Transcription
Please welcome President and CEO of Nvidia, Jensen Hong, in conversation with Institute Chairman Michael [Music] Milin.
Thank you guys. [Music] They're big fans of yours. I think that's for you. I think the hoohoo is for you. And I think your outfit is for me. So, welcome.
So, AI is this the next industrial revolution? Yes. The next manufacturing revolution? Yes. Let me let me uh let me explain. So, so all of us have have um uh been talking about the technology of AI, that it can perceive the world; it can uh generate content; it can translate; uh it can now even reason and um solve problems; use tools; use the web browser; read PDFs; um uh do research for you. And so so we know what the technology uh is able to do, and that's very exciting in itself. It's transformative completely in itself. Uh we understand that the technology is unlike uh any IT technology of the past. Remember, IT technology is a tool. You have to use it to make it effective. You have to sit in front of the computer and use it. Uh but now AI has the ability to uh automate, and the concept of of robotics and robots are very well understood, and so imagine a physical robot, but you know we understand that we can imagine that uh imagine a digital robot and it's in you know in the computer in your data center doing work for you. And so so this is it's exciting because for the first time it's no longer just replacing or the next generation of the IT technology that we know, but for the first time it actually could augment um and add to the digital workforce. And so the part of the economy that is part of uh is much larger than a trillion dollars; it's part of the hundred trillion dollars. And so that's that's the first layer.
The second the second layer is uh how do you uh how do you generate this AI? How where does the AI come about? You know, whereas the the last generation of computers uh was uh software written by hand and it runs on CPUs, what NVIDIA took some 33 years to build is this idea of a new type of computer that learns—the machine learns to write the software itself—and it runs on this this processor, this computing platform we invented called accelerated computing and GPUs. And so so so now the question is how does the AI get produced, and it gets produced in essentially what people call data AI data centers, but it's essentially a factory; it's unlike a data center; it doesn't look like a data center; um it's uh quite large in scale. Uh it does use energy, and it produces—you apply energy to it, and it produces these things called tokens, but they're basically numbers, and these tokens can be reformulated into numbers or words or images or pixels or videos or chemicals or protein combinations for drug discovery, um or even uh uh motor skills uh necessary to drive a robot or steering wheels to drive a self-driving car. And so that these tokens are being manufactured by this factory. And so what's interesting is that people are starting to understand that that there's a whole new industry that has been created. This new industry has a factory, and this factory—there are AI factories—and um uh how large can the these factories be? They you know they could be a gig. We're building ones that are about a gigawatt, and each gigawatt's about you know 50 60 billion dollars, and over the course of the next um you know call it 10 years or so uh I wouldn't be surprised to see tens of gigawatts of AI factories being built around the world, and so that's the second layer.
The third layer that's that's probably even more profound um is that for the very first time you have a capability, a technology um that affects almost every industry, from financial services to healthcare to manufacturing to logistics to you know retail to entertainment—you name it. And so so this this this infrastructure, if you will, this AI factory now becomes an infrastructure for a whole bunch of other industries, and and uh just like the last generation this infrastructure is kind of hard to understand, but the last generation we had the information infrastructure, and the generation before that we had the the energy infrastructure, and now we have the intelligence infrastructure, and the the internet was is the information in uh infrastructure. ffstruure and artificial intelligence is this one. And so so now you know I think when you look at AI from those different lenses, you could start to understand the impact of AI to the technology industry that we're in, to a new industry uh that that every country wants to be part of—anybody have excess energy, you're going to want to be part of this industry—to the infrastructure that affects every industry. So let's step back for a couple minutes and talk about the skill sets needed to interact.
Um we estimated a number of years ago that if you took the most modern agricultural technology that in the world you might eliminate a half a billion jobs of what's going on in farming, subsidized etc. There's significant cons, you know, questions today. Who's going to be disintermediated? Now, I had a substantial advantage uh in the 1960s that I could calculate yields in my head, and then in 1970 they came out with the calculator. So, I got disintermediated. Uh then looks like he's done pretty well since the invention of the calculator. Then I could remember millions of trades, and then computers started storing in those trades. What do you see in the concept of work and the interaction with the technology you're going to provide?
Yeah. Um so we've you all of you have heard a lot about about job displacement. Uh every job will be affected. Uh some jobs will be lost; some jobs will be created. Uh but every job will be affected, and immediately it is unquestionable. You're not going to lose a job your job to an AI, but you're going to lose your job to somebody who uses AI. Um but let me give you and those are those are fairly common sense things to to to uh uh to have observed, but let me give you the two extremes that you might want to consider as well. Uh computer technology computer science has benefited about 30 million people. There are about 30 million people in the world who knows how to program and use this use this technology to its extreme. And it's really benefited all of us that have been in this industry the last 30 years. Potentially one of the best and most most wealth creating industry you could have selected. I could I could have been a petroleum engineer. My dad was, and uh I could have been I could have been a doctor. My my mom thinks everybody should be a doctor. uh but I chose I chose uh to to go into computer engineering, and it turned out to have been quite a good good choice. Um and and how however however there are about 30 million people like in in this industry. Um and so we've created Mike in the last you know 30 40 years probably the greatest technology divide the world's ever seen. The the the instrument that we've invented uh we know how to use, but but the other eight seven and a half billion people don't. Um I I'll I'll put on the table that in fact artificial intelligence is the greatest um opportunity for us to close the technology divide. And let me prove it to you. You know, if we just look in this room, it's very unlikely that more than a handful of people know how to program with C++. Um and uh an equally equal number know how to program in C. Um and yet a 100% of you know how to program an AI. And the reason for that is because the AI will speak whatever language you wanted to speak. You could draw a schematic and show it to it. You could draw a picture and ask it what to do. Um you could, you know, obviously talk to it in in words. You could you could write a prompt. Uh you could describe your prompt in a very explicit way. Uh you could describe your promp prompt in a very implicit way. Uh and if you don't know how to program that computer using AI, you just tell the AI, I don't know how to program you. How do I program you? and the AI will tell you exactly how to program you and and program it. And so I think that that um and the number of people who are using ChatgPT and Gemini Pro and these AIs kind of demonstrate that in fact this is one of the easiest to use technologies in his in history. And so now all everybody could take advantage of this capability where there's a a a teacher or a student wanting a tutor, and every student should use it as a tutor. I use it as a tutor every day. And so I think the the ability for us to now use artificial intelligence to close the technology gap is incredible. So that's one extreme. The other extreme that I will say is that remember uh we're we have a shortage of labor. We have a shortage of workers. We don't have an abundance of workers; we have a shortage of, and for the very first time in history we actually have we can imagine the opportunity to close that gap, to put 30 40 million workers back into the workforce um that otherwise otherwise uh the world doesn't have, and so you could you could argue that artificial intelligence is probably our best way to increase the GDP, the global GDP, and and so those are you know two two other ways to look at it. In the meantime, I would recommend 100% of everybody you know take advantage of AI, and don't be that person who ignores this technology and this result loses you.
So let's talk for a moment. They're going to walk out of this room and Thursday after six days of the conference, they're going to want to learn more about AI. Do they ask their computer to teach them about AI? Is that what we're going to do?
Excellent. Excellent way to do it. Just pick up your phone, get yourself, you know, Perplexity is pretty good. Chat GPT is really excellent. Uh Gemini Pro is excellent. I use all three of them. And um just ask it whatever you want to ask it about AI, and it'll tell you as deep as you like it to be. And I sometimes I I um uh in areas that that are fairly new to me, I might I might say uh start by explaining it to me like I'm a 12-year-old and then work your way up into, you know, into a into a doctorate level over time. And so so I you could all do the same.
Let's look at it from another side. Johnson, you your family came from Taiwan, you went to Washington, then the eventually your parents moved to Oregon, and I've had a chance to finance many other entrepreneurs. Uh Bill Gawan comes to mind at MCI. So he had a company that had 99% market share in AT&T that he wanted to take on. In those early years that you often talk about, you didn't know if you're going to make it or not. He often was wondering where the payroll was coming from every month. Um what did the other companies miss who had more access to capital than you did at the time? What did they not see that you saw?
Um gosh, what in other words in Intel, what did Intel not see in the market? What what did they not recognize?
Yeah. The reason why I paused it to say it is is um because from the very beginning we we imagined so what we were trying to do as a company was to build to invent a new way of doing computing that solves problems that normal computers can't. In fact, if you just if you just wrote that mission statement out to do something that normal things can't, it's like to go I would like to build a new car to go places where normal cars can't. Well, usually what happens if normal cars can't go there, those places also aren't paved with roads, or they're not that desirable to go to anyways. And and um and so we we came up with this mission statement to solve problems that normal computers can't. And uh uh several problems with with that with that mission statement. Turned out it took us three three years to do, and it's we succeeded at it. But the first thing is that the whole economy, the whole industry, the whole ecosystem wants to go where problems can be solved. Nobody wants to go to where problems can't be solved. And and so where where we are was rather lonely. You know, there aren't other people solving this problem because it's hard to solve. uh there aren't many customers because they they tend not to choose problems like that. They want to have their problems be solvable, not unsolvable. Uh and and then the other thing is is that Intel watching us the whole time I had the benefit and you said they had they had um greater source of cap access to capital, and that's completely true because they were so successful doing what they were doing. They kind of rejected what we were doing, and that's in fact the good news over time. The reason why it took us so long is because it's hard. And the reason why we're here alone is because people left us alone for a long time. And and and there there there's a there was a book that was rewritten recently, and I I I picked it up and skimmed it. Uh Peter Teal's 0ero to 1 book. In a lot of ways is kind of a story about Nvidia 2. You know, we we chose we chose to do something that nobody thought was possible or very hard to do, um and very unlikely to succeed. Uh but to us it was very common sense. And so I think simultaneously because it was hard to do um and also because they were so successful doing what they were already doing, they kind of rejected the idea until until everything came together.
Well, and you're also trying to make sure your company doesn't go in the direction of Intel also. So you're the leader today. How do you get that culture of constant innovation? And if I wanted to talk about Captain Kirk to go where no one has gone before, um I think I think partly well first of all there's just no guarantees, but but um we have several things about our company that's really quite extraordinary and and and I appreciate it as a person. I I wish my I wish upon my kids and the people I love to have have the same same his same experience, which is which is that long long suffering that comes with struggle, and you never take anything for granted. You're super super efficient. Uh you're trying to you're trying to save everything you can save every penny you can because you don't you don't know when you know how long the struggle is going to last. uh you have incredible resilience uh because because it took a long time to do it, and so the company has has that in its character. Uh almost everything we undertake these even these days are you know 5 10 year endeavors. We're probably we're probably the the deepest in this new area called physical AI, which translates to robotics uh in the world, and um uh the fundamental technology necessary for the next generation of AI were probably the furthest along the deepest of anybody, and and so so I think I think those those characteristics of dreaming big on the one hand um and having the resilience and the character to to suffer until you see it happen. I I think that's that's very good. Um I think the the other part that's good is is um you're always going out of business for us for you know for 30 years we're always in a perpetual state of going out of business, and so you don't take anything for granted, and and I I don't you know when we when there's a setback it doesn't it doesn't it doesn't trouble me too much. Um when we make mistakes it doesn't surprise me too much. uh when we have success, I don't I don't take it for granted, and we don't celebrate it too much. Um and we really stay, you know, stay focused on on doing our work. And so I think part of that is is just how long it took to build a company.
Let's talk for most lay persons. How do you make a chip? What is required to make a chip? So we all like we'd all go out there and make chips. We have no idea how to go about it, but we'd like to. And as you remember, the US passed an act. We're going to invest 62 billion. And then they discovered six months later, there's no one in the United States that knows how to build a factory. And we need to get 7,000 people from Taiwan here.
Mhm. Okay. Well, with all things, I think I think craft craft and artistry matters. Um, if you want to learn how to build a chip, you know, I would start with YouTube and then and then Uh so it turns out uh we started we're very good at building chips, and the reason for that is because we build uh not not since IBM in the 60s has a company like us existed where we come up with a blank sheet of paper, design a brand new architecture, create the chips, create the systems, um create the networking, create the infrastructure, write all the software, take that software to the market, um have the world's developers and ecosystem develop for that computer, kind of like we develop for um for uh iPhones and you develop for Windows, you develop for Nvidia. And so so u not not since not since the 60s and 70s uh when IBM built everything from the ground up has a company like us exist. We build the chips, but we build the entire system, and we're really we're really an AI infrastructure company today. If you look at the systems we build, each one of our chips, if you will, it's a ton and a half. This is a ton and a half chip. It's $3 million each. We build these things in very high volume. Uh we we uh uh manufacture it, assemble it, and then we test it. We use a supercomputer to test the supercomput because you have to be smart to test if the computer you make is smart. And so we test everything is liquid cooled. Uh, and then we we test everything, assemble everything. We disassemble everything, put it on a plane, ship it to wherever the the the data center is, assemble it again outside their door, put it inside their data center. This entire process has 200 manufacturers and uh um supply suppliers working with us around the world. Uh we build a couple of hundred billion dollars of it a year. These at this moment, we're the largest technology company, chip company in the world, I guess, if you will. And so the it's it's it's uh incredibly hard to do. Our R&D budget, you know, per generation is probably about 2030 billion dollars. Um and so these are this is a this is a giant game. Um but but um uh we're we're working into an industry, you know, Mike, that that you know, the the uh the intelligence industry will likely be measured in trillions of dollars. And so that the amount of investment that we're making is is warranted for the the opportunities ahead.
So, we've all had a chance to read about potential restrictions on where you're going to be able to sell your chips. Uh there's pros and cons that people put forth in this debate. Lay out the issues as you see them. um Nvidia's technology um oftentimes described as a national treasure, and um uh the technology obviously uh is important to this new industry called artificial intelligence, and so uh one extreme uh in one side we want to we want to make sure that this technology is available only to uh the friends of friends of our nation; we want to ensure we want to ensure sure that that um uh access of this technology doesn't fall into the hands of people who might use it for military reasons. And and so those are those are the arguments uh for limiting the access for economic reasons, for national security reasons. And and um uh the the fallacy the fallacy of that is is uh uh no government, especially the government of of of uh our adversaries, are limited by uh the available capacity of computing in their country for their military reasons. uh our country is not—no countries are—if uh they they need it for military advance um they'll just secure whatever computing resources that they already have, and uh there are millions and millions of NVIDIA chips in in just about every country already, and so so it's it's not going to limit um shipping additional GPUs Nvidia technology into into uh whatever country is not going to limit uh their military. I think the the the reason why uh the pro the the reason for um leaning into uh uh the export of this technology is we want to build the world's AI uh whereas whereas American standards are being adopted around the world, the ecosystem of artificial intelligence will build on on top of our standard versus somebody else's standard, and we are not alone. You know, Nvidia of course is the world leader. Okay. Um, but in our absence, if we don't serve a particular market, if we leave a market altogether, uh, there's no question somebody else would step in. Huawei, for example, is very formidable, one of the most formidable technology companies in the world. They'll step in. And so, so the the reason why, uh, it would make sense is to to win in the marketplace. uh make the American standard uh the global standard, have AI be built on top of American technology, and and of course very importantly it's a giant market. You know, when we were uh limited to ship to China uh the Chinese market in a couple years is probably about $50 billion; the the market we've left behind is utterly gigantic—$50 billion—just so you so you have a feeling for that number—$50 billion is like Boeing—not a plane—the entire our company. And so so um uh that's the that's the business opportunity that I think we could we could enjoy, uh bring back tax dollars, uh create jobs, uh advance uh advance our technology.
Even further, well also your interaction with the customer, you're missing that chance to learn from that interaction from the customer—the most important thing for any business is the interaction with the customer—and what have what have your customer what have you learned from your customers over the last few years and their demand for one their demand for chips but their use of chips, and how is that feedback to you and Nvidia?
Well, we work with just about every AI developer in the world, and and um we learn everything from uh how our architecture how the the the nature of our technology is used is optimal or not optimal for the future of AI. And so uh when we understand what an AI researcher would like to do with um uh for example um the the AI model for a for a virtual cell, you know we we've now uh made very good progress with virtual proteins, and so now we're working on virtual cells, and if we can understand how cells uh interact and how how its pathways um uh can be expressed uh and uh understand basically the meaning of cells and and its dynamics, we could do that with AI. So that AI model is very different than AI model for large language models, and so understanding how people want to use that helps us change our architecture in the future to be better suitable.
So as I mentioned you Jensen you know many years ago I went to IBM and tried to convince them at what they called their super chip at the time that we should use it for medical research. And then they wrote me a Dear John letter. Thanks Mike, great presentation, but we're going with video games. So, where where does the demand you see in the future? Is it is is certain industries going to play a larger role such like bioscience and what you could do in that area? Um where do you see the demand from different industries?
Well, if you look at where AI is today, as large as Nvidia has been already and as large as the AI industry is already, we're serving basically the consumer internet market. If you just kind of take a step take a step back for a second, that's the entire that's the only tiny little sliver of the global economy that we're serving. Uh above this are uh healthc care industries, uh life sciences, uh manufacturing—the act the actual manufacturing in the future—the factory will be one gigantic robot orchestrating a whole bunch of robots inside working with people to build products that are robotics. So you have robots building robots building robots, and this nested layer of technology uh is nearly upon us, and so that in that application for manufacturing industrialization um plants
And factories and that entire area needs this new AI called physical AI. If we if we can solve that, we're talking about trillions and trillions of dollars of industries.
So, last question, we might have a lot of people after listening thinking, how do I get a job at Nvidia? Okay. What are the skill sets you're looking for today? You have in high demand?
Well, if you can't design a chip, if you tell me if you tell me that you you learn how to design chips on YouTube, I think that's going to that's going to tell me a lot. And so, no, no, I look, you know, we're we're um Nvidia is a is is a is the world's first chip company. Um, of course, that we, as I mentioned, we're building the entire AR infrastructure stack, but we have we have uh uh we have uh digital biologists, we have quantum chemists, uh we have computer graphics engineers, we have roboticists, we have uh language experts. Um we have we have expertise across a very large domain of science and we have a very large domain of industries and so uh you know we serve healthcare industry, we serve financial services industry. So if you have domain expertise, we love that. We love people with domain expertise. And so um we also love people with general b general basic intelligence. And if you love hard work and especially if you love to suffer uh you know you know you know exactly who to call.
You know, we have seen over the years that those—I don't know if my my microphone is working or not here. Uh, can you hear me? We have a famous professor, a friend of mine, tells me when he gets a new class, he tries to figure out who got in because of their ability and who got in because of their family relationships. And that is—There we go. Okay. So I was saying, when you talk about what built Nvidia, hard work, a lot of challenges, tough days. I was saying this professor friend of mine points out that he tries to figure out when the class comes in who got in on their ability and who got in because of relationships. That's easy, he tells me within a week. What's harder is how long will it be before the person that got in on the relationships is working for the person with ability. That takes a long time. He needs to get to know the students. Those hard days have paid off, and we look forward to seeing what you can accomplish in the future. Thank you for joining. Thank you very much, Mike.