Andy Cross (2:52)

And, and Steven and then Jose. We'll bounce back and forth our questions here. My follow up question is if you go back 10 years, Nvidia has gone through all of the cycles around different technologies it has supported with its GPUs, gaming, crypto mining, the list is almost endless. And then AI, of course, over the last really two or three years. But when you go back and look over the history of Nvidia, does that surprise you or is it and, and what is in the DNA that got Jensen and Nvidia where it is on the AI vanguard?

Stephen Witty (3:25)

Here's the link, and it's a little subtle. In the 90s, Jensen and his team realized that there was going to be infinite demand for 3D graphics rendering for video games. No matter how well you rendered the sprites, no matter how well you render the characters, the customer is always going to want it better. So you couldn't throw enough computing in it. There was no limit. Jensen saw that and for 10 years he looked for another application that had that profile, okay, and they did a bunch of stuff. They did like science stuff, they did high academic computing, as you said, they did crypto, but all of those. The demand was ultimately kind of satisfied very quickly. There wasn't a huge market for it until AI came along, intelligence came along and Jensen and his team said, here is the other application with infinite demand. No matter how much computing power we throw at AI, we think the customer will always come back to us and ask for twice as much or even 10 times more. And that insight turned out to be correct. That's exactly what happened. And so Jensen was always hunting for that. He didn't know it was AI specifically, and I think that came as something, as a surprise to him. But the moment AI arrived, he pivoted his whole company toward it because you could see just like 3D graphics. I'm never going to satisfy the customer here. No matter how fast these chips go, no matter how much hardware I build, the customer is going to come back to me and ask for 10 times more.

Andy Cross (4:44)

Mm.

Stephen Witty (4:46)

So that's the link.

Andy Cross (4:47)

Yeah.

Jose Najaro (4:47)

Thank you, Steve. Jumping to another question here. I'm pretty jealous, right. Because you got to. To spend a lot of time, it seemed, with Jensen and his team and something as Nvidia investor. I. I can only hear him during a earnings call or kind of like just random interviews he does here and there, but kind of jumping almost to the end of the book. You talk about kind of this last chat, the last interview you had. Yeah, in the book, and it was kind of not as pleasant as the others.

Stephen Witty (5:16)

Jensen can be kind of hard to.

Jose Najaro (5:17)

Be around, and it definitely seems a little bit more on, like, the negative side of the experience. Like you mentioned, he's tough to be around with. Did that have any big impact on the way you saw Nvidia in forms of management or the company after that experience you had with him?

Stephen Witty (5:36)

I knew that Jensen had this aspect to him from interviewing so many people around him. I just never thought I would see it myself because he's kind of guarded around the media. But basically, Jensen is different, in my opinion, than most Silicon Valley executives who are driven either by greed or maybe ego or just a vision for the future. Jensen is totally driven by anxiety. He's completely afraid that Nvidia will fail and that he will be disgraced. Like, it's the. It's totally how he motivates himself, totally from negative emotions. So he. Even when Nvidia was literally the single most valuable company on the planet, he's sitting there thinking, oh, my stock's going to go down and my firm's going to go bankrupt. And there were times in Nvidia's history where, in fact, this did almost happen. So it's not a completely academic thing, but it's just how he motivates himself. He's constantly beating himself up, saying, I'm not good enough, my company's not good enough, and what we're producing isn't good enough. And this is true even when they dominate the entire space. So it's kind of a very unusual thing. It's really a different way to motivate yourself. I think it's hard. I think it's hard on Jensen, and it's hard on the people around him because he has this kind of fiery temper. He will explode in anger if he thinks that his executives or even the people around him just aren't prepared. And he'll even do that at people outside his company. He's done it to me. He did it to, you know, I think he's done it to other executives in the field. He's got a bad temper. Now for Jensen's point of view, you know, he told me my mind is racing, you know, and I'm thinking faster than I can put words into. Into kind of like I think my can't put my thoughts into words fast enough. And that comes out as anger. And maybe that's true, but I must say it also seemed a little bit self indulgent.

Jose Najaro (7:22)

There's this interesting topic that in the most recent gtc, Jensen talked about a lot, Steven, and it's what he calls AI factories. So when he mentions AI factories, this is completely different from your traditional kind of cloud infrastructure where your typical cloud infrastructure is meant to host data, to host information, to run applications. This form of AI factories only have one job, is to just generate your tokens. That's all running 24 7, generating that thinking energy of AI. And this is something he mentions is completely new. The market, they're building it themselves. It's always hard to kind of think of something new being developed, especially kind of in this. I, I feel even though AI has been here for such a, for, for a While now, almost two, three years from, from the ChatGPT moment, it still seems like sci fi to, to some.

Stephen Witty (8:14)

Yeah.

Jose Najaro (8:15)

Your experience, what, what do you think about these AI factories? Is this something that's really, really going to be happening or.

Stephen Witty (8:20)

Oh, it's happening already. No, it's real. So what happens when you put a request into ChatGPT? What happens? So what happens is it takes your request on a data pipe, it shoots it off to some industrial warehouse on the edge of the city somewhere, which is filled with racks of Nvidia hardware. Right. It's filled with 10,000 GPUs or more, all acting in concert. It takes your request. Let's say we want to make ourselves look like a Princess Mononoke Studio Ghibli avatar. That's actually an expensive request. It requires a certainly large amount of computing power, enough to run probably a microwave oven for an hour. It processes that request inside the industrial warehouse, inside the kind of Nvidia computing stack. And then, yeah, as you say, it returns a bunch of tokens which are used to generate this image. So data goes in, requests come in and tokens come out. It really is a factory and they already exist. I've been inside one of these things. There's no humans on the floor, it's completely sterile. They're liquid and air cooled. The supercomputing equipment itself is inside kind of sealed pods. It's very loud because there's hundreds of thousands actually of fans spinning all the time. You can barely hear yourself think. These are the thinking machines, as I say in my book. I mean, these are the things that think for you and they take your request and they give it back. They are AI factories. Now Jensen might be compared to Thomas Edison. I think it's not an inaccurate comparison. That's who I think of when I think of Jensen and building these things.

Andy Cross (9:48)

Which is just very quickly. GTC stands for GPU Technology Conference. I think it's their big. It's become the big AI conference that, that Nvidia hosts every spring, I believe. Is that right?

Stephen Witty (9:59)

Yeah, that's right. Initially it was not an AI conference. They were actually looking for any kind of use for these GPUs. They didn't know. Imagine that you built a propeller with an engine attached to it and you were like, what do I use this thing for? For like 10 years before the airplane was invented. What is this for? Who could use this? And you're just going out to everyone with your airplane propeller and an engine attached to it. Does anyone want to build an airplane around this? That's kind of like what, what, what Nvidia did.

Andy Cross (10:24)

Stephen, tariffs are just such in the news. So I do want to spend a little bit of time about this. When you think about this as a, as an analyst investor yourself, but you look at Nvidia, what you learned from the book or just thinking about the environment with Nvidia, where do you see Nvidia into the tariff mix these days? Right now?

Stephen Witty (10:41)

Yeah. So I was just in Taiwan and they are terrified of these tariffs. It's a sense of dread. It would really have an incredibly negative impact for them. Now there was a carve out actually for semiconductors in the original tariff plan, but even so, if you look at Gensys presentations at Computex, the big Taiwanese computing conference, he'll list like 40 or 50 Taiwanese suppliers. Nvidia doesn't make anything, they don't manufacture anything. It's all outsourced. They're really just a very cutting edge R and D laboratory. Basically most of their components come from Taiwan and that's not easy to replicate in the United States. Personally, I think these tariffs are extremely misguided and bad for everyone involved, but who's going to argue with Trump about it? So it's not a good thing for this company if the tariffs are imposed. Nvidia earns a 90% profit margin on its most advanced equipment. So they can afford to pay and they will pay if they have to. I mean, they'll pay the tariff, I don't think. I, I think the cost of production and trying to onshore to the United States is just more than 20. Well, more than whatever the tariff is, 30 or whatever. So they'll just pay it. But it will hurt their bottom line for sure.

Andy Cross (11:48)

Do you see it from the investing side? Do you see that as the biggest risk that's facing Nvidia, let's just call it this year, or do you see it more around the AI data center spend?

Stephen Witty (11:58)

So I will tell you what I think the biggest risk was. The biggest risk was that Jensen could not sell his equipment to China. And this created an opportunity for Chinese vendors to create a second Nvidia. A low cost Nvidia. Right. That doesn't earn a 90% margin. That earns a 10% margin and destroys Nvidia's economics. And Jensen was terrified of that, I can tell you.

Andy Cross (12:20)

Yeah.

Stephen Witty (12:20)

Now, most recently, perhaps you heard, you saw this, Jensen paid $1 million or Nvidia paid $1 million to be at some kind of fundraising dinner with Trump. And what's the first thing that's announced after that? Oh, yeah, well, we're going to delay the export ban on these chips to China. That Jensen's biggest concern. He is terrified of someone in China building the stack again because they can do it. He's always had a fear of Asian manufacturers knocking off his equipment. This goes way back. And in fact, it was kind of the whole impetus for CUDA in the first place. He was making the state of the art 3D graphics hardware. But he said, you know, if we're not, what are we? What is this company? We don't make anything. We're just a bunch of guys in a laboratory. So if we're not constantly innovating new products, manufacturers in Asia are going to knock off what we're doing and compete our margins down to zero.

Andy Cross (13:07)

Yeah.

Stephen Witty (13:08)

And so they have to stay on the front lines of kind of that and they can't. It's bad for Jensen if he cannot sell his chips to the world's largest market.

Andy Cross (13:16)

Yeah.

Stephen Witty (13:16)

I think also the, the deep sea experience has shown there's no point to this ban anyway. Like, what are you going to ban? It's not going to stop Chinese innovation in AI. It had no real impact on that. So it's counterproductive and it could cost Nvidia its leadership position. I think that was his biggest concern. You know, Having said that, the tariffs are bad for Nvidia, they are a huge business risk. If Trump got his way and put a 35% tariff on Taiwan, even if semiconductors were exempted from that, Nvidia's cost of production would go up a ton. It would eat into their margins. Now if any company in the world can afford little margin reduction, it's Nvidia. But still, you know, the stock price would go down 100%.

Jose Najaro (13:59)

Discussing kind of current fears, one being the tariffs, the other I keep hearing is kind of just the fear. And I feel like we've heard this since this AI market boom spending started. It's just the fear of overcapacity being built. Is that something that maybe through your experience writing this book or just like you mentioned you were just in Taiwan, is this something that you would consider a yellow flag or a red flag at the moment?

Stephen Witty (14:25)

I've heard people talk about the capex cycle. I don't think so. I don't think anything is going to stop even a recession. I don't think, I mean yet it'll impact it somewhat but. But if you look at a 10 year projection, just about how much capacity computing power AI is going to need, how many power plants we're going to build. I mean Facebook is trying to co locate with a nuclear power plant to build its AI data center. I know, yeah. Maybe timing issues related to capex over a period of a few months might affect the stock price. Sure. But I think over the long term we're at the beginning of a gigantic capex cycle and trying to manage around order flow every few months for that capex cycle is maybe to miss the, miss the forest for the trees. I think the longer term trend is practically vertical for this stuff and I don't see any reason why that wouldn't be the case. It certainly could be the case now that a cheaper competitor could appear or that Nvidia could even miss a production cycle and flub something. But the scenario where the AI ultimate like kind of inference demand for consumer or businesses doesn't materialize for some reason seems relatively remote to me. I mean I think these products are delivering what people have said they will deliver. I think they are a pretty special product. Every academic computer scientist I talk to views this is like a civilizational advancement. If we were talking about kind of the dawn of the era of electricity or the dawn of the era of the Internet, sure, probably in a month to month basis there's some capex stuff to talk about. But if we're talking about the 20 year trend. It's, it's a meaningless blip in the longer term of what's going on.

Jose Najaro (16:03)

AI is just moving in so many directions and AI is just this huge umbrella that you can kind of trickle into all different types of markets. Three that Nvidia really kind of talks about are one kind of robotic space, AI and robotics being your autonomous humanoid robots, industrial robots, or even kind of autonomous vehicles. The second, I want to say, is the healthcare industry. The healthcare industry is one that Jensen and I forget. The VP of Health always says a lot of how that could be such a massive market opportunity for Nvidia. And then the third is just kind of like your software AI agency solutions.

Stephen Witty (16:42)

Yeah, first of all, you nailed it, Jose. This is exactly correct. The big question is, all right, where does the next $3 trillion in market capitalization come from? What could possibly be left to build? But I think Jensen's perspective is, yeah, there's areas that are just waiting to be revolutionized. One, as you mentioned in healthcare, and there they have like 30 initiatives, right? There's, there's diagnostic initiatives, metal, medical imaging, drug discovery, this kind of thing. I think the most interesting thing I heard about is basically using neural networks to essentially build almost a biological compiler that runs over. If we can think of the kind of elementary nucleotides of rna, I think it's ACG and U. I didn't go to medical school, so if I got that wrong, let me know in the comments. But I think those are, but those are the zeros and ones of programmable biology. And so many people want to build essentially a biological compiler that runs on top of these nucleotide bases and then builds bespoke custom drugs and maybe even like replacement tissue for damage. The possibilities are just incredible. Intermediating that making that happen is going to be some kind of neural network somewhere and it's going to require an incredible amount of computing power from Nvidia. So Nvidia wants to be. Just as they grew the AI platform around it, they want that vendor lock, they want to be right in there. When you're building a system like this, you go through the Nvidia toll booth, they're trying to be there. The other one, as you mentioned, huge, and this is, I think Jensen's biggest focus is robotics. So let's say Fei, Fei Li, Stanford academic who actually ran the neural net competition that really broke through Nvidia and neural nets, her new thing is robots. And so what she did is she created a survey. Survey had one question, how Much would you benefit if a robot did this for you and they took it to a bunch of people? And the number one task that people would benefit the most from if a robot did it, I think was washing dishes. And then number two was like washing the toilet or cleaning the toilet. Number three is cleaning up after a crazy party. So these are the tasks that people least want to do and would be the most profitable for a robot to do if you could sell a robot to do this to consumers. So how do you train a robot to wash the dishes? Well, if you try and do it in the real world using a neural network strategy with like reinforcement learning, you're going to break 10 million dishes along the way. You're just going to, it's going to be, you're going to, the sink is going to be a mess. It's going to be so expensive to do that. What Jensen wants to do is build a digital gymnasium, a kind of high fidelity physics simulation that the robot can learn to wash dishes in. This is called Omniverse. Perhaps you've heard about it if you followed Nvidia. This is what Omniverse is. It's a basically robotics training platform, a high fidelity reality simulator. In that simulator, the Robot can break 1 billion dishes, right? Who cares? It's all digital. Then when the brain is trained, the neural network is trained, we download that and stick it into a real world body and send that off to work at the sink. And it's not going to break anything because it learned in this digital gymnasium. Now Jensen will charge his robotics customers a very large amount to participate, gym membership, basically, which as you know, has fantastic kind of subscription economics. And he'll say, well, I'm saving you money. Look at all the money I saved you. If you didn't do it this way, you'd be, you know, have a laboratory with 10 million shattered dishes in it. You don't have to do that now. That's hard. Getting the physics right is hard. It's not just a game simulator. We've got to get very, very precise tactile physics. We've got to get like fluid dynamics right, because the dishes are going to be wet. We've got to get a whole lot of things right, you know, surfaces. It's not easy, but I think Jensen feels like, all right, so we do that and, and then we're going to be at the center of the robotics revolution. That's another trillion dollars in market capitalization for us because that's a trillion dollar industry and we're going to earn A multiple on it. So those would be the two things I would. I know Jensen is paying a ton of attention to these things. I asked him, what looks like Cuda? What's the bet you're making now that looks like Cuda used to look like 10, 15 years ago. And his first answer, write up one word. Omniverse. This is his next Cuda. I would pay attention to that, see if they're actually getting customers. I mean, right now it's, you know, it's an idea. I mean, it exists. They don't have a huge customer base right now, so it's kind of conceptual, but they're working very hard on it.

Andy Cross (21:09)

What about companies that are building their own silicon, especially clients of Nvidia? Where do you see that on the risk spectrum?

Stephen Witty (21:19)

Yeah, it's a great question. I at first was like, well, why don't I just knock this off? I'm gonna go with AMD and just make the same thing.

Andy Cross (21:27)

Sure.

Stephen Witty (21:27)

And J. Jay Prabhu, who designed circuits at Nvidia, was like, yeah, they can. They can make silicon just as good as we can. There's nothing that we can build that they can't just pry open the lid and look at with a metallurgical microscope and, and rebuild it. Right. There's no, there's no trade secret in the metal. The trade secret is these software development kits that we build. The trade secret essentially is that, you know, they're not as good as. As being on the front lines with the scientists and building them tools as Jensen is. And they don't have that same sense of, oh, my God, if I don't build this scientist a tool, I'm going to die, my company's going to fail, and I'll be disgraced. Jensen thinks about things, but do you.

Andy Cross (22:09)

Think there's an advantage just if they were going to build it just for their specific uses? One of my thoughts was maybe there's an advantage. And Jose, jump in here. Maybe there's an advantage if they are building it just for what they really needed to do. Where Nvidia is building it, tied to Cuda and tied to lots of other clients.

Stephen Witty (22:28)

Yeah, sure. But remember, if they're doing it just for what they need to do, that's not a large market. Maybe carve out 1 to 2% of the market. You know, Google had its TPUs.

Andy Cross (22:37)

Yeah.

Stephen Witty (22:38)

For a while. They haven't made a great amount of traction. AMD was trying its approach and mostly are not successful. So a whole wave of this cycle has kind of already happened. Once and Nvidia was really just almost completely unaffected by.

Andy Cross (22:52)

Yeah, I'm just thinking like Microsoft or Facebook Meta, Meta especially maybe you know.

Stephen Witty (22:58)

The integrate outside of Apple. These companies don't have the DNA of that kind of hardware manufacturing.

Andy Cross (23:08)

Okay.

Stephen Witty (23:09)

In a certain sense, one of the longer term lessons of computing Apple accepted. But one of the longer term lessons is it makes sense to separate software and hardware. It makes sense to have your hardware stack built by someone else and to focus on kind of like your competitive advantage. I don't think Facebook's competitive advantage is building microchips and I don't think that's going to change. I don't think Microsoft's competitive advantage is building hardware and I don't think that's going to change. Now. I can be wrong. And Apple in fact has built silicon. It works great and it is integrated tightly into their product. They don't use Nvidia stuff, so maybe someone else can do that. But you got to remember, most of what Nvidia sells is basically it's very easy to swap it in and out of the data center. They're all on racks. I can just pull out the rack and stick in another microchip. Right. So there's not really any kind of stickiness happening at the physical level like there maybe would be with it. With it. It's all modular hardware that's highly commoditized and can be, well, not commoditized, but it's modular hardware that can be very easily replaced. Right?

Andy Cross (24:13)

Yeah.

Stephen Witty (24:14)

Nvidia, though, operating that environment has done just fine. So I think the fact that you can pull the rack out of the data center and replace it with a different chip, very few people are doing that right now. So I don't, I don't see that in the immediate horizon. I think also Nvidia benefits for the same. Let me say this. Jensen learned a ton from TSMC and its founder Morris changing. And one of the things he learned is do not compete with your customers. So the reason TSMC succeeded with its foundry business and the reason Samsung did not do as well is people were paranoid that if they manufactured their chips at the Samsung Fab, Samsung was going to steal their idea. Yeah, but TSMC wasn't selling any chips, they were just fabricating them. Jensen's hardware platform. Jensen is not going to build the next version of GPT. He'll build one internally just to test it. But he's not attempting to compete directly with OpenAI because he knows if he does, then his customers will be incentivized to come back and compete against him if Microsoft builds a training chip. And at the same time they're backing OpenAI, it's not an open platform. And for other competitors, they're not going to want to use Microsoft's product. Right. They say, well, I'm going to train my thing on Microsoft's product and then Microsoft's gonna steal my idea and wrap it in a GPT. No way.

Andy Cross (25:37)

Yeah, right.

Stephen Witty (25:38)

And so I think this is part of Nvidia's competitive advantage as well.

Andy Cross (25:41)

What was your biggest shift in your thinking about Nvidia as you were starting thinking about the book and as you ended the book and finished the final word?

Stephen Witty (25:54)

My biggest shift in thinking was personal. So I was originally like, I'm toast, I'm cooked. Like, I can't be a writer anymore. ChatGPT is going to write in two, three years. It's going to write better than I am. It's going to just produce books that are better than mine on demand. In five minutes. It's going to take whatever I do. I'm going to feed my interview notes into the knowledge engine and it's going to produce a bunch of tokens and those tokens will be a fantastic best selling book. That still can happen and maybe even that will happen. But I've learned to think differently about that. So now I think about it this way. And this was from exposure to Jensen and watching how he thinks about things. I did something like, you know, 200, 300 hours worth of reporting it. Just interviews for this book, right? And maybe 1% of that knowledge actually makes it into the finished product and the rest is just sitting in some, you know, database somewhere, maybe what should happen. And I have to do this. Like I'm constantly having to decide what am I going to put in the book? What does the general reader want? Who is the general reader? I'm trying to guess what the reader wants to read about. But what if the reader came to me and the reader said, well, listen, I'm an Electrical Engineer with 10 years experience designing microchips or I manage a portfolio and I need to know more about Nvidia's stock price? Or I'm a teenager and I'm interested in this field, I don't know anything about it. And then the AI wrote the book on the fly to meet the demands of this particular customer. And so the book stops being this kind of static paper document and evolves them to something like, more like a knowledge base that you can query. But maybe my voice is in there in some way, too. And maybe there's still a compelling narrative that brings the reader through the book, but also customizes or tailors certain sections to meet the reads of the needer, kind of in real time. So this is how I think about this now. The other thing I have observed is that chess has long, long super surpassed humans. And yet very few humans are interested in watching two computers play chess against each other, which is completely incomprehensible. And in fact, weirdly, this has actually turbocharged the personality driven aspects of chess. It's become this kind of, like, actually more popular than it was before. The computers beat it. And now rather than kind of thinking like this and being boring, you have these, like, chess personalities or like, twitch streamers. They're fun, you know, Magnus Carlsen has leaned into this and become almost like a celebrity.

Andy Cross (28:15)

Yeah. Scandals, drama. Scandals, Drama.

Stephen Witty (28:18)

Yeah. I mean, they kind of always existed to some extent, the chess world, but now it's like. Or also like, you know, hot people playing chess. Like, come on. That did not exist before. I don't think too much. So maybe even if the computer can write a better book than me. Yeah, maybe no one would read it. Maybe they still want a person, an author behind the book. And I'm growing a little more comfortable with this to some extent.

Mary Long (28:53)

As always. People on the program may have interest in the stocks they talk about, and the Motley fool may have formal recommendations for or against. So buy yourself stocks based solely on what you hear. All personal finance content follows Motley fool editorial standards and is not approved by advertisers. The Motley fool only picks products that it would personally recommend to friends like you. For the Motley fool money team, I'm Mary Long. Thanks for listening and we'll see you on Monday.