NVIDIA founder and CEO Jensen Huang sits down with Joel Hellermark for a sweeping conversation at the cutting edge of AI, computing, and organizational leadership. Huang recounts his early conviction in deep learning, the transformative role of GPUs, and the rise of multimodal AI models. He shares his vision for democratizing intelligence, augmenting human expertise, and empowering billions through accessible technology. The discussion explores the reinvention of software, AI-driven chip design, the importance of domain expertise, and flat, transparent company structures. Huang’s practical and optimistic perspective highlights continuous learning, first-principles thinking, and creating environments where people can do their life’s work.
Watch the full episode above or read the transcript of their conversation below.
I'm so thrilled to be here with you. Jensen, It's an honour to get to learn from you. I wanted to take you back to the moment when you gained conviction about deep learning. Do you remember when that was?
Very, very well, I learned about deep learning at the same time. But you also did and maybe maybe slightly before that. It was because several pockets of researchers were simultaneously trying to submit for Image Net the big contest in 2012 and wanted two things access to the latest GPU, which was the GeForce GTX 580. It just came out yeah and also learned how to program it Yeah and for deep neural nets and so we were fortunate to have learned about them a little earlier but it was called the neural network and as you know, during that time, artificial intelligence wasn't very popular in neural networks were suspicious and a
questionable area of research. But nonetheless, nonetheless, we were helpful and we're the best we could. But the thing that really, really stood out for me was that that showed how effective it was. Yeah. And of course, the question the next question, when you see effective technology that's surprising is how with the scale, what other problems can it solve?
Yeah. And because of the nature of the neural network and each one of the layers isolated from the other, you and the backpropagation approach are so effective you could imagine scaling this tremendously. Yeah. And it turned out we were right. But the thing that the thing that the observation that we made was that I was deep learning was both an algorithm for solving problems that were difficult to specify.
It was also a new method of developing software. You know, imagine you had a universal function approximate or yeah. Of any dimensionality. Yeah. And irrespective of the dimensionality and the size of the problem, so long as you have a large enough model, you could back prop it and, and learn it. And extrapolating that insight I think was really important for us.
And, and we came to be very convinced of, its potential. And because we realized that it was going to be the new way of developing software. Yeah. And since then our capacity has been called software 2.0. However, we observed that it was a new way of doing software in the beginning and realized that maybe we need to change computing altogether.
Yeah. And, and, and, and since then you can look back and realize that maybe for the very first time in 60 years since the system 360 from IBM that accelerated computing using GPUs and deep learning really created reinvented computers that that we might have discovered quantum computing before. Quantum computing, if you will.
Exactly.
And I think that that was a really important moment. We were fortunate to have connected all the dots.
Exactly. Yeah. I remember reading the paper from Alien to folks and how they had started deploying it. And it was incredibly impressive because no one had really put it to that use previously. But ever since sort the model architectures have started evolving a bit. Where do you think they're heading next? You know, back then it was then and we're now seeing increasing bets on multi-modality.
What makes you excited about that and where do you think the model architectures are heading?
Well, let's see, Transformers and the work that researchers have done to reformulate just about every problem, every type of data into something that a transformer can learn, you could take, you could create a vision transformer, you could create an audio transformer, you could create, of course, a text transformer. You could create a transformer just about out of anything, it seems.
Multi-modality is really important for very, very clear reasons. They're they're they're higher performance. So for example, if I were to say, you know, you're trying to train a neural network for a vision and you've seen nothing about horses and you've never seen zebras, but if you add another modality to it like words and, and you understood that a zebra is a horse with black and white stripes, somehow the combination of images of a horse and the words and the understanding of black and white stripes exactly combined allows you to imagine what a zebra is without ever having seen one.
And so you've extended your performance. You could use multi-modality to enhance the robustness of perception. So we do that for cameras and radars and light hours. We combine multiple, multiple sensor modalities so that we can extend the capabilities of your perception to encompass, the superset of all of them. Yeah, you could also, I guess, enhance the capabilities with ambiguous things like for example, if I were to say what is that the combination, of my words and my gestures helps you understand what I meant.
Hard to do that How to do that with pronouns. Exactly. With words. And so modality is really important. And now we have the transformer model and we've, we've, we've expressed transformers in a way to understand multi-modality. And so I think that the next generation of, of A.I. models is going to be more performance, safer, more robust and do more things.
So I think this can be a big deal.
And I think what's really fascinating is just how much we have already been able to derive from language that was basically derived in this world model language that we can apply to a very wide context. But you also speak about teaching this model physics. Do you think that's a new need or could we derive it as these models become multimodal?
Will it able to learn that from the training that they get trained on video for example?
Yeah, I think you can let's see just about what you can imagine just by everything that has been described in words. Yeah, physics has been described in words. Newton’s law has been described in words. So you could imagine that we can learn through the world's corpus of words just about all of the physical effects.
If you've never seen the colour red, it's really hard to imagine what ‘red’ means. But I wouldn't be surprised if there's enough poetry that describes the beauty of a Red Apple and how it compares to something else that's red - a heart-shaped red emoji or something like that - And through all of those words you've connected, “oh this must be red” through compare and contrast.
And you must be able to learn that without having ever seen it. But I don't think you'll ever understand the sensation of something and the subtleties and nuances of it without having the ability to combine all of these different things. Now, I think I think it's possible to, you know, teach people the effects of physics.
But if you wanted to predict the laws of physics, things that have physical phenomena, you wanted to ground the AI in, in physics. And it's no different than grounding large language models today, using reinforcement learning human feedback. Yeah. In the future, you use reinforcement learning and physical feedback.
And the physical feedback, what would what instead of a human doing it, the physical feedback for a robot would be expressed in some kind of physics simulation. We created this system called omniverse that obeys the laws of physics so that we can give omniverse essentially the digital twin of the robotic, the embodied robot if you will, and that in body language model would then get reinforcement learning, physical feedback, simulate the physics feedback, digital twin feedback through omniverse.
And so I do think that there you want to be grounded in physical truths for some types of robots, you want to be grounded in ethical truths, and that's human. That's alignment. One way of helping, helping us create safe, safer chatbots. So I think the two ideas are somewhat parallel and somewhat sensible, and I think it's going to be helpful.
Yeah, that's probably how we all ended up in this simulation. It was you who started experiment doing this experiment. Exactly where do you think the cap cap goes? Do you think we can keep running with the existing model architectures, all of them slightly scale compute, or will we need to do radical breakthroughs and diminishing returns quite soon?
There are you know, I think first of all, I don't know the science behind it, but intuition would suggest that there are many, many aspects of our intelligence where we apply rules and rules that are either learned rules that are expressed. Yeah, there are many principles and rules in our company, in a particular society, and in particular cultures that are expressed simply through words.
You never didn't learn it into existence. Yeah, you know, Thou Shall Not Kill was not one of those things that we learned by trying everything until we discovered that thou shalt not kill. We just, you know, we just learned it. We were given that rule. And so there are many rules that that that that can be expressed without having it be learned.
And I think there's that form of symbolic reasoning that could be could augment these exhaustively learned models. And so there's I don't know what the science is behind it and whether it suggests one way or the other way. But it seems intuitive that we augment our intelligence through things that are learned through reinforcement learning and things that are learned through experience, things that are learned through just because that's the way it is.
One thing we're so excited about at Sana is it's really this possibility to augment the human intellect now as well. If we could provide researchers with the tools to run thousands of parallel experiments and get augmented by this rather than replace it, how do you view the role of humans in this over time, once we reach a point where this is orders of magnitude more intelligent than humans, what role do I have to play it anymore?
I don't know. I don't know. But I'm surrounded by people who are orders of magnitude more intelligent than I in certain skills. Yeah, and I have no trouble co-existing among them. And so I, I, you know, I am already living in an environment where I'm surrounded by super intelligence relative to myself. I feel like they're able to do things that I can't imagine doing.
And, and somehow, somehow I exist, you know, quite harmoniously within it. I, I, I do think that that we are going to elevate basic intelligence. We've elevated many, many things that over time have commoditized a very, very important human resource. We democratized the hunting of food by farming. We didn't have to, you know, run after or be chased after by food.
Yeah, we democratized, of course, the production of power. And so, you know, people who are smaller and weaker can live in a world where things are rather heavy and, you know, and so you don't have to be just gigantic and muscular to exist in that environment. And we democratized access to energy.
And as a result, you know, society elevated in its capabilities. And I think that we're now democratizing intelligence, the production of intelligence. And but I, I think the value will still go to people who have deep domain expertise, and some incredible, incredible passion, which is now elevated, as you mentioned, through the ability to try all of these future scenarios so that we can amplify our own passion and our own expertise.
The other dimension of it, which I'm hopeful for, is that for the last 40 years of my career, the vast majority of the population has not learned how to use this instrument that we call the computer. Yeah, only a handful. And, you know, I think you, you started learning how to program the computer when you were early teens.
And I for the vast majority of the world, they don't know basic they don't know Python, they don't know Pascal or Fortran or C or C++ or, you know, Java or they don't know how to program these computers. And even though it's very sensible to a small population of the world, call it a million or so, billions of people have no clue how to do this until now.
You know and now with with which AGP, as you know, the programming language of choice is human and you could program it and Swedish and I can program it in English and you can program it to write a program to program another computer. Yeah. And, so we democratized programming for the very first time in history.
And I can't help but believe that that's going to empower billions of people who currently see this instrument, this instrument of value creation and this instrument of productivity. You know, there's this incredible, incredible light travelling, you know, light speed travelling machine that we sit on, call the computer, you know, for the first time, everybody can have access to that.
And so I'm hoping that it's going to bridge the digital divide. I think you could make a case for that. Yeah. And I and you know, and you're seeing it happening right now in real-time with kids on the Internet, prompting these machines, writing amazing programs, generating beautiful images. And they're just kids. You know, they didn't know how to write a program, but they could do this.
And so I think that's really fascinating as well with how you're accelerating computing at NVIDIA because you're using your satellites to develop more advanced computing. Could you tell me what that cycle looks like?
Well, our current generation of chips is so large and so complex that all the employees in our company can't possibly design it. And so we use AI to help us design, design, explore the design space and as you said, experiment with hundreds of thousands of permutations of the design and find the one that has the best tradeoffs.
And the best tradeoffs are tradeoffs that we determine. There is there is no perfect optimization. There's just an optimization that's based on some tradeoff. And sometimes we might want to trade it off on speed, an all-expense. And the reason for that is it's the critical path. And you want to make that critical path absolutely as fast as possible.
Maybe it's because it represents it's something that you have to instantiate thousands of times. And now energy efficiency is really important. So you optimize for energy efficiency. And so we make those kinds of judgments and then we set the AI free to go and discover all these chronic conditions and all these different permutations for us. And it comes back with a design that no human can do.
And that's really quite amazing that we're looking at these designs that we've put into the latest generation Hopper that no human has ever designed before. And it's that incredible. And so or it's because we have to connect thousands of modules on a chip and you know the combinatorial explosion of thousands of combinations is more than all the atoms in the universe.
And so only AI could figure out what is the best optimization for that.
One other question I was really intrigued by your perspective on is now how should we think about defense ability? You know, last time we spoke a bit about the value of domain expertise and really integrating yourself deeply into workflows, but for companies executing on their application layer, how should I think about the sensibility as these models become increasingly commoditized, they're able to learn from smaller and smaller amounts of data.
A lot of the historical moats don't seem to apply anymore. How should we think about building a defensible business?
Well, I don't know that there's any evidence that understanding a domain of problems or understanding a segment of customers has ever lost its value.
Our technology, the world's computer technology and the number of well-educated people obviously growing, the new college grads that are coming out of universities these days are really quite amazing. You know, just about every single one of them was smarter than we were then, you know when we first graduated. And the type of problems they can tackle just from, you know, graduating from school were the type of things that entire companies used to do.
And yet there's every every every every evidence would suggest that that understanding deeply your customer's challenges is really, really quite valuable. And so I think that that that's not going to go away. We're going to have incredible basic capabilities and it could do amazing things. It could solve, you know, all the hardest math problems. And it'll pass, you know, bar exams and yet we know very well that somebody who's passed the bar exam when brought into a domain of a particular issue still has to learn about the fundamentals of that industry.
You know, what is the problem that needs to be solved? What are the complex relationships between people? You know, oftentimes customers are people only. They know they have complicated challenges that they're trying to try to solve themselves. And so understanding people, understanding the context and oftentimes those things are poorly expressed. And so those are social science problems.
And what Are you most excited about the application?
But Well, for us, there are three, I would say, if I can categorize it into three things, what are the things that that I can do that can enhance, revolutionize the work that we do to build our products? So that's one and we were just talking about one. The way that we design chips has completely changed.
The second is the way that we design software has changed. And the second, of course, is what are the things that I can now enable us to do so that our products are different now, revolutionizing not just the way we design products, but revolution, revolutionizing the products that we build. So for example, there are many, many gamers in the world and the way that we used to build a graphics card, we would, we would design programable shaders.
And of course, there were compilers and things like that, but that's it. We ship it and we ship a great processor with great compilers and we integrate them with games and such. But now you can't even ship a G force by itself because there's a supercomputer in the back learning how to make predictions of the missing pixels because we have to make noise and we have to infer pixels.
We infer about, you know, one out of every pixel that we render, we now infer somewhere between 8 to 16 pixels. I mean, it's like getting a jigsaw puzzle where you're given one piece and you're supposed to guess the other.
And so we taught anyhow, we have a supercomputer in the background just learning how to do that and improving the algorithm. Then we download the algorithms whenever we improve them. Yeah. And so now we use AI to revolutionize not just the way we design the GPU, but the way that the GPU even produces the images. And so it helps us create much, much more energy-efficient processors that exceed any capabilities that Moore's Law would have predicted.
And then the other thing that the third category I would say, is where your world is, which is to turn the whole company into an API. Yeah. And so that all of our employees are augmented by this, you know, a system that is running all the time.
And and so that so that we're not searching for information that we'll never find so that maybe we can connect dots and predict opportunities in a marketplace or maybe the supply chain has changed, the supply chain has changed, or maybe the market demand has changed, and it's impossible for us to see all the signals, but it's not impossible for an API to see all the signals.
Yeah. And so over time, from the way that the employees work with each other, the way we learn about the information that we need to use to the way that, you know, we forecast demand and work with our supply chain, all of that is going to be revolutionized by.
And what do you think is your most contrarian view on AI right now?
I don't know that I have anything that's particularly contrarian because, you know, if you look at the world today, most of the conversations about AI are either beyond enthusiast for its promise and and beyond the concern of its perils. So somewhere between the extreme conditions of promise and peril, which are being discussed in real-time all the time, probably is the truth.
There's no question that technology of this capability and throughout history, technology of any capability has brought enormous transformations and discontinuity in society and economy. And so there's a fair amount of parallel that we have to consider. Yeah, the type of who would have thought that there would be a whole bunch of people who are doing web design?
Yeah. You know, this is this is a this is a profession that didn't exist or programmatic ads. These are professions that didn't exist When I came out of school 40 years ago.
But they're entire industries today. And so somehow this enabling technology called the Internet has recommender systems. The early versions of AI have made it possible for us to create this new industry. I think that we're going to have to reskill, retrain, and so that so that jobs that are dislocated can be transformed. But I'm certain that there will be brand new industries that are created that we never thought of.
Like, for example, right now we are seeing this coming to coming alive in real-time in prompt engineering. Yeah, exactly right. Prompt engineering is like it's going to be a real thing. It's going to be a giant industry. It could be the most important programming industry.
It's actually it's.
And so how many people are going to be prompt engineers? It's incredible that you're seeing eyes that are helping write prompts right to prompt others I'm.
Sorry metal.
Spring indeed yeah and so so I, I think we're going to see all of that and I think I think there are a lot of good things that are being discussed. I love the conversation about safety. Yeah. We have to dedicate as much in, to technology investment in the capabilities of AI as well as the safety of AI.
Yeah, you know, we happen to be we have been working in the field of self-driving cars, and we probably have just as many resources dedicated to making the AC and the car safe as we do to making a drive. And so I think largely language models will have to do the same. And the technologies associated with building ice for guardrails so that you keep it in the operating domain to technologies related to alignment, reinforcement learning, human feedback, which is like reinforcement learning what physical feedback that we were talking about earlier to technology is of vector databases that reduce the ability or reduce hallucinations.
So you augment it with facts. With all of those capabilities, I think I think we're going to see just an explosion of new ideas to transform that core large language model and surround it with other A.I. technology and methodology use and best practices. All of that is going to turn that large language model into useful chatbots. Exactly.
One thing I admire so deeply about you is how technically current you are still. I think you must spend a lot of time reading or whatever, but you're always so incredibly thoughtful and well and every single detail of what's going on in the industry, how to stay so on top of things while running one of the world's largest companies.
Wow. Let's see. What's the answer to that? Well, first, I'm surrounded by amazing people. When I went to visit you, you were surrounded by amazing people.
And I and they're generous to to to teach me. And so you have to make the effort to learn. You know, people are people who are. I love teaching people who are great students. Yeah. And so I dedicate myself a lot to being a good student.
And, you know, of course, we're in, in a lot of domains from, you know, self-driving cars to climate research to digital biology.
And so the vastness, the breadth of impact that we can have in the world is great. But, but we also have to we have to learn to be able to, you know, and so you're in a tech you're in a tech very tech driven industry solving, solving and creating solutions for companies. Yeah, I'm in a very tech-driven industry.
And so for both of us, it's essential we understand the underpinnings of the technology. So you have an intuition for how the industry is going to change.
You have an intuition for how which one of the technologies is a bit of a little bit of a, you know, left turn and which one is fundamental. Yeah. To realize, that maybe the early works that we did with generative adversarial models to variational autoencoders to diffusion models were somewhat cousins of each other. Yeah.
And that realizing the impact of one could lead to a breakthrough in another which opens up the horizon for now diffusion models that are they're utterly incredible. Yeah and so I think having an intuition for technology allows you to better extrapolate Yeah and in our ability to extrapolate and see down the road is really vital because, because gosh, you know, technology is changing fast, but it still takes us several years to build a great solution.
And so how do you how do you, on the one hand, dedicate yourself to building something that's going to take years to do? Yeah, building it on top of technology that's utterly changing, you know, by a factor of a thousand every few years. How do you do that? Unless you have an intuition for it? Yeah. And so I think that the fact that you're so deeply gifted in technology and so you understand it and you have great interest and curiosity in technology is essential to running a technology-driven company.
And so I think it's it's I love that part of my job. It is surrounded by people who are who are generous to teach me. And I just have to dedicate myself to being a good student.
So it's an area I'm very intrigued by as well. This is how you run the run the company. I've understood you don't have one-to-ones, you know. Can you talk me through some of the sort of classic management playbook that you've challenged and evolved?
Well, first, with respect to building a company, the first thing that you have to do as with all problems and you do this very naturally, you start from first principles. Yeah. What is what is this? This this machine that we're trying to create and what is what is this output? What is this input? What is this output? What are the conditions that it's in?
What is the industry like? Is it a fast-moving industry? Is it a bureaucratic industry? Is it a highly regulated industry? You know, what kind of industry is it and what are you trying to what are you trying to build? And so I think you think about it from that perspective. There are several things that I wanted to do with the company.
I wanted to create something, a company that naturally attracts amazing people.
And the reason for that is because we're we're solving problems. Our company's mission is to solve computing problems that are barely possible. And if a problem can be solved by normal computers, we don't do it.
And so we have to go find problems that are that are impossible for normal computers to solve are barely possible. Yeah. And so you want to attract amazing people who want to invent this new form of computing and apply it to solving some really difficult problems. And so I want to be an amazing person. Second, I wanted a company that was smaller, not larger.
Yeah, you want a company that's as small as possible, Not as large as possible. Yeah. You know, it's it needs to be as large as necessary to do the job well but to be as small as possible. And so naturally, you want to empower people.
Well, if you want if you want an organization that obeys command and control, then you make it a pyramid. Just like. Just like the old military all the way back to the Roman Empire. And. But if you want to empower people, then you want to make it as flat as possible so that information travels quickly. And so in order to make something as flat as possible, the first layer has to be well considered.
Well, the first layer happens to be no senior senior people. You would think that they need the least amount of management. Yeah, nobody's coming to me. None of my management team is coming to me for, you know, career advice. You know, they made it and they're doing great. Yeah. And so, I have a whole lot of people reporting to me because I don't need to do what I want.
I don't have to do career coaching. I have to you know, they're all fabulously happy and they know what they're doing. They're experts in their field. And so those one-on-ones are really not necessary. And and if if there's a strategic direction, why do you tell one person? Yeah, you tell everybody. And so after we're swimming in the soup of, of strategizing and how to formulate the path to the future, you know, when the time comes, I just, you know, I send it out to everybody at the same time or I'll tell everybody at the same time and people will give me feedback where we find it.
And because the company is so flat and you've empowered the organization so much with knowledge of the company and and their their access to information, the company is also agile. And so it turns out it turns out that that by having a lot of direct reports and not having one-on-ones made the company flat information travel quickly employees were empowered which made it possible for me not to do one-on-ones that algorithm was well conceived.
Yeah, and the architecture is well-implemented. We also don't have business units. We don't have divisions. Everybody. Everybody works us one. Yeah. And two, the company is shaped in a way that allows us to build accelerated computing fast. You know, if you ask me to go do fried chicken, we'd have a hard time doing fried chicken. Swedish meatballs.
No chance. But accelerated computing very well.
That's. I think you have 40 direct reports, right?
Yeah, something like that. Yeah, yeah, yeah. The challenge is getting everybody together. Yeah. You know, when I want to get ready together, but either somebody is out or somebody is on vacation or somebody. Somebody doing something.
Exactly. Yeah.
The odds of everybody sitting at the office is approximately 0%.
How did your leadership style change over time? You've been going on for four decades now. How did that evolve as you learned?
Well, I don't really have a style. It's just just me. You know, there are a lot of things that that I want to do better. And you know, if something's happening at work and I don't like its direction, I'll just say it because I don't take anybody aside. Do you know 1 to 1 coaching? If something's not right, I'll just say it.
If I have a different opinion, I just say it. And so. So it could be a little too direct. But if people just realize that I'm not trying to do anything except be direct, and then I spend a lot of time reasoning through my decisions, which empowers employees because they learn how leaders think through problems.
Yeah, And just by meeting I'm in, I'm explaining how do I think through this. Yeah. Let me reason through this. Yeah. Let me explain that you know why and how do we how do we compare and contrast these ideas. That process of management, I think is really empowering and we also we also don't don't do just vice president meetings or just director.
And, you know, the meetings I have, there's new college grads in there. They're people from every different organization. And we're just sitting Yeah, we just all sitting in there. Yeah. They're kind of like your office. Yeah, yeah, yeah. But yeah, everybody's just kind of sitting in there.
Exactly. Like, that's actually one thing that I found very intriguing because that's one of the playbooks, right? You have like a very clear leadership term and you have the leadership to remember things and so on. And that's something I've always struggled with because you'll have a lot of the best individual contributors and of course, like they should be in that building.
It shouldn't be the just like vice presidents, you know, not knowing that that the craft it's it's fascinating that you can that's.
Exactly that's exactly you got it You want the person who is most informed or best skilled or just had the most experience. Yeah. They actually made the mess or they actually confronted the situation. You want you want ground truth. Yeah. You want you want ground truth and experts the best you can.
And you have some model for folks to communicate their top priorities. I've heard something about sending an email. What's that all about?
We have. We don't do status reports. Yeah, And so I don't read any status reports. And the reason why I don't is because status reports are are you know, they're meta information by the time you get it, you know, and so they're they're they're barely informative.
You know and so I distilled and refined and bias has been inserted perspective has already been added and you're not looking at ground truth anymore and so I, I tend to appreciate information on that that anybody presents. Yeah. So you're allowed it if you send out an email and it's called top five things and just whatever happens to be your top five things whatever observed or whatever you did or what have you learned or thinks that just thinks, really.
Yeah. Top five things, whatever it is. You just went to a great restaurant. Who doesn't want to hear that? You know, that's important information. And so I just had a baby. That's important information. Yeah. You know, so. So whatever these things are, top five things. If you send it out. If you send it out, I'll read it. Yeah.
And so I read I read every single morning probably, you know, 100 or so. And I do it every day.
And it's one giant thread that everyone in the company senses, you.
Know, everybody has their own version of the top five things and they just send it out. You know, if you send it, I'll read it.
What are your top five things?
The top five things are not meant to be from the centre out. They're meant to be from the out. Yeah, they're good. That's right. Yeah. Think of it as IoT. Yeah. If I take the top five things that I sent it out. Yeah. Then I contaminated the system. Yeah. Yeah, that's the reason why I don't do it. But I have my own tough, like, things that I keep to myself.
And how do you balance that with? With planning so sort of bottoms up, be having the best, best engineers in your term decide what to work on combined with, you know, sometimes you also have to execute on, on, on the plan. How do you balance those two things?
First of all, strategy is strategy not words. Strategy is action. Yeah. And so if the company has a set of strategies, but the people's actions, their top five things are not that, then they're obviously not executing the strategy. And so the strategy, it turns out, isn't what I say is what they do. Yeah. And so it's really important that I understand what everybody's doing.
And you do that by just getting a feel for every stop point thing and you don't have to read all of them. You have to read them all every week. You don't you just, you know, it's sporadic and, you know, sarcastically sampling the system you have a feeling for where whether the company is going in the direction that you want it to go.
Yeah, that we all agree we've gone and so that's one-second planning we don't do a periodic planning system. And the reason for that is because the world is a living, breathing thing. And so we just plan continuously. Yeah, yeah. There's no five-year plan, There's no one-year plan, there's no plan. There's just what we're doing.
That's that's, that's really that's really exciting, to hear. I think one thing as you're executing on first principles and you come to some of my ideas, it can also be hard to trust your intuition if you're doing something that's, you know, contrarian to what that the playbook is. What do you think made you trust your intuition on some of these things?
Well, you know, most most everything that that you dedicate the company to go after should be reasoned through first principles.
You know, there's a there's a foundation of what are the assumptions, the important assumptions that led to you believing that the computer has to change or the chip architecture has to change or the way the software is developed or how a data centre has transformed. You know, a data centre used to be a place where we stored all of our files and we would go retrieve them.
But every company in the future will have to date more data centres, but one of the data centres will not be a centre for data. But it's a factory. Exactly. It's a factory for producing intelligence. And data comes in. It's refined through the computer. And what comes out is this invisible thing, which is the most valuable thing in the world called intelligence.
And this this building is going to be, you know, driving this this thing continuously. Right. And so you and I, we're all going to have factories and how would you reason through that and you kind of back your way out, you know, And before you know it, you formulate a view of the world based on first principle thinking and and and then the next part are you go after it with enough dedication, you know, with conviction, so that you could realize it.
Oftentimes it's really hard. But if you're wrong, you change your mind. And that's that's the thing that's really great about about modern leadership. You know, if I'm if I'm wrong about something, I just say so I'm just, you know, that was wrong. Yeah. You know, that was goofy. Yeah. And then you say, I changed my mind and then and because you're.
You're adapting. Yeah. And literally replanning constantly. Yeah. It's interesting that over time people might not even notice that you've adapted, you know, 17 times in the last year. Yeah. You've changed your mind maybe 35 times. Yeah. You know. Yeah. And so if you don't do these giant five-year plans.
Which I think, I think five-year plans are just horrible for technology. First of all, it's just ridiculous. And these continuous planning systems could, could maybe just lead to easier leadership.
Yeah. One thing we were as obsessed about as the product we would build was the company we would build. And I think there are very few companies that are truly focused on empowering folks to do their life's work. And that has also been, you know, a key passion of yours. And how do you enable that? What are some things that you put into place to empower folks to do their life's work at Nvidia?
That is the mission of leaders to create an environment for others and to empower others to do their life's work. And now there are a couple of ways you do that the most important way of of realizing that mission is to not cause people to have to do commodity work. So for example we never talk about market share in our company.
Yeah. And the reason for that is because why are you talking about I have 23% market share. They have a 27% market share. Why are you fighting people for market share? Yeah, because the whole concept of market share says that there are a whole bunch of other people who are doing the same thing.
And if they are doing the same thing, why are we doing it? You know, why am I squandering the lives of these, these incredibly talented people to go do something that's already been done? And so unless unless we just enjoy the competition, which, which I tend not to we tend not to go fight people for market share, fight people for for markets that that that are already commoditized.
And so that's one way of thinking to go do something that's never been done before. The Other way. The other way to demonstrate that is to walk away from businesses that have been commoditized. Yeah. And and either either through through our own initiative or otherwise. We've walked away from many businesses in the past. And so that that demonstrates very clearly to your employees that we're not going to go do commodity work.
And so the combination of choosing the right work and walking away from the wrong work is the best way to create the conditions and the rest of it is what you and I were already talking about, which is empowering people with information. Exactly. Whereas some companies are very siloed and information doesn't travel outside of organizations. I encourage our company to be rather transparent.
And if you ask me a question about our company secrets, there aren't there wouldn't be that many secrets. Yeah, you know, and so, so on that that empowers people. Yeah. And the rest of it is, is how you conduct yourself during work. Yeah. There's a sense of hierarchy in the company then obviously that that's not very empowering.
But if anybody can come into a meeting and contribute, including a new college grad. Yeah, that's very empowering. And so, so I think empowerment is a big deal. AI - Multimodality AI - physical feedback model AI and humans - optimistic view AI - Upskilling/ Emerging Industries AI - AI Safety Leadership - First principles leadership - flat hierarchy leadership - leadership style leadership - five year planning leadership - five year planning/ being agile business - finding your niche