Nvidia CEO Jensen Huang On How His Big Bet On A.I. Is Finally Paying Off - Full Interview
Jensen, thank you so much for being here today. And I want to get started just with a question that is really basic, but I think chips have been in the ecosystem a lot more lately, and there's people who probably didn't even really know what a semiconductor was a few years ago. Can you give me a very basic definition of what Nvidia is? Wow, that's an easy question. Well, we are a technology company that processes software. For applications and domains of science that are barely possible without us.
And so because of what we do, we can make what is barely possible, possible. Or we can make something that is very energy consuming, very energy efficient. Or we could turn something that costs a lot of money and make it much more affordable. And so we created this thing called accelerated computing, and that was what we pioneered about three decades ago.
And it's taken until now to really take off. In the early days at Denny's with Chris and Curtis, the dream was probably simpler. Can you explain what it what your first dream was, what the vision was, even though now it's come so far to be this accelerated computing company? Well, at the time, if you go back 30 years, at the time the PC revolution was just starting. The microprocessor was starting to take off.
The CPU was starting to take off. And there was quite a bit of debate about: what is the future of computing and how should software be run? And there was a large camp, and rightfully so, that believed that CPU or general purpose software was the best way to go and it was the best way to go for a long time. We felt, however, that there was a class of applications that wouldn't be possible without acceleration. Or you couldn't make it affordable enough for everybody to enjoy without acceleration.
And so we started this accelerator company, this accelerated computing company, to solve those problems. In the beginning, there weren't that many applications for it, frankly, and we smartly chose one particular combination that was a home run. It was computer graphics, and we applied it to video games.
And that combination turned out to have been a giant industry. And now video games is the largest industry in the world and the largest entertainment industry in the world. And it drove our technology for three decades because making video games more and more realistic, making it available to more people, took a long time. And we're still in that journey and frankly, probably early in that journey. There are now probably, you know, over a billion gamers in the world, but there are 8 billion people. Someday everybody's going to be a gamer.
And so it's going to be the largest by far entertainment industry. And so it turned out to have been a fantastic technology driver for our company. And we step-by-step added more and more things that we could do, to today, artificial intelligence. Beyond gaming and graphics, Nvidia has grown immensely. I think that there's a lot of things people might be surprised to hear are powered by Nvidia. Can you just give a very simple list of some of the use cases and big name customers that people might be surprised to hear are powered by Nvidia? People would probably be surprised that the most powerful and energy efficient supercomputers in the world, that are used for molecular dynamic simulations to climate science research to material science research to quantum computing research, are powered by Nvidia.
All the way to the other extreme: a whole bunch of robots that are powered by Nvidia in manufacturing lines. Self-driving cars that are powered by Nvidia to the Nintendo Switch that I'm very proud of that's powered by Nvidia. So we're in very powerful systems and we're in very energy efficient systems. And probably one of the most talked about systems today are the systems at the Microsoft Azure data centers that are powering ChatGPT. And the work that we did with OpenAI in the very beginning to now that Powers ChatGPT.
I think those are really quite exciting. I'm going to come back to ChatGPT for sure, but first I wanted to ask you about betting it all. This is something that you have not shied away from in the 30 years since you started the company. It was maybe seven times that you've been reinvented and faced, you know, success or utter failure. What is the lesson here? Well, we're in a really fast moving industry. You know, technology is incredible in the sense that such enormous challenges and problems could be solved by computing, on the one hand.
On the other hand, the technology changes. And there are so many great companies in the world and we're pursuing very similar aspirations. We want to solve the world's greatest challenges. And so every now and then, a technology revolution comes along. We were started in the PC revolution. After that, the Internet revolution came and
all of a sudden the companies before it, some of them didn't make it to the revolution. And some great new companies like Google and others got invented during that time. And then the cloud computing revolution came.
And then the mobile cloud computing revolution came. And now we're talking about the AI revolution. And so each one of these transitions, it's very unlikely that the companies that were great before it are still great after it. And there are some companies that have made the ability to, because of their adaptability and agility, reinvented themselves along the way. We had to reinvent ourselves in each one of those technology revolutions.
And, you know, agility is just really, really, really important to companies. And one of the things that I'm really proud about our company is, at the core of our company is incredible technology. We have incredible technologists. You know, if you're pioneering one of the most important computing platforms in the world, from use for scientific computing to genomics to digital biology, all the way to video games, well you're going to need incredible computer scientists. So on the one hand, we're incredibly technology rich. On the other hand, we're in an enormous, we're in a giant sea of technology companies.
And so the ability for us to adapt and reinvent ourselves and continue to be relevant and from one generation to another generation was really important. And I'm very proud of that. It hasn't always been success. Can you talk to me about some of the biggest stumbles that you've had to overcome in the years? Well, you know, every company makes mistakes and I make a lot of them. And, you know, some of them puts the company in peril, especially in the beginning, because we were small and were up against very, very large companies and we're trying to invent this brand new technology. And, you know, when you invent something new, you have to convince customers to use it. You have to convince the ecosystem it's the right thing to use. And you've got developers, you know.
We're a computing company, so developers matter a lot to us. And so we're trying to invent something new and we're barely, we barely know exactly what we're doing, you know? So when you're doing something that's never been done before, you're not exactly sure what you're doing. And yet, on the other hand, you have these giant companies who would like you not to disrupt the industry. And so early on, there were product mistakes that we made.
There were, you know, execution challenges that we had. There were some strategy mistakes that I made. And, you know, there's just so many of them. And, you know, one of the skills of resilience is
the ability to forget the past. You know, just as coaches tell you, don't worry about the last down, worry about the next down. And so I tried to make sure that the company remembers our learnings from the mistakes. Most founders would be very satisfied being at the helm of such a huge industry with gaming graphics. What signaled to you, and when, that it wasn't enough? Well, our ambition was always to be a computing platform company.
We selected computer graphics and video games as our first market combination: technology, market, product technology and market combination. But we always believed that accelerated computing was going to be impactful for many, many different industries. We expanded from video games into design. And today just about every product that's
designed or every digital asset or movie or, you know, almost anything that's designed in 2D or 3D digitally uses Nvidia somehow. And then we extended that into scientific computing, into physical simulation. And it started with seismic processing, as a field called inverse physics, to particle simulations, molecular dynamic simulations, and so on and so forth, and fluids. And just about every field of science we're in today. And so I'm really proud of that.
And that led us to a much more general purpose type of accelerated computing that we created. Which then, one day, artificial intelligence found us. You know, this is one of the things that's really amazing about a computing platform. You have a vision about what you want to create. And for whatever reason, you differentiate in your computing approach.
And maybe you made it super convenient in the cloud. Maybe you made it possible for you to keep the computer with you all the time: mobile cloud. And in our case, accelerated computing makes it possible for you to solve problems that were impossible before, or much more energy efficient than before. And so there's a fundamental reason that makes a new computing architecture successful. And at some point, the positive feedback system starts to work. You know, you reach now a lot of
different customers and different applications. We're in every cloud, made by every computer company, and then all of a sudden one day a new application that wasn't possible before discovers you. First you discover them, and then pretty soon they discover you. And this positive feedback system
starts to feed on itself. I assume you're talking about the moment with AlexNet and CUDA powering that, and sort of the big bang of AI, if you will. I'm curious how much of that you feel like was luck? I mean, what you're talking about is it finding you.
It sounds a bit like luck. And how much of it was foresight? Well, it wasn't foresight. The foresight was accelerated computing. The foresight was making this architecture exactly the same for everybody.
Having the discipline of staying true to that platform for generation after generation after generation, believing that eventually our install base would be so large that not only would we have reach, but applications would therefore be enabled by us. New entire applications that weren't possible before would discover us. This is the nature of cloud.
This was the nature of PC. This was the nature of mobile cloud. And each one of these revolutions and generations of technology. In the beginning there was some
fundamental reason it was successful, and then at some point it achieves a bit of a escape velocity and it becomes exponential because these applications start to be enabled by you and they come and discover you. And so we made a lot of great decisions. And the great decisions associated with the architecture and discipline of the platform and evangelizing it to everybody. And we reached out to research universities all over the world. And we just believed that some day something new would happen. The rest of it requires some serendipity.
But the part that was really wonderful was when we realized that AlexNet is not just some neural network, but it's a whole new way of doing software. AlexNet is profound in that way. Not only was it a giant breakthrough in computer vision, it was also a profoundly new way of doing software. Some people call it software 2.0, where the machine augments the software programmers and the data writes the software. Instead of humans typing in a software program, the data creates the software. That way of using experience or data to cause a software to be able to make future predictions was so profound, and we had the good wisdom to go put the whole company behind it.
We saw early on, about a decade or so ago that this way of doing software could change everything. All of the software that we've wanted to write that we didn't know how to write, we can now do. And that was a great decision.
And we changed the company from the bottom all the way to the top and sideways. Every chip that we made was focused on artificial intelligence. We built a wonderful research organization dedicated to artificial intelligence. Our entire software stack was invented for AI and and then all the things that we did to create large systems and networks. Which then became this thing called an AI supercomputer.
And I remember delivering my very first AI supercomputer. I hand delivered it myself. I delivered it to OpenAI. The world's very first AI supercomputer was delivered to OpenAI. What year was that? Well, I guess it's like five, six years ago, I guess. Five years ago.
Yeah. And now here we are and OpenAI has taken the world by storm. Do you think that your products, Nvidia, is at the very center of this and has become the must-have products to power this next big step? Well we're the world's engine for AI. Because of the decisions we made a decade or so ago, and we put so much of our might and expertise into it. We're now in every cloud. We're in every country and every field of science.
35,000 companies use our AI computers to develop and advance this field. Giant companies like cloud and internet companies, all the way to startups. Thousands of startups.
They're in all kinds of areas: consumer internet to digital biology to robotics. I'm really happy with the diffusion of the technology. I'm really pleased with how we've democratized the technology so that anybody can access it. You can't ignore the incredible vision and dedication to the work at OpenAI. From the very first day I saw them, they were dedicated to wanting to do this and they've been focused on it for five years. And of course in research, even longer than that.
I'm incredibly proud of the work that they've done. Yeah really terrific team. Here in Silicon Valley, there's a bunch of CEOs and founders who've started bringing up the A100 and kind of publicly competing with each other about who bought more when and who saw this coming. Sort of competing for bragging rights around the A100. What would you want to say to them?
There's more. Come get them. Everybody should win. You know, winners to all. In the past, when you start a company, a software company or technology company, you need a lot of software engineers. It is still true and you need amazing computer scientists. But today, startups - and there are some amazing startups that we're working with right now - where they're 25, 30 people.
Backed up with a large data center of AI supercomputers powered by A100s. If you want to start a startup today, it's you and AI. And you're supercharged by the AI supercomputer and the algorithms that you have inside and all the data that you're going to teach it with. And so it's really quite a transformation in how startups are going to get built in the future. Now we're onto something even larger than that, you know, built on these AI supercomputers, these large language models. It's definitely a watershed event for the AI industry.
It feels very much like the iPhone moment, when mobile cloud really took off and all of the environmental conditions feel exactly the same way, just larger and much, much more industries. Right now, generative AI is still extremely expensive to accomplish. How do you think it'll really take off if only a couple big companies have true access to do it at scale? Well, it turns out it doesn't cost that much. And the reason why there are so many CEOs with bragging rights on so many A100s is because it's really quite democratized.
We took what otherwise would be a $1 billion data center running CPUs and we shrunk it down into a data center of $100 million. Now $100 million is, when you put that in the cloud and shared by 100 companies, is almost nothing. If you take a look at how much it costs to design a chip, so you put that in perspective, it costs us about $2 to $3 billion to design A100. When I hit enter and asked TSMC to help us make it, that email is $100 million. And then it populates these AI supercomputer data centers. And when you train a large language model,
let's say it costs $10 million. So a chip, and there are 3,000 chip companies in the world, taping out a chip is like $100 million or $50 million, $30 million, depending on the size, but nothing less than $10 million. And now you could build something like a large language model, like a ChatGPT for something like $10, $20 million. That's really, really affordable. And so I think the the ability for every industry to create their foundation model: there's going to be a protein foundation model, a chemical foundation model.
There will be a robotics foundation model. There'll be foundation models for science, for finance, for all kinds of different applications and different industries and different countries. I was just in Sweden and the Berzelius supercomputer there, we helped them with. We built an AI supercomputer. It's a Swedish foundation model
supercomputer. And with just tens of millions of dollars, you can build the most powerful supercomputer in Sweden. And so these are really, really accessible technologies now.
There are always skeptics and people who are alarmed, perhaps, by how fast AI is taking off and how powerful it's become with capabilities like deepfakes, fake eye contact, for instance, that I've seen an example of. What do you say to them? Well, the first thing that everybody should do is to take advantage of the technology and to boost their own capability. There's no question that the interest behind ChatGPT has been so great. It is the fastest growing application in the world, and it's been used in all kinds of different ways. The thing that's really amazing about artificial intelligence is that what ChatGPT has shown is that it has eliminated the digital and the technology divide. Everyone is a programmer now.
Everybody could program a computer. During my generation, the way that you program a computer was: started with Basic and I learned Fortran. Then you learn C and then you move to C++ and Java and now PyTorch or Python.
And each one of those languages, there was Ooc, and these are really weird languages and they're hard to learn. And the whole time that we've been making computers more and more capable, the technology became harder and harder to use. And the technology divide arguably has been growing, until artificial intelligence. And you hear about cucumber farmers who are teaching a robot how to sort cucumbers. And a high school student did that for his mom. And now 150 million people are programing the computer, instead of programing the computer with C or Python, you're now programing the computer with anybody's plain language.
And you tell this computer what you want to do. And this computer goes off and does it. Or you tell the computer you'd like to write a Python script, and it goes off and does it.
And so this capability has democratized computing for the very first time. It's put technology, very powerful technology, in the hands of anybody who would like to use it. And so I think this is really genuinely the first time in my generation that we've created something, or contributed to creating something, that made our technology accessible to everyone.
Not just to use, but to harness. Not just to use, but to program. And so I think every domain expert in the world will be able to do that. And I recommend everybody just, number one, take advantage of AI and augment your work . Make yourself more productive. Lift yourself, you know, power up.
Power up your own career, power up your own capability. And then from there, you know, increase the productivity of society and move everything along. How do you stay ahead in an industry where some of your customers could become your competitors? You know, speaking about Google's TPUs and Amazon has their own internal chips as well. How do you stay ahead in that landscape? We stay ahead by, number one, doing it very well. But also we do it very differently. The first thing that I would say is that every data center in the world should accelerate every workload they can. And the reason for that is because, as you
know, the world's data centers consume a lot of power now. And it used to be the case that because of Moore's Law, even though we required more computing throughput every year, the amount of power that the world's data centers consume didn't grow that fast. And the reason for that is because Moore's Law. But now that's changed.
That has ended. And as a result, if we want to increase the amount of computing throughput we want, and there's no question that's happening, then the amount of power that the world needs in the data center will grow. And you can see in the recent trends, it's growing very quickly and that's a real issue for the world. The first thing that we should do is: every data center in the world, however you decide to do it, for the goodness of sustainable computing, accelerate everything you can.
Now, an ASIC is designed to be application specific. It does nothing, it does exactly that and it does it very well. What Nvidia does is a general purpose accelerated computing platform. So we could, on the one hand, simulate climate science. On the other hand do robotics. On the other hand, do large language models or computer graphics and play video games and such.
And so our ability to be flexible, versatile and also extremely performant lets us increase the versatility and the utility, the utilization of it, inside data center. When you build an infrastructure, the most important thing for you is utilization. You can't afford to have hotels that are occupied 30%. You would like the data center even more so because it cost billions of dollars. Nvidia's accelerated computing platform lets you have versatility and utilization.
So our TCO, our cost, is actually the lowest of all. And that's the reason why people use it: because they can use it on so many things. The second reason is we're in every cloud. And so if you're an enterprise customer or a developer or a startup company and you would like to have the ability to operate your service in every cloud or any cloud across the world, we make it possible for you to do it in every cloud: on prem, hybrid cloud, all the way out to the edge. One architecture.
What do you say to gamers who wish you had kept focus entirely on the core business of gaming? Well, if not for all of our work in physics simulation, if not for all of our research in artificial intelligence, what we did recently with GeForce RTX would not have been possible. We invented the GPU and programable shader 25 years ago, a quarter of a century ago, and it's remained basically the same for the last 25 years. About five years ago, we came to the conclusion that in order for us to take computer graphics and video games to the next level, we had to reinvent and disrupt ourselves.
Change literally what we invented altogether. And so we invented this new way of doing computer graphics: ray tracing, and basically simulating the pathways of light and simulate everything with generative AI. And so we compute one pixel and we imagine with AI the other seven.
It's really quite amazing. Imagine a jigsaw puzzle and we gave you one out of eight pieces and somehow the AI filled in the rest. Pretty amazing.
And so as a result, we increased the performance of what made possible ray tracing. We increased the performance by probably a factor of five. Or another way to think about that: we reduced the amount of energy consumed by a factor of five. And so that great invention completely revolutionized video games. And the next 25 years, because of what we
did, I think we have 25 years of amazing future. Just a couple of questions about the state of the industry. Experts seem to say the worst of the chip shortage is over. How did Nvidia weather that storm? The chip shortage was a strange one.
On the one hand, there was chip shortage. On the other hand, about the same time, you know, this is now, we're now coming out of it. But some two or three quarters ago, we had supply challenges and demand challenges at the same time . But not at the same customer. Not in the same industry. Not in the same market. And so that was very, very challenging: to have your foot on the gas and your foot on the brakes at exactly the same time and full pressure on both. Our company weathered it just fine.
We're a strong and resilient company. Our financial performance wasn't as good as our technology and contribution performance. We did some of our best work ever in the history of our company. A100 was replaced by H100, which we're in full production now.
All the work that we did with AI supercomputing and RTX ray tracing and all of that came out during this time. Meanwhile, our financial performance wasn't very good. And so I think the lesson there is: focus on doing your good work and things will work out for itself. And so I'm really, really pleased with the company and the work that everybody's done. And going forward, I think it's starting to ease up now. I think we're starting to have a lot less inventory
in channels. And the industry has more capacity and more flexibility and we're moving nicely into the next generation nodes. And so almost everything is starting to to get better.
What about a price slump? Does that worry you? Everything that we build is rather singular. And the markets that we serve aren't commodity markets. You know, right now, more than any time, the investment needed in AI is just off the charts. Generative AI, this is the moment that we've all been working for in the last ten years. And now AI is about to be used to revolutionize digital biology and genomics and transportation and retail and all these different industries. search. And everything about the situation we're in right
now is really about growth and really about getting into the next phase of computing. And AI is at the center of that. So I'm super excited about the moment we're in.
I want to make sure that we take advantage of it and capitalize on it. The vast majority of your chips are made by TSMC. How have you insulated against geopolitical risks of the region in the case that the "Silicon shield" doesn't hold.
As a company, our first priority is to make sure that we're as resilient as possible. And in every area that we can, to be as resilient through diversity and redundancy as much as we can . In semiconductor design tools, the manufacturing of our chips, packaging, memory, systems. The systems that we build, AI supercomputers, these things are like cars. They weigh 350 pounds per computer. They're the heaviest computers that humans make. And it's complicated.
It's got tens of thousands of unique parts. And so we try to engineer and design, into everything that we do, diversity and redundancy. The fact of the matter is TSMC is a really important company. This is a really special company. And the world doesn't have more than one of them. It is imperative upon ourselves and them, also invest in diversity and redundancy.
And the move that they made recently in building the fab in Arizona is a very big deal. Will you be moving any of your manufacturing to Arizona? Oh, absolutely. We'll use Arizona. Yeah. Yeah, absolutely. The thing that's really great about TSMC is every mask runs everywhere and so they have the ability to use all the various fabs for the masks that we have. And so I'm excited about the investments that they're making so that the entire world can count on them for diversity and redundancy. Yeah, it's a really special company.
About a quarter of your revenue comes from mainland China. How do you calm investor fears over the new export controls? Well, Nvidia's technology is export controlled. It's a reflection of the importance of the technology that we make. The first thing that we have to do is
comply with the regulations. It was a turbulent, you know, month or so as the company went upside down to reengineer all of our products so that it's compliant with the regulation and yet still be able to serve the commercial customers that we have in China. We're able to serve our customers in China with the regulated parts and delightfully support them. And so I think we're going to be just fine in the ability to serve the customers there.
The customers that we have there are consumer companies and consumer internet companies. And the regulation is going to be just fine. We're going to be able to work through it. You are famous for reinvention.
What's the next one going to be? The next big reinvention is probably where AI meets the physical world. And today, all of our AI experiences are related to digital. It's in software. It's, you know, it's information.
It's all digital-related Al. The next generation of AI, and where AI meets $100 trillion of the world's industry, that's in the physical world. And so it could be transportation. It could be robotic surgery . It could be warehouses and manufacturing plant and energy plants and fabrication plants and so on and so forth. And in order for us to bring digital technology and artificial intelligence technology into that physical world where humans are, and safety is important and resilience is important, and all of those kind of physical world physics-related challenges, we need a new type of software.
And we created this thing called Omniverse that allows us to connect the digital world and the physical world. And Omniverse is going to be a phenomenal success. And we have 700-plus customers who are trying it now. And from car industry to logistics warehouse to, you know, wind turbine plants. And so I'm really excited about the progress there. And it represents probably the single greatest container of all of Nvidia's technology: computer graphics, artificial intelligence, robotics and physics simulation all into one.
I have great hopes for it. This is the last one. Just on a personal note, you are the longest running tech CEO. Is there any end in sight? Well, as you can tell, I'm sprightly and quite enthusiastic and energetic yet. I'm surrounded by amazing people.
They keep me inspired and I feel that we could do great things together. They give me so much confidence in what we can do and the impact we can make. And I feel that I'm making a real contribution to the company and to them, and to create an environment where we can make really amazing contributions. And so I think for so long as I believe I could do that, and I don't know exactly how long that's going to be. But three or four decades, I would say. In another four decades, I'll be robotic and, maybe another three or four decades after that.
And so hopefully, I'll get to enjoy this for a very long time. Wonderful. Well, thank you for today's conversation. Thank you, Katie.