NVIDIA: Jensen Huang. From near collapse to becoming the world’s biggest company

Introduction to Success

0:00Success is a journey, especially when it comes to your finances. One of the most common obstacles in our financial journey is dealing with high-interest debt. It can often leave people feeling trapped, but it doesn't have to. If you're dealing with high-interest debt, there is a way forward. A SoFi personal loan could consolidate all your high-interest debt into one low-interest monthly payment, helping you craft a roadmap to paying down your debt. It even comes with no fees required, getting you a financial win right away.

0:35Keep pushing on your journey to financial success. You could even get as soon as the same-day funding. View your rate for a SoFi personal loan at SoFi.com slash Guy Raz. Loans originated by SoFi Bank N.A., member FDIC. Terms and conditions apply. NMLS 696891. Fast funds term apply at SoFi.com slash Guy Raz.

Apple Card Promotion

1:02This message is brought to you by Apple Card. For a limited time, when you get a new Apple Card and purchase AirPods Pro 3 at Apple, you can earn back the cost up to $250 daily cash. New AirPods Pro and up to $250 bonus daily cash back? Now that's music to my ears. Subject to credit approval, limitations and spend requirements apply. Apple Card is issued by Goldman Sachs Bank USA, Salt Lake City Branch. Terms and more at apple.co slash AirPods.

Cleveland Clinic

1:35When it comes to your health and well-being, the right care can change everything. That's why Cleveland Clinic has been elevating world-class patient care for over a century. From the most specialized heart, neurology and cancer treatments to the latest surgical innovations and beyond, Cleveland Clinic is here for every care in the world. Whether you're exploring advanced care or just looking after your health, all the info you need is waiting for you at clevelandclinic.org.

NVIDIA's Journey

2:15We sort of have the outlines of all of the troughs, and there were a lot of troughs. And I'm trying to even imagine quarterly earning calls in 2007, when your stock price is in the toilet, and you're putting all of this money. I tell you, it's embarrassing. It was embarrassing. It was humiliating. Your employees are probably embarrassed for you. In fact, right now, it is really quite hard for me to resurface those feelings. And the reason for that is I spend all of my time, all of my life trying to forget yesterday.

2:46Right. What do they teach athletes? Forget the last point. Yeah. It's about forgetting.

How I Built This

3:01Welcome to How I Built This, a show about innovators, entrepreneurs, idealists, and the stories behind the movements they built.

3:12I'm Guy Raz, and on the show today, how Jensen Wang went from making graphics chips for gamers to powering the AI revolution and building the biggest company in the world.

NVIDIA's Value

3:31If NVIDIA were a country, it would be one of the five richest in the world, just behind the U.S. and China. NVIDIA's value is now more than the entire economic output of Japan or the U.K. or France. That's how big this company is. And it's also probably the single most important company in the world right now. It is the biggest player powering the AI revolution. But NVIDIA didn't start out that way. The company actually began by selling graphics processors for video games.

4:04And for a good 20 years, gamers were NVIDIA's main customers. But back in the 2000s, the company's co-founder, Jensen Wang, made a pretty important bet. A bet that those graphics processors, known as GPUs, could actually do a lot more than just power video games. He believed these processors could be the cornerstone for the future of supercomputing. This was a very bold and very expensive idea. Now, bear with me for a moment here,

4:35because if you're not super familiar with the technical terms around AI and computing, I will do my best to explain. So think of a computer like a kitchen. The CPU, or central processing unit, is like a master chef. It's really smart, it can do almost anything, but it can only cook a small number of dishes at a time. The GPU, or graphics processing unit, is like a kitchen with thousands of line cooks. Each one isn't as talented as the chef,

5:06but together, these guys can crank out thousands of simple tasks all at once. Now, to make that kitchen do even more, Jensen poured billions of dollars into developing a platform and software layer called CUDA, which is basically the instruction manual that lets you use all those line cooks in entirely new ways. So what CUDA basically did is turn the GPU from a video game tool into a general-purpose supercomputer.

5:38The problem with it, though, is that it was way ahead of its time. Only a tiny number of users, mostly university researchers, had any need for it. And so every single NVIDIA processor sold to a teenager playing video games was actually very sophisticated, and yet that teenager had no need for it. And so for nearly a decade, NVIDIA was stalled. The stock price stayed flat. Sometimes it actually fell. And at one point, there were rumors of a hostile takeover bid.

6:10Many investors were losing faith. A lot of them were dumping the stock. And people outside the company were questioning Jensen's obsession with CUDA. But he kept going. He withstood massive pressure to move away from it because he really believed in the power of this platform. Which may be the most remarkable thing about this story, because a decade after NVIDIA started this experiment, the bet started to pay off. And when it did, it was like every slot machine in the casino hit at once.

6:42NVIDIA's chips found a massive new market in the emerging world of artificial intelligence. And today, NVIDIA is at the center of it. Their chips dominate the computing power needed to power AI. So how did Jensen Wang see it coming? What did he believe that others didn't? And what does he think about where all this is going? That's what I wanted to find out when I sat down with him at NVIDIA headquarters in Santa Clara, California.

Jensen Wang's Background

7:10Jensen Wang was born in Taiwan in the early 1960s. He spent some of his childhood in Thailand. And when he was about nine years old, his parents started to get worried about political unrest in Thailand and decided to move to the U.S. They sent Jensen and one of his brothers ahead of the rest of the family to what they thought was a normal boarding school. And there weren't that many boarding schools in America that would take international students, but somehow, Oneida Baptist Institute in Kentucky,

7:41Clay County, Kentucky, little tiny town, no stoplight. And there was a school on this small mound, you know, kind of a small hill. But probably the most important feature was that it was a boarding school and it was incredibly affordable. Because my parents, you know, didn't have much money. And so we went there and every kid that lived there had to work. And so they had no custodians. And so we were, you know, we took care of ourselves.

8:11And my job was to clean the bathrooms. And apparently I was the youngest kid that ever went there. I still am, I think. Because you were 10. I was nine when I went there, yeah. I've got great memories of it. I loved the place. It was a tough school because they invited kids from all walks of life. And so, you know, I was nine and my roommate was 16. And so none of the closet had doors and none of the drawers. Because everything had to be out in the open, you just don't know what the kids are going to have.

8:41So welcome to America. Yeah, right. But I loved it. It was incredible. I was on the swim team and the soccer team. And then afterwards, the coach would take you out to give you a treat. And I remember recording, we used to record a tape and we would send it back to our parents. You know, we didn't, long distance phone calls cost too much money. And so we never spoke to them live until they came to the United States almost two years later. And so we would record, you know, what happened this last month. And I remember telling them that after the swim meet, the coach took us to this incredible restaurant.

9:17And lights everywhere is like from outer space. And all the food was in boxes. And it was McDonald's. It was McDonald's, yeah. Which was magical. Yeah, incredible. Yeah, yeah. You eventually reunite with your family in Poland, which is where your dad settles. And you grew up there. And you went on to, this is the early 80s, late 70s, early 80s. You go to Oregon State. And this is like the beginning of what would become the computer revolution. You are studying electrical engineering because I don't think they had a computer science program there at the time.

9:52And that's also where you would meet the person who would become your wife. Yeah, so when I was in high school, I wasn't very outgoing. And so I didn't have that many friends. All of my best friends were in two clubs. One club was the math club. And then the other club was the computer club. Nice. It was the same, you know, four kids. The same guys, yeah. The same four kids, yeah. And my best friend in high school, I asked him, you know, where he was going to college. And he said, Oregon State, and he asked me where I was going to college and never crossed my mind to go anywhere out of town for, you know, the idea of going to a great university never crossed my mind.

10:30And so he said he was going to Oregon State. I said, yeah, that sounds great. I'll go to Oregon State. And so we were roommates. And there we both enrolled in electrical engineering. And our lab, the class had 200, must have 250 kids. It was a big class. And there were three girls, you know. And so I noticed Lori is a super pretty girl. And I was probably the youngest kid in class then as well. I skipped two grades in school.

11:01So we were in lab together. And I found a way to arrange myself into the same lab as her, same group as her. And so we became lab partners. And here we are.

Jensen's Career

11:13Both of you, after you graduate, moved to California. This is like 1985. 1984-ish, I think. Which is the center of the revolution, right? I mean, your first job was with AMD. Yeah. And I think she also had a job. Silicon Graphics. Silicon Graphics, yeah. But what did it look like to design computer chips in 1984? I mean, was it CAD? What did that look? What did that actually physically entail? Yeah. Well, at the time, I was the last generation of chip designers that did it by hand.

11:44And the first generation of chip designers that used software to design chips that ran into computers, that ran software. And so it was an incredible time. You know, kind of to put in perspective, at the time, the chip I was working on had a few thousand transistors. And now we design chips that are 200 billion, a trillion transistors. Insane. Yeah. So the scale of the problems that we work on now compared to where we started is incredible.

12:17But anyhow, my office mate went off to work for a startup company called LSI Logic. And she called me and said, hey, you know, this is really quite a special company. You ought to come take a look. And so I went to take a look at LSI Logic. And it turned out, this is an extraordinary company and completely revolutionary in what they were doing, really, in a lot of ways, invented the modern way of doing designing, the way of designing chips. And so I joined them. I guess I was, you know, probably employee number 150 or 170 or something like that.

12:50And you were like 22 probably at the time. Yeah. I was just a kid. I was just a kid, yeah. And from what I've read, I mean, they gave you a lot of responsibility. I mean, they, I mean, you were eventually, you would be in charge of tooling, right? Like that was sort of what you oversaw. Well, the way LSI Logic's business worked is they had all the technology, the tools, and they made the chips. But they would make it for somebody else who created computers. And so some microsystems and silicon graphics and incredible companies at the time in Silicon Valley who were systems companies.

13:25But they needed a chip company to help them build the chips. And so the CEO of LSI Logic, you know, recognized my talents and put me in front of all these companies to help them. And I met some incredible people. I mean, you were, this was sort of like the PC boom, the Clone Wars. I mean, there was a lot of demand for what LSI was producing, right? I mean, it was an exciting time to be here, to be young and to watch, you know, probably the next most exciting time what happened with the internet boom, right?

13:55Well, LSI Logic was an extraordinary company because it happened at precisely the time, exactly as you're saying, Guy, that many computers, many supercomputers, supercomputers, all of those types of giant systems were being created by companies like digital. And then the PC revolution started. And so I was really in the right place at the right time. And I saw new industries being created. I saw a lot of great startups. And I was able to see firsthand, you know, companies being built.

14:25Yeah. And great technology, bad strategy. Moderate technology, excellent strategy. I saw it all. It was quite an incredible thing. And then, of course, I met Chris and Curtis. Those people, Curtis Prem and Chris Malachowski, were Edson. The story is that these guys wanted to create, Edson Microsystems, a chip for specifically for games. For computer graphics. For computer graphics for games and were rebuffed. And so, as often is the case, they decided to start their own company.

14:56Yeah. And they approached you and they wanted you to work with them. I mean, this is around 1992, 93. And from what I understand, I mean, you had a great job. You were well paid. You were well on the path to upper management. Like, this could have been a great, stable career. You already had, I think, at least one kid at the time. Yeah, Spencer Madison, yeah. And your wife was looking after the kids. So, everyone depended on your salary, you know. And I think we had a whopping $30,000 in the bank. Um, when they initially approached you, I mean, did you think that, I'm not going to take that risk?

15:33I'm not going to leave this amazing job. Yeah, I mean, they, um, but they didn't give up asking me. Because they wanted you to run their, this business that they had an idea for. Yeah. But I think, in the end, what we got excited about was probably just a piece of revolution in the end. That, here's the first time a computer was going to be general purpose, and you're going to use it for all kinds of different applications. And if it had computer graphics, like what people were seeing with Jurassic Park and Silicon Graphics.

16:05And if we could figure out a way to make it affordable and architect something that would fit into a personal computer architecture, you know, maybe there's a company here. But, I mean, at the time, there were, like, 60 companies trying to do this, right? Yeah, right. And so, was there any part of you, do you remember any part of you thinking, God, what if this doesn't work out? No.

16:26I should have.

16:29It didn't, it didn't really. You were, you were prepared for whatever happened would happen. Yeah. Maybe that's what's called vision. Maybe, maybe that's, you know, determination come from. But NVIDIA manifested in my mind that, and an industry manifested in my mind that, that was so crystal clear. Never once did it cross my mind it wouldn't work out. So, NVIDIA is the company, which I guess comes from the Latin word envy, NVIDIA.

NVIDIA's Name

16:55And you guys start working on the first product, which is going to be a revolutionary graphics processing chip that it's going to make. We were the first 3D graphics company in the world that started with the idea that we bring 3D graphics to consumers. Yeah. And so, 1993, it's founded, and you start to work on this product, and you guys raise a little bit of money to do it. And this is going to be a game changer, right? I mean, this is going to be, like, the coolest thing.

17:26Yeah, we thought so. The technology that we created allowed us to generate images, but using, using a lot less electronics, a lot fewer chips, and a lot more affordable than these giant supercomputers that Silicon Graphics was making at the time. You know, at the time, the computer that generated the images for, for Jurassic Park would cost, you know, a million dollars at a time. And we needed to get something to fit into about 300 bucks. And so, that gap was so large that we had to reinvent the algorithms altogether.

18:03People got very excited about it. This is the envy one, a chip. Yeah, we sold 250,000. I've now, you know, I believe we could sell 250,000 of almost anything, but the algorithm didn't really work. And so, we received 250,000 back. I mean, it all came back. It was not just our lesson, it was a disaster. Yeah, it was probably a technology disaster as great of a technology disaster ever seen. You know, I think the right architecture, the right algorithm is inverse texture mapping.

18:33Ours was called forward texture mapping. At this point, there were probably 20, 30 3D graphics companies, and they were all doing it the right way. And we were the only one doing it this weird way. And Microsoft is about to announce Windows 95. And Windows 95 has a API called DirectX. And DirectX does it the right way. And we were incompatible with DirectX. And so, anyways, we chose some approaches that were just fundamentally wrong. I've read that that understandably created a lot of tension with you and your co-founders because it was two and a half years of work.

19:12And from what I also understand, all the architecture that you'd created for the next NV2 and NV3 was based on NV1. And so, imagine, I mean, do you remember the three of you just going at it? Yeah, well, the argument was pretty stressful because we had a contract with a company called Sega, the video game company. And they contracted us for $12 million in 1995, 1994.

19:43And they were guaranteed to buy the NV2 and the NV3. Yeah, to use it for our game console, the Sega game console. And so, we were contracted to do that already. And we had invested two and a half years.

19:55And so, the question is, how do we deal with this contract? If we cancel the contract, how does the company stay alive? And if we don't cancel the contract, you know, the company doesn't go out of business. And so, there's the argument of, let's not cancel the contract. Let's keep on going. Then, of course, there's the argument, the contract is based on an architecture that is fundamentally flawed. And so, why finish something the wrong way? And it would have burned another two years. And in two years' time, 30 other companies were so far ahead of us doing it the right way.

20:29We'll never catch up. And so, this was the dilemma. And if we decide to change the architecture, we've got to go cancel the contract. And so, I went to Japan. And I contacted the CEO. Of Sega. Sega. His name is Iramadri. And I told him, my recommendation for you is that Sega finds another partner to build the 3D graphic system for their next game console. I have a request, however, that even though we're not finishing the contract, I still needed the money.

21:03Because if we didn't have the money, the company would be out of business. And I really believe the company deserves to succeed. And so, I asked him if he would convert the rest of the contract to an investment in our company. And he says, but Jensen, you know, your company has 30 competitors. You're most likely going to go out of business. I don't even know what your business plan is. And I said, can I tell you, I'm not sure what my business plan is. But my first job as the CEO is to make sure we don't go out of business. And we've got to get the technology in the right track.

21:33And if you could help me with this, I think we'll figure out a way. And I think it's going to be a great investment. Anyways, he talked to his board. And they turned the rest of the $5 million of investment or contract into our investment in our company. I took that $5 million and we came back. And I was incredibly grateful. Came back and here we are. The company was a little bit too big. I had to cut it back in half. You were like 100, 250 people or something. Almost, yeah. And so, I cut it back to it. Because you hired all these people anticipating NV1, NV2, NV3.

22:06It was going to be successful, right? And we're going to get all these Sega games, all these console games into the PC industry. So, anyways, I laid off two-thirds of the company. And it was incredibly hard to do. And also, you now had to focus on making the NV3, which was a new chip that would actually work. And, I mean, to be clear, you needed to do this. Not only to compete with all the other companies out there, but basically to save yourselves, right? To save NVIDIA. Yeah. Well, it's scary. It's scary even now.

22:37You're asking me. You know, these are traumatic experiences 33 years ago. And so, we decided, okay, first, let's just decide to do things the right way. But we don't actually know how to do it the right way. And then the next problem I have is that we have $5 million. But by the time that we were done designing the chip, the company would have ran out of money before the chip comes back. And usually… And the chips were being made in…

23:07TSMC. By this point, you had already started working with TSMC in Taiwan. Because initially, your chips were made in Europe, I think. That's right. Right. But the capacity wasn't there. So, you had to move it. So, this is like around 97, you start to shift production to Taiwan. That's right. And you would run out of money before they were able to produce them. That's right. And so, the problem was, back in the old days, you would design the chip. The chip comes back. You would write the software for the chip, fix whatever bugs you found, and then you would make the chip again.

23:39And that would go around a couple times. And back in the old days, it would take about a year and a half to two years to design a chip and get it to work and ship into production. Well, we didn't have a year and a half. We had about six months. We were on fumes. So, you had no choice. You had to do it. The chip had to be right. Which means you couldn't do all of the processes. We couldn't iterate. Yeah. Yeah, I couldn't prototype. So, how did you know it was going to work? And so, we heard about this company called Icos was building an emulator. And this incredible machine, this giant machine, it would pretend to be your chip.

24:13And so, you would take all of your software and put it into this machine. And this machine would pretend to be the chip. And you would plug this machine into a PC and run the software. So, you could see how it would work. I was super excited about it. And it was airtight accurate. Well, supposedly. And so, I called the company and I said, hey, listen, I heard you guys have this machine called an emulator. I would like to buy one. And they said, well, that's terrific. But unfortunately, we had no customers that were going out of business. But we had this one that's leftover.

24:46If you want, you could buy it out of the creditor who now owns it. And so, we did. We bought this leftover piece of machinery from this company that ultimately went out of business. And I took half of our company's money. So, we were already running on fumes. And so, I cut that life short even further, took that money, bought this machine that nobody else wanted to buy, brought it to the company. I said, here we are. We've got to put NVIDIA's chip, NV3, Riva 128, into this machine.

25:21Did anybody say this is crazy? I mean, this is nuts. You're taking half the money. I mean, you're the CEO. We don't even have – that was – the money we had was already not enough. And we took half of that and spent it. But these engineers who are used to a process, right, and used to a way of doing things, you're basically saying we're going to put it in the emulator and that's going to be it. And then we're going to fabrication. I mean, did anybody say this is not a good idea? It was the only idea we had. I got to tell you, I don't – I didn't remember anybody objecting to it.

25:54But I also didn't remember asking too many people whether they objected to it. But, you know, I think smart engineers reasoned about things. I guess we always knew it was existential. We always knew we were going out of business. But I just remember us being super calm, super focused. And we just step by step by step reasoned that it was the only chance for us to be – That had to work. It had to be. You're – this is – I don't know if it's apocryphal or – but I think it's true. You are known to say at that time our company is 30 days away from going out of business.

26:28Yeah. And it kind of became a meme and a mantra for a while. But it was true. I mean, you were – so this had to work. This thing had to work. And we – I'm not revealing any secrets. It did work. It did actually work. Mostly. But enough to save – Good enough. To save the company. Yeah. Good enough to save the company. When we come back in just a moment, Jensen identifies a whole new market for NVIDIA, a market that does not yet exist. Stay with us.

26:58I'm Guy Raz, and you're listening to How I Built This. You know that moment when you order food and suddenly everyone around you gets very interested in your dinner? Yeah, that's what Grubhub does. Gives you deals so good, you'll have to guard them. Gold Days of Grubhub Plus is here. Four weeks of Grubhub's best offers all month long in May only for Grubhub Plus members.

27:29And if you're not a member, you can sign up now for just 99 cents a month for six months. That's 90% off. Grubhub Plus membership auto-renews and terms apply. Sign up now on the app or at grubhub.com slash plus slash gold. Don't miss it. You know, the challenge I always had with Bitcoin was, well, how do you spend it, right? Bitcoin is often talked about as an investment, but it was built to be used. With Cash App, you can actually do that. Send Bitcoin instantly, pay at local Square businesses that accept it, or move it to your own wallet whenever you want.

28:05It works like real money and less like something locked in an account. For a limited time, new customers can get $10 added to their balance. Just use code CASHAPP10 when you sign up. And don't forget this part. Send at least $5 to a friend in the first two weeks. Terms apply. Cash App is a financial services platform, not a bank. Banking services provided by Cash App's bank partners. Bitcoin services provided by Block, Inc. brand. For additional information, see the Bitcoin disclosures at cash.app slash legal slash podcast.

CUDA Launch

28:47Hey, welcome back to How I Built This. I'm Guy Raz. So it's the late 1990s, and NVIDIA has escaped extinction by using an unproven machine to test its latest chip. And as it turns out, that new chip, the Reva 128, is a hit, which means the company can now write its next chapter. We're 100% focused on the gaming industry. We were the first computer graphics company ever created to focus on one application industry fully, which was video games.

29:19So the way we saw the world, that the chip is important, but ultimately what makes people happy and what really creates industries is the applications that were on top of the chips that make new things possible, recognizing the importance of application developers or game developers so that we can help them realize the full potential of our chip while we make their application as wonderful as possible. I know, I want to just pivot for a second because I, you know, here you are, you're still a young guy, you know, in your late 30s, mid-late 30s, running a company.

29:56It's the first time you are the CEO of a company. It's not like you have any management training or you need to go to business school, which is very normal, right? And you kind of had to teach yourself because you have and had a reputation for being like a hard-ass, yelling at people, demanding excellence, and really just tearing into people when they, when you feel like they didn't do great work. But I read that you stumbled upon this book by Christian Claytonson. Clayton Claytonson, forgive me, he's no longer alive, hard with the score. That's right. It's a book called The Innovator's Dilemma.

30:27Incredible, yeah. Very important book. And it had a huge impact on you. Like, you read this book because he has this example of, like, Honda. They were making motor scooters for kids and no one was paying attention to them. And so when they started making cars, they had this advantage because they could really scale quickly and get out there. And you read this book and you were inspired by it. What do you remember about that book that you thought, that's it, this is the thing that I need to think about? The single most important thing about technology is the moment it becomes good enough, when you over-serve the market, you're ready for disruption.

31:03In fact, at the NVIDIA journey, every single step of the way, we were always the disruptor. So I think the large lesson of Clay Christensen is really about disruption and how technology emerges out of thin air that apparently looks like toys, but went off to disrupt large markets. If you looked at NVIDIA's Reaver 128, the quality of it was okay, but it was good enough that it disrupted an entire market. If you looked at NVIDIA's first GPU that was designed for high-performance computing, it was not perfect, but it was good enough.

31:41And so example after example after example, the way that we went off to revolutionize large industries, initially it looks a little toyish, but the outcome has always been the same. All right, so I want to jump ahead a little, because in the late 90s, NVIDIA developed a new technology called parallel computing. And this basically gave your chips the ability to perform multiple calculations at the same time, multiple tasks, right? But this was another gamble, because I think a lot of other companies had tried and failed to produce parallel computing chips, right?

32:18During that time, there were all kinds of different processors being created, and people were trying to come up with new ways of doing computation. And then we realized that computer graphics, if it was just beautiful, but the world was static, it was hard to create beautiful and immersive worlds. And so you really need to find a way to bring physics into that virtual world so that, you know, waters would flow and, you know, leaves would blow in the wind and explosions would look like explosions.

32:51And so we would try to use the processor, which was incredibly parallel, to express, you know, the types of algorithms that represent real-time physics today. And so that was really the beginning of our journey down that world of general-purpose programmability. Meanwhile, scientists around the world noticed that NVIDIA's processors were super powerful, lots and lots. You could do multiple things. Lots, yeah. Because you were thinking, okay, this could be used by game designers, maybe in the film industry, which it eventually would be.

33:23But still, it was like you were thinking beyond people playing video games on their consoles or on their PCs. We would use it for fluid dynamics and image processing and particle physics. And one of the areas that really caught my eye was the whole field of inverse physics or imaging. And two doctors at Mass General were using our graphics chips to do CT reconstruction. And it was during that time other types of techniques for general-purpose computing was coming along, which ultimately led to what we now call CUDA.

33:57All right, so let's jump into this because you launched this project called CUDA or C-U-D-A in roughly 2006. And to put this in very basic terms, CUDA is this platform that makes your graphics chips a lot more versatile, right? And originally, you thought this would be great for scientists and researchers who had to process tons of data. But meantime, your customers were gamers. And so they were getting these really sophisticated chips that they didn't really need or use, right?

34:30Yeah. And so you just go back to the inspiration that we had at the time that it could be used for a whole bunch of other types of general-purpose computation, parallel computing nature computation. And image processing could be one of them. You know, there's a whole bunch of them in early experimentation, but none of that paid the bills. Most of those applications were in universities, and the researchers, you know, didn't buy too many of them to justify it. Yeah. And so the only thing that really paid the bills was video games. You understood that the GPU could do something, lots of things, but that we just had not imagined those things yet.

35:07Yeah, we knew of some things, and during that time, I was constantly looking for algorithms that required parallel computing. I was always looking for algorithms that somehow only ran on supercomputers, but if we could just figure out a way to bring it into a personal computer, its exposure or its ability to reach more people could be incredible. Constantly looking for things like that. I'm trying to understand, because basically, from 2006 to 2012, your stock price was just either collapsing at times or wasn't moving.

35:43You had to refund hundreds of millions of dollars to people who didn't like the processors. And still, there was this conviction in continuing down this path, this CUDA path, because you and the people around you believed that there was something there that you couldn't really know exactly what it was, but it was going to be something, so let's keep investing in this. And I'm trying to get into just this space of how you were dealing with, you know, your publicly traded company, your stock price was terrible, you probably had a lot of pressure from investors.

36:20What made you and the people that you worked with say, we're going to keep our heads down and keep going? Because it was a long time. We're talking six, seven, eight years on this thing that nobody understood and had zero, not zero, but very little commercial success. Well, that's when CEOs had to be CEOs. We believed on first principles, this should be quite useful. And I had to believe that it's quite useful. Now, the question is, what's the strategy for creating this new architecture for computing that everybody would be able to enjoy?

36:55And the problem with computer architectures is this chicken or the egg problem. Let's say you created a brand new architecture. It's incredible. It's the most amazing thing in the world. But computers are built to run software. And if your install base is not large enough, it doesn't attract software developers because developers want to program on large install-based computers like iPhone and PC. And so the problem is, even though we believed that this architecture was going to be incredible, that CUDA was going to be everywhere or could be everywhere, how do you get it everywhere?

37:30And if you don't get it everywhere, how do you attract the developer? And if you had no developers, who would write the killer app? And if there was no killer app, then why would people buy it? And so the answer was very simple. It was literally sitting in front of us. And it just required enormous sacrifice. The answer was, let's use GeForce, which is the GPU that is now everywhere in the world used for playing video games. And let's have GeForce carry on its back CUDA to every single computer in the world.

38:01Now, of course, by doing so, our gross margins would go from bad to horrible. Okay, well, let's get past that. Let's not worry about the fact that our final… It's going to be years. I know. Well, you never think that it's… Nobody ever thinks it's that long. And I thought it was going to be next year. It's going to be okay. And then next year, it's going to be okay. And next year, it's going to be okay. But nonetheless, the cost was incredible in carrying CUDA. And we couldn't charge anybody for it. And so we gave it away. Meanwhile, we started up programs to reach out to every single university.

38:35We evangelized CUDA. We flew everywhere around the world to pitch CUDA whenever you can. Meanwhile, GeForces were taking CUDA out to the world. Yeah. And so the hope was that someday, some software company or some researcher or hopefully a lot of them eventually would know how to use CUDA and they would take advantage of it because it's sitting right there on GeForce. Was there ever any thought in your mind that maybe we're too early? We might be the ones who actually make this revolutionary technology, but we might not be the ones to

39:09bring it to the promised land. Yeah, all the time. But that's, you know, then there was 900 other smart strategies and, you know, the list of new ideas that we came up with to keep the company alive and, you know, successful for just another few more days. And it was countless. And so you're solving the problem both in making sure that you stay alive long enough to proliferate this technology everywhere, looking for every possible way. You're making it easier and easier for people to use this technology.

39:41You're teaching people to do it. You're talking to software developers and you're saying, hey, you know that imaging software that you had? Maybe Photoshop, for example, or some video imaging system for broadcast. Can we modify that so that it runs on CUDA? And we're live, I mean, but were enough people adopting it to make it seem viable? G-Force kept the lights on and just didn't do it very well. You know, we were always under pressure because unlike other, our competitors, they didn't have to carry CUDA on its back.

40:13G-Force carried CUDA on its back for literally 25 years now. And people were using this product and not even knowing what it could do. Not one day. Yeah. It was just all sitting there except for the scientists, except for researchers and universities that heard about CUDA and said, oh, all you have to do is go buy G-Force. But then at this point, I mean, everything was about to change for NVIDIA because in the early 2010s, a group of researchers in Toronto bought two NVIDIA gaming GPUs, right, or G-Forces.

40:46They plugged them into a computer in a bedroom, and then they trained a neural network to recognize images way better than any computer had ever done before. And this was essentially the beginning of the AI boom, right? So when you found out about that, what do you remember? I mean, do you remember thinking, this is what we've been waiting for for 15 years? Well, we, at that time, several different groups reached out to us to ask us for help on using CUDA to accelerate deep learning.

41:19And the reason for that was because there was a contest coming up for computer vision called ImageNet, and they were all developing similar techniques that wanted to use CUDA instead of using CPUs, which would have taken thousands of CPUs. CPU is one task at a time. One task at a time, and they could use our GPUs with just a few GPUs, maybe a couple of them running simultaneously on CUDA. Maybe they could train these deep learning models a lot faster and a lot more cost effectively. And so they reached out all at the same time, and the leap over previous generation algorithms

41:54were so significant, it really caught my eye. And so we asked ourselves, you know, what is this thing about? And where can it go from here? What is the implication to our chips? What's the implication of the computers? And so, like we do with everything, you know, reason about the so what's, which led to a lot of other good decisions. There was a weekend, I guess, in 2013, where you sent a note out and essentially said, we're now an AI company. Like, Friday, you were focused on the gaming.

42:24I mean, because the majority of your business was still gaming, maybe people who were doing 3D animation and doing graphics for films. That was a pretty significant source of your customer base. But now you are saying, well, we are becoming a different company. Yeah, and the process kind of goes like, you know, what I was explaining earlier, as I do with everything, I break things down to first principles. And you ask yourself, what did I learn?

42:54Why is it impactful? What are the foundations of this technology that made it effective? Can it do more than this? How far can this algorithm go? What is the implication to our computer industry? So you just got to go through all of that. You know, it's no different than somebody writing a business plan about something, except, you know, this is kind of how my brain's wired for almost everything. I'm good at connecting dots, good at system thinking. And, you know, I came to the conclusion that this could be a really significant future direction

43:25for the company. And long before I would send out an email that declares something, I had already through tens of engagements with different groups. Each one of the groups have already been brought along. So by the time that I sent an email to the company integrating everything, everybody's already been brought along. And so it just kind of tends to be my management style. But at some point, I will, you know, set a direction for the company.

43:54I'm curious, again, about stress management and conviction. Like, I wonder how you personally dealt with that. I mean, there were periods of time where you and obviously other people believe in the potential of this thing. Like, the capacity is going to be enormous. But we just have to stay the course. And nothing is really moving for a while. I mean, it's going to take a long time. But you don't know that. You don't have a crystal ball. I mean, are you just comfortable with that level of stress?

44:28I mean, do you feel – because you don't strike me as somebody who was saying, we're going to be the biggest company. We're going to be the masters of – Not going out of business is always high on my list. Yeah. Yeah. Showing up in life matters a lot. Showing up every single day matters a lot. Guy, I guess to me it's not that complicated. There's a lot of unknowns. And being a CEO, you're dealing with mostly unknowns. And I'm very comfortable with unknowns. But the first thing you have to do is, to the best of your ability, reason about what

45:02it is that we're doing and what do we believe in. I deeply believed. And because I deeply believe it, I help everybody else believe it. And I really believe they believed it. And then, of course, we don't want to dedicate our lives to go work on something that the world already have. I don't want our company working on things to capture share from somebody. They built a market. We want to take their share. And so let's go and fight to the bitter end. And when your share point goes up by a point, you celebrate with joy.

45:33I mean, I just don't find any joy in any of that. That wasn't interesting. It was about going in a completely different direction. Creating something that's really, really hard. And I believe if it works, it would make such an extraordinary impact. It was not hard to stay the course. It would have been harder to give up. You know, I think that because you believe so deeply in something, it's already manifested fully in my head. I imagine everybody using it. G-Force has carried it to everybody's computer. Everybody has CUDA inside their PC.

46:03And so the question is not to me whether we would succeed. The only question is when and who is going to be the first application. When we come back in just a moment, NVIDIA becomes a major player on the world stage. And Jensen reflects on the future of AI and on his own reluctance to reflect on himself. Stay with us. I'm Guy Raz and you're listening to How I Built This. How I Built This Hey, welcome back to How I Built This.

46:51I'm Guy Raz. So it's 2013 and NVIDIA is now pivoting away from making chips for gaming to making chips for AI. And as the technology takes off, NVIDIA starts to soar right along with it. And I'm going to jump ahead to the part of the story that's familiar to all of us because we've all lived it. By 2022, ChatGPT and Claude start to transform the workplace in our daily lives. Massive data centers go up around the world, and most of them are full of NVIDIA chips.

47:24NVIDIA becomes the biggest company in the world and has as much influence on world affairs as some countries, maybe most countries. But just to state the obvious here, a lot of that can feel kind of disconcerting. You know, you're well aware of lots of people who have a lot of worry about this. I mean, Jeffrey Hinton, one of the earliest users of your GPUs to train these networks, they've signed letters saying, you know, there's a famous letter came out May of 2023 saying,

47:56look, to quote, mitigating the risk of extinction from AI should be a global priority alongside other societal scale risks such as pandemics, nuclear war, signed by Bill Gates and Sam Altman and Dario Maure and many, many others. You didn't sign this letter, and you have been really clear that you are not worried about it. You think a lot of this is alarmism, a lot of the conversation around AI and how it's going to take over our jobs and can manipulate, you know, all kinds of things. Um, why do you think the people who are making those warnings wrong?

48:30Yeah, I think the first thing that everybody should do is to take the science fiction and the Hollywood versions of artificial intelligence and set that aside and come back to a sensible understanding of what a computer is. It's computers, it's software, it is not conscious. Having said all that, I think we all, everybody wants the same thing, that the technology evolves in a safe way. I am worried about it.

49:01And so the question is, how do you channel your worry? What are you worried about? Well, we have to build the technology safely. And so could you imagine if the auto industry every single day told you that the car is going to kill lots and lots of people, it's their job to make the car safer, channel their worry towards making the car safer. I don't think it's helpful for the airline industry to tell you that every single day if the plane fails, it would crash out of the sky. I think all of those narratives are not helpful.

49:33But they're coming from the people who are making some of this technology. I guess I've got a different perspective on why it is that they're doing that. Maybe it's because they want the world to realize the technology they invented was so powerful. It's a celebration of their own achievements. Maybe they are genuinely scared themselves and that they believe that only they can create a technology and keep it safe, ignoring the fact that there are millions of innovators

50:06around the world who are thinking about AI safety, AI security, are building technology for AI safety and AI security. There were even attitudes that we had to slow down AI innovation, which I believe is exactly the opposite. We had to speed up AI innovation because through innovation, we can make the technology safer, more secure. And obviously, it's proven to be true. The fact of the matter is the AI technology of today is more grounded on truth. It hallucinates less.

50:36It lies less. And all of that technology needed to be invented. And so my single greatest worry is that the United States doesn't take advantage of the technology because we scared everybody. What do you mean? Right now, the sentiments for AI in the United States is lower than most countries. Because the message is it's going to take your job? It's going to take your job. It's going to be existential threat. It's going to go into a singularity and it'll be the end of the world. And you think that conversation is only happening here or in the West?

51:07Oh, by far. By far. I travel the world by far. In the West, it's really out of control. And I think we're doing ourselves a disservice. Let me give you an example. So a decade ago, somebody said, listen, the first profession that's going to be wiped out is radiology. Yeah. And the reason for that is because the first model that was invented was used for computer vision. Computer vision is used, would be trained to read radiology scans better than any humans can. And therefore, no radiologist would be necessary. Well, it turns out they were absolutely right.

51:38A hundred percent right. AI has now completely revolutionized radiology. Every single radiologist uses AI to study the scans. Right. It does it incredibly fast, incredibly accurately, better than a human can. However, radiologists' demand has gone up. Yeah. The number of patients that we're able to take care of, the number of scans has now grown. Because it's cheaper. Because it's cheaper and faster. And therefore, you want to do more scans so that you can do a better job diagnosing disease. What is fundamentally missed is that a job has a purpose and a task.

52:13The task is to study the radiology, the scans. But the purpose is to diagnose disease. Turns out hospitals are now seeing more patients, generating more revenues, which enables them to support more patients and they need to hire more radiologists. Now, where's the harm? The harm could be that by well-known experts declaring the end of an entire profession, who's going to go into radiology? So, young people who wanted to be in radiology decided that this profession is completely obsolete

52:46and therefore don't come in. What happens? The world doesn't have enough radiologists. As a result, we've brought harm. So, you're saying the conversation we're having here is very binary in the United States. We're saying it's going to be this or this. And so, the argument you're making is that there's a lack of imagination. So, for example, you know, I've got a son who's going to go to college next year, right? And so, a lot of the conversation is, what are these kids going to do in five years? What are the jobs going to be? I mean, if law firms are using AI for discovery, if consulting firms are using AI for, you know,

53:19what an entry-level job would be if finance firms, et cetera, et cetera. Would you argue that that's a lack of imagination, that we actually are reaching those conclusions because we can't think beyond that scenario? Partly. I believe the opportunities, the potential for a new college grad in computer science or computer engineering or software engineering or chip design today is far, far greater than the opportunities and potential when I came out of school. The tools they have to work with is a billion times more capable.

53:51It's super highly automated already today. And yet, they're busier than ever. And the reason for that is because we have ideas of the things that we want to build that we didn't conceive of at the time without the tools that we have. AI is going to help this next generation of new college grads achieve greater things, build greater things, not take their jobs. For us to scare them into not even want to go to college, okay, not even want to be a computer scientist is a disservice to society.

54:22You're not saving anybody. You're talking a whole bunch of people that are out of professions that we need in the future. And that conversation is not happening in China? They're not buying it. They're just not buying it. For me, the most harmful thing we could do to our society, our closest families, is to scare everybody so that we don't benefit in the next several decades the other countries will. We can't allow that to happen. Do you think that there is a possibility that in 10 years from now, NVIDIA will have an

54:54order of magnitude more employees?

54:59Before AI or after AI, I don't think it would make any difference. The difference is that with AI, along with my employees, we're going to have hundreds, thousands of AI assistants helping them doing amazing things. And so our expectations of our company will be different. Yeah. Things that take 10 years, I think it'll take one year. And so is that good or bad? It's thrilling. It's exhilarating, I think. Yeah. You know, there's just so much of the universe that we don't understand.

55:30For the first time in history, we have the technology to do so. We will do in the next several decades what it took humanity several hundred years to achieve. Well, as a dad of two boys, I hope you're right. Yeah, I expect to be. I know that you don't like to talk about speculation, but I read an excerpt from a new book that's coming out. It's called Defending Taiwan by a writer named Ake Freeman. And he paints a scenario of a possible conflict, hot conflict, where, you know, all of a sudden, Taiwan is somehow invaded or occupied, whatever you want to call it, right?

56:09And we've seen what, you know, we've got some challenges geopolitically, right? I mean, what part of that, let's just say that scenario is not realistic, that it would be destructive for China as well, because our economy really depends on the U.S. market. Does any part of the fact that such a high percentage of advanced chips, not just NVIDIA, but advanced chips are produced in Taiwan, is any part of that, should people be concerned about that at all?

56:41We should always have a resilient supply chain. That's part of building a strong company. And a resilient supply chain has diversity and redundancy, but sometimes you don't have the benefit of diversity and redundancy, so you can make the best you can. I think there are a lot of conflative questions that are put into that concern. You know, if you ask me, am I concerned about a hot conflict, I'm less concerned than most.

57:12But depending on the actions of the leaders, we may cause the hot conflict. And so I am of the opinion that we all must be more long-term minded if we can. We ought to have a more balanced and nuanced set of policies, not all or nothing. Don't push your adversaries and don't push your competitors to the wall where they have no choice or where there's no cost to a strategic alternative.

57:43And so I think that, you know, in your question, do I believe it? The answer is I don't. Because I believe, can it happen? Of course, it can happen everywhere, obviously. However, I believe in the wisdom of the people involved.

58:00What have you had to unlearn as a leader? The bigger this company has gotten and the more experience you've gotten, I mean, you are, you know, known for being hard demanding, hard charging. You demand the best out of people. You have yelled at people, all those things. Would you say those things are still how you manage or have you unlearned some of those things or have you changed some of those things in your style? I don't think people care about style.

58:30I think people care about your values and what you care about. So long as you're tough on the same things every time it comes up. So long as the moment that it's over, it's over, and they know that they're safe, you love them, you want them to succeed, this is their life's work, and you want it to be as good as it can be. And everybody knows that I have enough of everything. Whatever it is that anybody thinks they need more of, I've got plenty of it.

59:00And yet, I work harder than ever, and the reason for that is because I want everything that I have, they have. I want them to see, to realize their dreams. So long as you're pure and high integrity and completely there for them, you could be quite challenging to people, and they know that you're on their side.

59:21You have been on the record as basically saying, you've been asked, would you do this again? Would you do this all over again? And I'm sort of paraphrasing, but you said a version of no. I wouldn't do it again. Yeah, no way. Yeah. So can you explain that? I mean, we've sort of had the outlines of all of the troughs, and there were a lot of troughs. Yeah. And I'm trying to even imagine like quarterly. It's rough. Yeah. I'm trying to imagine quarterly earning calls in 2007 when your stock price is in the toilet, and you're putting all this money. I've got to tell you, it's embarrassing.

59:51I'm trying to figure out how you can do that as a publicly traded company, withstand the pressure, and, as you say, the embarrassment. It was embarrassing. It was humiliating. You're the only face that everybody hates. Your employees are probably embarrassed for you. Your question about doing it again, you know, most people, I just think they're being dishonest.

1:00:14So let me just tell you why. When somebody asks me, would I do this again? If your question is, knowing how NVIDIA turned out. Yeah. Knowing the contribution we've made to the world. Knowing the consequence of the company today. How it impacts so many different industries. Yeah. All of the benefits that we have accrued as a result of our success. Do I love those things? The answer is yes. But that wasn't the question.

1:00:44You know? Right. The question is, suppose I knew everything then that I now know how hard it is. And all of the pain and suffering and all of the pain and suffering and all the embarrassment and humiliation and all the setbacks. And you compress all of that and you just told that 30-year-old kid, listen, this is going to take a lot longer than you think. However fast you thought your company was going to be successful, it's not going to be the person who delivers most of the most horrific financial return news that anybody's ever explained.

1:01:20You know, so on and so forth. You'll be going out of business. You'll be, you know, you have to lay people off. Would you start again? The answer, absolutely not. And so I think a lot of people forget that the pain and suffering necessary, the endurance necessary to do something great, is because you're always looking forward and forgetting the past. Yeah. I spend all my time forgetting yesterday. Yeah. In fact, you know, right now as we're talking, it is really quite hard for me to resurface those memories and resurface those feelings.

1:01:56And the reason for that is I spend all of my time, all of my life trying to forget yesterday. Right. So that you can get back on that horse. That's, in fact, what do they teach athletes? Forget the last point. Yeah. It's about forgetting. Yeah, I understand. I mean, the sacrifice, right? I mean, you've got a family. Like, you probably missed a lot of things. You probably did 12-hour days, seven days a week, four years. And after you sort of really began to focus on AI, it doubled again.

1:02:28I mean, it was all those things. All true. All true. And during the time when the kids were still young, I was finishing my master's at Stanford. And so I was busy on multiple levels. And, yeah, I missed all their karate tournaments. I missed a lot. I missed a lot. I don't know that I've ever asked them what it was like on the receiving side of it. But if you're going to build a company, you need somebody strong and you need somebody amazing like my wife.

1:03:00And she took care of everything. Lori took care of everything. And I don't remember her complaining even one time, one nanosecond or even one instance. That's the gift that she gave me. And both kids love NVIDIA to their core. And both work here. Both work here. They both went off in their own careers for a while. And we were lucky to attract them back. And both kids have read every shareholder letter, gone to every shareholder meeting, been to every single conference that I've ever done.

1:03:35I've missed most of theirs. They've been to all of mine. And Spencer's 35, Madison's 34, and the company's 33 years old. And that kind of puts it in perspective. I mean, they've had NVIDIA in their lives the whole time. And despite the fact that it was gone a lot, you know, they found a way to always love the company. And I'm really lucky.