Gavin Baker - AI, Semiconductors, and the Robotic Frontier - [Invest Like the Best, EP.385]

0:00 I know firsthand how complex the tech stack is for asset management firms. And seemingly every new tool and data source makes the problem even worse, adding more complexity, more headcount, and more risk. Ridge line offers a better way forward, one unified platform that automates away the complexity across portfolio accounting. Reconciliation, reporting, trading, compliance, and more, all at scale. Ridge line is revolutionizing investment management, helping ambitious firms scale faster.

0:25 Operate smarter and stay ahead of the curve. See what Ridgeline can unlock for your firm. Schedule a demo at ridgeline.ai. Hello and welcome everyone. I'm Patrick O'Shaughnessy and this is Invest Like the Best. This show is an open ended exploration of markets, ideas, stories, and strategies that will help you better invest both your time and your money. Invest Like the Best is part of the Colossus family of podcasts, and you can access all our podcasts, including edited transcripts, show notes, and other resources to keep learning at joincolosis.com.

1:00 Patrick O'Shaughnessy is the CEO of Positive Sum. All opinions expressed by Patrick and podcast guests are solely their own opinions and do not reflect the opinion of Positive Some. This podcast is for informational purposes only and should not be relied upon as a basis for investment decisions. Clients of Positive Sum may maintain positions in the securities discussed in this podcast. To learn more, visit psum.vc.

1:29 My guest this week is Gavin Baker. Gavin is the managing partner and CIO of A Trade's Management, and he has been on the show many times before. He is one of my favorite investors to talk to, and this may be my favorite conversation with him. Gavin first started covering NVIDIA as an investor at the turn of the century, making him the perfect guest to discuss all things AI and investing. There's so much detail in this conversation, and I'm incredibly grateful to Gavin for sharing his wisdom with us again.

1:56 Please enjoy this fantastic conversation with Gavin Baker. All right, Gavin, you and I have actually not done this in years, even though we do it offline much more frequently than that. So I'm really excited to do it again. On the record. I have a list of fifty things I want to talk to you about. So we'll see how many we get to.

2:15 But a really fun opening framing was something I saw you put out into the world, which was Back in nineteen sixty, the magnificent seven were led by Yule Brenner, and ultimately only three of the seven survived the shootout. We've got a new Mag Seven today, and we could probably spend the whole time talking about them. We won't. But I thought it would be a fun opening moment. Just to hear you riff on Why

2:38 These massive companies might actually be in some form of business shootout. now with a lot of what's happening in the world of technology. So I think these companies were all In their own discrete swib blades for a long time, competitive swib blades. Ноли плейс

2:56 Where they really overlapped was Cloud computing. Where you had Google, Amazon, and Microsoft all competing, but that was a very stable oligopoly. Google cut prices aggressively, something like two thousand fourteen, two thousand fifteen. Episoded.

3:12 And that actually materially impaired Amazon's revenue growth and fed into all of these fears that Cloud computing was gonna be a commodity, which was a real fear, hotly debated topic, looks deeply ridiculous now. But I think that set the stage for like a we're effectively gonna agree between the three of us on Some

3:29 Mark up on cost. But they try to differentiate in other ways. Amazon had e covers. Facebook had advertising that was, you know, higher up the funnel from Google, Google had search. Apple obviously had the device the OS of the app store.

3:44 Yeah, minor competition from Android. Netflix is doing streaming video. Again, a little bit of competition with Google there, maybe. But they're pretty distinct competitive sets and then Microsoft while they had cloud. They also had this massive enterprise software business really focused on productivity.

4:01 It just with JID AI. With LL libs to the day, but I mean Jet AI beats generative AI. But the G a GPT means purpose. It's such a general technology. That they're all of a sudden in the same swim blade.

4:17 And They all feel like it's existential. Mark Zuckerberg. Tchau. It suited, or just told you a different ways.

4:25 We are not even thinking about our. And The reason they said that Is because The people who

4:33 actually control these companies, you know, the founders. There's either super voting stock or significant influence in the case of Microsoft. believe they're in a race to create a digital god. And if you create that first digital God.

4:46 We could debate whether it's tins of trillions or hundreds of trillions in value, and we could debate whether or not that's ridiculous. But that is what they believe. And they believe That If they lose that race.

4:59 Losing the race is an existential threat to the company. So Larry Page has evidently said internally Google many times. I am willing to go bankrupt. Rather than lose this race.

5:12 So everybody's really focused on this ROI equation. But the people making the decisions are not. Because they so strongly believe The scaling laws will continue. And there's a big debate over whether these emergent properties are just in context learning, et cetera, et cetera, but they believe ceiling laws are gonna continue.

5:33 the models are gonna get better and more capable, better at reasoning. And because they have that belief. They're going to spend until I think there is Irrefutable evidence. that scaling laws are slowing.

5:47 And The only way you get irrefutable evidence Why has progress slowed down since GPT four came out? There hasn't been a new generation of NVIDIA. GPUs.

5:57 It's what you need. A nable. that next real step function change in capability. It is actually gonna be interesting. So everyone The Blackwell delay plays into this.

6:08 It's really, really hard to create what's called a coherent trading cluster of tens of thousands of GPUs, and coherent just means Beach GPU, we could say knows what the other is thinking, technically it's more like they have a shared memory space. And that cluster has to be coherent to train. And the biggest coherent cluster in the world until very recently was thirty two thousand. So you had thirty two thousand

6:29 H one hundreds. probably only using like at most fifteen, sixteen thousand of those H one hundreds at once. Because of Efficiency problems. But X AI

6:40 decided they were going to build a hundred thousand GPU cluster. And I would say with Elod's unique. physical engineering mind, which we've seen play out at SpaceX and Tesla. And that would be a link.

6:52 Where he figured out how to miniaturize everything. Along with really capable teams. I think he re architected the XAI data center from first principles. It be fuss. It's very different than other data centers.

7:05 And because of that, they're able to get enough density that they could effectively Make it one hundred thousand GPU cluster coherent even with Hop or without next generation networking technologies. For MidVide, Broadcom, and others. And they started

7:19 trading on that. And that means that I think we're gonna see Probably, you know, scale. the first GPT four and a half class model sometime, I don't know, in the next six nine months. I don't know the timing. It did. After that you'll have Blackwell.

7:35 I did you'll go up to three hundred thousand GPUs. In a cluster. 'Cause you have next generation networking technologies that make that easier. And That is gonna be a massive step function change. And then that's where you can get maybe a

7:48 GPT five, five and a half, sixth class bottle. I think the reason it's existential is just like if JIT AI is so generalizable and you have this ASI You're like, what's the value of content in a world where AI can make content that's infinitely better? Did any human. What's the value of Netflix in a world where I could say I wanna watch a mashup of Star Trek and Star Wars tonight?

8:09 Search might just go away and be replaced by agents. It's important to have a lot. Of humility. About this. Chase Colbyn, who runs Tiger.

8:19 exceptional investor had a really interesting statistic. Which was Chat GPT came out in two thousand twenty two. And was to AI as Netscape Navigator was to the internet and nineteen ninety four. Only less than one percent of current global internet market cap.

8:37 was founded. Pound it. In the two years after Deathscape Navigator came out. And nobody could imagine the companies that were going to be founded. So it's just the biggest companies were founded

8:47 Many years later. several years later, five or six years later. It's just we're very early important to have a lot of humility, but as long as scaling laws continue. And the only way they can be disproven. is if you have a new

9:01 Generation GPU. That has the traditional whatever it is, three to five X performance improvement. And then you get better networking such that you can link three to five X more of them together. And did you get that TIDX? the oh the order of magnitude of proof but a compute.

9:16 And you Don't see a massive improvement. And model quality? Then Scaly laws will stop and that'll be a catastrophe for the

9:25 Entire data center infrastructure. But the people close to this All believe the scaling laws are gonna continue. Last night I my daughter

9:33 On the way home from dinner said to me, Hey, can you ask Chat GPT some question? And I said, Do you have chat G PT on your iPad? And she said, No, not yet. You have to get it for me. I said, Well just Google it. She goes, What's Google? Wow. She she goes, Google's not a real thing. That's what she said. They just use chat G P T for everything. My son uses it for everything. She's eight, he's ten. Wow.

9:55 It's just fascinating to think about. Wait, what? What did you just say? And they use it all day. They use Dolly all day, every day. Search in their mind is chat GPT. Which is just a fascinating thing for kids that could use both if they wanted to. And rather than a search engine, they just want the answer engine. And I just find that

10:12 Completely fascinating. Absolutely. What does it mean for Humidity. The open internet has been really, really good.

10:20 And yes, we have these walls cotted gardened now, but they're still like a very robust vibrant Open internet were When you Google you do get divergent opinions. Generally, you know, even though we're all of these well known filter bubbles. But if you just get one answer.

10:36 I think we're at an important Mobile. It's very rare that to me investors can really really contribute To the world.

10:46 We contribute. In aggregate. In an indirect way, in a massive way, because we fund all these new technologies that you know are solving cancer. It directly funding AI. But I think it's rare that you can have like a more direct

10:58 impact as opposed to kind of a diffuse impact. I think it is supremely important for humans that we do not get up in a world. Whether it's just What? Dominant.

11:10 Bottle. That is the most dystopian. Future. I can imagine. Does that model then?

11:16 If kids all over the world follow your kids' behavior. Whatever values that model has will be imbued to the rest. Oh Timberry And I think

11:28 We are at a time in history where for a lot of reasons the idea of objective truth Is under attack. For a lot of younger People Their feelings are facts.

11:38 The idea of science and objective truth is really under attack. And I think it's really important to have AIs that are dedicated to the idea of objective truth, no matter how unpopular that may be. And I think the best way to accomplish that is to have them compete. You only have one domina day. That's very scary and forget nineteen eighty four.

12:00 Hello, who knows what kind of future. But whereas if you have three, four, five of these That's very different. They'll compete. They'll have different value systems. But I think

12:11 We had as investors by lowering and improving the cost of AI infrastructure, you know, whether it's breakthrough networking. Storage technology. software that improves the utilization rate of GPUs.

12:25 And directly funding I think some of these Labs that are Competing. With Google and better. Like I think it's very important for the world.

12:35 Can we talk a little bit about the world of data centers and semiconductors because you are one of the most season investors in both of these spaces you've been I think you started your career covering semis twenty five years ago or something, and you just know more about it than just about anyone else I've talked to and I

12:52 I often find myself asking you or calling you with a question about these two areas. And most of the attention has been on the models. and less on the world of semiconductors. Everyone knows NVIDIA, obviously, but they just sort of assume like yeah, it makes these amazing chips that power the rest of it. And there's just so much else that's going on. You've hinted at it with like the new clusters that are being built.

13:12 But just give us like a state of from your perspective. How this has evolved and what aspects of it are most interesting to you, whether that's the data center or the individual chip or the systems or however you want to approach it, I just think you have probably the most interesting and valuable perspective of anyone I know on these two topics. And have been thinking about them for a long time. And now it's like the main event. It's for sure the main event. I started covering NVIDIA in January of the year two thousand.

13:35 I would say watching the early days of NVIDIA has a public company. Tesla has a public company, it being a substantial It's for both.

13:47 have been by far the greatest privileges of my career the most Exciting. thing I've ever done as an investor. And Elon of Jitson.

13:57 Are for sure the Two best CEOs I've ever seen. With Lisa Sue right behind them. At AMD. And I just say actually took AMD from company five times levered.

14:08 Literally four years behind Did tell. To essentially utterly Dominating Intel in every way. I think they have twenty days of cash when she took over. Everybody talks about Satya and he's very impressive.

14:19 He took over a monopoly that had been extremely poorly run with recurring revenue and high margins. Anyways. I have been at Civis for a long time. It is my first love. Where are we? So first, you know, I think it's been touched on a lot of your podcasts. The number one thing we have all heard is tech investors. One reason the tech has been so amazing for thirty or forty years.

14:37 Is that software has zero marginal costs. These companies all have extremely high gross margins, recurring revenues. And AI is the exact Opposite AI has extremely high marginal costs. Because scaling laws literally mean

14:54 That the only way you get an improvement in quality Is By spitting a lot more. Full stop. That is what scaling laws beat. If you believe in scaling laws, you believe AI will have

15:06 Very high marginal costs. 'Cause of a variety of things that we're gonna talk about, those marginal costs. go down really, really quickly. but they're still really, really high at the leading edge. For bottles.

15:17 Especially for trading. And for inference. But much less so for inference. Inference and trading are two totally different markets. So

15:25 What this means that it has really high marginal cost. І за інфраструктур. Efficiency and excellence. I think Is going to emerge.

15:35 As the single most important success factor. Particularly for the model companies themselves. And this has been measured today in something called MFU bottle flops utilization. That July runs around thirty five to forty percent. Yeah, that's Literally the percentage of compute.

15:52 Theoretical compute flops. that you are actually applying to trading. So of the theoretical Well, most companies that have been publishing people stop publishing this'cause it's so competitive. But you know the technical papers for GPT three

16:07 I think what was called Google's Lib uh in NVIDIA's Megatron. All showed their MFU. And it was between high twenties But high thirties for all of them. Google had the highest.

16:18 If you have a higher MFU It means That you can choose for the same amount of money you are spending. Yeah, the same amount of GPUs and the same amount of power, presumably. You could choose between faster time to market.

16:30 If you run a fifty percent M F U and your competitors running forty. for equivalent amount of trading flops, you could be in market Forty five percent faster. You can choose between Better quality.

16:41 You might just do the trading route run as long as possible, or you can make the model lower cost in a variety of ways. Most of which relate to quantisation and we could get into that, but it's very technical. It's almost like if you have a twenty five percent higher quality model because you have a higher MFU and you build in quantitation. Did you get actually get a almost a fifty percent reduction in inference cost because if you could quantize

17:04 one level lower than competitors. It's a profound advantage. And so I think M F U If I were to pick one metric to evaluate a lab success, cause now there's been five GPT four class models from

17:16 It's Google, open AI. Anthropic. X AI better. There's five GPT. Does Mr All have one that good?

17:25 Maybe right under arguably just as impressive because it gets really good results. In all the evals with a much smaller parameter count. By the way, these models our commodities today. But I am suspicious once we get to

17:40 Scaly laws continue, GPT seven or eight. Literally cost five hundred billion dollars to trade. I Don't think they're gonna stay commodities. Scale is the

17:49 most powerful barrier to Idry and that's a lot of scale. So B. is the most important metric'cause it gives you all of these advantages and ways to differentiate yourself. Amongst five people, six people who've trade these GPT four class models. But I think there's something better than MFU and I actually came up with it this morning.

18:08 It it adds to MFU, it decomposes MFU. I would think of it as maybe like a unified AI efficiency equation. So MFU I would decompose into two. Thanks.

18:21 The first thing is something called Baba. Maximum. Achievable. Matrix multiplication

18:29 Plops Babuff. Batbowl. And what this measures is software efficiency. And this goes to CUDA. This guy Stan Bexman came up with it on X.

18:38 What he did. Is each Yep. has a theoretical maximum performance, which is just easily calculatable, you know, flops. And then he looked at what can you get in practice.

18:50 And video GPUs run it. For his testing, eighty three percent. And I'm sure that there are some labs that are running them closer to ninety percent, but this is just a single GPU. One reason it took A of D so long to break into this market, beyond the fact that it almost always had semiconductors.

19:07 It almost always takes you till your third generation chip to really, really hit it. That's what it took with Google with TPUs. You know, it's D by three hundred is good chip. Because they're Rocky open source software. Was terrible.

19:20 They didn't have the internal capabilities to really improve it. Nobody in the community took time to improve it. So the NC Tby three hundred first came out. It ran at something like twenty five or thirty percent. Be both. Now per Stance tests it's up to sixty percent, but because it has more flops.

19:38 It's actually slightly ahead of the Nvidia GPUs. And this was actually really I thought it was brilliant and it was a way of conceptualizing and quantifying the CUDA advantage. They can run an eighty three.

19:52 A D's running at twenty five and it answers question. They buy two fifty was actually a pretty good chip. Nobody used it. Why? It's maybe was probably ten percent. So first there's Babus, and that is How good is the software. For your chip. And then how well do you

20:08 as somebody who's using that software optimized it. Netflix used to say that we know how to use AWS more efficiently than Amazon. That a lot of people have said Some of these labs. Know how to use CUDA.

20:21 More efficiently than Nvidia. That's a big secret sauce. But they're running at ninety five, right there, that's ten percent. If NVIDIA GPUs are running at eighty three percent efficiency on a per GPU basis.

20:34 Why are we doubt it? Thirty five, forty percent for MFU. There is something I would call SF U. system flops efficiency. And this really captures

20:45 networking Storage and beverage. In each one of those, and we could decompose SFU into each of those. But I think SFU is like a helpful way to think about it. That's not all.

20:57 So you need to multiply beabuff. Times SFU, system flops efficiency. And then you need to Multiply that. By the percentage of type

21:07 That is spent at a checkpoint. Sure. Right. These things

21:13 Um Compound. And they can pound out to some wild things, it's just in case. Should explain checkpointing or now. Because as I referenced earlier, trading cluster has to be coherent.

21:23 To function and that means each GPU. needs to be aware of what every other GPU is thinking. If any one GPU fails You lose Everything.

21:34 For the last time he saved the model, which was called the checkpoint. The GPUs fail all the time. Not just GPUs, but GPUs melt. If you read the Lava Three technical paper, I mean it's like The list of reasons the GPUs fail is like astonishing.

21:48 An optical league goes down. A switch goes down. There's so Mini Please a failure. Is a chain of storage, networking, and memory that feeds each GPU.

22:00 That there are innumerable ways for them to um fail. The GPU without storage memory and networking is worth us. And if you want to really, really understand this is something I highly recommend to everyone, build a game PC. Because Every computer.

22:15 Whether it's the iPhone. Whether it's a laptop. Oh, there's a data center has the same I would call four fundamental elements. Three are

22:24 Memory? Storage and compute. And then the networking that connects all those things. And when you assemble a gave a PC, you Literally connected. Yeah, the PCI Express.

22:37 Into the GPU and then into the CPU, and then you you have to connect slot into D Ray M so that'll connect. And then The DRAM, you know, has to connect to the flash storage and the Drab's obviously memory. I highly recommend this. But I think the way to think of a data center Or any computer. It's just imagine

22:55 That it's a restaurant. In this restaurant. Um Primary unit of compute, it at an AI server, it is the GPU, is the head chef. A head chef.

23:05 Can do nothing. Without food and ingredients and utensils. I would conceptualize storage as like the delivery truck. That brings food. To the restaurant.

23:18 The way storage connects to the rest of the server. It's a always over PCI Express. That's a networking technology. And that is literally the guy who Moves the food.

23:30 From the food truck. It's the restaurant's refrigerator. We'll call the restaurant's refrigerator. This isn't a perfect analogy. Maybe the memory. And then you have to move The data

23:41 From the memory. Actually this case. Into the CPUs. Into the big pool of D RAM. The CPU can do its job.

23:49 But then you move it. To the GPU's memory and then The GPU can finally do its job. Maybe the GPU Vivery is like the stove and

23:58 What flow between the stove The sous chef is the connection between the CPU and the GPU. And we we can go into this, but just Unless that chef. As a stove. Cooking utensils and food, he could do nothing.

24:12 Yeah, the big problem at the data center is over the last Five years in particular. These numbers are gonna be directionally accurate. GPUs have gotten fifty times faster, and the rest of the data center has only gotten four to five times faster. And that is why MFU is so low.

24:28 Because the GPU is sitting around waiting for all those things to do their job and it's doing nothing most of the time. So I do think it is sensible, really sensible. To invest. in next generation networking, storage, and memory technologies.

24:42 Particularly in networking. Because if we're gonna get to a million GPU cluster That'll be a million Rubens, which is the generation after Blackwell. We're gonna need Or food.

24:55 Breakthroughs. It every step of that Pay. Every single thing. All new stoves, all new refrigerators.

25:03 Robotic sous chef's New utensils. Everything. Otherwise it's gonna be wasted and MFU's gonna be three to five percent.

25:12 But anyway, it's coming back to checkpointing because there's so many points of failure in that shape. The head chef. And Data centers being the GPU. Goes up in flames a lot.

25:23 They literally belt. And then every other component breaks a lot. The cluster is thirty two thousand of these metaphorical kitchens. Вер і степ

25:33 If it fails. Breaks down the whole cluster. So because of this, people check point frequently and that means save the bottle. Now if you threw better networking topologies or

25:45 better cooling technologies such as GPUs don't melt, if you have a lower failure rate, if you have a more reliable cluster, You need to check pointless. So we've had Baba. S a few. Then checkpointing frequency.

25:58 A that gives us like let's just say there's one company that runs it. Ninety percent Mamma U. And then it runs at fifty percent. SFU. Now you're at 45% utilization.

26:11 And they have to check point, I'm just gonna make something up. A third of the time. You're down to thirty percent M F. You have a different company They can run it close to a hundred percent bear buff'cause they're

26:23 Cuda Basters. And they run it, let's call it. Sixty percent SFU. Now you're at sixty, the other We're forty five, so it's already a massive difference. A thirty three percent difference.

26:35 And that if you have to check point only Ten percent of the time. You're at fifty four percent, your competitor's down at thirty. That's not even All of it.

26:45 The last thing that we need to multiply it by is P U E which is power utilization efficiency. The power has a cost. And it's going vertical for these big clusters. There's only basically three places in the United States.

26:58 Today. Where you can get a gigawatt of power. to a single data center that's reliable enough, i.e., that's nuclear. I think it's like eight cents per kilowatt hour. On average, the United States.

27:09 What they're gonna be charging for that gigawatt? It's like Tid X that. But your PUE then really matters because your cost of electricity really matters. The things you do

27:22 To optimized. SF U might actually increase your PUE in a really negative way. And so I think if you're I just thought of this this morning. Maybe it's super obvious that every lab already is doing this.

27:36 But I just think if you leak it all into an equation and then you do Dollars that it takes to get it. You can really make trade offs. Oh wow. If I spend two X more on networking, I get much higher SFU.

27:50 Which is great. But did I have a much higher PUE, which is bad. And what we ultimately want is Not just um actual exaflops per second.

28:03 But we want exaflops. Per second. Per dollar of capex Hm. What of electricity?

28:12 And the equation I just described, which I'm gonna write out at least for myself. Captures all of that. But everybody's gonna be making Different. Design decisions.

28:22 A data center architecture These design decisions you make are gonna be Amidely important. And I think you're gonna see, particularly with these hundred thousand clusters. Some of these companies are going to have greater than a one hundred percent advantage.

28:41 It exaflops. Per dollar of capex. Huh. What consumed. And this is when you're gonna really

28:50 Separate these labs. Because that is the difference between GPT seven or eight costing a trillion dollars or five hundred billion. The next class costing Two hundred billion. Or four hundred billion. And then for inference, it's the same.

29:04 It's the exact same. Not only is it cost to serve, but it directly affects user experience in terms of tokens per second. Which we know for Google is one of the most important things.

29:19 For search UX. So all of this is also gonna apply at inference. Inference is a lot easier. 'Cause it really, really just comes down to memory bandwidth and on check memory. This reminds me so much of the Vakloff's meal history of energy stuff where

29:33 You would get a new he called them prime movers, some source of energy, fossil fuels, wind, whatever. And then you get this long period of the gains all coming from the efficiency. So if you're spinning a turbine or something. In the early days of coal, the turbine only captured ten or fifteen percent of the available energy.

29:51 From coal itself, and now we're at like 98 or something like that. And it basically sounds like that's the exact same thing that you're describing here. The GPU is the coal. And those will keep getting better, which is cool in technology, I'm like Cole. But it sounds like that's basically the story. Which is so interesting. That is like turning sunlight.

30:09 And to usable energy. Via different fossil fuel mechanisms and motors. And this is turning sunlight into compute. Literally. And now that sunlight can come in the form of actual sunlight.

30:21 They can Cover the form of artificial sunlight, that's nuclear. It can cover the form of stored sunlight, that's fossil fuels. But this is the efficiency at which you turn Sunlight. into compute.

30:33 Yeah so cool. And the dollars you pay for that ratio. Yeah. It by the way, it is just interesting, like it's one reason like I was never that excited about wind. Because turbines were really efficient. Whereas you could look ten or fifteen years ago, solar was terribly inefficient.

30:50 I actually think it's awesome, you know. It's a little sad to me, you know, like all these kids are like really worried about global warming and you have all these Things about twenty year olds, oh, I don't want to bring children into a warming world. Global war being is a big problem. It is a solved problem. And it is solved. Because photovoltaic cell.

31:07 Efficiency is compounding Just under ten percent a year. And battery efficiency is compounding two hundred fifths less than that. And if you compound that out Over the long term. Not even the long term.

31:19 The world is gonna run outside directly. Fossil fuels are just gonna go away. It's'cause of economics. But it's because the latest storage is gonna be cheaper. Did every other way of providing power.

31:30 It's just not going to be a problem for the next generation now. Maybe we hit some tipping point and it's irreversible. This is that. Emissions are going to Collapse.

31:43 In my lifetime. Ліри колапс Obviously like pre Industrial They eat humans.

31:51 We're actually massive polluters. Because the pollution per unit of firewood is really high and you needed fires, otherwise you'd get eaten by a saber tooth tiger. Yeah, there weren't that many of them. It is possible

32:04 that we're gonna be back to like Neolithic. levels of emissions. In my lifetime, assuming I could live a little bit longer'cause of AI. So That's an awesome and encouraging thought. That's good and I just wish Somebody was shouting that from the rooftops.

32:20 Well, here we are. I'd love to keep going up the stack here because Every single level is interesting. So the semiconductors themselves and the potential innovations there are interesting to me. The data piece is really interesting to me, all the way up to the application layer. And I'm just curious what you think about all of this stuff. So one of the things that we're assuming in all this is that

32:39 We're gonna have more data to train these things on at GPT five, six, and seven. And I think there are some interesting discussions about what available data there will be, or how we'll get more data, or how we'll create synthetic data, and whether or not that will work to create a better model. Like I'm interested in what you think about all of this stuff that is required if we're gonna have the big arms race that you were describing earlier. So I do think

33:02 This was a real bear case baby died months ago. But I think it was hinted at in the Claude Three point five technical paper. Maybe a little more explicit in the NVIDIA Debo tron technical paper, you know, it's awesome. NVIDIA, whatever. It feels like there's gonna be a rate limiting factor for AI, they solve it. And that's what Nebrotrod did.

33:21 But I do think for reasons that no one understands And no one understands these models. Don't understand how they work, why they work, why scaling laws. There's all sorts of theories. And maybe we're getting a little better at understanding them, but no one understands them.

33:35 So no one understands point one. But it does look like synthetic data works. No one understands why. But it looks like it works. Now again.

33:45 Will it continue working? I don't know. Nobody knows. No one knows. People who have seen

33:52 Kevin Scott just did a podcast and he basically said, Look I've seen some early checkpoints of GPT five. If Gaily Laws are continuing. I think in a lot of ways that's the best. indication we have that they're scaling. And I actually think largely

34:08 With respect to X AI, I do think open AI is The combination of what I say is doing and black well delay. The Blackwell delay means that if you're waiting for Blackwell and you're Trying to get a hundred thousand Fluster.

34:19 XAI is gonna have arguably a one year advantage, which is untenable to these other labs. So they're all now frantically working at stating up their own hundred thousand cluster, but They don't have Elon designing the data center, redesigning the data center from first principles. data center architecture was always like kind of nice to have. Now it is.

34:36 Must have it's existential. I think synthetic data Does look like it's gonna work. This goes to Where will the value for these models come from? Why is Meta so comfortable open sourcing?

34:49 You may ultimately see everyone up. Open source. Yeah, X AI has open source cro Crockwad. But the reason is the value may not come from the model. Now look. If you're running that equation I described and you have

35:02 A one to two hundred percent advantage on exaflops. Per capex dollar. Per what? It's getting well told. you're gonna have such a massive advantage of model quality that you're never gonna open source it. You can still

35:14 Effectively cut of steel bottles, distill them. If you have that computer van scale laws hole, wow, you're gonna have Something very valuable. The value clearly comes from distribution and unique data. Meta is open source lava.

35:29 They haven't open sourced all their data. Yeah. That data. is just gonna be for their version of Wallace. So it'll for sure be better. Google.

35:38 They have YouTube, but then all sorts of other data sources that they've developed. where they tried to boil the ocean to create the knowledge graph, which are those little knowledge panels that kind of appear when you do searches. So between YouTube and the work they did for the knowledge graph. And Google Maps.

35:53 They have crazy data. They don't care. 'Cause they could use that data to monetize Their bottle. XAI?

36:00 Whether they open source or not, they will always have For sure. Access to X data. In a way, no one else does because XO's twenty five percent of XAI. And then I would think over time XAI will kind of be an intelligence layer that cuts across Elon's ecosystem of companies.

36:16 By the way, we should talk about AI and robotics, which I should think may be the biggest disruption in our lifetime comparable to like artificial superintelligence of these digital gods. If these bottles Unless someone develops a compounded advantage of they have a

36:33 SFU checkpointing frequency and PUE. Such that they have a dramatic difference. It exaflops per dollar of capex per watt. I think they're probably all gonna converge to roughly the same intelligence.

36:47 And the reality is, given the way that I think at least Google and Meta are thinking Even if someone else is way ahead of them and more efficient. They will try to solve the problem with money. Oh wow, it only cost them three hundred? Billion, no problem. We're happy to spend a trillion.

37:01 Although ultimately Economics will apply. these May Apex stock prices, like and I think it's not inconceivable some of these Magnificent seven. Forget buybacse and dividends. They may Eliminate their dividends, stop buybacks, and start issuing stock.

37:15 To find this. could for years we could be in a really wild world in a lot of ways. But these intelligences will probably converge on kind of a Similar to what I'll call IQ. And then it's just Who has the most differentiated

37:28 Real time data about the world. It's these unique data sources. Upl'd. With

37:37 Every time you rate an answer. from one of these models, you're helping it improve. And so if you can couple unique data with internet scale distribution Then You're gonna have a winning formula.

37:51 And there's only a few companies that have that. You know, it's X AI. It's Google. It's Microsoft, although internet scale maybe. And OpenAI gets that through Microsoft. Probably Amazon drop it gets it through that, you know, obviously better.

38:05 And that Apple is the big wild card. We should talk about that. Because I think one of the biggest things is where is the inference gonna happen? And compute tends to cycle in between Centralization or the decentralization. And we're at the end of like a long period of centralization in the cloud.

38:24 Or a lot of compute right in the cloud and these big data centers. is just'cause you could get much higher efficiency, these big data centers. Now clearly for trading That is gonna happen to giant data centers. that will be in the cloud, although something that I think is underappreciated about these AI data centers is why we shouldn't worry about them. over the long term crushing power demand is you can put them anywhere.

38:45 You put'em in Wyoming, like they don't need to be near a big city. I think we'll eventually see giant data centers. In shell gas fields. With power plants all those fields. long away from many humans.

38:58 That will probably be an intermediate term solution to the lack of nuclear power in America. Yes, we do we have all the best models. And you ask what's these bottles trade to America for better or worse? Obviously the Straw is French, I don't know where their models for trade.

39:12 We have these cycles of centralization and decentralization. based on where can you get the lowest cost compute at the highest utilization, kind of a variant of the equation I Described earlier. And I do think the subcomponents of that equation

39:25 Not only are they maybe helpful to labs But they're really helpful to investors. You can Improve. Sf twenty percent.

39:35 It it all comes down to millimeters of silicon for only a few More millimeters of silicon or maybe even less millimeters of silicon. They're like, Oh my God. Does millimeters of silicate are the ultimate did cost. Wow, you have like a massively witty formula. And so you could just almost go look.

39:52 At each step of that long chain. And where do you have really high cost per square millimeter of silica. Got opportunity. To really optimize.

40:04 And whether it's at the infrastructure software, the data center hardware, the semiconductor Later, it's all there. But I think You're gonna see inference increasingly done on phones. And this is clearly Apple's play. It is one reason that they are in such a uh

40:22 advantage position. All these other Companies are trapped. But this prisoner's dilemma. I am sure they would like, given how much it costs.

40:33 They're all economic inputs. If they can just like reach an agreement. It's hey, you know what? Nobody is gonna stand up a black well cluster until two thousand and twenty six. Like they probably all cite it. But it it is literally a classic

40:47 Prisoners to Limba. And that would be a Nash equilibrium. But we're not gonna reach a dash equilibrium in the race to create digital god where the states are existential. So there is prisoners to Apple's not. But the same way Google spins Collectively, cumplibly spit, I don't know, hundreds of billions of search between CapEx, OpEx, all this stuff and

41:07 Apple monetizes it. Almost as well as Google. Because they have This distribution choke hold. Toll booth.

41:15 In IOS. They're clearly gonna do the same thing. With Apple intelligence, then Google will obviously do that with their phone. The Droid.

41:24 And so you're gonna have a small model running on your phone. That for simple questions. Think of it as like a One hundred IQ bottle that with like Really, really sophisticated knowledge.

41:35 There'll probably be two of them. They'll check each other. And a lot of inference will then happen at the edge. And if if it's happening at the edge, then I do think we're gonna see superphones. Because the advantage to you as a human What bounds inference at most times is memory.

41:53 And you will be willing to pay. Today iPhones are sold. Based on how much Flash storage they have. What you will most care about.

42:02 is the amount of D RAM. you have on your phone. 'Cause that will determine the perimeter count of the model you could run locally. That is therefore the quality of your own local intelligence.

42:12 Yeah. has access to all of your data in a privacy safe way. And when you Ask it for something if it could B dot odd.

42:21 On compte it will, and I think from networking like a tourism is route when you can, switch when you must. But I think the next thing For inference, it will be Local when you can, cloud when you must.

42:33 So if you could inference locally on your phone. You will always do that'cause it's free. And clouded for it's definitionally cost money. 'Cause you're burning GPU hours. No, it's very efficient.

42:46 But the inference on your phone is effectively free. It didn't. Increasingly our competitiveness as humans, you know, it's very funny. A lot of really smart people are a little skeptical about AI. I get it. If like you have a hundred and twenty five IQ

43:00 Or a hundred and thirty five or whatever it is. AI particularly domain. Isn't at all impressive to you and it's like okay, it was a little bit for search for me learning about other demands. But man, a lot of humans don't have one twenty IQs. That what people are missing is that

43:15 There's a lot of humans with like I don't know what I I I have no idea. Yeah. So let's just say it's a hundred IQs. They can use AI. And all of a sudden they're like a 115. And it's like Holy shit it's cool. It that's just gonna keep going.

43:31 Until these ASIs have IQ of a thousand and it's like Why as a human about bothering to do anything other than work with an AI to create art. Anyways, if I could have like a three thousand dollar iPhone It has.

43:45 four times the D Ray about it. The the thousand dollar iPhone has It may be. More storage. Does that local model could do rag.

43:56 And this is my model. It likes me. I pick the voice that talks to me in, it knows me, it's my friend, it's my agent. All it wants is thumbs up. For B. in that process of improving these models.

44:10 Yeah, I think they will ultimately be There will be a way that they could be early staff on an individual basis on the fly. Sure, we're gonna need a lot of breakthroughs to make that happen. And it's gonna be my model that likes me, it knows me, and knows what I want. And then this is where you get this agent future that everybody's talking about. Where it's like I have an agent on my phone.

44:30 And if I have an agent on my phone, because I have a super phone. That has, you know, an IQ of a hundred fifteen or one hundred twenty and someone else's agent only has an IQ of a hundred. I'm going to be profoundly advantaged as a human. So then that continues all the way up. Because I think a visually

44:48 Apple will monetized this and the way Apple will monetized this, clearly. Is with the odd device oh it was farted off. They're gonna send you to the cloud. They're gonna use something called a router, which is what Every um

45:00 A application company uses. You just want to route. to the best model. For the query. Per dollar.

45:07 Apple will charge companies to be in their router. That oh, if you wanna be routed to eventually you're gonna have to pay a bigger, bigger toll and everybody will pay that toll the same way Google paid that toll. But if I as a human I could have a Smarter model locally. And then I could opt in into

45:25 Maybe Apple will just make it easy. They'll say, Okay You could have cloud superintelligence, or you could have cloud intelligence. And I pay Sixty dollars a month for cloud superintelligence. Now or whatever it is, you know, a thousand dollars a month, ten thousand dollars a month.

45:40 What would you pay for that? So I have 40 points of IQ on my phone relative to a lot of people. That I'm paying ten thousand dollars a month for cloud superintelligence. Uh hyperintelligence. super intelligence just a thousand dollars a month, and then regular intelligence is twenty dollars a month. Like I'm gonna be really advantaged as a human.

46:00 And That seems dystopian to me. But it's also kind of hard. For me to not see that happening. And middy, biddy

46:10 Investment conclusions. flow from this in the same way that they flow from like hey Okay. Maybe if it's ninety percent. Okay, so that's probably not a good place to invest. S FU is at thirty percent? Wow, that's a great place to invest. P U E

46:25 Is it one point eight and it could theoretically be at one point three? That's a great place to invest in that equation. You kinda wanna invest at the most inefficient chain. A lot of investment implications flow from Where does it first happen? And then a lot of implications for humidity flow from that as well.

46:42 I'd love to talk a little bit about the commoner part of this whole equation. We've talked about like Mount Olympus and like the game of kings fighting each other for dominance. We haven't talked at all about What about just like a normal new company that's being started to take advantage of these superintelligences at the application layer or to do something else. Oh then we'll get to robotics after that. But if I force you to just be uh early stage series A, series B investor in the world of startups. And they don't have these kind of resources that all these magnificent seven and others have.

47:12 How would you think about the aspects of companies and opportunities that would get most attractive in the world where GPT X keeps scaling and getting better. First thing I would just say it's always what are people doing? What I am doing is really focusing on companies that approved that equation.

47:29 elements of that equation I described earlier. 'Cause I think that's the choke point. That's the highest likelihood of success. And it whatever you find to constraint. If you can invest in something that alleviates that constraint, you usually do well. And right now I profoundly believe the constraint Is it SFU?

47:46 checkpointing, which goes to reliability and then PUE. That is where I am targeting By dollars. I think the application layer is really Really

47:57 Hard. There are people who are kill yet. I think you have Sarah, who I've never met, but I admire on your podcast. And she's like absolutely

48:05 from my perspective, crushing it at the application layer. But just geez investing there today feels really hard to me. And I just think anyone who has a lot of conviction about that needs to be reminded of that one percent Chase Coleman stat. Yeah.

48:21 I was just thinking that. All these VCs had really strong views of like There are all these companies that were funded in ninety five, ninety six, ninety seven It they seemed like they had got in the net. But it just takes time.

48:34 I guess they did go with the debt for the V Cs. They would public and they were able to cash out. But they actually didn't go into the net with the fullness of time. So I just think it's important to have a lot of humility at the application layer. And maybe it's just

48:48 that I am so comfortable with this infrastructure layer. That is where I have Concentrated to date. There's really only been one application company that I've been excited about. We've seen a lot of them.

49:03 But yeah, I'm like too aware of the Chase Colbed stat. And I'm too at a being too careful in how I approach this application layer. But I'm gonna miss out on a lot of C M GI was incredible. You know, what were some other besides Yahoo, what were Lycos. You know, there are all these companies that we don't even think of today.

49:22 But there are incredible venture outcomes. Maybe I do need to be a little more mindful of that, but there Clearly. people, whether it's pitch mark. Well it's serious firm

49:33 Who are succeeding, and when I was saying the application layer, I heard this from Vishria. But it really resonated with me. And I think it also goes this whole Joy. AI ROI debate. Not only is it an ROI debate about

49:47 Creating a digital god. But I think there's a super clear ROI and you can actually do math to show that. And a lot of people are conflating CapEx and Op X. In ways that are just not helpful. Metal went down like eighty percent.

50:00 Partially'cause they were overspending on the metaverse, but partially because Apple took away their ability to target through IDF A. That is a five X, so the revenue growth has reaccelated. Why is it reaccellerated? 'Cause they spit a faster about a body out of AI. To figure out how to target

50:15 It's like maybe The meta Return alone. Justifies. All of the spending.

50:21 They call that meta advantage, but meta advantage is just part of it. That's just where as an advertiser you can let meta do the targeting. You need that Google has performance back. And all of that is and that's where the There used to be ads were created. And we're gonna come back to applications, but ads were created. We're gonna pay some humans to figure out that we need to show this creative to like

50:40 White forty old guy bastard and did this type to this Type of demographic category of this city and we're gonna show up at this time of day. Oh, we're gonna show after the local sports team is what and it's sunny, because that's actually the best time to see it add. But now AI can do all of that and we're gonna have a million creatives and be optimized on the fly.

51:00 And that's what AI is enabling Google and Meta and other firms to do. It probably Just on that. Just that justifies it, yeah. There has been a massive ROI on AI. What are we talking about? And then the other thing that's really funny to me about this whole AI ROI debate

51:17 Is okay, we're gonna have this abstruse debate. About ROI. These companies are all public. There is something called return on invested capital. And ROIC has gone up.

51:28 For all of these companies. Since they raped Cap X. What are we talking about? If you're an AI ROI skeptic Why is ROIC

51:37 Got up at these companies. Well yeah, CapEx is up. No pad is up more. Why is that? Because they're doing exactly what you would expect to happen in a world of AI. They're trading.

51:48 Of human labour. Against GPU hours. That's why their RYC has gone up and these GPUs are really efficient. Let's have an AI ROI debate.

51:58 With the ROIC. Not the big AI spenders. starts to go down. Until then it's just The height.

52:07 Of intellectual ridiculousness. I mean. Ridiculous. What Vishrias said that I thought was so powerful for the application layer. There's all these SaaS metrics.

52:19 After your first year you wanna be at million to be on pace. After your second year, you wanna be at five million. After third year you want to be at like ten, and if you're above that with reasonable Cash burn, you'd probably run these curves, you're gonna be a successful SaaS company. And all those curves, which everybody knew is one reason SaaS multiples got so inflated because it became almost like quantitative. Oh wow, this company is at fifteen million in year three. That means we can pencil them in for a billion dollars in year eight. A that led to multiples.

52:46 Expanding ridiculously. B it led to The industry being overfunded. Such that the degree of competition in each vertical went up such that the curves no longer held. And this is one reason why if you made a lot of SaaS investments.

52:59 particularly 2021, you're in a world of paid. And then on top of that, here AI comes along. And it fundamentally changes the paradigm. For application software because what application software fundamentally does is makes humans more efficient. And today we're in a state with AI where it's making Human's more efficient, but you don't need that big of a wrapper around it. That's when we're gonna come to the fishery comment.

53:21 And then ultimately if it's replacing humans. And you're selling application software on a per seat model and those seats start to go down, you have a problem. These companies are super aware of it. They were going fast in one direction. And now they have to contend almost at every vertical.

53:36 With this next generation of AI first application companies. Which are just these often very thin wrappers around GPT. You'll call it an L L M wrapper.

53:48 And what Fishery said to me that was so funny is like all these AI companies Are blowing these traditional SaaS metrics. out of the water. And all this isn't being counted in those AI ROI papers. Just that there's all these companies that are going like zero to thirty million in nine months.

54:05 And even though AI is really high marginal cost. They're being more cash flow efficient relative to software company. So to almost every vertical. There's multiple companies. They're AI first.

54:16 That are just really I think what Vishri said was they're just Paper thread wrappers. Around pick your LLIM of choice. Then yeah, they use a router to find the best LLI. But they're like

54:27 Magic to their customers. They aren't going after software budgets, they're going after labor budgets. It's very intellectually interesting. To me to watch. How do you get conviction that one of these companies

54:42 could use what is Initially magical to their customer. To build defense ability in. All while allowing for the fact that The big bottles are gonna continue to compound at a really high rate.

54:55 And this goes to things like Maybe you just find a really good salesmotion that works. Maybe you find A really easy integration point. Maybe you build some defensibility around that integration. maybe you make it really easy for a small business

55:10 To do rag. On a really unsophisticated system. How are you building differentiation into that wrapper. And hopefully

55:19 You're not fai tuning'cause that does extensively five tuning'cause that locks you into a model generation. How are you creating a compound AI system such that your Serving each query. at the lowest cost possible and you're starting with a small bottle and treating it to a big bottom. There's a lot of things you could do and all of those are really

55:37 Really important. But I just think at that application layer, there's just It seems like For any category. Any category it feels like they could be imagined. There's multiple startups that have exploded.

55:50 That are defying all traditional SaaS metrics. It is not clear to me at this point how much defensibility each one of them has. Some of them.

56:01 are gonna have a midst to fitsibility around one of the axes I described. But I think a lot of them may not. And I think that's what makes it so hard and why I've been approaching that space so carefully. You know the amazing thing? I think the framing around labor, I mean just take

56:17 Me isn't a stupid simple tiny example. I have a team of three people building One of these We'll call them like a really light LM wrapper for doing research on private companies. And

56:28 If you just think about how often I use this thing. In a way that I normally would have used an analyst. It's multiple times a day. And it's returning work. As good.

56:37 So there's so much undifferentiated heavy lifting in every white collar labor market, I guess is like one major Learning. Hundred percent. If you can have the answers.

56:48 And for effectively free. It's crazy how much you use these things. We're just getting started. Yeah and I have the same thing with like And I'll limit my firm. It was actually very interesting. We had an intern who was really talented.

56:59 You said, Listen. I think this internal AI tool is amazing. And I use it every day and more every day. It's gotten much better this summer. We work on it for a long time. It did. I was like, show me how you use it. 'Cause I've been using it a certain way.

57:12 It this literally this twenty one year old kid was like Well, this is what I do. And now all of a sudden my usage of this tool is two hours a day, three hours a day, four hours a day. And I do think Jack Welch. had this term called scut work.

57:26 Just like hard, unpleasant. work that had to be done really well. In a white collar setting. But I think it's All these little wrappers are

57:37 Mini wrappers, whatever we're gonna call them. are gonna replace a lot of that scut work and initially it's gonna be But combination with humans. But science fiction, Neil Asher's world, there's something called a high band. Which is a humid AI hybrid.

57:50 And it's where They're linked through something like a neural link to like a computer. Server. This Supported by like an exit structure, robotic exit structure that they walk around on.

58:02 There's like fusion chess. Whatever we wanna call it. We'll have that for a while and that goes to the you know a hundred IQ humans. performing like one twenty IQ humans and then performing like one thirty IQ humans then one thirty IQ humans performing like one sixty IQ humans. But then eventually it feels like as long as scale we must continue, which is a bigger F, it's just gonna be

58:21 The AIs. Maybe we could talk about robotics. I had a really interesting conversation. Call it April of this year. with an investor that has been investing in lots of these same things for long periods of time.

58:33 privately and publicly and has big positions in lots of the companies that we've been talking about today. His observation to me was the big underestimation that's happening. Over let's say five years. is the role that robotics and robots will have. combined with all of this technology we've spent all of today talking about.

58:53 And I would love to hear you riff on that because in the near term it feels like a little bit quite a bit of frothiness, like some Crazy funding rounds for these companies that You don't really know what the robots you're being designed to do, sort of general purpose humanoid type robots.

59:06 There's all sorts of interesting, more specialized ones that are cool too. But what do you think about all this? It does seem kind of under discussed relative to just all the foundation model and semiconductor stuff. I agree. I think it may end up being a bigger Near term disruption. Yeah.

59:21 What we were just discussing. the automation of a lot of white collar labor. I think the first robot The first robot that's really gonna impact the world. is every test of the car with what they call their AI four hardware.

59:34 'Cause from my perspective, there's a publicly sourced miles between Disengagement. So you have to remember for Tesla Tesla's gonna get the same miles between disengagements. Like if you built a new city On bars.

59:48 And it was populated. by entirely different looking cars and streets and everything. You could drop a Tesla in that city and it would have the same miles between disengagements. Like it gets it in the other city. Or something like Waybo.

1:00:00 It's geofenced. We're really only using it in cities that have like nice grids and good weather, et cetera, et cetera, et cetera. It is clear. To me, looking at the crowdsource data and miles between disengagements. With different versions of FSD.

1:00:17 Yeah, when they cut over to Well three. Which is effectively All deep learning, I think eliminated Almost all human code.

1:00:26 Something. dramatic changed in the rate of progress. And then when they cut over to twelve dot five. Which runs best on the AI four. which used to be called H W four. It's just like the local computer Oh, the Tesla.

1:00:39 It is now rolling out to AI three. was another step function. And these going to that same scale you law. Those step function improvements were made. With

1:00:51 A fraction of the comput. The Tesla is now installing. Publicly. in their data center at the Giga Factory in Austin. They've actually failed.

1:01:02 Sometimes I wish they as an investor, I wish they wouldn't file so many of these patents. But they filed some Really innovative Patents. For data center cooling related to what they're doing.

1:01:13 With what looks like. has been publicly said it's gonna be, you know, over fifty thousand H one hundreds or H two hundreds. FSD Is now on the same scale in law. But arguably on a faster scaley law,'cause they have a lot of catch up to do.

1:01:27 That GPT is a bit odd. So I think twelve dot five is like GPT three. And it can consistently

1:01:35 drive me most places with no interventions. I'm a seasonal driver. I really only Drive by Tesla the suburb. And actually

1:01:43 My wife Becky tends to do most of the driving because she likes the board of the B. So we kinda get like a seasonal Look. It's almost like Every May we check in.

1:01:53 There was just always Continuous progress. This year It's like Well we turned on twelve dot three.

1:02:00 It's like All the progress. over the last ten years was it that one release. From the first time I had that Tesla. And then we had that again.

1:02:09 When I went twelve dot three to twelve dot five. And they're at probably like a GP two level of compute. I think they're gonna go really fast to GPT four point five compute. Which means you're gonna get using these orders of magnitude, you're gonna get like a one hundred X improvement.

1:02:25 Really fast. So I think there's all these people who have been skeptical. They're all in for object humiliation. They just are. And then

1:02:34 Unlike GPT two. Oldie Tesla. has access Two A visual trade data set.

1:02:42 That is based on miles driven. We could argue whether it's a hundred X. A thousand acts? Ten thousand X? Bigger.

1:02:51 The second biggest trading data set. Which is Waybo. So it's like people, oh, how are they gonna make money? Well It's like in this case. In the world of self driving, from my perspective, it's like they owed YouTube.

1:03:03 They ought to. All of Meta's properties. And the open internet. Adex. And then

1:03:10 Other people are like trying to do it using Yahoo. Yeah, using Yahoo. Like good luck. Like who's gonna win? Now obviously that could change. Important to have humility. There may be an algorithmic breakthrough.

1:03:24 that reduces the importance of that tradey data set. But for sure Waybo is gonna try and brute force it. It just throwing. Whatever amount of dollars they need to get the data. To compete.

1:03:37 And they have a different approach using LIDAR, Tesla doesn't. We'll see. Like I don't think it's a foregone conclusion. Nothing about the future is certain. But just if I look at How amazing. Well that five is.

1:03:49 Odd AI four hardware. And think about the tiny amount of compute that that was trade on. In the mega cluster. That they are standing up at Austin. Using known techniques.

1:04:02 We're gonna skip I think twelve dot five at GPT two. We're gonna skip really quickly to GPT four. Then look, you know, I'm sure Waymo will reinforce it. There may be algorithmic breakthroughs such that there are other people. We'll see. But then the other big thing It's just Using it L for FSD.

1:04:17 One of the best followers at X is Dr. Jim Fayette, who's Nvidia's head of robotics. But he's had a lot of posts. About how there's a fascinating shape between h him and heal out on X. It is amazing the extent to which AI happens on X.

1:04:33 The Jack's team at Google or the PyTorch team. I'm better got into this bitter fight. And it went to Meba. Which framework was better for Memo. Visually the heads of each lab had to step in publicly on X.

1:04:48 They make peace, but like wow. You know, you learned so much just following that fight, like every AI researcher is active about X. AI Happens on X. And it's such a great form for for using it but

1:05:02 Jim Fayette had this fascinating exchange with Elon. Where Jim Vaya talked about how LLMs could massively improve FSD. And Elon replied, yes, the only two data sources that will scale infinitely are synthetic data. And real world video. And I thought that was interesting.

1:05:18 And then that goes to I think maybe the biggest risk In which This view that I just described of Tesla's Autonomous future.

1:05:26 It's wrong. It's just a synthetic Video. data can be used in the same way that synthetic data can be We know that synthetic written data works. We don't know if synthetic video data works. Nobody knows.

1:05:40 And obviously, there's a very high bar for regulator. You know, I think it's something like if whatever it is, like fifty thousand or hundred thousand people die in car crashes every year globally. It might even be a million. Obviously we could take that down dramatically using AI. But we're much less willing to tolerate traffic fatal accidents from AIs than humans. That is what it is. So you know it's gonna be heavily regulated. But Dr. Jim Fan.

1:06:03 posited that the reason Libs were gonna be able to really help With F S D is'cause the following This is the way my relative to some of the people working on these problems.

1:06:13 my comparatively low IQ braid conceptualizes it. Anything that's been trained on real world data. Just knows. what to do, what a really good human driver would do in that real world situation. If there's a novel situation.

1:06:27 It may not know what to do. And that's where From my perspective, the L L M can really help. 'Cause one of the emergen properties of GPT four. And we could debate.

1:06:37 Whether or not it actually is at a merchant property or just in context learning. It has What's called a world model. And that means I'm sure you know this, but if you ask GPT three

1:06:48 Okay, what happens if you stay in a champagne bottle upside down and put like a basketball covered in soap on top of it. T V three. No idea. GPT four will often get

1:07:01 questions like that right. I should actually see if he gets that exact question right. A three year old human will say That's gonna fall, the shear paid bottle's gonna shatter. It's really hard for GPT three and that goes, you know, this jagged for two that people talk about. So

1:07:14 If you What a really Speed optimized small L. And Locally on each Tesla.

1:07:23 There might be just enough reasoning capability to unlock another step function. An FSD capability. Now look. Wavo will have that too, as will lots of other people, but they won't have Tesla Vision.

1:07:38 all of the proprietary data set. Just think this is going to be a reality. in a way that is abjectly humiliating to everyone who is an FSD skeptic. In the next twelve to eighteen. Maybe in the next six months. And I haven't never

1:07:54 been willing to make a prediction like that before. So then you take that and the same thing goes for humanoid robots. You Google show this with research called sensor RT two. We're dropping it on LLM. It to a humanoid robot.

1:08:09 with a world model that understood what things were and what to do. Just made it so much easier. Instead of trading that cubinoid robot how to pick up a tennis ballot, a basketball and a football. How each one is different. You could reason. And so this is why putting LLMs into these humanoid robots, I think is gonna be

1:08:29 so transformational for the world and make a lot of blue collar labor. Optional. I do think politicians and political systems are utterly unprepared for what may be covenant. The one thing I would say that Elon Egyitson profoundly agree on.

1:08:43 publicly. Is that humanoid robots are the future. Not the specialized robots. The reason is just Of course a specialized robot could be better than a humanoid robot at any given task.

1:08:55 But the humanoid robot could do any task that a human can. The world is optimized for humans. And There are massive scale efficiencies in manufacturing So because You can make

1:09:07 It's almost like humanoid robots are gonna be to the field of robotics. Has GPT was to AI. GPT was a generalizable type of AI

1:09:18 And these humanoid robots are gonna be a generalizable form of Robotics. And because of that, they're gonna be Manufactured At such a scale that they have a cost advantage and then the physical world is gonna start to be optimized.

1:09:31 With that. And that's why they're gonna win. So Good luck to all these non humanoid startup robot companies. I hope you get Lycos or C of G or My space type venture outcome. But I don't think any of them are gonna be Google.

1:09:45 And in the same way that So much. of GPT advantages encumbants. Whether that's meta, Google X, X AI, Microsoft.

1:09:57 These humanoid robots The reason it advantages encumbs is'cause they have the raw ingredients of data, compute, and capital, which is what you need. to effectively monetize these, and that's why their ROIC is going up even as they read Cap X. I do think that encumbered manufacturers

1:10:14 We have expertise and Battery design. Actuators. Motors with big data sets are gonna be advantageous.

1:10:22 Yeah, reasonably bullish. On octopus. Not just such a giant market, there's gonna be so many competitors. And whether It evolves

1:10:31 F S D, I could see a world Where there's just two or three companies, maybe it's Tesla, Waybo. And some open sourced variant. Or it could be

1:10:43 Synthetic real world data works. LLibs really improved the efficiency of that specialized Visual algorithm. And so there's like thousands of them. I think that's unlikely, but it's possible.

1:10:55 Like robotics may end up in ventage encumbrance in the same way I think FSD and Oh, the GPT have advantage. Generally advantage to complex. But that startups willing to take like a new AI first approach.

1:11:08 But anything robotics is gonna change the world. It's super exciting. Like I can't wait to have my own personal robot. Or ten. Yeah, or ten. Like it's what one everywhere. What time is it? It just tells it has an AI that's loaded into my phone is also running on that and it likes me and is friendly to me and it laughs at my jokes. Tide me up. Yeah. Yeah. I'd love to ask a couple of closing questions that are

1:11:31 big picture, big arching questions. The first is responding to something you said earlier, which was having watched Jensen and Elon and Lisa at AMD operate for so long. And rating them as these like exceptional CEOs.

1:11:46 It just seems like Such a small handful of leaders. at these companies. Can have such a massive impact on the world. And so they're character and way of operating is really important for all of us.

1:11:57 Of course this will recycle and we'll get new ones and up and coming ones or whatever. But what is it about that group of three and maybe throw into the recipe a few others that you've learned a lot from? What is it about those three that match them so well with the modern world and way of company building? I mean, like the crazy stuff like Jensen having forty direct reports or Elon working on seven companies at once. These are unusual human beings. Maybe just riff a little bit on

1:12:20 Why those three and sort of the nature of leadership in this era of technology? I guess I would say Although beyond having clearly Unusual IQs and ranges of domain knowledge. Yeah, but I do think it's hilarious that Lisa and Jims are cousins. It's like Crazy. Like I just what a bet.

1:12:39 On everyone in that family going forward. Who's like a first cousin. Did we have to bet on those jeans. What a crazy Coincidence. Yeah.

1:12:49 In her genetic advantages. I actually think the modern American Corporation In the same way, like almost any large organization with its hierarchies Is set up to reward people who are charismatic and political.

1:13:06 And because those people are charismatic. And Political. They're not always Really smart.

1:13:12 Because they're charismatic and Political They get big egos'cause people like them. They get used to having their way. The same way every I think child under four is like a functioning sociopath. Most people who become

1:13:25 Multi billionaires. are really powerful politicians. The physical part of the brain that deals with empathy shrinks. Changes you as a human. It is a long way of saying that I think the way that modern corporations have been evolved

1:13:39 To be rugged. You end up with a lot of Not COs who are Excellent.

1:13:46 But objectively terrible CEOs. Who were amazing at rising to corporate ranks. Battery Whatever it was took to get there. Then by the time they got there Their ego.

1:14:01 They've gotten so big. Sense of empathy had gotten so small. They're no longer capable of functioning effectively. And often with people like that, you find there's like One or two people behind them.

1:14:12 There's like a key leader. who run the division or whatever. Who's kind of arisen with that. CEO through each division or group of people. Those are the people with the

1:14:23 good judgment and skills to make high quality decisions. So Elon and Jitson in particular R Nothing like that.

1:14:32 Not only are they the front person But they are always working on the most critical. problems at the company. Try but you're off a Which is each.

1:14:44 It yeah, it was a different conversation'cause I wasn't working. It he just said it And maybe this has changed, but the time he said, I have no fixed schedule. I have no standing meetings. I just find out what is the most important problem at the company.

1:15:01 And I go. And I sit my desk down in that area. And I pull the best resources to work on that. Problem. And I love it.

1:15:12 So he's working on the most important problem. That is also what I do. Uh each of his companies. Whatever is the most important problem is what he is working on. Well, the Raptor engine was in the critical path for starship.

1:15:25 I'm going to get the details wrong, but I think there was like a C and D what I A Tuesday boarding or Monday boarding. Me dig. But Raptor Engineering. Only the like twelve or eighteen or whatever the number is.

1:15:37 smartest people were allowed to be there. Everyone at the company wants to be there. So I think that is one Something that ties them together. Working on the problem. Second is loving to hear bad news. My dad was a bankruptcy attorney.

1:15:52 He always said that up on one thing all bankrupt companies had in common was the CEO who didn't like to hear bad news. Just say it's evil to just sort of Love hearing bad news. That those companies If there's bad news.

1:16:04 It must immediately Go to them. And that's very differentiating. No hierarchy. Wherever in the company the problem is, that is who Jetson and what to work with, the subject matter expert.

1:16:16 Whether they're like twenty three. Or fifty. There's no hierarchy. You know, it's like the same way like there's this apocryphal story when I think it's true with JP Borgard was by Bear Surds. The best modeler in the company was twenty four years old. So they set up a desk for him. Side by side with J B.

1:16:33 Yeah, Jamie would say, Change that, change this. You know, twenty four year old this kid was like the guy because he was the best. You know, Javid, another exceptional CEO. He did ask for that guy's boss or boss's boss or boss or whatever. He's like This guy's the best? I want to work with him. The last thing that I think really, really ties together, particularly Jets in in Elon, is a mission orientation.

1:16:55 Jidson started out about like, hey, let's make photorealistic video games in virtual worlds, because that'll be exciting. remove what someone called reality privilege and humans will eventually be able to be in The virtual world and by the way, the metaverse is still coming, I think it's just clear. That it will happen first with augmented reality. And then with BCI.

1:17:14 Just VR goggles. C'est skeptic, never gonna work. Argument reality glasses. If it be it computing and then BCI. It's kind of

1:17:23 The end state. Yeah, smart. Metabot that company. control C Gro labs it is working on this'cause I think That may actually be

1:17:32 the next truly disruptive form factor. We're gonna have superphodes. The phones are gonna stay as the compute layer for Student reality. But BCI may be the next true way we interact with computers.

1:17:44 After phones, the next true compute platform and maybe the ideal way for AI going back to that high end thing. But that was a compelling vision, photorealistic graphics, a lot of technical People are love to play video games. So he jets sort of hired a lot of really smart people based on that.

1:18:00 And then juts it in a lot of ways is one of maybe the single most I don't know if single ball he's very important to the history of AI. Because very early ten years ago you started hanging out with people like Jeffrey Hitta the the Allah Kud, I remember him saying those names to me.

1:18:17 And talking about the famous Image that competition would res that fifty one in this means Effectively. I would paraphrase it, but we've turned intelligence into an engineering problem the way we're gonna solve intelligence. It's just by gluing boardboard GPUs together.

1:18:32 He saw that? It dedicated all of NVIDIA to that. And then it became about intelligence. That's a mission. And then for Elon, all of his companies

1:18:41 Are really mission oriented. PayPal was um about making local At the time commerce frictionless and reducing the tax from, you know, all this crazy fintech systems and payment flows, reducing that tax all the world. And I think that tax has been massively reduced, not necessarily because of PayPal, but because of PayPal and a lot of other companies like them.

1:19:01 And then Tesla. was really about making the world sustainable. And I think between battery storage and pulling EVs forward. Tesla, the world was always gonna run out sunlight just because of

1:19:13 Economics. I mean, forget emissions or concern about the environment. Economics were gonna dictate an emission free. World. Apart from rockets, because you cannot get literally out of the earth gravity well without chemical propellant just from a

1:19:26 physics of energy density perspective. Tesla's done a lot to I think make the world more sustainable. And It was just so striking to me when I first started meeting with Tesla executives back in Two thousand eleven, two thousand twelve. They were at our so mission oriented about making the world more sustainable and environmentally friendly. SpaceX, too. It's incredible if you talk to all of them. I mean it's just wild. Yeah, the bartender at SpaceX. If you ask the bartender at the Tiki bar,

1:19:51 You're right I know each engineer's drink. have it waiting for them at the end of a long Day not that they're having drinks that often like maybe that makes sense. Yeah. But if everybody in each company and all this company they can say The mission.

1:20:06 And they all say it with this messietic zeal. I think you get better employees, and then XAI. Is about being dedicated to objective truth and new scientific advances. And if you have these missions. And you're competing A lot of these The world's best minds have spent twenty years trying to make people slightly more likely to click on one link than another.

1:20:26 And if you have these missions, you get better employees. So I think that mission focus is a really And exceptional teams. And then those exceptional teams wanna work with people like in Ilan and Jinnsen. Because they know if you're twenty five years old and exceptional, you go to one of those companies. And you happen to be in a critical path of a critical problem, you're gonna work directly with them.

1:20:47 And there's probably no other company. Where that is true. That has more than, I don't know, pick a number, five thousand employees, a thousand employees, I don't know. So I think that mission orientation There's no need for hierarchy.

1:21:00 You get really exceptional points. I cannot tell you how exceptional the teams are. And Elijah Jetson's companies are. Yeah, I mean I've met A lot of the space X People

1:21:11 for a bunch of different reasons and All of them have the same zeal, intensity. mission focus. It's absolutely remarkable. And a lot of them left and then came back very quickly because there was no other environment they could find quite like it. Maybe like the most fun place to close our conversation today. I could literally do this with you for seven straight hours. I got to like one third of my questions and your passion for markets and technology is just so palpable and amazing.

1:21:37 My last question is about Our business. How do you think The investing process. Investing firms

1:21:45 Edge Alpha will evolve. Against this backdrop that we spent the last two hours discussing. whether it's the internal tool you talked about or the one we're building, I'm sure everyone else is building something. It just seems like Holy cow.

1:21:58 And this has I guess been the history of our businesses. It keeps getting more competitive, but How do you envision it? five, ten years from now as such a passionate practitioner. Public equity, so we must play video games. The meta of any given competitive video game is always changing based on how the game deciders balance it.

1:22:14 So if it's a PvP game, you might go from a Cyper meta to a shotgun beta. sports are similar, but there's a fundamental truth, which is you want to have the highest K D ratio in a Sure based PvP. A more P VE environment, you want to have The highest or the fastest clear of the most difficult

1:22:30 Apex it game activity. And the meta of how to accomplish all those is changing. Sports is the same. The N BA evolves from a mid range jumper meta when Jordan was in it to a three point meta. And it'll evolve back again. And this is competitiveness in the meta, but there's

1:22:46 Still an underlying truth of the BA, which is you win by scoring more points than your competitor. The meta of investing, both public and private, I think is always changing. Teaches much faster in publics. There have been all these huge meta shifts since I started the business. R O I C

1:23:02 was a revolutionary concept in the mid nineties. Everybody knows about our IC. ROIC was just a big improvement on ROE. Which in some ways focusing on our OE But argue is

1:23:14 Buffett's greatest contribution to the field of investing, more than the four filters, et cetera, et cetera. But ROIC was a big meta shift. Reg F D and how that changed. Company communication, that was a better shift, but one that I think was awesome. And from my perspective, Reg F D means

1:23:31 There's Almost no reason to talk to companies. 'Cause I had kind of a unique experience. Like I was at Fidelity for eighteen years. It is very funny to me when hedge funds are small asset managers and I define small as let's say at under five hundred billion.

1:23:45 Say oh we have an access advantage with public companies. No you don't. Oh you don't. If you add up every hedge fund's access. All of it.

1:23:55 It's a fraction. of what firms like Fidelity Capital did T Row have. It is many orders of magnitude. And what was awesome to me, I had you're off. And I just realized

1:24:05 They never say anything that is not in a transcript. And I went back and read all the transcripts and things that I thought were like big insights through transcripts. Human beings, we have a big IO problem. It's BCIs. And The I for me of reading, the I being the input my input speed for reading is probably something like

1:24:24 five to ten times faster than humans can talk. And my input when listening is bound by the output speed of my partner, so extremely low bandwidth form of communication, literally. In terms of bits or bytes of information per second. Like I think that and having transcripts of everything, that was a big better shift. Credit card data was a big better shift.

1:24:42 Expert transcripts have been a big better shift. I think all the libs are gonna be the biggest meta shift. That I actually think they may be hardest. or purely quantitative investors. Because I think there's going to be a period of five to ten years. It's very clear that ribazads, bridgewater

1:25:00 They probably realize the importance of data compete before do a lot of other people. They're the only people who can compete with these tech companies. You can throw twenty or thirty billion or forty or fifty billion. at like a leading AI researcher per year. And they quartered a lot of data sources.

1:25:16 They have data no one else has. And they're throwing more compute at it than anyone. But I am optimistic that human fundamental investors.

1:25:25 The tool I'm using is It's called Intel Pro, by the way, if anyone's curious. I think right now the only alpha left in the market for fundamental investors. at the edge of probability because These quantitative investors. Why is there stop being why is there no dispersion anymore in value strategies?

1:25:42 Well, because value strategies, the alpha from them was based on human emotion. And people being embarrassed to buy stuff and afraid to buy stuff and taking career risks. Well algorithms, they don't have any of those. And that's why there's not as much alpha. They're just really simple value strategies that worked amazingly well until oh twenty years ago when quantitative investing took off and they squeezed the alpha out of value strategies.

1:26:05 That's not to say that value As a factor. Kid art. perform really well, it'd be the best performing factor. There's just not a lot of dispersion within.

1:26:14 the value factor. Because of algorithms. I am hopeful that there is a five to ten year period where if you're a fundamental investor like me who can get a slight edge on maybe future probability states. Over what's discounted in the market.

1:26:29 to deep to me knowledge and try not to have any biases that I can could buy what I have With this tool. with years of data that you know, I'm just so grateful we have a vector database with years of data.

1:26:43 I think it is going to enable me and other fundamentally based portfolio managers to probably benefit from a lot of the things that these firms like Rizzots have been benefiting for this just a long way of saying what LLMs do, what AI does. Is it means

1:27:00 The human language is the programming language. And now because of these LLMs are very sued. I'm going to be able to program Effectively from an effective investment effectiveness. like the same level as one of these fifty million dollar a year people are close enough. And then combine that with my own. Unique.

1:27:18 set of data and domain knowledge and Bye bye. And I'm hopeful that there's like a five to ten year Period here. Where

1:27:27 Fundamental investors share of the alpha in the market. goes up. At the expense of quantitative investors. That's my hope. We'll see if that happens.

1:27:38 I don't know if it's gonna happen. Going back to that highband. Future chests, sit our chest, like I hope I have a five to ten year period here. You know.

1:27:46 Where I could prosecute that. Very quickly in the world of venture, I think ventures could evolve outside of pure series A specialists. Victor is gonna evolve serious see it up is gonna evolve the way of private equity. Mainstream private equity. There's no sourcing advantage. There's no pricing advantage. In fact, there's a pricing disadvantage. Because the highest bidder wins. You're just the highest bidder. Everything's an auction.

1:28:09 Literally every deal that these big firms do is an auction. So where they compete to differentiate is an operational value add. And I think that is where C And uh

1:28:21 growth equity is heading. And it's not just it it's true. operational value add. You know, I've spoken about this, you know like I think Valor does this. Baller does it, yeah. really do it, you know, really hard operational problems.

1:28:34 It's not just helping with HR, helping with PR. It's not the LP window dressing operational support teams. It's real operational support. And I think that is where the world of growth equity is heading it evolving. But that's almost because of It's somebody's a combination of the rise of crossover funds and mega funds that would have happened absent Libs. I do think L Libs are just gonna make um

1:28:58 The knowledge part of venture. So much more democratized, you I think about IQ, EQ. And then there's KQ, knowledge quotient. Yeah no It's gonna really help to have a really deep differentiated

1:29:11 Domain knowledge database. But you're gonna have to work hard with an L. L to take your IQ for whatever it is up thirty points because people You had but twin your combination of IQ and KQ. But y you had thirty points on'em, now they're at your level unless you use an LM. But I just think it will really for a while.

1:29:28 Place the emphasis on JQ judgment quotient. Almost maybe for A's and B's the most important skill will simply be assessing team quality. And it maybe that's that five to ten year window where VCs can still differentiate in the same way fundamental. investors maybe I hope will be able to take some alpha share from quantitative investors'cause of L Libs.

1:29:48 But just it will be all about JQ at the C, the A, and the B, and some combination of JQ and EQ. Is this person exceptional? But I I don't know. It's just a hypothesis. Fascinating stuff. Gavin, you're one of my favorite investors to talk to. I have loved today's conversation. I love how specific and detailed it was. You're also one of the most passionate investors that I've ever met about your craft. This has been such a total pleasure. Thanks for your time. Thank you, Patrick. This is awesome.

1:30:14 Um If you enjoyed this episode, check out Join Colossus.com. There you'll find every episode of this podcast complete with transcripts, show notes, and resources to keep learning. You can also sign up for our newsletter, Colossus Weekly, where we condense episodes to the big ideas, quotations, and more, as well as share the best content we find on the internet every week.

What you see above is a preview of the first minutes. One unlock costs 10 credits and covers this episode forever: full segment and word-level timestamps on this page, plus .txt, .srt, .vtt and word-level JSON downloads, as many times as you like.