz: Now, if you've been involved in the AI sphere, generally, you kind of had an
8:14
Ejaaz: idea that this shift was happening. But Gavin is all in on this thesis.
8:18
Ejaaz: And if you have a model, it still needs to understand new information that comes in, new data.
8:24
Ejaaz: It needs to update it. Just because you pre-train it on a specific data set
8:27
Ejaaz: doesn't mean it's going to be a genius for the rest of its life.
8:29
Ejaaz: It still needs to learn new information.
8:31
Ejaaz: That happens in the post-training layer, and it requires a lot of inference.
8:35
Ejaaz: Secondly, if you need the AI model to actually think about the problem more,
8:39
Ejaaz: in the same way that we take in new information, we're like,
8:42
Ejaaz: hmm, I wonder if this angle makes sense or if another thesis maybe applies.
8:46
Ejaaz: That's known as reasoning. You need a lot of inference. Now,
8:48
Ejaaz: the estimates are the cost or the revenue opportunity from inference alone is
8:53
Ejaaz: worth around 5 to 10x more than the amount of compute that is being put into pre-training.
8:58
Ejaaz: So, AI labs and chip makers are suddenly making this big shift.
9:02
Ejaaz: Like you've seen NVIDIA create a bunch of different GPUs that are aligned to
9:07
Ejaaz: inference to allow agentic kind of exposure.
9:10
Ejaaz: And so we're seeing Gavin express this through his different investments on inference alone.
9:13
Ejaaz: And the final point that I'll make is Gavin made a really cool point,
9:17
Ejaaz: and we're talking about this before we started recording, Josh, on China specifically.
9:21
Ejaaz: It's been very much like China versus the US when it comes to the AI race.
9:25
Ejaaz: China has a very unique kind of setup where they have an abundance of energy
9:29
Ejaaz: and the ability to scale chip manufacturers. That's something that the U.S.
9:33
Ejaaz: Currently struggles with. That's why they outsource a ton of stuff to the likes of TSMC on Taiwan.
9:38
Ejaaz: And what he basically explains is that China has a unique opportunity to create
9:42
Ejaaz: infrastructure or chips specifically that are going to look very different to what the U.S.
9:46
Ejaaz: Is creating because they're focused so much on inference.
9:49
Ejaaz: So Gavin, you could say, is leading the charge in the U.S. through his investments
9:53
Ejaaz: on building or taking a bet on the U.S.
9:56
Ejaaz: Infrastructure setup for inference. And I think that could be a huge opportunity in the future.
10:00
Josh: And it's worth noting that this bet also isn't only for the upside.
10:03
Josh: There is a large put position here in an ETF named QQQ.
10:07
Josh: Now, for those who are not familiar, it covers the NASDAQ 100. It's a basket of stocks.
10:11
Josh: It's the second most traded ETF in the United States. And it's been performing incredibly.
10:17
Josh: In 2023, it was up 55%. 2024, 25%. 2025, 20%. And so far in 2026,
10:24
Josh: already up 17%. So QQQ, as an index fund, has been doing incredibly well.
10:30
Josh: It's easy. It's a basket of the top 100 stocks. Gavin is saying,
10:33
Josh: I'm shorting against that. I think that's going to go down. I think you're wrong.
10:37
Josh: And what this tells me is that he believes strongly in the AI play in the sense
10:42
Josh: that he's going to invest in the key makers who are solving these bottlenecks.
10:45
Josh: But as a market-wide general sentiment, he doesn't appear to be very bullish.
10:49
Josh: And this is a hedge against that downside protection, where if the market does
10:54
Josh: start to fall apart in ways that are less favorable, even though AI still wins,
10:58
Josh: he has that hedge with QQQ.
11:01
Josh: We can kind of break this down into a few bottlenecks that he believes are going
11:05
Josh: to be most important when it comes to investing.
11:08
Josh: These key things that he is looking at in terms of what the world of AI is going
11:12
Josh: to need as we progress forward. What are the actual constraints?
11:15
Josh: What do they look like? And then how do you invest? How do you convert these
11:17
Josh: into actual dollars that you could put into companies to earn you money?
11:21
Josh: And there's four of them.
11:22
Josh: The first one is verticalized small language models.
11:25
Ejaaz: If you think about LLMs in general, like the chatbots that you talk to,
11:30
Ejaaz: such as Claude and ChatGPT, they're generalized LLM.
11:33
Ejaaz: So they have a wide understanding of the world in context, and they'll be able
11:37
Ejaaz: to answer like specific questions.
11:38
Ejaaz: But it's another thing, training a model around a specific vertical or specific
11:43
Ejaaz: problem that you're trying to solve. Where do these specific problems exist?
11:46
Ejaaz: Well, in enterprises that are deep on solving a particular problem or companies
11:50
Ejaaz: that have made or formed a niche in their particular subsector.
11:54
Ejaaz: Now, verticalized small language models basically address exactly this.
11:58
Ejaaz: They're frontier models, but highly optimized towards running efficiently on
12:01
Ejaaz: specific enterprise data or locally on device.
12:04
Ejaaz: Now, we have spoken about on device or locally run models before,
12:07
Ejaaz: purely for the case that there's a bunch of data in your phone or the devices
12:11
Ejaaz: that you use that are highly personable to you, but you don't necessarily want to give up.
12:16
Ejaaz: And companies don't necessarily have access to that. For example,
12:20
Ejaaz: medical records, financial details.
12:22
Ejaaz: I saw OpenAI release a financial AI agent that can get access to your bank account,
12:26
Ejaaz: But it can't actually act on it because there's a lot of personal identifiable
12:30
Ejaaz: information that you don't want to share, such as your social security number, banking details, etc.
12:34
Ejaaz: Now, locally run models or these SLMs can solve that kind of a problem.
12:38
Ejaaz: And Gavin is making a huge bet that these are going to become huge in the future.
12:42
Ejaaz: One company that I've noted that Gavin is hugely bullish on is Apple.
12:46
Ejaaz: Although he doesn't express an investment interest, he knows or thinks that
12:50
Ejaaz: Apple is going to be the device maker, one of the major device makers,
12:54
Ejaaz: Allows for these locally run models to run on their devices.
12:57
Ejaaz: Now, in a world where that is the future, you can start thinking of a world
13:01
Ejaaz: where maybe Claw doesn't need to be the model that you need to interface with every day.
13:05
Ejaaz: Maybe you need a personalized AI agent trained on your own data,
13:08
Ejaaz: and that's what these SLMs end up eventually becoming.
13:11
Ejaaz: Now, that's the generalized version of it where you can run it on your own phone,
13:15
Ejaaz: but a bunch of enterprises will run these highly optimized and specialized models
13:19
Ejaaz: to train on their proprietary data, which ends up helping them sell a product
13:23
Ejaaz: or market a product much better.
13:25
Josh: Oh man, Apple's in such a good position there. So good. I can't wait for WWDC. It's coming.
13:29
Josh: We are just a few weeks away from Apple's developer conference where they're
13:32
Josh: going to unveil all of this new AI software that's coming and what that looks
13:36
Josh: like integrating the hardware.
13:37
Josh: That's going to be huge. We will be covering that. I'm so excited to talk about
13:40
Josh: that. In terms of the next pillar of this, it is sovereign infrastructure.
13:44
Josh: We always talk about this, that the speed of bits is so much faster than the speed of atoms.
13:49
Josh: I mean, When you think about AI infrastructure, the quality of models has gone purely exponential.
13:54
Josh: The amount of intelligence we could generate per watt, the amount of intelligence
13:57
Josh: per token, is up into the right only.
13:59
Josh: What isn't up into the right at nearly the same rate is the speed of physical
14:03
Josh: deployment, because that itself is the mode.
14:05
Josh: It's very difficult to take hardware that is incredibly complicated.
14:09
Josh: We're talking about transistors that are down to the atomic level of precision
14:12
Josh: and deploying them at scale in a world in which our infrastructure is already
14:16
Josh: suffering. I mean, with the acceleration of electric cars, the grids have already
14:20
Josh: been feeling a little bit more restrained. They're kind of at max capacity.
14:23
Josh: Now they have the energy problem. Now they have the chip problem.
14:26
Josh: Gavin is very strongly betting on the fact that infrastructure is hard.
14:30
Josh: It's going to take many, many days to months to years to do.
14:34
Josh: And he's betting on the people that can compress that into weeks.
14:36
Josh: So the speed of the physical deployment itself is the moat.
14:39
Josh: He's kind of narrowing in on who the companies are that are able to deploy as
14:43
Josh: fast as possible. When I think about this, my first thought is SpaceX.
14:46
Josh: And how quickly they've been able to build Colossus and then rent that out to
14:49
Josh: Anthropic and I'm sure companies in the future.
14:52
Josh: But that infrastructure pillar is one of the key ones that he's looking at.
14:56
Josh: I think everyone, we looked at.
14:58
Josh: Leopold's portfolio as well, that was a core component of that.
15:01
Josh: It's just, it's really hard to build things. And whoever can build things,
15:04
Josh: they can sell it for a lot of money.
15:05
Josh: SpaceX, their largest line item now in terms of revenue is the data center that
15:09
Josh: they're renting out. It has nothing to do with rocket ships.
15:11
Josh: And I think that's a testament to how important this pillar is.
15:13
Ejaaz: So it's speed that he cares about and he thinks that it's important,
15:16
Ejaaz: but it's also like the cost, right?
15:19
Ejaaz: He keeps referencing this metric, which is performance per watt.
15:24
Josh: Perf per watt.
15:25
Ejaaz: Yeah, perf per watt. And what he's talking about here is companies are increasingly,
15:30
Ejaaz: companies being AI labs, are increasingly caring about how many tokens can you
15:34
Ejaaz: generate per watt, right?
15:36
Ejaaz: Because if you think about spending billions and trillions of dollars, which is what...
15:41
Ejaaz: Five companies are currently spending this year alone on GPUs and compute to
15:46
Ejaaz: kind of like power these things or electricity to power these things,
15:49
Ejaaz: You want to get a lot of bang for buck, especially if you're scaling to the
15:53
Ejaaz: size that most of these hyperscalers are doing.
15:55
Ejaaz: So if you think about it, if I prompt Claude and if I prompt ChatGPT and Claude
16:01
Ejaaz: gives me an answer that cost me two cents and ChatGPT gives me an answer that cost me $1,
16:06
Ejaaz: I'm probably going to end up using Claude, even if it's like hypothetically
16:11
Ejaaz: like say 95% of the intelligence hypothetically that ChatGPT has.
16:17
Ejaaz: Because the point is you can prompt
16:18
Ejaaz: it even more and eventually get to the answer for a lot less of cost.
16:21
Ejaaz: So cost becomes a huge thing. This week alone, Microsoft and Uber announced
16:24
Ejaaz: that they are effectively reducing their exposure to clawed code because their
16:28
Ejaaz: annual budgets kind of like got sequestered in about four months.
16:32
Ejaaz: So the point is like the cost of getting access to this intelligence matters a lot.
16:37
Ejaaz: And you see this across Gavin's investment portfolio with Cerebras,
16:41
Ejaaz: with Alexa, Positron, and with Astro Labs, which basically, like,
16:44
Ejaaz: what I've noticed is he identifies these really niche infrastructure bottlenecks,
16:50
Ejaaz: and he basically makes one simple bet.
16:51
Ejaaz: He's like, yeah, if this company solves that, the performance per watt gets
16:56
Ejaaz: to this specific level, which means that AI Labs are probably going to buy more
16:59
Ejaaz: of these GPUs or more of these companies or more of these things,
17:02
Ejaaz: and you end up with a bottleneck being resolved by one of these different companies that he's betting on.
17:06
Ejaaz: So his thesis actually is quite simple, although it's quite complex,
17:10
Ejaaz: which is I'm just going to focus on the AI bottlenecks at the infrastructure level.
17:14
Ejaaz: And if I can identify a company that can increase performance for what by this
17:17
Ejaaz: amount, that can make tokens this amount cheaper, then my bet is those companies
17:21
Ejaaz: are going to be valuable and will either IPO or get acquired for a large sum of money.
17:25
Josh: And the thinkers to know for the section, for those looking to copy trade,
17:28
Josh: Gavin, Estera Labs, we have Cerebrus, we have Sci-5, and we have Positron.
17:33
Josh: Those are the four companies that are really critical in this sector.
17:36
Josh: Now, the fourth and final is a combination of energy and space.
17:40
Josh: Because, I mean, like we talked about in the previous point,
17:42
Josh: the terrestrial grid very much limits energy.
17:45
Josh: And it's very difficult to earn new energy. I think a statistic like 40% of
17:50
Josh: new data centers have very strong petitions against them, or lobbying against
17:54
Josh: them, people protesting they do not want these data centers.
17:57
Josh: There's a lot of resistance to them. And the way that we solve this is twofold.
18:02
Josh: One is creating energy out of the box. It's kind of portable energy you could
18:06
Josh: bring to these data centers, power them with a smaller phone.
18:08
Josh: That's companies like Bloom Energy, who Leopold is very bullish on.
18:11
Josh: But then there's also the orbital compute part of this, which is what Gavin
18:14
Josh: has really shifted his focus to.
18:16
Josh: Now, the first largest company in this sector, of course, is SpaceX.
18:20
Josh: They're the only one who is capable of being the highway to space,
18:23
Josh: to delivering actual payload to orbit, to getting maths and data centers in
18:28
Josh: low-end orbit that can generate enough intelligence and power that can funnel this.
18:32
Josh: I think the space stack is a little bit bigger than just SpaceX.
18:35
Josh: I was surprised to see there wasn't more allocation of space stocks in his portfolio,
18:40
Josh: given the fact that he believes this is such a huge industry.
18:43
Josh: Perhaps the reality is that it's just too early and that SpaceX is the linchpin
18:47
Josh: to unlocking this industry and just kind of closely evaluating Starship V3 launches.
18:52
Josh: We had a Starship launch last week. It performed very well.
18:55
Josh: Without Starship functioning, we don't get energy in space. We don't get racks
18:59
Josh: to orbit. It is required because the amount of payload is so large.
19:02
Josh: So I'm sure SpaceX is the one to watch. There's a lot of other second order
19:06
Josh: companies that can be impacted by this.
19:08
Josh: I think we want to get to the end of this with the question that a lot of people
19:12
Josh: are going to be asking, which is why is this not just the dot-com bubble again?
19:16
Josh: And Gavin was asked this question many times. He has some pretty strong answers
19:19
Josh: and I kind of believe him. It seemed his case that he made was pretty convincing.
19:25
Ejaaz: Okay, so the way I think of his case is as follows.
19:29
Ejaaz: In the dot-com bubble in the 2000s, it was debt-fueled.
19:34
Ejaaz: You had people borrowing a hell of a lot of money for unproven theses and products
19:40
Ejaaz: that people didn't actually care about or use.
19:42
Ejaaz: Now, if you compare this to the current AI super cycle, as he describes it,
19:47
Ejaaz: just from OpenAI and Anthropic alone, they're on track to reach $200 billion
19:50
Ejaaz: of ARR this year, just two companies.
19:54
Ejaaz: And this isn't made up money. This is money that they've signed through contracts
19:58
Ejaaz: that are already prepaid in large part, I think 40 to 60% from a bunch of enterprise
20:02
Ejaaz: and retail customers that are funding this. So this is real money exchanging hands.
20:07
Ejaaz: Now, if you look at the GPU computer, so let's not look at the model labs,
20:11
Ejaaz: let's look at the infrastructure, the people who are buying the goods from NVIDIA.
20:15
Ejaaz: Google, Microsoft, Amazon, and Meta are all paying from their own cash reserves.
20:21
Ejaaz: So they also haven't borrowed any money at all.
20:24
Ejaaz: They're just spending free cash flow on this.
20:26
Ejaaz: Amazon just came to the end of their free cash flow. Now, if they start borrowing
20:28
Ejaaz: money, then we can start to get worried. But the point is, they are not levered up either.
20:32
Ejaaz: Also, this is like five of the top companies in the entire world who are arguably
20:39
Ejaaz: like some of the smartest companies because they are where they are in terms
20:41
Ejaaz: of like market cap, value and size.
20:43
Ejaaz: So the fact is like back to the dotcom bubble, you had a bunch of like no-name
20:48
Ejaaz: companies that had raised a ton of money that was spending money in ways that
20:50
Ejaaz: didn't really make sense.
20:52
Ejaaz: In this cycle, you have like some of the smartest companies in the world spending
20:56
Ejaaz: money that isn't levered at all, that isn't kind of like compressed into like
20:59
Ejaaz: various different margins.
21:01
Ejaaz: And all the quarterly reports that we've spoken about on previous episodes very
21:04
Ejaaz: recently over the last couple of weeks have proven that profit is optimizing
21:07
Ejaaz: through a bunch of these different movements that models are progressing,
21:10
Ejaaz: they're getting more intelligent.
21:11
Ejaaz: So the single argument that Gavin has is, this isn't a dot com bubble,
21:15
Ejaaz: because we're not levered on money, but also because the bottlenecks themselves
21:19
Ejaaz: that we're speaking about is constrained by physical atoms.
21:22
Ejaaz: It's one thing being like, OK, I'm going to buy a bunch of memory chips and
21:25
Ejaaz: I'm going to buy a bunch of GPUs.
21:27
Ejaaz: But it's not like NVIDIA can oversell GPUs. It's not like Micron can oversell
21:31
Ejaaz: AI memory chips because they just don't have the chip production facilities to be able to do that.
21:36
Ejaaz: And so his simple argument is it's not a bubble if you can't oversupply the entire market.
21:40
Ejaaz: We are constrained by the fact that we don't have enough picks and shovels to
21:43
Ejaaz: do the thing. And that's what he's investing in.
21:45
Josh: And there's one great point. If you scroll down just a hair on this artifact,
21:48
Josh: you can see the $2 to $3 trillion number.
21:51
Josh: This is something that Gavin believes is a reality, that NVIDIA could sell $2
21:54
Josh: to $3 trillion of GPUs this year and next year if only TSMC could supply them.
22:01
Josh: So he's saying that TSMC is actually one of the major linchpins of bubble territory.
22:07
Josh: And I'll explain why for a second, because I found this really interesting.
22:10
Josh: If TSMC were able to supply the amount of demand that is required from these
22:15
Josh: companies to actually provide them with that many chips, it will cost them a
22:19
Josh: tremendous amount of money.
22:20
Josh: And in fact, EJS, if you scroll up just a little bit more, you could see the
22:23
Josh: earned income to the CapEx.
22:26
Josh: There's like this kind of bar chart there. Yeah, right here,
22:28
Josh: where you could see AI CapEx is not much different than the operating cash.
22:32
Josh: And so far, companies are generating enough cash to fund the build out of these things.
22:36
Josh: In the case that TSMC came to NVIDIA tomorrow and said, actually,
22:39
Josh: we could triple our capacity overnight.
22:41
Josh: NVIDIA wouldn't say no. And they would start spending an unbelievable amount
22:45
Josh: of money to buy those chips.
22:46
Josh: Other companies would then have to borrow money to fund the purchase of all these chips.
22:51
Josh: And therefore, we would start to see that CapEx bubble really start to grow
22:55
Josh: and separate from the operating cash of these companies.
22:57
Josh: But because there are these supply constraints across the board,
23:00
Josh: we have them in memory, we have them in chip making, we have them in energy,
23:03
Josh: particularly as it relates to TSMC with the chips, we're not actually able to build out that fast.
23:09
Josh: And therefore, TSMC is creating a blocker on the rate of acceleration of this bubble.
23:14
Josh: And so long as TSMC stays limited and constrained to the amount of chips that
23:18
Josh: they could produce, and so long as companies like Samsung and other chip makers don't.
23:23
Josh: Actually overtake that market share, which seems improbable because it's going
23:26
Josh: to take an incredibly long time and it's very difficult to do,
23:29
Josh: then we're at a pretty sustainable rate of growth where it feels fast,
23:32
Josh: but there's still this overwhelming amount of demand that can't be satiated
23:36
Josh: because we just can't build it out fast enough.
23:38
Josh: And so long as that dynamic stays intact, I think we're probably good for now.
23:42
Ejaaz: Well, there's also this other thing, right? Because like you could assume that
23:45
Ejaaz: like demand stays static, but that also doesn't happen.
23:48
Ejaaz: You have an exponentially increasing demand side for all this AI stuff,
23:52
Ejaaz: which is outpacing the production supply that we have for all of these different chips.
23:58
Ejaaz: So the only way that I see this thesis kind of being unproven is if somehow
24:04
Ejaaz: someone miraculously recreates ASML and we suddenly have ASML competitors all around.
24:11
Ejaaz: By the way, those of you who don't know, ASML produces these $400 million machines,
24:15
Ejaaz: which basically tsmc and every major chip fab
24:18
Ejaaz: manufacturer needs and asml is only
24:21
Ejaaz: one team in i believe norway that creates
24:24
Ejaaz: these things and it takes ages and they're backlogged for like five
24:27
Ejaaz: years right now or if we recreate a different type of llm that doesn't require
24:33
Ejaaz: as many gpus or doesn't require as much memory but we're just simply not seeing
24:36
Ejaaz: that i saw a story break today about sk hynix which is the number one memory
24:42
Ejaaz: manufacturer and supplier for nvidia gpus they are basically the top dog when it comes to AI memory.
24:47
Ejaaz: And they're currently courting, I think it's like $50 billion,
24:51
Ejaaz: $50 to $100 billion worth of offers from the likes of Google and Microsoft.
24:56
Ejaaz: So two companies alone, just to pay for the equipment that they need to scale
25:01
Ejaaz: up future supply over the next three years to lock in supply that they're going
25:06
Ejaaz: to create three years from now.
25:07
Ejaaz: This is how desperate some of these big companies are for memory.
25:11
Ejaaz: This is one subsector of the AI component.
25:14
Ejaaz: And SK Hanox is instead just saying, no, I don't want to give you guarantees of our supply.
25:19
Ejaaz: Instead, we'll just hike up the price. They have like 70% operating margin,
25:23
Ejaaz: by the way, which is just unheard of in semiconductor companies at all.
25:26
Ejaaz: So it makes complete sense why Gavin is just going all in, because it doesn't look like a bubble.
25:31
Ejaaz: It might seem like a bubble. The market might react to it. We opened our stocks
25:34
Ejaaz: portfolio today before we started recording and everything was like down.
25:38
Ejaaz: But it's just reactionary because the directional goal of all of this is that
25:41
Ejaaz: we're just going to need more GPUs. going to need more semiconductor chips and
25:45
Ejaaz: we don't have enough supply we don't have enough manufacturers for this
25:49
Josh: So in conclusion, watts and wafers, that's it. Those are the two brick walls.
25:53
Josh: Those are the two limiting factors, the limiting constraints that are going
25:57
Josh: to prevent us from accelerating too fast.
26:00
Josh: And so long as those watts and wafers stay valuable, stay in high demand and
26:03
Josh: stay limited in their supply, there will be good times ahead.
26:07
Josh: Now, if you are looking for a TLDR on Gavin's portfolio, I will read out some
26:11
Josh: of the largest holdings for you.
26:12
Josh: That way you can just, again, not financial advice. This is what Gavin's in.
26:15
Josh: This is not what we're in.
26:16
Josh: I have no idea if any of these are gonna go up, down or in circles.
26:19
Josh: But his largest position, ironically, is that QQQ put position.
26:22
Josh: He's generally speaking bearish on the market, which I think is very noteworthy to talk about.
26:26
Josh: Second to that is Astera Labs, 7.4%. Alab is the ticker. Third is Unity, the 3D software.
26:33
Josh: Then there's a whole bunch of others. Sienna, Micron, NVIDIA,
26:37
Josh: Amazon, Lumentum, Alphabet, Coherent, Roblox, EchoStar, Twilio, Wayfair.
26:42
Josh: Wayfair is a furniture company. This guy's in everything. There's a lot of investments.
26:46
Josh: I think if you're interested, you can take a look at a 13F online.
26:50
Josh: We can link to it in the description. But that is the Gavin thesis,
26:52
Josh: that the bottleneck is Watson wafers.
26:54
Josh: So long as these things stay intact, we are up only. Ejaz, how are you ingesting
26:59
Josh: this information and what are you doing with it?
27:01
Ejaaz: So the market's been pretty rocky since the Leopold 13F.
27:05
Ejaaz: And I'm starting to realize as we're recording this episode that Gavin's kind
27:09
Ejaaz: of like an older, wiser Leopold.
27:11
Ejaaz: He's been around for a while. He may not have $13 billion in AUM,
27:16
Ejaaz: but I have a feeling he's going to be around a decade from now.
27:18
Ejaaz: And so if you're listening to this and you're like, listen, I don't want to
27:21
Ejaaz: keep up to date with AI every single minute of every single hour, every single day.
27:27
Ejaaz: And I want to just kind of park my money and kind of like see how it grows over
27:31
Ejaaz: the next couple of months or years.
27:33
Ejaaz: Gavin's portfolio probably kind of makes a lot of sense. Again,
27:36
Ejaaz: this is not investment advice, but he takes a more cautious,
27:39
Ejaaz: long-term, futuristic approach to a lot of these things. and if his trends indeed
27:42
Ejaaz: end up playing out like he early backed Nvidia and Cerebrus,
27:46
Ejaaz: you could end up having like exponential gains over the next couple of years.
27:50
Ejaaz: But again, it's all based on his one thesis, which is we are not in a bubble.
27:54
Ejaaz: I'm curious whether the listeners of this show agree with this.
27:57
Ejaaz: Obviously, most people aren't as technical or in the weeds as Gavin,
28:00
Ejaaz: but just generally speaking, after you've heard this episode,
28:03
Ejaaz: do you think we're in a bubble?
28:04
Ejaaz: Do you think we're not? What are the arguments for and against?
28:07
Ejaaz: Is there anything that we particularly missed?
28:09
Ejaaz: I don't know, Josh, do you think we're in a bubble as we round up?
28:12
Josh: I think we're certainly in a bubble. Where we are in that bubble is to be debated.
28:15
Josh: It seems like it's in the earlier stages.
28:17
Josh: So hopefully it continues to be that way. According to Gavin,
28:19
Josh: so long as TSMC continues to limit their ability to produce chips, it will be fine.
28:24
Josh: But that's the outlook. We have Leopold now that we've covered,
28:27
Josh: whose success is measured in quarters.
28:30
Josh: We have Gavin, whose success is measured in decades.
28:32
Josh: And perhaps somewhere in the middle is where a lot of people think of themselves.
28:35
Josh: So if you did enjoy, please don't forget to share this with your friend.
28:38
Josh: Let us know which ones you are most. If it's not a thesis, perhaps there's a
28:42
Josh: ticker that we should be looking at. I think it's exciting time because everything's moving quickly.
28:46
Josh: Whether it be up or down, there is a lot of movement. There's a lot of volatility.
28:49
Josh: It's fun to get involved in.
28:50
Ejaaz: See you guys tomorrow.
28:51
Josh: See you tomorrow.