How NVIDIA Became The Kingmaker of AI
Stephen Witt is a Los Angeles-based investigative journalist and author whose work has appeared regularly in The New Yorker.
His latest book, The Thinking Machine, chronicles the remarkable rise of Nvidia and its enigmatic CEO Jensen Huang—from immigrant dishwasher to leader of the world's most valuable company.
Published in April 2025, the book reveals how one man - Jensen Huang made modern AI possible.
Stanford's former computer science chair Bill Dally noting, "Without Jensen we'd be ten years behind."
Witt's previous book, How Music Got Free, explored the digital revolution that transformed the music industry and was later adapted into a television documentary. As both an author and television producer, he specialises in uncovering the human stories behind technological upheavals that reshape our world.
TIMESTAMPS:
00:00 Trailer
01:00 How Stephen Got To Jensen Huang
06:35 Reading In An Age of Social Media
08:15 How NVIDIA Became So Valuable
12:25 Jensen Huang's Early Career and Its Impact
13:48 Jensen Huang's Early Career
17:15 The Secret of NVIDIA's Success
18:06 Jensen Huang's Unique Management Style
20:41 How LSI Made Him
21:51 Why NVIDIA Didn't Make Financial Sense
24:52 How NVIDIA Made The AI Transition
30:58 How Jensen Leads
34:35 How Jensen Evolved Into A Business Leader
37:38 How Jensen Knew To Double Down on AI
43:53 NVIDIA's Competitive Edge
47:36 Current A.I Bottleneck
48:23 AMD vs NVIDIA
50:52 NVIDIA in China
52:25 Intel's Downfall
56:05 Should We Be Worried About AI
60:43 Advice for the Next Generation
This is the 50th episode of The Front Row Podcast.
Transcript
Keith 00:00:00
Today I have the pleasure of speaking with Stephen Witt. Stephen wrote one of the best books of 2025, a book about the man of the hour, Jensen Huang. It's called The Thinking Machine. And the biggest takeaway I had from the book was that in a gold rush, the best position to be in is not just to sell shovels, but to actually have a monopoly on the shovel production that you will eventually sell to the shovel merchants. And that's the key thing that stuck in my head - to invent the shovels.
Stephen 00:01:28
Yes, to invent the shovels. And with that, Stephen, thank you for coming on.
Keith 00:01:33
Of course. There are a lot of aspiring authors out there, journalists that will want to do an interview series or even do a biography on Jensen Huang because he seems to be the man of the hour. And part of me wonders, how do you get such unprecedented access?
Stephen 00:01:53
Great question. So, a couple things happened. One, I was writing for The New Yorker, which is a very prestigious magazine here. And so, that got me in the door, but at first, I only had one hour with Jensen.
I like to do a lot of research before I go talk to anybody. But in Jensen's case, I did like four or five times as much research as I normally do. I read basically every news article I could find on him. I went on YouTube. I watched videos going back 10 or 20 years, sometimes to like 2009 or 2008.
And I said to myself, Jensen's time is probably the most valuable commodity on the planet. He's the CEO of the world's most valuable company. So the amount of money that he directs each year, it's like $250 billion in capex and in long-term planning. So if you do the math, roughly like one minute of his time is worth $1 million. So I said to myself, I just can't waste his time. I can't ask him any question any other journalist has ever asked him before.
And so I made a list of questions that were all things by doing my research that I could tell that Jensen had not been asked before or had seemed not to have been asked before. One time in 2009, he'd mentioned on a Taiwanese news program that he once almost died in a car accident, but then he didn't follow up on that and the interviewer didn't ask any more questions. So my first question to Jensen when I sat down with him basically was, "Tell me about the time you almost died in a car accident." And he'd been kind of sitting like this, it was clear he didn't want to be interviewed. And then he sort of perked up and he looked at me and he was like, "Oh, you heard about that one, huh?"
For the next hour, I only asked questions of Jensen that I was pretty sure no one had ever asked him before. And this turned out to be the perfect thing to do because Jensen's time is extremely valuable. So, you can't waste his time asking him questions he's already been asked before that you can find the answers to elsewhere. And it showed that I had done my homework. And I didn't know it at the time, but this is the thing that Jensen values the most is doing your homework, coming extremely prepared into every situation.
So, that first interview went very well and I was very lucky and I was like, "Oh, it seems like Jensen likes you. Would you like to come back for a second interview?" And I said, "Yes." And so, I kept doing more and more of this. And it was only toward the end when I kept asking him one particular question over and over that he actually lost his temper at me because he felt he answered it already. So the idea was just simply not to waste his time. The idea was simply to ask him questions that were new that only he could answer that I couldn't get from anyone else and that had never been answered before. And in this way I maximize the efficiency of the interview with him. And for an engineer like Jensen, that's just the perfect way to approach that.
Keith 00:04:37
Throughout the whole book, what I've come to appreciate was the depths that you go into, not just interviewing him, but like the entire Nvidia ecosystem. And that part of it is really striking because it's almost as if you try to encapsulate the world that he was living in. So in a sense, this book is not just about Jensen Huang, but it's about the Nvidia story.
Stephen 00:05:02
One question I ask myself constantly is what did the world look like to Jensen at say 1995 or 2008. Remember Nvidia did not become an AI company overnight. Nvidia hired its first AI employee full-time after it had been in business for over 20 years. They were not an AI company originally and they didn't really have that as their focus. They were making cutting edge microchips for video game designers and video game programmers. And it turned out that AI was a use case that they had not originally anticipated. And so Jensen was sort of very smart and very far-seeing in this way that he wanted to open his platform to more uses but he didn't know that AI was coming along.
I would even go to chat GPT and this was something that actually I didn't use chat GPT to write any of the book but one question I asked it was stuff like what was the competitive dynamic of the semiconductor industry in say 2006, what would the world have looked like back then to them not knowing what was on the horizon? And so I framed the book in such a way that you're on the journey with Jensen.
I got very lucky as well because it turned out to be just an incredible story, a great business story, but also a great story of technology, feats of engineering, and even a great personal story. So in that way, I got very lucky as a writer because that doesn't always happen.
Keith 00:06:30
I would say that this is what really got me hooked. Well, this is something I very much did on purpose. And in fact, I did it in my last book, too. I very much do this on purpose because I don't think my competition is other books or other writers. I mean, there are other books and other writers out there, but I think the main competition is like TikTok. It's Instagram.
Exactly as you say. People are, especially if they're reading on their phone, but even with a paper book or listening to it, they're constantly their attention is constantly being pulled in some other direction by some TikTok or by YouTube or some distraction, whatever it is. It's hard to get people to lock in and engage with the book. So, I borrow strategies from thrillers when I write these books. I use cliffhangers and I set up like a central mystery that gets resolved. For this book, it was a little tricky because you know what I had established the first central mystery of the Nvidia story is will this company succeed? Will this crazy scheme that Jensen has to build these scientific microchips with no customers at all? Will this work?
What happened though is that that question got answered in 2012 or 2013 and I had to reset the tension. So then I did a part two of the book was okay, it's going to work, you know, is it going to get too powerful? Is it going to kill us all? And that was the central theme of the second part of the book.
My first book was about the MP3 and that was easier because when I wrote the book, I published it in 2015. That chapter was closing. That era was at an end. For this book, the Nvidia story, the next 20 years could be more amazing and more unbelievable than the first 20 or 30 years. It's almost impossible to imagine what this company's going to look like in 15 years.
Keith 00:08:13
If you ask someone in like 2005 for example that Nvidia would be a $4 trillion dollar company today I don't think many people had that kind of vision. I mean you don't think you would have thought that maybe maybe actually Jensen would have but but inside the company there were some true believers.
Stephen 00:08:31
And the true believers believed the following thing. The classic model that we've used for microchips where we feed it one math problem at a time is not good enough. We have to convert it to what we call accelerated or parallel computing where we're feeding the processor hundreds of problems at the time. And there were visionaries inside Nvidia who were convinced that this was going to change the world, but they were viewed as eccentric and weird and mavericks and they couldn't get any funding for what they wanted to do. And so this is an incredible story. That change made everything possible.
Simultaneously in places like Toronto and France, you had other mavericks who were like, "We think the computer shouldn't follow this logical, orderly instruction set. We think it should work in a more connectivist paradigm, more like the human brain." And neither one of these sides knew that they needed the other one. This is the fascinating thing. The guys building the new kind of microchip didn't know that they needed this new kind of connectivist software. And the guys building the brains didn't know they needed to run it a million times faster. And it was only when they met that they turbocharged one another's capabilities.
It gave us the age of AI, the connection between this GPU parallel computing paradigm which had nothing to do with AI when they were building it and this neural networks that run on GPUs which had never heard of GPUs or Nvidia before that connection was made. So amazingly you have one side building the software of the future. You have the other side building the hardware of the future and they don't know each other. They don't know they need each other. I found that to be the most fascinating part of the story.
Keith 00:10:15
Crazy thing to me was that it seems that especially the scientists they hacked the GeForce GPUs in order to try to actually attempt maybe the first few versions of what we now know as neural network computing or what later became the large language models. And that to me is quite fascinating because you would never think that such a fundamental innovation really comes out of people hacking things in a dorm room despite the fact that we kind of know that already that has happened before.
Stephen 00:10:51
This was so the software platform that Jensen built for scientists was called CUDA. And basically what CUDA was is you took your graphics card, you installed CUDA, and it was like you flipped the switch on the graphics card and it turned from being a video game component into a low-budget supercomputer. And they put this software on every card that they shipped. They put this architecture on every card that they shipped.
And you might ask, well, who is that for? Who is CUDA for? Well, it's not for established research scientists because those guys can already get time on a supercomputer. It's for marginal scientists who can't afford funding for their research who need a cheaper supercomputer because their research is out of favor. Really, it's for like mad scientists with crazy ideas that aren't supported. And that's what neural networks were. It's hard to imagine now, but 15, 16 years ago, neural networks were totally unpopular. Nobody was using them. Even within AI, nobody was using them. They were viewed as a bizarre and dysfunctional technology that only a few eccentric people pursued. And that was true of parallel computing which people found too difficult to program and it was like an overengineered solution.
So the connection of those things, it's really two groups of outcast mavericks finding one another and then building this world conquering technology and now we are living through the technological revolution that has produced.
Keith 00:12:25
When I was reading through your book as well, it isn't just a story about Jensen Nvidia as well, but it also alludes to the earlier days before a lot of these AI pioneers came to be, right? There was a little bit of that when we look at the Geoffrey Hintons and the Ilya Sutskever of the world, you see them today as titans, but you kind of bring us to this earlier time where they were just fringe scientists.
Stephen 00:12:51
They were totally outcast. They were doing like image recognition, right? They would have like a contest of who was doing the best image recognition and you'd have 15 or 16 entrants and you'd only have one neural net entrant. So the other 14 were approaching like different more mathematically complex approaches that had nothing to do with a connectivist approach that was within AI which was already considered like a career graveyard in like 2012. The amount of venture capitalist investment in AI was like I mean it was closer to zero than any other meaningful number. It was nothing. There was no AI industry whatsoever. It was a way to burn money.
Now parallel computing was seen the same way. Nvidia's stock price was in the toilet for decades as Jensen pursued this hardware solution. But once they found each other, I mean it turned into a money printing machine as you say they invented the shovel in the gold rush.
Keith 00:13:49
Maybe bring us back to his early beginnings. If you were to look at the kind of person that Jensen portrays himself to be today, I mean if you look at his LinkedIn profile, people talk about him working at Denny's and that gave him like the work ethic that shaped him to who he was. But there was a 10-year interim where he was in LSI and he was in AMD and he became a top class world-class computer scientist. Can you help me understand how did the 10 years in between shape him?
Stephen 00:14:21
So what Jensen did and what made him special in his 20s was that he would go to the most demanding customers. This was before Nvidia. He was a microchip designer at a company that is long gone called LSI Logic but what at the time was one of the leading microchip companies. And he would say, "Bring me your most demanding, your most difficult task, and I will attempt to engineer with my team the tools that you need to succeed." And what this meant, and this is critical for understanding what Nvidia later became, is that Jensen was always talking to the leading scientist in the world in something. He was talking to someone where if only they could give this guy a thousand times or a million times more computing power, it might unlock some whole new universe of functionality that didn't previously exist.
Now, at that time, that meant 3D graphics. It meant games, but also animation, 3D rendering, Pixar, a lot of that stuff, too. Nvidia has their hands in all of that. They were in the movie industry for a long time. They still are, actually.
So, they would always go to them. Then when Nvidia started, Jensen kept doing that. He would go to quantum physicists. He would go to people doing climate modeling. He would go to people doing like complex nuclear research and he would say, "Okay, what's the hardest problem you have from a computational perspective? Imagine you had a million times more mathematics you could do per second, a million times more arithmetic per second. What would that solve? What kind of functionality would that unlock?"
And then this was Nvidia's real secret. He would lose money to build them that tool. He would build tools that had 10 customers in the world and cost $10 million to develop. It seemed crazy. But the gamble and Wall Street didn't understand it. Only Jensen really got it. The gamble was maybe one of these scientific applications. We will not only have a scientific breakthrough, we will actually unlock an entire new universe of functionality. He called this the zero billion dollar market. I refer to it in the book. And that's what AI was. He didn't know AI was coming. But once he saw that AI was going to be built around Nvidia's platform, around Nvidia's microchips, he actually pivoted the whole company overnight just to sell AI hardware.
And so that is if you can think of what he's doing back at LSI in his 20s where he's like, what's the hardest thing I can do? He's still doing that today. And he loves doing hard stuff because you know competitors won't follow and that's the secret. He doesn't want his hardware to become commoditized by competition as has happened so many times before in this industry so he will lose money to go find some hard scientific research project just because he knows no one else is going to do that. That's kind of his secret.
Keith 00:17:15
I would say it becomes apparent to me that not only was he in a sense a visionary that he was able to try to solve these hard problems but the financial structures disincentivized others from supporting him. So there were I mean cases where you kind of detailed attempted takeovers, people trying to boot him out and I think there was one point when I was reading the book was that the high cost of the R&D wasn't just on the financial performance like there was a deeper problem, right? It created dissension within the company. It distracted him from his core focus on his consumers back then which were the gaming market. So then how did he persist through and not only persist but I guess unite his team to follow him on this charge?
Stephen 00:18:03
Great question and it's the secret of Nvidia. The reason he was able to do this is because Jensen is a world-class computer scientist. An MBA CEO from Harvard no matter how smart he was would not be able to do this because he just wouldn't understand the technology well enough. It is because Jensen understands the microchip from the transistor up that he was able to do this.
In about 2003 or 2004, and Jensen referred to me many times as this was one of the most important hires this company ever made. A guy named John Nicholls approached Jensen and he said, "Listen, for years we've been operating in a paradigm where Intel is just packing more and more components on the microchip and they're relying on Moore's law to double the speed every year. But I have done the math. I have done the physics and I know the electricity of these things. Once they get down to a certain level where they're too small, they won't be able to pack more components on the chip because those components will be so thin like 10 atoms in width that they will start leaking electricity into the surrounding circuitry and the whole architecture will be compromised."
So if you can follow the logic of this, the electrical engineering logic from first principles, then you can become convinced that you have to do Nvidia's approach sooner or later because you're running into the laws of physics by making the components on your microchip so small.
Now, Intel had their heads in the sand about this. They did not think this was going to happen. They thought they were going to find some way around it. But Jensen and Nicholls became convinced because they were world-class engineers that in fact they were hitting the physical limits of what was possible and that made the investment in parallel computing make sense.
At this time Intel's CEO was not an electrical engineer. He did not have that physics background. He was not a dumb guy. He was very smart. He checked all the boxes. He had the MBA. He managed the company profitably. But he did not understand the physics of the component. And Jensen's a stubborn guy. If he gets it in his head that the first principles are correct and he knows the physics and he's a world-class computer scientist and he designs microchips from the transistor up and he's done that his whole career, then he can have the expertise and the clout to take that kind of risk. But only a domain expert can really do it. Not some cowboy, but somebody who understands every part of the chip. And that was Jensen.
Keith 00:20:31
His transformation into this world-class computer scientist was really backed by his that 10-year interim, would you say?
Stephen 00:20:38
Yeah. I mean, I think at LSI, that was his real education. He was getting night classes at Stanford, but his real education was and still is actually, I think he's learning every day, how do I build the world's most advanced microchip for the world's most advanced customer? Even if I lose money doing it, how do I do it?
Then I know I'm always on the physical frontier because the risk for so many microchip companies is not actually that the tooling or machinery or whatever costs too much. It's that they miss some trend. It's that they miss some important scientific concept or trend. And Jensen who is constantly on the front line where the world's like literally personally and to this day will table everything to talk to a sophisticated customer. He'll push his entire calendar away just to have a conversation with some research scientist who has some complex computing problem because Jensen never ever wants to be left behind. He doesn't want to be commoditized. He always wants to be building some high value bespoke technology that he can sell for 90% profit margin because only a few people in the world can even use it.
Intel who used to do that got away from it because to the accountants and to the investors and the money people that looked like it lost money in isolation.
Keith 00:21:58
I guess to a certain extent if you just have a purely financial perspective you can't tell really the difference between an investment and a loss. From a financials perspective makes no sense and for many years made no sense at all. I mean it was and it was reflected in the stock price. The stock price didn't go up for 15 years. People thought Jensen was a lunatic. But he had to have that engineering background to see or to understand what he was doing. The stock prices did not reflect the true reality of what Jensen was seeing.
But at the same time, he was able to attract and concentrate a ton of engineering talent. It doesn't seem to me apparent that those two statements are congruent to each other. Because if I put myself in the shoes of a software engineer or a technical talent and maybe the top 1% my first instinct would maybe be to be working at a company whose stock is increasing over time for the past 15 years and not necessarily join a company like Nvidia say back in 2008 or 2009.
Stephen 00:22:57
There's a great anecdote about exactly this in the book. So Bill Dally who is the chairman of the Stanford University computer science department, probably the single most important academic post in the world at least at that time, chairman of computer science department at Stanford University left to join Nvidia just as their stock price was at like $6. I mean, he was absolutely in the toilet.
And Bill Dally can go to any company. He can go to Google, he can go to Apple, he can go to Amazon, he can go anywhere he wants. He can go to Intel. And he chose to go to Nvidia. And people were baffled by this. And I asked him about it. I was like, why did you, you know, at this time Nvidia, I mean, like Intel's like 40 times larger than it by market capitalization. And Dally was just like, you know, I conquered Jensen and he had the vision. So Jensen's very charismatic and he can convince these scientists. Come and do your life's work with me. I will give you the tools to work on the hardest problems in computer science.
Like yeah, maybe the financials aren't great right now. I'm working on that side of it. I think this can be a bigger company. I think we're undervalued. So if you come in now too, maybe this is an undervalued play and you can get discount stock options when we do go up. And that did happen.
But the most important thing for you I think is not I think the most important thing for you is to work on world-class technology and for someone like Dally it's 100% true and I will give you the opportunity and the runway to do that more than Intel. You won't have to deal with the bureaucracy. I'll let you build the most important and advanced hardware tools in existence and you'll be granted all sorts of patents and you'll get all sorts of accolades and in time if you believe in my vision we will make money over off this. It's just going to take longer than we think.
Keith 00:24:49
If you're a gaming company I mean if you you've been building gaming chips for so long because like what you said the AI transition only happened within the past decade or so. How do you actually make that transition? Especially when you survive the blood bath when you've already cornered the market, right? And you've become like one of the leading chip providers for gaming consoles.
Stephen 00:25:21
Around a couple things came together actually in like 2002 or 2003. There were three strands. One was you had these discount commodity manufacturers in Asia who were knocking off Jensen's chips and he knew there was no real copyright or patent protection against them doing this. You remember the microchip was an open book. I can peel the lid off the thing and look at it through a metallurgical microscope and if I have the expertise I can tell you exactly how it works. There's no trade secret in the metal. Anyone can look at it.
So that made him concerned and his concern was if we aren't constantly innovating, we're just going to be commoditized. If we're not constantly taking all of our profits and jamming it back into R&D, then that actually is its own kind of risk. If I take this cash that's coming in and just pay it out to shareholders, in 10 years, this company won't exist because I haven't built anything new. There is a risk, particularly in the semiconductor industry, of not innovating. And that risk is much higher than the risk of innovating or losing money on some project. That's how Jensen thought about it and he was correct.
The second thing that was happening is they packed all this complex accelerated computing architecture to do lighting in video games, but scientists were attracted to it. And at first they were actually hacking the chips to get at it. And when Jensen saw that, he very wisely said, "Well, why are these guys hacking it? We should turn these guys into our customers." He built a whole platform for them so they wouldn't have to hack it anymore. And in fact, the chief hacker, Ian Buck, he hired that guy. He brought the hacker on board. So that was very smart.
And I think the last thing, as I said, was the principle of physics that determined that you would have to make this change as the components got too small. So they unrolled this thing in 2006 and they were like, well, this is it. This is, you know, CUDA. This is our new thing. It's going to save the company, etc. And at first there was a fair amount of customers, but actually it took a long time for CUDA to succeed. And for the first six years, it was kind of a bust. In fact, in 2010, 2011, and 2012, CUDA downloads dropped each consecutive year. It looked like it was failing or at least that they had saturated the market for people who wanted this kind of thing because how many research scientists are there in the world? It's not a huge market and once you've printed enough cards, that's all they need.
So they got very lucky that AI came along and saved them. They got very lucky that in late 2012 AI came along and people started using the cards, the GPUs for this purpose. The next year downloads tripled and then they tripled again and then I think they even tripled again and now we're in a world where everybody uses CUDA all the time. It's the default platform for any kind of AI training, any kind of AI inference. And that is just massive amounts of money.
Now because people have so much demand for AI, we're building these giant data centers. It's like the size of the island of Manhattan. It costs $50 billion. There's land costs, there's construction costs, there's water and power costs, but basically most of that $50 billion is being spent on NVIDIA microchips. That's the number one cost in building a data center.
Keith 00:28:30
There seems to be an element that I think remains underappreciated which was that Jensen Huang was both world-class computer scientist like you said but he wasn't a natural entrepreneur. I mean if I'm like a science guy that wouldn't be my first inclination and he made the transition to becoming a world-class entrepreneur in the sense of not only going against mainstream MBA financial analyst Wall Street thinking but also being able to make really good bets and knowing when to cut losses really well as an executive.
Stephen 00:28:57
He has as we would say in the United States balls of steel like it is he has diamond hands these bets I mean they didn't pay off for 10 years sometimes and I should say Jensen made a lot of bets that just didn't work at all. I mean, he was doing north bridge chips for a while. That was a disaster. He was doing modems for a while. That didn't work either. I mean, his portfolio has more failures than successes. But his thinking is I only need one thing to really succeed to make this work.
So, in some ways, it's almost more like a venture capitalist where he's especially now, I mean, he's funding like 600 things inside Nvidia. It's crazy. But he's paranoid. It's almost like defensive investing because he's so afraid that he'll miss some technological wave that he has to be present for.
And so he's the other thing about Jensen and I think this is more true of Jensen than almost any other tech executive. If he realizes he is wrong, he will change course very quickly. No sunk cost fallacy. If he makes a mistake, he will just close the division, shutter it, move all the employees to some other part of his firm and start over. And he's done this multiple times. So I think this is the other thing. He's very willing to, as we say in the US, kill his darlings, like kill his most favorite project if suddenly he perceives that it doesn't make financial sense. And he's done that a great number of times.
On the other hand, he will stick with what looks like a bizarre or losing bet for a decade if he still thinks from first principles it might work in some way that opens up a huge market for him.
Keith 00:30:30
You made a point earlier and I want to double click on that which is you said that he would move his employees. That's not a very intuitive executive move. Most people would just fire them. And if you think about Jensen Huang as the manager, he's famous for saying that he has 60 plus direct reports. He's famous for being the anti-MBA person and he wants to push people to greatness. Can you talk a little bit more about the kind of management style that he has that not only can be acrimonious at times but at the same time seems to engender a great loyalty amongst his colleagues?
Stephen 00:31:09
Yes. He so first of all he pushes people incredibly hard and if you don't deliver he's going to yell at you but he never fires anybody. I mean unless they do something really egregious. If they steal from the company he'll probably fire you. But for in terms of like failing at a business initiative you won't get fired. Even if you underperform at your job usually he'll just reduce your responsibilities until you're at capacity. He really never almost lets anyone go. And that's a very unusual trait in Silicon Valley. Certainly Elon Musk, you know, they're out there firing people all the time.
I think I'm not totally sure, to be honest with you, why Jensen does this, but I think he doesn't want to make people afraid to fail or afraid to experiment in this company. I think he always wants people to feel like they have the willingness to try something weird and if it works out, they don't lose their job. He wants them to be courageous and not timid. And if they feel that they have job security and the ability to pursue unusual goals, then he will give them that runway.
Of course, the flip side of that is you have to work as hard as you've ever worked in your entire life. And that's what Jensen demands of his people in return. But, you know, Nvidia attracts a certain kind of person. It's not for everyone. I think I would struggle to work there because I need work life balance. They don't have that at NVIDIA. Everybody's working 12 hours a day all the time. As an American, it's harder. I think in Taiwan and China, actually, this is more common, but in the United States, it's not. So I think people do struggle there sometimes with just how much workload is put on them. But you know the flip side of that is you get to do your best work and you're compensated for it and you feel like you're working on something very important.
Nvidia is not hierarchical. There's really no org chart. It's just 60,000 people basically reporting right to Jensen. He's like the guy in charge and then you do what he says. And often, for example, with AI, he took a guy who was a researcher who'd never really made any kind of production software in his life, who had never managed anyone, and put him in charge of what he considered to be the single most important company product's future. And he does that all the time.
So, one thing that happened to Jensen when he was young, and I think this conditions his thinking, when he was in his 20s, he built a very successful division at LSI, but he was like 28 or 29. He was managing people who were much older than him. And I think the CEO of LSI at that time brought in someone else to co-manage the division, a senior guy, so that there was like an adult in the room. And Jensen just hated this. He was like, I built this. I built this. Why is this old guy coming in, you know, when I built all this?
And so I think within Nvidia, if you're in charge of something, you can be young, you can be inexperienced, but if you have an idea and you're willing to commit with the work, with the hours to make it work, he will put you in charge of a division. He really promote you in the field to a very important role and then really push you to your capacity. So I think this is what he means when he talks about pushing people to greatness. He'll put you in a role you never thought you would have so quickly that is so demanding it almost looks impossible and then he will push you to succeed and make the impossible real.
Keith 00:34:22
I couldn't help but think about it as LSI's most expensive mistake because he described the decision as political and you detailed that it was that as the catalyst that pushed him off to really join Nvidia.
Stephen 00:34:35
Corian was the CEO of LSI. He was a great guy, legendary figure in Silicon Valley. And LSI was a very innovative firm and my impression was that Wolf actually Nvidia co-founder Chris Malachowsky told me he thought Wolf was grooming Jensen to be the next CEO of LSI. But Jensen defected to do his own startup and I think that made Wolf very sad because he really wanted this guy even in his 20s people were like this guy is amazing this guy that was spotted in Jensen basically very early even as a teenager people were saying this about him I never seen anybody like this guy this is before he even did computers but like with table tennis that's what people were saying Jensen's ability to learn new things is just exceptional he learns incredibly quickly he's just very very smart guy very high IQ and I think that impresses people around him constantly.
But with the LSI case, you know, I do think a lot of what we considered to be the DNA of Nvidia was present even in the 80s when Jensen was just designing microchips for customers. I think the thing that changed that was more original to Nvidia was that, you know, Nvidia had all these bankruptcy scares.
He went from being a smart guy who could design a good microchip to a smart guy who could design a good microchip, who could fight to the death, who could win a basically a battle royale with 60 other companies, who could dig deep within and be the one guy left standing in this kind of like radioactive crater that had killed everyone else in his industry. He was ruthless. And I think he found that inside himself probably in his 30s. And that was during the period where everyone was trying to fight for GPU. They wanted to be the GPU king, I guess, and here to survive.
My theory is it scarred him because I think he was like, "Wow, that was hard." And we got really lucky. You know, I don't want to be in an industry with 60 competitors. I don't like I won, but I didn't like that at all. I had to like, you know, it was a knife fight. I want to be in an industry with no competitors at all. And he spent the rest of his career avoiding situations where he'd have 60 competitors. So, I think you found that very stressful and difficult and never wanted to repeat that experience.
Keith 00:36:45
He kind of encapsulates what Peter Thiel thinks of as like the zero to one as the ideal strategy for every startup entering into the space of technology.
Speaker 2 00:36:56
Yeah, I think he embraced that you know I think he embraced that concept. He would have done that even well. I don't know when Thiel's book was published, but he would have done this even before that book came out, but he must have been thinking along the same lines as Thiel most likely when Jensen was coming into that field.
Keith 00:37:13
I wanted you to speak a little bit more about that moment of convergence between the scientists and Nvidia. Can you give us some light into how that moment actually transpired? What were the flashpoint events that drew Jensen Huang's attention where this was maybe a small thing and then he was like oh I need to double down on this now?
Speaker 2 00:37:38
So the flashpoint event was Jeffrey Hinton's group in Toronto and Hinton recently won the Nobel Prize in physics as well as the Turing prize and every other prize but at that time nobody took him seriously at all and what his group was doing was you know neural networks it was an attempt to use software inspired by the biological brain to solve real world problems. People were skeptical of this approach. They mostly did not believe in it. And the results from neural nets were okay, but they weren't anything too special.
Hinton had been experimenting with Nvidia cards and GPU technology as early as 2009 or 2010, but he actually could not get Nvidia's attention. Jensen would hold these technology conferences for scientists and they would list all of the many possible use cases for the GPU cards. AI was not on there. Not in 2010, not in 2011, not in 2012. They didn't make the connection.
In 2012, Hinton got a brilliant graduate student named Alex Krizhevsky and another brilliant graduate student who you will know named Ilya Sutskever. And those two programmed a super neural net on a GPU, just two GPU cards. And then they trained it in I think Krizhevsky's childhood bedroom for a week. And the thing that they produced was by far the best image recognition AI of all time at that time. And that proved that the neural net had great value.
They published the results in late 2012. Jensen went on stage in 2013 at his technology conference and this is like early 2013. He didn't talk about AI at all. So, it took some time in there. He hadn't quite figured it out yet. He went back on stage in 2014 and it was all he talked about. That's all he talked about the whole time. One hour, nothing but AI.
So, sometime between early 2013 and early 2014, Jensen caught wind of this and he started to research it and investigate it and he said to himself, "This isn't just image recognition. Everything is going to change. This is the new paradigm for computing. And if we move fast right now, we can be the dominant hardware provider in this space."
And sometime between 2013 and 2014, Jensen wrote an email to his whole company saying, "We are no longer a graphics card company. We are no longer even a scientific computing company. We are an AI first company. From now on, we are betting the company on AI."
Now, today that looks very smart, but at the time it looked wild. I mean, this was based on a handful of promising scientific research papers, but it was Jensen reasoning forward from first principles about what these neural nets could do today and what they were going to do tomorrow. And more specifically, the vast amount of computing power they would need to accelerate and grow. And that was maybe like it was certainly the best business investment idea of the 21st century and one of the best business ideas of all time.
Keith 00:40:34
The crazy thing to me was that if you really went back to the zeitgeist of that era, AI was nowhere in the conversation. Even Jensen for 10 years had been cool. He didn't talk about AI. AI had a terrible reputation. It was a career graveyard. If you went into AI in the first decade of the 2000s, kiss your career goodbye. You were going to be a professor. You were going to be in academics. You were going to be at quiet conferences. Nobody was calling you. No venture capitalists were calling you. People didn't want to go into that field had disappointed many times in the past and people just didn't think progress was going to happen from there.
And honestly what they were missing was the GPU. That's the thing they were missing. I mean, that's the key finding of the Krizhevsky and Sutskever and Hinton paper from 2012 is that we can train neural nets 1,000 times faster per dollar per flop on a GPU than we can on a classic Intel CPU. That was the thing that unlocked the current paradigm. It was a hardware play more than anything. It was a hardware innovation. And that part is so that part is still not well understood. I mean, it's well understood among among the general public. People still don't really know what Nvidia does, you know.
Keith 00:41:53
But to make that claim so boldly, like you said, the whole company and you write about it where he was putting on the whiteboard, right? The once in a lifetime opportunity.
Speaker 2 00:42:00
Yeah. Just like that. And I talked to people at the company and they're like, he just got it. He saw it before anybody. He saw before anyone what it was going to be and what it could be. As close as 2014 or 2015. They were talking about building giant data centers that are actually now just coming into being. They were talking about that 10 years ago.
I think he told me he just reasoned forward from first principles. He was like, "Look, I saw what it could do and then I built out what it was going to do and I just saw the runway and I just didn't see any reason why it would stop improving, stop getting better or stop demanding more computing power." And all of those things turned out to be true.
It solved basically image recognition, which is an unstructured problem. It's a complete unstructured problem. It had no there was no special image recognition programming in the net. It just learned from scratch. And so if you can learn that unstructured thing from scratch, then it should be able to learn pretty much anything. And that insight turned out to be 100% correct. We're living in that world now.
And I think that was Jensen's kind of deep and abiding insight. I think also Bill Dally from Stanford saw this and was whispering in his ear. Brian Catanzaro, his leading AI guy, was telling him this as well. And I think Hinton told me right away the moment they got this thing working they were like from this point forward we're gonna sell so many GPUs. He saw it too even before Jensen did and starts but Jensen had you know Jensen doesn't hedge his bets. Let's say if he sees a good idea he is all in immediately. He's not going to wait around and be like oh we'll allocate a certain amount of the company to this. It's just not the way he thinks. He's a guy who swings for the fences. He goes for a home run every time and that's what he did.
Keith 00:43:55
Now that everyone has gotten an insight that GPUs are the key or the unlock to everything AI related, basically he's the shovel monopoly. But now there are other shovel makers that want to come in. How does he maintain that competitive edge?
Speaker 2 00:44:07
Yeah, it's a great question. I think there's several paradigms that could unseat everybody. Their stock price has gone down 85% before, 90% before and that can in fact happen again. The biggest risk would be some competitor just comes along with a CUDA clone that does the same thing at 1/100th of the cost. The other risk in the opposite direction is that somebody comes on with a CUDA improvement that goes basically 100 times faster.
Incrementally better things than CUDA are not going to dethrone it. It has to be orders of magnitude better for people to switch. Jensen from the start has been engineering what one of his engineers called vendor lock. He has been engineering switching costs into the CUDA platform from the beginning. So it is not easy to switch away from CUDA.
For most designers right now trying to save money on microchips, that's not really what they're thinking about. At the Frontier Labs, the cost of microchips are in some ways the least of their concerns. Look at Mark Zuckerberg. He's spending a billion dollars on research talent, research scientists. That's the pipeline concern. If he's spending a bunch of money on Nvidia microchips, he's just going to eat it for now. Then maybe down the road, he'll look for a cheaper, more cost effective solution. But I have not yet seen anyone who can compete with the Nvidia computing stack. But it could happen. It could definitely happen.
The other risk that maybe we face is that in the late 90s early 2000s we built out all this fiber capacity for the internet. All these predictions that there was going to be streaming stuff and we need all this fiber. And that actually was correct over time. We did need all that fiber, but they overbuilt it at first and then there was a big slump that lasted for years. It was a dark fiber era, at least in the United States, where there was all this fiber optic cable. Cost a ton of money to build and nobody was using it.
Maybe it's possible that we build all this AI capacity and then customers don't show up. Doesn't seem like it's happening yet. Growth is exponential, but at some point maybe that could level off. And so you've built data centers around the planet and you've overbuilt capacity and nobody needs it. And now Nvidia doesn't have the next generation of microchips to sell. They've saturated the market and their stock price goes down. That can also easily happen.
People ask me a lot for investment advice about Nvidia. And at this point, Nvidia is the single most analyzed stock in the entire world. So the idea that anyone's going to have any kind of alpha generating insight about Nvidia seems unlikely to me. But maybe they will. I don't know. It's not going to be me.
Keith 00:46:50
A funny thing I've noticed is bragging about how many H100 chips or how many Nvidia chips you're buying from Jensen seems to be like the go-to move of all the leading AI labs.
Speaker 2 00:46:57
It's totally become also like a luxury consumption good for a certain kind of AI person. They're bragging about how much money they're spending on this. And the thing is these companies are not currently at least constrained by capital in any way. If I have a decent idea for an AI startup, don't worry, it's getting funded. Like in the current environment, the biggest risk they face, as I say, their bottleneck is talent. It's not chips. The bottleneck for all of the frontier labs is talent.
And if I'm not a frontier lab, right? Let's say I'm a bank and I've actually talked to people who are doing this kind of work. They want to do an AI, an LLM like lending model to replace their classic lending standards. It was supplemented with AI. They can't even hire anybody because that's going to be the highest paid employee at the whole bank, the guy who develops the LLM. So, I think the constraint here for most AI companies is getting the right people and the right talent and worrying about trying to engineer a switch from say Nvidia microchips to AMD microchips or Huawei. It's just not worth their time right now. It's just not worth their time. It's not where the constraint is. It's not what they're worried about.
Keith 00:48:22
If you look at AMD, which is what many consider to be a rival or a close rival to Nvidia and there's a story about him and his cousin Lisa Su, right? What is what how will the competitive dynamics play out?
Speaker 2 00:48:35
AMD did a big push. So their competitor CUDA is something called ROCm and it's open source. I think that might be a mistake. I think that the kind of high-tech research lab that pushes CUDA to the frontier of computing, I think you need money to do that. I think you need paid employees to make that happen. So I think what's happening is ROCm is always a generation or two behind as they implement as they implement things that are already at CUDA, but they're not an innovative frontier R&D lab for this kind of inference or engineering. And I think that hurts AMD a lot. People I have talked to do not have a great opinion of AMD. And I'm talking about people in AI Frontier Labs who are releasing this stuff.
Now you might say downstream of that, well I don't need the world's most powerful AI for certain kind of customers. Maybe I just need some AI, or maybe I don't need to be a frontier lab. I just need an AI to do this one task for me and I don't want to pay $3 per hour per GPU for Nvidia. I can pay 50 cents to AMD, and so maybe that does end up being an important use case for AMD. But in terms of taking the throne, they actually took a shot at it over the last year and they were unsuccessful.
Now we have Amazon building chips. We have Meta talking about building chips. We have Google that's built its chip. But I think Jensen is most worried about Huawei in particular. This is why he was so adamant and was so successful in the end of getting China's market back open for his H20 chips. He does not want to compete with SMIC and Huawei. He does not want to compete with an indigenous Chinese microchip stack. He wants them to sell his chips so that doesn't form in the first place.
It's a really tricky balancing act. He's done a successful. He's done a fantastic job of it so far this year. All of his wins this year have basically been political, which is new territory for him. I mean, he was not a political guy as far as I could tell. He'd never made any kind of political statement at all. And he was very successful in getting convincing Trump to come around to his point of view on this kind of thing.
Especially if you think about the fact that in China, you could easily create an alternative tech ecosystem. The golden grail of getting a vendor lock in is if he's crowded out of the market for two to three years, it's possible that everyone would have switched out and he would have lost their market permanently.
Lutnick, Howard Lutnick, who's the commerce secretary and a close Trump adviser basically said in public, the idea is to get the Chinese researchers addicted to American technology. And I'm sure that line came from Jensen. I bet probably that.
Keith 00:51:25
He seems to also now have emerged as like the next emissary or sort of in terms of the US China relations doing like independent private jet diplomacy.
Speaker 2 00:51:30
Chang built TSMC in Taiwan and part of his objective or part of the idea there was to create an economic asset that would guarantee peace. If China is totally reliant on this TSMC factory perhaps they're less likely to invade and I wonder if Jensen and speculating here, I should say. I haven't talked to him about this, but I wonder if Jensen isn't taking a page from Chang's book and saying, well, if we can encourage tighter economic cooperation, integration between the United States and China, maybe that diffuses the tension a little bit and maybe we can go back to the more harmonious relationship that actually we enjoyed in the first decade of this century, which was mutually beneficial, at least in my opinion, for both countries. And China has a bunch of frontier research labs. They're very smart. They're doing incredible work and I think Jensen wants to be present for that.
Keith 00:52:22
Intel for the longest time was the leading by far the leading company. How did things go so wrong for them that now what was once an upstart that they tried to kill essentially has eclipsed them by leaps and bounds?
Speaker 2 00:52:40
Oh yeah. Bad management. They made a number of errors. I think we could say they made two major errors. The first was not separating their foundry business out. They so they should have spun off their foundry business decades ago. They were attached to an old business model where they designed all the chips and then they manufactured all the chips but they didn't have the capacity to do both those things and it allowed these asset light upstarts like Nvidia. Nvidia basically owns no manufacturing facilities whatsoever. They outsource all of it. Intel is completely integrated. That's a flawed model. That's a flawed model and they just institutionally they just couldn't accept it. It was just the way they had always done things and they couldn't accept that it was a superior model to break off into two companies like TSMC and Nvidia.
They couldn't innovate at the pace Nvidia innovated at because they didn't have a relationship with TSMC which was better than building the plants. They were trying to do it all themselves and that was a totally failed model. So I think that really hurt them a lot.
The second thing is they missed the parallel computing paradigm that I was talking about the physics problem. The problem that the trend they they weren't able to maintain Moore's law. They just had they just had always been able to engineer their way out of things before. And now they were facing this problem of the components getting too small and leaking electricity. And I think the mindset at Intel was we'll find some way around it. Something we'll find some kind of way around this. We've always found a way in the past. this time we will too. We're not going to get stuck on this. But they did get stuck.
I talked to Brian Catanzaro who was an intern at Intel in like the mid-2000s and one of his projects was design a road map for a microchip that fires at 10 billion cycles per second which is very fast. And he did the math and he came back and he's like well it must be a trick question because that's impossible because the components will leak too much electricity if we try and build this. And the manager is like what do you mean it's a trick question? this is part of our road map. It's part of our road map here internally to build this thing. They still haven't built it.
So I think their road map was flawed. I think because they didn't have a world-class computer scientist in charge of the company like Jensen or for that matter like Lisa Su who is also a world-class computer scientist. I just think they started to not understand how to react to the technical challenges in front of them. And I think that has resulted in where they are today. They got, you know, they got used to being the boss, and they got lazy maybe or complacent and they thought they could overcome physics. I guess the limitations of physics.
If you read about the history of Intel, they did work miracles. They did always find a way. They had always found a way in the past. And so I think it wasn't so crazy to say, "Hey, we're just going to do it again. We know we'll find a way to do this. We've always found a way. Moore's law is not really a law. It's just something they did. It was actually an extraordinary accomplishment that they were able to speed up the CPU by whatever it was 2x every 18 to 24 months for 20 30 years, but there was no law. In fact, they were always trying different tricks to do that. And I think finally they just ran out of tricks to do that. And and I think that's what really cost them in the end. That and the foundry business. So kind of two things.
Keith 00:56:05
Jensen Huang remains optimistic about the future of AI. And he thought that the idea or the reality where the marginal cost of computing converging to zero would unlock a great abundance for the world. And he I think you alluded to it that he took offense that you were maybe worried or cautious about it.
Speaker 2 00:56:23
He yelled at me for like 20 minutes because I kept bringing up these AI risk issues which were grounded in science. I mean I was conveying things that Jeffrey Hinton and Yoshua Bengio who won the Turing prize were telling me. I wasn't just making stuff up but Jensen really objects to this and he really feels like Hinton in particular and Bengio are holding back scientific progress by raising these concerns and that they're even tarnishing their scientific legacies.
Bengio, who I talked to just yesterday or two days ago, is the single most cited computer scientist alive. And in fact, he's the single most cited academic research scientist of any kind alive. And he is absolutely terrified about the capacity of what AI is going to do in the future. Jensen is not. Jensen thinks we're safe. He thinks we're going to build these systems and it's going to unleash a new era of prosperity and human flourishing.
And they're both two of the smartest people I have ever talked to in my entire life, if not the smartest people I've ever talked to. And they can't agree on this issue. So, I find this to be an important question. I'm actually continuing to write about this right now. I'm talking now for another story to a bunch of red teams who penetration test Frontier models. So they'll go to OpenAI independently and they'll try and feed it jailbreaking prompts to get bad outcomes out of the LLM and they're successful all the time. And that's an early red flag because as these systems get smarter, as they get more intelligent, will we maintain control over them or will they start to serve their own ends? I think it's an important question. Jensen thinks it's ridiculous. He thinks it's science fiction.
The biggest difference between him and Musk is in the way they operate or the way they perceive they're operating from like Musk he he says that it's operating from science fiction. He operates from a basis of reality and then he works forward he works backwards.
But it's not clear to me like why he's so certain that AI is an unalloyed good that there can be no real drawbacks. And I noticed this also with other people I talked to who were very close to the microchip, who were very close to the metal. So to Jensen and to a lot of his staff, they're like, "What are you talking about?" To them, AI is a series of mathematical operations that they run. It's not machine intelligence even. It's just math. That's all Jensen said. It's just math. I'm just doing, you know, 10 trillion math operations. If I do an 11th, who cares? 11 trillion, who cares? Like, it's not risk. That's not risky to me.
So Jensen's point of view is he's so close to the microchip. You could say maybe he's even too close to the microchip. It's sort of like asking, well, what is a human brain capable of by looking at brain cells under the microscope? Well, that's not really going to tell you the answer. And in fact, even if you understand the system holistically, you're not going to see the answer. You have to interact with humans to sort of understand what they're capable of. And I think this is true of the AI as well.
We it's a you could say well it's just a gigantic grid of numbers and it is it's just a series of weights it's a just a giant file of numbers that's what the AI is but obviously the file of numbers can do something just as a bunch of cells in your head can do something so we have to probe or test the capabilities of the AI to really see where it's at.
To me this is actually the single most important and interesting story in AI right now. I am these AI red teams are so interesting because they really no matter how much protection they put in, they can crack the AI open and get it to output all sorts of stuff that violent imagery or instructions on how to build, a nuclear bomb or just all sorts of scary stuff. As the LLM gets more capabilities, this becomes a greater and greater risk. And I don't think that we've solved or even begun to address the control problem yet.
Keith 01:00:32
My last question is a hard pivot. If there was one piece of advice you give a fresh graduate entering the game world, what would it be?
Speaker 2 01:00:37
Yeah, I don't know honestly because I have I'm facing this question myself. I've been a writer. It took a long time to become a successful writer. I mean I'm only marginally successful and it's hard. It's a ton of work and it took a lot of skill development and time to make it succeed. I think in the next three or four years all those skills might turn out to be worthless because an LLM can already write better than I can instantly.
So what happens to my job or career in that scenario? I'm not sure. I noticed this did happen with like chess. So the supercomputers are way better at chess than humans, but nobody's interested in watching a supercomputer play a supercomputer. They want to see two humans play. They like the drama. They like the personality of it.
So I would say in the coming era, you have to attach something human to whatever you do. I would tell young people to engage with and indeed master these tools. Everyone I talk to who does the AI pen testing, they're all like 23 years old. They're so young. I can't believe it. It's a totally nascent industry of people in their 20s who are building this industry from the ground up. So in that sense, I would advise you whatever you study, make AI a critical part of your components. If your professors object, frankly, just don't listen to them because they're dinosaurs. Like, we're going to be using AI all the time and no matter what we do.
And the last and most irreplaceable thing, I think, is to pursue something that you're genuinely passionate about because if you try and fake it, you might succeed, but your life isn't very pleasant. Jensen actually interestingly said not to do that. He said, let them find something you're doing and learn to develop passion for your work. But I've struggled to do that in my life. I've had to pursue things that came from within and so for me writing was really important always I think I always knew I wanted to do this in some capacity and when I didn't do it I became unhappy so so try and figure out what that thing is and pursue it you know with the knowledge that the world is going to shift under your feet like it's guaranteed this technology is going to revolutionize almost every aspect of business every aspect of technology of society the world will shift in ways is that we cannot predict.
And there's no real way to prepare for that. But if you develop passion and you develop skill, you should be able to navigate it. Also, think for yourself. I think we know last night.
Keith 01:02:56
Well, that's solid advice. Stephen, thank you so much for coming on.
Speaker 2 01:03:02
You're welcome. Thank you.
Keith 01:03:03
Thank you for listening to today's episode. If you are watching this on YouTube, please consider subscribing and turning on notifications for whenever new episodes are out. If you're on Spotify or Apple, it would help us greatly if you could leave a five-star review on those platforms. Once again, thank you for tuning in.