저자: JensenHuang · 외국인
요약 보기
AI는 에너지·칩·데이터센터·앱 등 5계층 스택으로 봐야 하며, 각국은 자국 산업에 AI를 도입하는 인프라를 구축해야 한다. 젠슨 황은 AI가 소프트웨어를 넘어 로봇·자율주행 등 물리 세계로 확장된다고 주장하고, 안전 논의는 필요하지만 과도한 규제보다 실질적 피해 규제와 빠른 발전이 낫다고 강조했다. 일자리는 대체되기보다 역할이 바뀌며 AGI 근접 전망과 함께 1GW 데이터센터 투자(약 500~600억 달러), 100GW 목표 등 대규모 투자 확대를 제시했다. 엔비디아 GPU는 모든 모델을 지원하는 범용 아키텍처라 투자 내구성이 높다고 주장했다.
다듬은 글
structure. The laws of physics have structure. So, each region in the world specializes in different areas of AI, and AI encompasses all different languages, technologies, and information theories. So AI permeates all of this, and there are many models of AI. But the most important part of AI is the fifth layer, the one above this, namely—data and applications.
To benefit, you know, the United States is investing more deeply in the invention, creation, and development of this whole five-tier pie. However, each region, each country, must decide which of these levels they want to invest in. You don't have to win every one of them. You don't have to develop all of them.
Most importantly, each country should advocate for the spread or implementation of AI in its own industries. And the industries we're talking about encompass everything from education and healthcare to manufacturing and science, basically every industry in between. So, the way to think about AI is to simply think of it as a five-layer stack.
The United States will be a leader in AI development at all five levels, but we want to collaborate with everyone so that everyone can reap these benefits. Nvidia has built everything to work for every model, every field of science, and every modality. So, it launches American models, international models, models in biology, chemistry, physics, and robotics.
So when I go to dinner, Yes. and I say "AI", everyone thinks it's software. Yes. But what we're seeing now is really AI on the periphery. This is hardware, robotics, advanced manufacturing . So how do you see AI and the difference between the physical infrastructure of factories and the general perception that AI is software?
Well, AI is intelligence, and it starts with software. And actually, if you look at AI over the last year, we started by talking about large language models, these LLMs. You know, the fact that the world now knows what an LLM is is a pretty amazing thing. So, big language models. What made AI useful was creating an exoskeleton around LLM.
This exoskeleton is called an agent harness. It essentially gives the “brain” the components needed to acquire knowledge, work with memory, use tools, collaborate, and so on. So, solve problems. So, first of all, the digital version of AI is what's called an agent. When you take this digital version, you can make it use not a program like spreadsheets, PowerPoint, Excel, or web browsers, but a tool, essentially a mechanism.
And then this agent would be embodied in a physical body, and suddenly it becomes a robot. You can take this agent and put it in a four- wheeled vehicle—a self-driving car. You can put this agent in a manipulator and it will become an assembly manipulator or an industrial robot. You could put this agent in something that is essentially a food delivery vehicle, a logistics vehicle , a surgical robot, or an autonomous drug development lab.
You know, all these different versions are essentially the same idea. Large language models with an agent system around them; you can use this agent system to work with both digital and physical tools. Ultimately, every country must understand that this is the same infrastructure as water, roads, electricity, or the internet.
Every country needs to build infrastructure to support its own local economy. This becomes a capacity, an intellectual capacity, a digital intellectual capacity to support your researchers, students, society, industries and startups. And every country we've worked with, and there are many that we've already worked with, as soon as we create infrastructure in place, startups and researchers immediately step up.
So this is a really fantastic investment. We in the United States have had the advantage of building some of the largest infrastructure, and thanks to the energy growth stimulus, the speed, and the regulatory environment in the United States, we have been able to build it out quite significantly. And if you look at what's happening in the United States, AI is now creating jobs in every sector.
From chip factories to computer factories to AI factories, the number of jobs we are creating is in the hundreds of thousands. And, as you know, this year alone, we're going to invest almost a trillion dollars in the infrastructure of the United States. And this generates huge economic growth and, of course, many jobs.
So change makes people nervous. Yes. You know, it makes different people say all sorts of different things. You know, these changes brought about by AI, how do you rate them compared to other changes since the Industrial Revolution? Do you know what place you would give them? Are they small? Do you think they are big?
And what is the best way for our friends here to embrace these changes? Well, as with all new technologies, it seems magical at first. Just imagine, at the beginning of electricity, it seemed like you could be somewhere very far away, and suddenly the light turned on. It seems like magic. And at the beginning of any technological innovation, everything seems like magic.
But in reality, all of us who create this are engaged in engineering. It involves computers, software, and mathematics. There are probabilities here, and in general, there are a lot of things involved here. As in any technological field , when we create something, whether it's a car or an airplane, developers are responsible for ensuring safety and a responsible approach, as well as for improving the technology so that it becomes safer as it develops.
I prefer a car from today with a bunch of modern technology to a car from a century ago . I prefer to fly a modern plane with all the technological advances rather than a plane from a century ago , so in many ways we want to accelerate the development of technology so that it performs its functions better, becomes safer to use, and the same is happening with AI.
We need to remind the industry that discussions about safety and security are great. However, if you think about each of your countries, what is the worst outcome that this technological industrial revolution could bring ? The worst outcome is not taking advantage of it. It's to be left behind. This is the single worst outcome.
And such an outcome is possible if we talk about technology with fear, with uncertainty, if management talks about it with excessive concern, fear, and focuses only on conversations about safety. This needs to be balanced by conversations about prosperity and what it is capable of. So I don't think it's a choice between one or the other.
We just have to make sure we're balanced when we talk about it. You probably know that in the history of all technological advancements, one of the stories I read about recently has to do with the aviation industry. There was a time when they all spent, you know, a lot of their advertising budgets and marketing campaigns on the idea that " my plane is safer than everyone else's." It turned out that no one wanted to hear about it.
They wanted companies to take care of it themselves. What they wanted to hear was that airlines could take them to new places, expand horizons, make the world smaller so we could learn about each other's cultures, create new economies. This is what they wanted to hear about. They wanted it to be our responsibility to make everything safe, not a burden for them to think about safety all day long.
And so, I think the same thing is happening in the automotive industry, in the energy industry, it happened in the aviation industry . So, I think it's the responsibility of all of our leaders, all of us, to make sure that we take responsibility for safety and security in building technology properly.
And cooperating with regulators, and of course, there are already so many regulations and laws. Intelligence is the digital version of what we do, so you can think of it as the digital version of certain tasks that humans do. In fact, there are already regulations for people in many different areas, whether it's the FDA or the NHTSA or whatever.
There are so many different agencies that already have their own regulations. I'm sure there's a way to integrate artificial intelligence into this. Therefore, I believe that the balance of fine-tuning regulation is not to regulate hypothetical or theoretical harm, but to regulate real and practical harm.
And to give this technology, which is in the early stages of the innovation S-curve, the opportunity to evolve quickly so that it can become safe. Some people , if you look back and think about the idea of slowing down technological progress three years ago, when everything was just taking shape...
In fact, the progress made it safer. This progress forced her to not hallucinate, to base herself on real truth, and to iterate until the answer was scientifically sound. All of these technologies have been made possible by technological advances. So we make it safer by continuing to develop it further.
So, let's move on to your business for a moment. Um, your chips , your GPUs, they're different from what others hope to create. Isn't that right ? So can you tell the audience about your products and why GPU is different from any other letters before PU? M-hm. If you wish. Well, we invented a technology called GPU.
Essentially, a GPU performs parallel computing tasks, solving problems in fluid dynamics, particle physics, quantum chemistry, computer graphics, image processing and reconstruction, and, of course, it does an incredibly good job with artificial intelligence. Our architecture and technology are universal, meaning we support any closed models, we support every open model.
We support languages, large models, small models, we support physical and biological models. So our architecture is interchangeable in the sense that it runs any AI model, performs all data processing and parallel tasks. As a result, we are everywhere. We are in every cloud, on local servers, at the edge, in robotic systems and driverless cars.
The company has been around for a long time, we have been working on this for a long time. We are already on the 14th generation of our architecture, and so Nvidia is now virtually everywhere. One of the advantages of this is that such infrastructure is very expensive. 1 gigawatt costs about 50-60 billion dollars.
Yes. You know, previously, building a $ 25 billion plant was considered a gigantic investment. And now 1 gigawatt costs twice as much. And we're building 100 gigawatts by the end of the decade. Because we want to generate intelligence for the whole world. So when investing in such expensive infrastructure, you want to be sure that it is interchangeable and durable.
Because if you build something too specialized, and the architecture and algorithms in the world change, and AI models and technologies evolve very quickly, then your investment can instantly depreciate. Therefore, Nvidia's growth is accelerating, as people are starting to understand that we are the safest architecture for investment.
It is interchangeable, durable, constantly improved through software, and runs everything in a row. We launch every American model, every international model, and every modality of models that exists. So people are always interested in where the world is heading. If you could give us your vision of the big picture, where, where are we headed?
What should we think about that we don't always talk about ? And you see everything, really, all the companies in the world come to you and tell you where they are going. How could you describe to us the big picture of ideas that we don't usually hear about? Well, there are things we hear, and I'll compare them to what I expect.
For example, in the next couple of years we will achieve what people call AGI. And that doesn't seem illogical to me. In fact, I would say we're almost there today. Of course? So the question is, what does this mean? It either means a lot or it means nothing . Let me give you an example. If we say that in the next few years we will achieve artificial general intelligence, there is no doubt that this will affect everyone's tasks, what we do—whether it is learning, or work, or any other field.
This will benefit everyone. However, what was not well thought out was the thought experiment itself. Let me just suggest this experiment to you. Let's assume, and you know, that a few hundred years ago, people, in order to have a more civilized society, created one of the most important things in the world—education, intelligence.
We produce intelligence and ensure stability on a massive scale through universities and schools. How would we have civilization without it? And now we're creating a digital version of this. We are digitally producing intelligence at scale, so that even countries and people who don't have access to the highest levels of education can now reap its benefits.
Of course? This is the easiest way to make sense of what is happening now. So the question is, what is the benefit of this to society? Let's do a thought experiment. Let's assume that every young professional we hire after graduation is incredibly well-educated. Well, that's pretty much the same as the ones I hire.
They come out of Stanford, Harvard, MIT or, say, ETH, and they are incredibly smart young people. However, when these kids come to our company, any company, any industry, we still have to invest a lot in providing them with context. So it turns out that even when we achieve AI of the general level (AGI), it doesn't mean that every company's problems will be solved.
It won't be like this, that suddenly in 2 years I'll say: "Okay, I'll connect this to this AI service, sit back , and my company will instantly become more productive ." This simply won't happen. You're going to spend all your energy, as you are now, on onboarding a new MIT PhD, which means surrounding it with context, purpose, relevance, access to all these other things, which is essentially a framework for this AI, an environment around it, so that it can do productive work.
All this framework creation, all this environment creation —that's what companies do, that's what leaders do. As it turns out, our work won't change. Our work is defined by purpose, context, and content. This is work. This job involves a lot of typing, talking, and, you know, things like that. These skills will be automated by AI.
Tasks will be automated, but the purpose of our work will remain. So I think what's going to happen is that we're all just going to get a " supercharge," just like we're getting right now. At Nvidia, we're using AI across the company today. We use off-the-shelf AIs like Anthropic, OpenAI, Cursor, and we build a lot of our own specialized AIs, and I encourage everyone to take the same approach.
You should use off-the-shelf AI wherever you can, but don't forget that you should invest in building your own intelligence. Every country, every company. You can't outsource all your intelligence to someone else. So, you should use as much as you can and build as much as you need. So we do this in the company, and that's the first thought: we'll just be much more efficient.
The idea that all jobs will be eliminated or eliminated is simply nonsense. Of course? And the reason for this is because of everything I just said. Now the other thing that is a fact is that all our ambitions will be reinforced. Today we are more ambitious than at any time in the history of human civilization .
We conceive things that people 100 years ago would have called crazy. That would be science fiction. And yet, for us, this is completely understandable and expected. So imagine how ambitious we will be in 5–10 years. This is an exciting time. You know, I've been working for a long time, and I do n't have to work anymore, but I don't want to miss—I definitely won't miss the next 10–15 years.
The ambitions of the industry, the ambitions of my colleagues, and my own ambitions have truly received a powerful boost. This is another way of saying that all of you, all countries, should be similarly ambitious . This is a time when you should feel enlightened, inspired, energized, and uplifted, because now you have incredible technology behind you.
So, you know what I expect in the next 5–10 years? What took 10 years will take one. What took a year will take a month. And you know, if you think it's going to cost a million dollars—well, not so long ago, just a few years ago, a million seemed like a big investment. But today, think about the level of investment we're talking about— trillions of dollars.
And the reason for this is this . Now, ordinary people like you and me can think about influencing an industry worth hundreds of trillions of dollars. This is the first time you're hearing from people, right? For the first time, you hear innovators, people, companies, and countries talking about the rise of a hundred trillion dollar industry.
That's what ambition is. To think that you would add 20, 50 trillion dollars of economic benefit worldwide. That's what ambition is. Never before in history have we been able to talk about something like this. So my goal today and our shared goal is growth. Isn't that right ? This is our goal. So that we can all grow , benefit our countries and society as a whole.
And I think what you did, Jensen, we really appreciate. I mean, you're an American treasure. Thank you. Is n't that right? Your company is leading us down a path that none of us even dreamed of in our youth. Maybe we dreamed about it a little. This is something that was kind of science fiction, but now we're here.
I think what you've outlined for us is a vision that using AI, both by contextualizing it and directing it, is how we're going to succeed. That each of the countries here should invest in their infrastructure to get this for themselves, to make sure that their citizens and society benefit from it . They need to make these infrastructure investments.
They should build data centers. They must use the capabilities of AI themselves. This is true. Because you said that education has become democratized. It has become accessible to everyone. Now you can have a PhD in anything. This is true. At hand, and you don't have to look for such a person because you helped find him or her and, in essence, absorbed this education into a single source.
Now you are that person. True. Now everyone is that kind of person. So, I think, you know, our goal is to have this idea of growth, this ambition for growth. And what you did—you showed us that our dreams should be higher. Because they can be higher. Because if you have all this intelligence at your fingertips, how can you not dream of more?
This is true. And I think for the company and its leader to have such an extremely positive vision for our future was the best way to start today. So I want to thank you for coming. Friends, I would like to thank Jensen Huang for coming and joining us. Thank you all.
Comments 0