英伟达™（NVIDIA®）首席执行官黄仁勋揭示人工智能和领导力的关键所在

哥伦比亚商学院的这次活动邀请了NVIDIA的联合创始人兼CEO黄仁勋，并由Siamak Malemi和院长Costis Maglaras主持。黄仁勋分享了NVIDIA的旅程和对人工智能、加速计算以及公司战略的见解。讨论范围从创业、领导力和商业哲学，到未来人工智能趋势和地缘政治考量。

关键要点

NVIDIA的基础：黄仁勋和他的联合创始人在早期PC革命期间创立了NVIDIA，专注于“加速计算”，补充而非取代CPU，这从图形处理引领到了人工智能的进步。

CEO和创业者视角：黄仁勋强调了战略的重要性以及选择对团队来说具有突破性、挑战性和激动人心的项目的重要性，避免商品市场以培养一个吸引伟大人才的环境。

增长和管理：NVIDIA经历了快速的增长，这得益于出色的管理团队，并保持对长期目标和在创新技术领域的投资的关注。

加速计算与人工智能：黄仁勋区分了摩尔定律和“黄氏定律”，指出理解应用领域以创建架构的重要性，这些架构能够实现相比CPU的显著加速。

人工智能趋势和应用：预测人工智能不会取代工作，但那些有效利用人工智能的人将更加成功。预见人工智能在包括蛋白质工程以及识别各种领域内在结构方面的重大进步。

地缘政治问题：确认NVIDIA认识到国家安全和经济安全考量，表达了对继续全球合作以及国际学生和外国人才在美国的关键作用持续发挥希望。

教育鼓励：鉴于技术的持续和快速发展以及战略思考对领导力的重要性，提倡在冒险出发前尽可能多地从教育机构获取知识的机会。

结束语：院长Maglaras以感谢黄仁勋的见解和展示他为何被认为是顶尖执行官作为讲话的结束。

全文（英文）

Good evening, everyone, and welcome. My name is Siamak Malemi. I am the William von Muffling Professor of Business in the Decision, Risk, and Operations Division here at Columbia Business School. I want to thank you all for joining us this evening for tonight’s program, which features Jensen Wong, co-founder and CEO of NVIDIA Corporation, as well as our own Costis Maglaras, the Dean of Columbia Business School, and the David and Lynn Sifflin Professor of Business. Our two speakers tonight have much in common. In fact, they both graduated from Stanford in electrical engineering, NVIDIA. He has revolutionized first the graphics processing unit industry, and now, more recently, the artificial intelligence industry. He has been named the world’s best CEO by Harvard Business Review and Brand Finance, as well as Fortune Magazine’s Business Person of the Year, and one of Time Magazine’s 100 Most Influential People. Our fireside chat today has been made possible through both the David and Lynn Sifflin Leadership Series, as well as the Digital Future Initiative, and additionally, I serve on the leadership of the Digital Future Initiative, the DFI, here at CBS. The Digital Future Initiative is CBS’s new think tank, focusing on preparing students to lead through the next century of digital transformation, as well as helping organizations, governments, and communities better understand, leverage, and prosper from future waves of digital disruption. Now I would like to hand it over to Dean Maglaras. Thank you, Siamak. Thank you all for coming. So this is an exciting topic, and a topic that is near and dear, certainly, to my heart. And it’s a topic where the school, everything that we do at the school, is changing so fast. Trying to keep up. Trying to change.

change curricula, trying to create opportunities for our students to actually learn about technologies and how they’re changing the world, and to be honest, prepare for the future. And there is no better person to be having to talk about AI than Jensen Huang. Jensen, thank you so much for making the time and coming here. Welcome. I’m delighted to be here. Is this on? Yes. I just love hearing that I’m the best CEO in the world. Can we end the talk here? I think the expectation is going to be pretty high to say something smart. Well, good luck with you. So I want to start by having you walk us through a little bit the history of NVIDIA, and then I want to talk a little bit about that leadership thing that we just mentioned. But you’ve been, you launched that company 30 years ago, and you have led it through a transformation, different applications, different type of products. Walk us through a little bit that journey. Yeah. One of the most proud moments, I’ll start with the most proud moment that happened recently. The CEO of Denny’s, where I used to work, was my first company. And they learned that NVIDIA, not only was I a dishwasher, a busboy, and worked my way up the corporate ladder and became a waiter at Denny’s, and that they were my first company. And that I still know how to take, I still know the menu well. Superbird’s excellent, by the way. Anybody know what a Superbird is? What kind of college student are you? Denny’s, it’s America’s diner. Go. And that…

That NVIDIA was founded by Chris Curtis and I outside our home in San Jose at one of the Denny’s there. And so they contacted me recently. And the booth that we sat at is now, you know, the NVIDIA booth. And my name is NVIDIA. This is where a trillion-dollar company was founded. And so that’s a very proud moment. NVIDIA was founded during a time when the PC revolution was just starting. And the microprocessor was capturing the imagination for just about the entire industry. And the world properly saw that the CPU, the microprocessor, was going to be revolutionary. And it really reshaped how the IT industry, how the computer industry was formed. Companies that were successful before the microprocessor revolution, the X86 revolution, and companies that were successful afterwards were completely different. We started our company during that time, and our perspective was that general-purpose computing, as incredible as it is, can’t sensibly be the solution for all kinds of problems. And we wanted to, we believed that there was a way of doing computing we called accelerated computing, where you would add a specialist next to the generalist. The CPU is a generalist, if you will, it could do anything. It could do everything. However, it can’t, obviously, if you could do everything and anything, then obviously you can’t do anything very well. And so there are some problems that we felt that were not solvable, not good solutions for not good problems to be solved by what we call the normal computer. And so we started this accelerated computing company. The problem is, if you want to create a computing platform company, and I don’t know how many computers.

or sentences. But if you want to create a computing platform company, one hasn’t been created since 1964, a year after I was born. The IBM system 360 beautifully describes what a computer is. In 1964, IBM described that the system 360 had a central processing unit, IOS subsystem, large memory access, virtual memory, binary compatibility across a scalable architecture. It described everything that we describe computers to this day, 60 years later. And we felt that there was a new form of computing that could solve some interesting problems. At the time, it wasn’t completely clear what problems we could solve, but we felt that accelerated computing has a future. And so, nonetheless, we went out to start this company, and we made a great first decision that frankly is unfundable to this day. If somebody were to come up to you and said, one, we’re going to invent a new technology that the world doesn’t have, everybody wants to go build a computer company around a CPU, we want to build the computer company around something else connected to the CPU, number one. And the killer app, the killer app is a video game, a 3D video game, in 1993. And that application doesn’t exist. And the companies who would build this company doesn’t exist, and the technology that we’re trying to build doesn’t exist. And so, now you have a company that has a technology challenge, and a market challenge, and an ecosystem challenge. And so, the odds of that company succeeding is approximately zero percent. But nonetheless, we were fortunate because two very important people, frankly, that I had worked with.

and Chris and Curtis, the three of us had worked with, were incredibly important people in the technology industry at the time, called up the most important venture capitalist in the world, Don Valentine, at the time, and told Don, Don, and his name’s Wilf, Wilf Horrigan, one of the founders of the semiconductor industry, told Don, give this kid money, and then figure out along the way whether it’s going to work. And fortunately, they did. But that business plan, I wouldn’t fund myself today. And it just has too many dependencies. And each one of them have some probability of success. And when you compound all of these together, when you multiply all these together, you’ve got 0%. And so nonetheless, we imagined that there would be this market called video games. And this market would be the largest entertainment industry in the world. At the time, it was zero. And 3D graphics, we postulated, would be used for telling the stories of almost any sport, any game. And so in virtual worlds, you could have any game, any sport. And as a result, everybody would be a gamer. And so Don Valentine asked me, so how big is this market going to be? And I said, well, every human will be a gamer someday. Every human will be a gamer someday. Also the wrong answer, quite frankly, for starting a company. So these are horrible habits. These are horrible skills. I’m not advocating them for you. But nonetheless, it turned out to have been true. Video games turned out to have been the largest entertainment industry in the world. 3D graphics took off. And we found the first killer app for accelerated computing, which bought us the time to use accelerated computing to solve a whole bunch of other problems, which eventually led to artificial intelligence. This is a fantastic story. So before we go to AI.

I would like to ask a little bit about the crypto period. So gaming was a huge, obviously, journey for NVIDIA. And then at some point in time, the killer app became crypto and mining. What was that chapter? Yeah, accelerated computing can solve problems that normal computers can’t. And all of our GPUs, even though you use it for designing cars, designing buildings, use it for molecular dynamics, use it for playing video games, it has this programming model called CUDA that we invented. And CUDA is the only computing model that exists today that is as popular as an x86. It’s used by developers all over the world. And so anyways, one of the things that CUDA can do is parallel processing incredibly fast. And obviously, one of the algorithms that we would do very nicely on is cryptography. And so when Bitcoin first came out, there were no Bitcoin ASICs. And the obvious thing is to go find the fastest supercomputer in the world. And the fastest supercomputer that also has the highest volume is NVIDIA GPUs. It’s available in hundreds of millions of gamers’ homes. And so by downloading an application, you could do some crypto mining at your house. Well, the fact that you could buy one of our GPUs, one of our computers, and you plug it into the wall, and money starts squirting out. That was the day that my mom figured out what I did for a living. And so she called me one day, and she said, son, I thought you were doing something about video games. And I finally figured out what you do. You buy NVIDIA’s products, you plug it in, and money squirts out. And I said, that’s exactly what I do.

I do, and that’s the reason why so many people bought it. Bitcoins, of course, led to Ethereum. But the idea that you would use a supercomputer, use a super processing system like NVIDIA GPUs, to either encode or compress or do something to refine data and transfer it, transform it, into a valuable token. You guys know what that sounds like, to generate valuable tokens? Chat GPT. And so to this day, really one of the things that’s happened, if you extend the sensibility about Ethereum and crypto mining, it’s kind of sensible in the sense that all of a sudden we created this new type of industry where raw data comes in, you apply energy to this computer, and literally money comes squirting out. And the currency is, of course, in tokens. And that token is intelligence tokens. This is one of the major industries of the future. Now, I’ll describe something else, and it makes perfect sense to us today, but back then it looked strange. You take water and you move it into a building, you apply fire to it, and what comes out is something incredibly valuable and invisible, called electricity. And so today, we’re going to move data into a data center, it’s going to refine it, and it’s going to work on it, and it’s going to harness the capability out of it and produce a whole bunch of digital tokens that are going to be valuable in digital biology, it’ll be valuable in physics, it’ll be valuable in IT and computer, all kinds of computing areas and social media and all kinds of things, computer games and all kinds of things, and it comes out in tokens. So the future is going to be about AI factories, and NVIDIA Gear will be powering these AI factories.

into the neural networks and I want to, and we talked about parallel computing, how we render graphics, let’s say, on a monitor, how we play games, how we solve cryptographic problems for Bitcoin. Talk to us a little bit about how the GPU is useful in training neural networks, but then what I wanted us to do for this audience, tell us a little bit about what it takes to train a model like JAD-GPT, what it takes in terms of hardware, what it takes in terms of data, what it takes in terms of the size of the cluster that you’re using, the amount of money that you need to spend, because these are like huge problems and I think giving us a glimpse of the scale will be fun. Well, everybody wants you to think that a huge problem is super expensive. It’s not. It’s not. And let me tell you why. It costs our company about five, six billion dollars of engineering cost to design a chip. And then at one point, two years, three years later, I hit enter and I sent an email to TSMC and I FTP basically a large file to TSMC and they fab it and that process cost our company something along the lines of half a billion dollars. So five and a half billion dollars, I get a chip. And that chip, of course, is valuable to us. But it’s no big deal. I do it all the time. And so if somebody were to say, hey, Jensen, you need to build a billion dollar data center and once you plug it in, money will start squirting out the other side, I’ll do it in a heartbeat. And apparently a lot of people do. And the reason for that is because who doesn’t want to build a factory for generating intelligence? Now so a billion dollars is not that much money, frankly.

The world spends about $250 billion a year in computing infrastructure, and none of it is generating money. It’s just storing our files, passing our email around, and that’s already $250 billion. One of the reasons why we’re growing so fast is because after 60 years, general-purpose computing is on decline because it’s not sensible to invest another $250 billion to build another general-purpose computing data center. It’s too brute force in energy. It’s too slow in computation. And so now accelerated computing is here. That $250 billion goes to build accelerated computing data centers, and we’re very, very happy to support customers to do that. And in addition to that, that accelerated computing, you now have an infrastructure for generative AI for all of the things that we’re just talking about. Basically the way it works is you take a whole lot of data and you compress it. You compress it. Deep learning is like a compression algorithm, and you’re trying to figure out, you’re trying to learn the mathematical representation of the patterns and relationships of the data that you’re studying, and you compress it into a neural network. So what goes in is, say, trillions of bytes, trillions of tokens. So let’s say a few trillion bytes. What comes out of it is 100 gigabytes. And so you’ve taken all of that data and you’ve compressed it into this little tiny file. 100 gigabytes is like two DVDs. Two DVDs you could download on your phone and you can watch it, right? And so you could download this giant neural network on your phone, and now that all of this data has been compressed into it, the data, this compressed neural network model is a semantic model, meaning you can interact with it. You could ask it questions, and it would go back into its memory and understand what you meant and generate text for you. Have a conversation with you.

So at the core is kind of like that. It’s it’s not you know, it sounds magical but you know for all the computer science and scientists in a room, it’s very sensible and Don’t let anybody convince you it costs a lot of money. I’ll give you a good break Everybody go build a is go build a eyes the The scale if I press you a little bit on that scale like how do you need a A computer that is essentially a data center to to estimate these models 16,000 GPUs is what it took to build GPD for which is the largest one that anybody’s using today. That’s a billion dollars Okay, and that’s a check It’s not not even a very big check. Yeah, don’t be afraid Don’t let anybody talk you out of building a company Yeah, build your dreams Let me ask you a question about the billion-dollar check and the growth that you’ve been experiencing I Think you you were named the best CEO by HBR That’s entertainment. I’ll keep repeating it and then eventually that and then eventually we’ll end with that line. But But in some sense You’re leading a company right now through a period of extreme growth hyper growth Something that most companies have not experienced in their life and I want to perhaps it tell us a little bit about What does it look like? I mean, you know sort of doubling in size In under a year or you know sort of managing supply chains managing customers managing growth managing money How do you actually do that I Love the management money part of it just counting it

It’s fun. You just wake up in the morning, just roll around in all the cash. Isn’t that what you guys are all here to do? My understanding is that’s the end goal. That’s the end goal? Yeah. Let’s see. The building companies is hard. There’s just no, there’s nothing easy about it. There’s a lot of pain and sufferings, a lot of hard work. If it was easy, everybody would do it. And the truth about all companies, big or small, ours or others, in technology, you’re always dying. And the reason for that is because somebody is always trying to leapfrog you, so you’re always dying. You’re always on the way of going out of business. And if you don’t internalize that sensibility, if you don’t internalize that belief, you will go out of business. And so I started at Denny’s, as you guys know. And NVIDIA was built out of very unlikely odds. And it took us a long time to be here. I mean, we’re a 30-year-old company. And when NVIDIA first founded, the PC, Windows 95 hadn’t come out, 1993. And that was the first usable PC. We didn’t have email. And so there were no such thing as laptops or smartphones. None of that stuff existed. And so you could just imagine the world that we were started in and the world we are. We didn’t have LCD. Everything was CRTs. And so the world was very, very different. CD-ROMs didn’t exist. I mean, just to put it in perspective, all this stuff, that was the era we were founded in. And it took this long for our company to be recognized as having reinvented computing.

for the first time in 60 years. Growing fast is all about people. Obviously, companies is all about people. And if the right systems are in place, you’re surrounded by amazing people like I am, and the company has craft, has skills, it doesn’t really matter whether you ship $100 billion or $200 billion. Now, the truth is that the supply chain is not easy. People think, does anybody know what a GeForce graphics card looks like? Can you just show me as a hand? Anybody knows what an NVIDIA graphics card looks like? And so you have a feeling that the graphics card is like a cartridge that you put into a PC, the PCI Express slot of a PC. But our graphics chips these days, what is used in these deep learning systems, is 35,000 parts. It weighs 70 pounds. It takes robots to build them because they’re so heavy. It takes a supercomputer to test it because it’s a supercomputer itself. And it costs $200,000. And for $200,000, you buy one of these computers, you replace several hundred general-purpose processors that cost several million dollars. And so for every $200,000 you save, for every $200,000 you spend with NVIDIA, you save $2.5 million in computing. And that’s the reason why I tell you, the more you buy, the more you save. And apparently it’s working out really well. People are really lining up. And so that’s it. That’s what we do for a living. And the supply chain is complicated. We build the most complicated computers the world’s ever seen. But how hard can it be, really? It’s really hard. It’s really hard. But at the core of it, if you’re surrounded by…

The simple truth is that it’s all about people and I’m lucky to be surrounded by a great management team. And then the CEO says things like, Make it so, number one, something like that. Make it work. Make it so. Make it so. I want to go back to AI trends and what you think about the future. You mentioned the word platform earlier on. You mentioned your software environment. So you have the hardware infrastructure, you have a software environment that is actually pervasive in training neural networks right now. You’re building data centers or you’re creating environments within data centers that are sort of clusters of NVIDIA hardware, software and telecommunication between these resources. How important is it to be a whole platform solution versus a hardware play and how core is that into NVIDIA’s strategy? First of all, before you can build something you have to know what you’re building and why, what is the reason, the first principles for its existence. Accelerated computing is not a chip. That’s why it’s not called an accelerator. Accelerated computing is about understanding how can you accelerate everything in life. If you can accelerate everything in life, if you can accelerate every application, that’s called really fast computing. And so accelerated computing is first understanding what are the domains, what are the applications that matter to you and to understand the algorithms and the computing systems and the architecture necessary to accelerate that application. And so it turns out that general purpose computing is a sensible idea.

Accelerating an application is a sensible idea. So let me give you an example. You have DVD decoders. You play DVDs or H.264 decoders on your phone. It does one job and one job only and it does it incredibly well. Nobody knows how to do it better. Accelerated computing is kind of at this weird middle. There are many, many applications that you can accelerate. So for example, we can accelerate all kinds of image processing stuff, particle physics stuff. We can accelerate all kinds of things that includes linear algebra. We could accelerate, you know, we could accelerate many, many domains of applications. That’s a hard problem. Accelerating one thing is easy. Generally running everything under a C compiler is easy. Accelerating enough domains such that if you accelerate too many domains, supposedly you accelerate every domain, then you’re back to a general-purpose processor, right? What makes them so dumb that they can’t build just a faster chip? And so on the one hand, on the other hand, if you’re, if you only accelerate one application, then the market size is not big enough to fund your R&D. And so we had to find that slippery middle and that, that is the strategic journey of our company. This is, this is where strategy meets reality and that’s the part that Nvidia got right. That nobody, no other company in history of computing ever got right. To find a way to have a sufficiently large domain of applications that we can accelerate, that is still a hundred times, five hundred times faster than the CPU. And such that the economics, the flywheel, the flywheel of number of domains, expanding the number of customers, expanding the number of markets, expanding the sales, which creates larger R&D, which allows us to create even more amazing things, which allows us to stay well ahead of the CPU. Does that make sense? That flywheel is insanely hard to create.

Nobody’s ever done it. It’s only been done just one time. And so that is the capability. And in order to do that, you have to understand the algorithms, you have to understand a lot about the domains of applications, you have to select it right, you have to create the right architecture for it, and then the last thing that we did right was that we realized that in order for you to have a computing platform, the applications you develop for NVIDIA should run on all NVIDIA. You shouldn’t have to think, does it run on this chip? Is it gonna run on that chip? It should run on every chip. It should run on every computer with NVIDIA in it. That’s the reason why every single GPU that’s ever been created at our company, even though we had no customers for CUDA a long time ago, we stayed committed to it. We were determined to create this computing platform since the very beginning. Customers were not. And that was the pain and suffering. It cost the company decades and billions of dollars getting here. And if not for all the video gamers in the room here, we wouldn’t be here. You were our day jobs, and then at night we can go solve digital biology, go help people with quantum chemistry, go help people with artificial intelligence and robotics and such. And so we realized, number one, that we were accelerating computing as a software problem. The second thing is AI is a data center infrastructure problem. And it’s very obvious because you can’t just train an AI model on a laptop. You can’t train it on a cell phone. It’s not big enough of a computer. The amount of data is measured in trillions of bytes. And you have to process that trillions of bytes billions of times. And so obviously that’s going to be a large-scale computer distributing the problem across millions of GPUs. The reason why I say millions is 16,000 inside the 16,000 are thousands. And so we’re distributing the workload across millions of processors. There are no applications in the world today that can be distributed across millions of processors.

works on one processor. And so that distribution, that computer science problem was a giant breakthrough, an utterly giant breakthrough. And that’s the reason why it enabled generative AI, enabled large language models. So we observed two things. One, accelerated computing is a software problem, algorithm problem. And AI is a data center problem. And so we’re the only company that went out and built all of that stuff. And the last part that we did was a business model choice. We could have been a data center company ourselves and be completely vertically integrated. However, we recognized that no computer company, no matter how successful, will be the only computer company in the world. And it’s better to be a platform computing company, because we love developers. It’s better to be a platform computing company that serves every computing company in the world than to be a computing company all by ourselves. And so we took this data center, which is the size of this room, a whole bunch of wires and a whole bunch of switches and networking and a bunch of software. We disaggregated all of that. And we integrated into everybody else’s data centers that are all completely different. So AWS and GCP and Azure and Meta and so on and so forth, data centers all over the world. That’s an insane complexity problem. And we figured out a way to have enough standardization wherever it was necessary, enough flexibility so that we could accommodate, enough collaboration with all the world’s computer companies. As a result, NVIDIA’s architecture is now graphed, if you will, into every single computer company in the world. And that has created a large footprint, larger install base, more developers, better applications, which makes happier customers who buy them more chips, which increases the install base, which increases our R&D budget, so on and so forth. The flywheel, the positive feedback system. So that’s how it works, nice and easy.

and I wanted you to explain to us why, if you haven’t invested in fabricating your own chips. And why is that? That’s an excellent question. The reason for that is as a matter of strategic choice, the core values of our company, my own core values, the core values of our company is about choosing. The most important thing in life is choosing. How do you choose? How do you choose, well, everything. You know, how do you choose what to do tonight? How do you choose? Well, our company decides to choose projects for one fundamental goal. My goal is to create the environment, an environment by which amazing people in the world will come and work. An amazing environment for the most, the best people in the world who wants to pursue this field of computing and computer science and artificial intelligence to create the conditions by which they will come and do their life’s work. Well, if I say that, then now the question is, how do you achieve that? So let me give you an example of how not to achieve that. Nobody that I know wakes up in the morning and say, you know what, my neighbor is doing that and you know what I want to do? I want to take it from them. I can do it too, I want to take it from them. I want to capture their share. I want to pummel them on price. I want to kick them in the, I want to take their share. It turns out no great people do that. Everybody wakes up in the morning and says, I want to do something that has never been done before, that’s incredibly hard to do, that if successful, makes a great impact.

in the world. And that’s what NVIDIA’s core values are. One, how do we choose? Do something that the world’s never done before. Let’s hope that it’s insanely hard to do. The reason why you choose something insanely hard to do, by the way, so that you have lots of time to go learn it. If something is insanely easy to do, like tic-tac-doe, I wouldn’t fuss over it. And the reason for that, obviously, is highly competitive. And so you’ve got to choose something that’s incredibly hard to do. And that thing that’s hard to do discourages a whole bunch of others, all by itself, because the person who’s willing to suffer the longest wins. And so we choose things that are incredibly hard to do. And you’ve heard me say pain and suffering a lot. And it’s actually a positive attribute. People who can suffer are ultimately the ones that are the most successful, number one. Number two, you should choose something that somehow you’re destined to do. Either a set of qualities about your personality or your expertise or the people you’re surrounded by, your scale, whatever your perspective, you’re somehow destined to do. And then number three, you better love working on that thing so much, because unless so, the pain and suffering is too great. Now, I just described to you NVIDIA’s core values. It’s as simple as that. And if that’s the case, what am I doing making a cell phone chip? How many companies in the world can make a cell phone a lot? Why am I making a CPU? How many more CPUs do we need? Does that make sense? We don’t need all those things. And so we naturally selected ourselves out of commodity markets. We naturally selected ourselves out of commodity markets. And because we selected amazing markets, amazingly hard to do things, amazing people joined us. And because amazing people joined us, and because we had the patience and let them succeed to go.

when we think about this decade. Are these right answers, by the way? I don’t have an MBA and I didn’t get a finance degree. I read some books and I watched a lot of YouTubes, I gotta tell you. Nobody watches more business YouTubes than I do, and so you guys have nothing on me. But are these right answers, professor? You’re asking the wrong person. I also didn’t study business, but yes, they’re the right answers. Best CEO, yeah, right. What do you think about AI? When you’re thinking about AI applications and where we’re gonna see change in our lives, let’s say over the next three, five, seven years, where do you see that going? And in places where we will all potentially be affected in our.

and daily experience. Yeah, first of all, I’m going to go to the punchline. AI is not going to take your jobs. The person who use AI is going to take your job. Do you guys agree with that? OK, so use AI as fast as you can so that you could stay gainfully employed. Let me ask you the second thing. When productivity increases, when productivity increases, meaning we embed AI all over NVIDIA. NVIDIA is going to become one giant AI. We already use AI to design our chips. We can’t design our chips. We can’t write our optimizing compilers without AI. So we use AI all over the place. When AI increases the productivity of your company, what happens next? Layoffs? Or you hire more people? You hire more people. And the reason for that is, give me an example of one company that had earnings growth because of productivity gains that said, guess what? My gross margins just went up. Time for a layoff. So why is it that people think about losing jobs? If you think you have no new ideas, then that’s the logical thing. Does that make sense? If you don’t have any more ideas to invest your incremental earnings, then what are you going to do when the work is replaced, is automated? You lay people off. And so join companies where they have more ideas than they can afford to fund. So that when AI automates their work, it’s going to shift, of course. It’s going to change the style of working. AI is going to come after CEOs right away. Deans and CEOs, we’re so toast. Sounds good. So I think CEOs first, Deans second. But you’re close. And so you join companies where they have more ideas, more ideas.

its relationship to every other word and you learn all you know read a whole lot of sentences and paragraphs and you try to figure out what is the best number vector what’s the best number to associate with that with that word so mother and father are close together numerically oranges and apples are close together numerically they’re far from mom and dad dogs and cats are far from mom and dad but closer probably to mom and dad than they are from oranges and apples chair and tables and chair and hard to say exactly where they lie but those two numbers are close to each other far away from mom and dad king and queen close to mom and dad does it make sense imagine doing this for every single number and every time you test it you go son of a gun that’s pretty good you know and when you subtract something from something else it makes sense okay that’s basically learning the representation of information imagine doing this for English imagine doing this for every single language imagine doing this for anything with structure meaning anything with predictability images have structure because if there are no structure it’d be white noise physically be white noise and so there must be structure that’s the reason why you see a cat I see a cat you see a tree I see a tree you can identify where the tree is you can identify where the coastline is where the mountains are where the clouds is right and so we could we could learn all of that so obviously you could take that image and turn it into a vector you could take videos and turn the vector 3d into two vectors proteins

into vectors, because there’s obviously structure and protein, chemicals into vectors, genes eventually into vectors. We can learn the vectors of everything. Well, if you can learn everything into numbers and you know its meaning, then obviously you can take cat, the word CAT, translate it to the image CAT, the image of cat. Obviously this is the same meaning. If you can go from words to images, that’s called Majorny, stable diffusion. If you can go from images to words, that’s called captioning, video, YouTube, videos to words, right, underneath videos. And so what if you went from, what do you call it if you go from, say, amino acids to proteins? That’s called a Nobel Prize. And the reason for that is because that’s AlphaFold. Incredible breakthrough. Isn’t that right? And so this is the amazing time. The amazing time in computer science where we could literally take information of one kind and convert it, transfer it, generate it into information of another kind. And so you can go text to text, a large body of text, PDF, a small body of text, a summarization of archive, which I really enjoy, right? And so instead of reading every single paper, I can ask it to summarize the paper. And it has to understand images because in the archive, the papers have a lot of images and charts and things like that. So you can take all of that, summarize it. And so you could now imagine all of the productivity benefits and, in fact, the capabilities you can’t possibly do without it. So in the near future, you’ll do something like this. You could say, hi, I would like to design, give me some options of a whole bunch of cars. I work for Mercedes. I really care about the brand. This is the style of the brand. Let me give you a couple of sketches and maybe a couple of photographs of the type of car I’d like to build. It’s a four-wheel SUV.

4WD, SUV, let’s say, you know, so on and so forth, okay? And all of a sudden it comes up with 20, 10, 200, completely fully 3D designed CAD. Now the reason why you want that instead of just finishing the car is because you might want to select one of them and you say iterate on this one another 10 times and you might finally select one and then modify it yourself. And so the future of design is going to be very different. The future of everything will be very different. Now if you gave that capability to designers, they would go insane. They would love you so much. They would love you so much. And that’s the reason why we’re doing this. Now what’s the long-term impact of this? One of my favorite areas is if you could use language to describe a protein and you could use language to figure out a way to synthesize protein, then the future of protein engineering is near us. And protein engineering, as you know, creating enzymes to eat plastic, creating enzymes to capture carbon, creating enzymes of all kinds to grow vegetables better. All kinds of different enzymes could be created during your generation. And so the next 10 years is going to be unbelievable. We were the computer, the chip engineering generation. You’ll be the protein engineering generation. Something that we couldn’t imagine doing just a few years ago. Okay. I think we’re going to open it up for Q&A to the audience. So questions and maybe I’ll point and we have some mics that will be running. Okay. Over there. We’ll start there. Thank you for coming tonight. Thank you for having me. I have a question. So are you worried that Moore’s law- Business schools, students are so serious. I understand. I understand that the graduates of Columbia ends up being-

investment bankers and stock traders. I’m actually a computer scientist. Is that right? Is that right? And one computer scientist. You’re going to be a quant. You’re going to be a quant. And so that’s what I understand. So I’m here selling stock. Okay? And so in the future, in the future, if somebody asks you what stock to buy, NVIDIA. Yeah, go ahead. Yeah, question for you is, are you worried that Moore’s Law might actually catch up to GPU industry as it did for companies like Intel? And can you also explain the difference between Moore’s Law and Huang’s Law? I didn’t I didn’t phrase Huang’s Law and it wouldn’t be like me to do so. The very simple thing is this. The the Moore’s Law was twice the performance every year and a half, approximately. The easier math to do is 10 times every five years. So every 10 years is about 100 times. If that’s the case, if general purpose computing is microprocessors, if general purpose computing was increasing in performance at 10 times every five years, 100 times every 10 years. Why change? Why change the computing method? 100 times every 10 years, not fast enough. Are you kidding me? If cars would go 100 times every five years, wouldn’t life be good? And so the answer is, it’s in fact, Moore’s Law is very good. And I benefited from it. The whole industry benefited from it. The computer industry is here because of it. But eventually, general purpose computing, Moore’s Law, it’s not about the number of transistors in computing. It’s about the number of transistors, how you use it for CPUs, how you translate it ultimately to performance. That curve is no longer 10 times every five years. That curve, if you’re lucky, is two or four times every 10 years. Well, the problem is, if that curve is two or four times every 10 years, the demand for computing and our aspirations of using computers to solve problems, our aspirations of using computers to solve problems,

our imagination for using computers to solve problem it’s greater than four times every ten years isn’t that right and so our our imagination our demand the world’s consumption of all exceeds that well you could solve that problem by just buying more CPUs you could buy more but the problem is these CPUs consume so much power because their general purpose it’s like a generalist a generalist is not as efficient the craft is not as great they don’t they’re not as productive as a specialist if I’m ever going to have an open chest wound I you know don’t send me a generalist you guys know I’m saying you know if you guys are around just call a specialist all right and so yeah he’s a vet you know it’s a generalist look or the wrong specialist so so so generalist is too brute-forced and so today it cost it cost the world too much energy cost too much to just brute force general-purpose computing now thankfully we’ve been working on accelerated computing for a long time and accelerated computing as I mentioned is not just about the processor it’s really about understanding the application domain and then creating the necessary software and algorithms and architecture and chips and somehow we figured out a way to do it behind one architecture that’s the genius of the work that we’ve done that we somehow found this architecture that is both incredibly fast it has to accelerate the CPU a hundred times five hundred times sometimes a thousand times and yet it is not so specific that is only used for one singular activity does that make sense and and you so you need you need to be sufficiently broad so that you have large markets but you need to be sufficiently narrow

so you can accelerate the application. That fine line, that razor’s edge, is what caused NVIDIA to be here. It’s almost impossible. If I explained it 30 years ago, nobody would have believed it. And in fact, very few did, to be honest. It took a long time, and we just stuck with it, stuck with it, stuck with it. And we started with seismic processing, molecular dynamics, image processing, of course, computer graphics. And we just kept working on it, working on it, working on it. Weather simulation, fluid dynamics, particle physics, quantum chemistry. And then all of a sudden, one day, deep learning. And then transformers. And then the next will be some form of reinforcement learning transformers. And then there will be some multi-step reasoning systems. And so all of these things were just one application. And somehow, we figured out a way to create an architecture and solve all that. And so will this new law end? And I don’t think so. And the reason for that is this. It doesn’t replace the CPU. It augments the CPU. And so the question is, what comes next to augment us? We’ll just connect the next to it. We’ll just connect the next to it. And so when the time comes, we’ll know that there’s another tool that we should be using to solve the problem. Because we are in service of the problems we’re trying to solve. We’re not trying to build a knife and make everybody use it. We’re not trying to build a plier and make everybody use it. We’re in service of accelerated computing is in service of the problem. And so this is one of the things that all of you learn. Make sure your mission is right. Make sure that your mission is not build trains, but enable transportation. Does that make sense? Our mission is not build GPUs. Our mission is to accelerate applications, solve problems that normal computers cannot. If your mission is articulated right and you’re focused on the right thing, it’ll last forever.

Thank you. Okay. Up there somewhere. Yes, guys. That guy right there is Tony. Go ahead, Tony. Am I Tony? Yeah. No. All right. Fine. I’m Tony today. Where’s Tony? Tony was that guy in the middle, right? Yeah, see? I met him just now. I’m just kidding. All right. You’re just shredding my memory. I’ll take my chance. I wasn’t. All right. All right. All right. All right. All right. All right. All right. All right. All right. I wasn’t. I wasn’t trying to give Tony the mic. I was just demonstrating my incredible memory for Tony. Go ahead. Thanks again. Now there’s a push for on shoring the supply chains for semiconductors. Then there are also restrictions on the export of high tech to certain countries. How do you think that would affect NVIDIA in the short term? But also how would that affect us as consumers in the long term? Yeah. Really excellent question. You guys all heard the questions, I’ll repeat it, it relates to geopolitics and geopolitical tension and such. The geopolitical tension, the geopolitical challenges will affect every industry, will affect every human. We deeply, we, the company, deeply believe in national security. We are all here because our countries are secure. We believe in national security, but we also simultaneously believe in economic security. The fact of the matter is most families don’t wake up in the morning and say, good gosh, I feel so vulnerable because of the lack of military. They feel vulnerable because of economic survivability. And so we also believe in human rights and the ability to be able to create a prosperous life is part of human rights. And as you know, the United States believe in the human rights of the people that live here as well as the people that don’t live here. And so the country believes in all of those things.

simultaneously, and we do too. The challenge with the geopolitical tensions, the immediate challenge, is that if we’re too unilateral about deciding that we decide on the prosperity of others, then there will be backlash. There’ll be unintended consequence. But I am optimistic. I want to be hopeful that the people who are thinking through this are thinking through all the consequences and unintended consequences. But one of the things that that has done is that it has caused every country to believe to deeply internalize its sovereign rights. Every country is talking about their own sovereign rights. And that’s another way of saying everybody’s thinking about themselves. And as it applies to us, on the one hand, it might restrict the use of our technology in China and the export control there. On the other hand, because of sovereignty and every country wanting to build its own sovereign AI infrastructure, and not all of them are enemies of the United States, and not all of them have a difficult relationship in the United States, we would help them build AI infrastructure everywhere. And so in a lot of ways, this weird thing about geopoliticals, it limits the market opportunities for us in some way. It opens market opportunities in other ways. But for people, for people, I am just really hopeful I really hope, not hopeful, I really hope that we don’t allow, that we don’t allow our tension with China result into tension with Chinese. That we don’t allow our tension with the Middle East turn into tension with Muslims. Does that make sense? We are more sophisticated than that. We can’t allow ourselves to fall into that trap. And so that, I worried a little bit about that. I worry about that as a slippery slope.

One of our greatest sources of intellectual property for our country, as you know, foreign students. I see many of them here. I hope that you stay here. It is one of our country’s single greatest advantage. If we don’t allow foreign students, the brightest minds in the world, to come to Columbia and keep you here in New York City, we’re not going to be able to retain the great intellectual property of the world. And so this is our fundamental core advantage, and I really do hope that we don’t ruin that. So as you can see, the geopolitical challenges are real, and national security concerns are real, but so are all of the other economic, market, social, technology matters, technology leadership matters, market leadership matters. All that stuff matters. The world is just a complicated place, and so I don’t have a simple answer for that. We will all be affected. So we’ll take one more question over there. Yes. But in the meantime, stay focused on your school. Do a good job. Just study. Hi there. So I actually started off working as an engineer at a semiconductor company out of Brixton in entrepreneurship, and now I’m here. As someone like yourself that is fundamentally a technologist, an engineer, started a company very successfully, learned finance from YouTube videos, what do you think of MBAs? Oh, I think it’s terrific. First of all, you’ll likely live until you’re 100, and so that’s the problem. What are you going to do for the last 70 years or 60 years? And this isn’t something I’m telling you. It’s something I tell everybody I care about. Look, to the best of your ability, education, when you come here, you’re force-fed education. How good can that be? After you leave, like me, I got to go.

scour the planet for for knowledge and And I’ve got to go through a lot of Junk that gets it to some good stuff Here in school you’ve got all these amazing professors who are curating the knowledge for you and present it to you in a platter My goodness, I would stay here and you know pig out on knowledge for as long as I can If I could do it again, I still be here, you know Dean and me sitting next to each other. I’m the oldest student here I’m Just preparing for the big, you know step function when I graduate just go instantaneously the success but I’m just a little kidding about that. You have to leave at some point your parents would appreciate it And and so so but don’t be in a hurry. I think I think learn as much as you can There’s no one right answer to getting there. Obviously, I have friends who never graduated from from college and they’re you know Insanely successful and so there there are multiple ways to get there But statistically I still think this is the best way to get there Statistically and so if you believe in stat in math and statistics stay in school Yeah, go through the whole thing and and so I I got a virtual MBA by working through it not because of choice I was when I first graduated from school. I thought it was gonna be an engineer. Nobody says, you know, hey Jensen Here’s your diploma. You’re gonna be a CEO, you know, so I I didn’t know that So I when I got there, I think I’ll learn it MBA and there’s a lot of different ways to learn business Strategy matters obviously business matters. They’re very different things finance matters It’s very different things. And so you got to learn all these different things in order to build a company But if you’re surrounded by amazing people like I am they end up teaching you along the way And so there’s some things that that that that depending on what role you want to play. That’s critically yours Okay, and so for for a CEO

or a sentence. The use of punctuation between statements is a critical part of leadership. It’s not only my job, but it’s critical that I lead with it. That’s character. There’s something about your character, about the choices that you make, how you deal with success, how you deal with failure, enormous setback, how you make choices. Those kind of things matter a lot. From a skill and craft perspective, the most important thing for a CEO is strategic thinking. There’s just no alternative. The company needs you to be strategic. The reason for that is because you see the most. You should be able to look around corners better than anybody. You should be able to connect dots better than anybody. You should be able to mobilize. Remember what a strategy is, action. It doesn’t matter what the rhetoric says. It matters what you do. Nobody can mobilize a company better than the CEO. Therefore, the CEO is uniquely, uniquely in the right place to be the chief strategy officer, if you will. Those two things, I would say, are, from my perspective, two of the most important things. The rest of it has a lot of skills and things like that, and you’ll learn the skills. Maybe if I could just add one more thing. I do believe that a company is about some particular craft. You make some unique contribution to society. You make something. If you make something, you ought to be good at it. You should appreciate the craft. You should love the craft. You should know something about the craft, where it came from, where it is today, and where it’s going to go someday. You should try to embody the passion for that craft. I hope today I did a little bit of embodying the passion and the expertise of that craft. I know a lot about the space that I’m in, and so, if it’s possible, the CEO should know the craft. You don’t have to have…

And I have found the craft. But it’s good that you know the craft. There is a lot of craft that you can learn. And so you want to be an expert in that field. Those are some of the things. You can learn that here, ideally. You can learn that on the job. You can learn that from friends. You can learn that by doing, you know, a lot of different ways to do it, but stay in school. So before I thank the best CEO, I want to thank the Digital Future Initiative, the David Silfen speaker series, but mostly, thank you, Jensen, for coming. We all understand why you were voted the best CEO now. Thank you very much. Thank you.