黃仁勳APEC演說：AI不只是ChatGPT而已！產業進入獲利正向循環！｜下一代實體AI競賽 南韓是少數具備成功關鍵的國家？｜分享AI新知識 一定要看！ #中英對照 #英語聽力

Nice to see all of you. Nice to see all of you. Kombu chicken is very good. I enjoyed Chimek with my friends. And it is the best way to enjoy Korea. I enjoyed it very much. Thank you for the opportunity to address you today. And first of all, congratulations on such an incredibly successful APEC. It tells you something about the ability for Korea to attract, to unite, and of course, Chairman Tony Hsieh's incredible work to organize what is a historic APEC. And so I want to congratulate you for that. The industry, as you know, the computer industry that NVIDIA is part of is going through a fundamental transformation. And every single industry, because computers affect and impact every single industry, it is the single most important instrument of humanity today, every single industry is being impacted as a result of a platform shift we call artificial intelligence. Our journey started over 30 years ago to invent a new way of doing computing. The computing approach that we created and invented is called accelerated computing. We acknowledged, we recognized years ago to create a new computing model to solve problems normal computers cannot. It has another benefit that when Moore's Law runs out of steam, it could be the path forward for the computer industry. 33 years later, as we continue to be able to create more transistors with every generation of process technology, and even though transistors' performance and power benefits have slowed tremendously, we can now continue to expand computing for as far as we can see because of accelerated computing. We are now seeing a platform shift. This platform shift was the reason for creating the GPU and inventing the model we call CUDA. However, a GPU alone is just a chip. This new computing approach requires tremendous amounts of algorithm software to be developed. We call it CUDAx libraries. Very few people talk about this, but this is NVIDIA's treasure. These libraries make it possible for us to apply accelerated computing to computational lithography, the work that we do with Samsung, or in deep learning and robotics, the work that we do with many companies here that I'll talk about. It allows us to do quantum physics, it allows us to do quantum computing, molecular dynamics, fluid dynamics, robotics, of course, artificial intelligence. These libraries open new markets for our company. These libraries are precisely the reason why NVIDIA's computing approach is now available all over the world in so many different industries. But it started from a very humble beginning. It has been 33 years. In fact, NVIDIA came to Korea 30 years ago. We invented the video game industry together. I introduced GeForce here to Korea 25 years ago. And last night, we had a big celebration with the esports gamers and PC gamers here in Korea. It was incredibly fun to celebrate GeForce 25, 25 years of GeForce. It is incredible how far technology has come. And what you saw just now is fundamentally a new computing approach. One of them is called accelerated computing, but you are also seeing artificial intelligence. Artificial intelligence is, of course, affecting every single industry. And most people think artificial intelligence is a chatbot, as they should. GPT, chat GPT revolutionized artificial intelligence and put it into the hands of everybody. However, artificial intelligence has many things. And I'm going to tell you what it means to you, what it means to the industry that we're in, so that you have a feeling for how you can engage artificial intelligence. First of all, artificial intelligence is a whole new computing platform. The old way of doing software with handwritten software, hand coding to new way of doing software, where the computer learns by itself, where we train the computer, we train the model for it to perform the task we want it to perform. Artificial intelligence and AI runs on GPUs, whereas hand-coded software runs on CPUs. This entire software stack, from the needs of energy, chips, the infrastructure, all of the software associated with the systems, the AI models and the applications on top, every single layer of computing has been fundamentally changed. Just think, the computer industry has been largely the same for 60 years. And now, with AI and accelerated computing, every single layer of the computing stack is being changed. All of the computers we've created in the past, a trillion dollars, maybe more of computers, needs to now be transitioned, shifted to the new computing platform. The second thing to know about AI is that, although Transformers is incredibly successful and very important, there are many AI models. There's AI models for languages, but remember, there's all kinds of information we want to use AI to solve. It could be AI for chemicals, proteins, physics, quantum physics, different fields of science, even AI for robotics, AI that understands the physical world. AI is quite diverse. The number of algorithms necessary to encompass all of the world's different AIs is incredibly vast. ChatGPT and ChatBots is one of them. But the beautiful thing is, with all these different AI models, we can therefore revolutionize many applications in many industries, which is the reason why we are seeing AI affecting literally every single industry in the world. There are no industries where intelligence is fundamentally not necessary. The observation, and this is a very important idea, that the technology of the past, software of the past, is a tool that humans use. Excel is a tool, web browsers are tools, PowerPoints are tools, these are very important tools, but these are tools used by humans. It's no different than a car that is used by humans, it's no different than screwdrivers and hammers that are tools used by humans. But AI is work. AI performs work, which is the reason why everybody realized that AI has the ability to address the largest technology industry in history. The tools industry, the IT industry we know of the past, is a few trillion dollars. AI can address 100 trillion dollars of the world's industries, make it more productive, help the GDP grow. And then one more idea. Artificial intelligence, unlike software, artificial intelligence needs a factory. This is the first technology that requires, that transforms energy, and on a continuous basis, applying a computer, computers that NVIDIA builds, these GPU AI supercomputers, to generate intelligence for everybody to use. Each single token is generated based on the context of the time and based on the questions that you have for the AI. What that means is, the world is going to have an enormous number of these AI factories. And because AI is going to be integrated into every single application, into every industry, AI infrastructure will be built by every single country. Just like electricity, just like internet, and now artificial intelligence, AI factories will be built all over the world. It is one of the reasons why this is so consequential to every single country. It is the reason why it is so consequential to the conversation we're having today. AI is an industrial revolution. It changes the computing software stack, the software is work, not tools, it addresses extremely large industries, and it requires factors. Well, in the last couple of years, we've seen extraordinary progress. And this is the system that makes it possible. This is the latest NVIDIA system called Grace Blackwell. This is one giant GPU. It weighs two tons, has one and a half million parts, it consumes 120,000 watts, 120 kilowatts, and it produces tokens, or produces intelligence at an incredible speed. This is what a GPU looks like today. People think that NVIDIA GPUs are like a gaming GPU, but these GPUs are really massive. Entire supercomputers interact. Probably the question that most people ask me is what changed this year? You might have seen that although AI has been growing quickly, NVIDIA has been growing quickly, but in the last six months, it seems the growth has really accelerated quite substantially. There's a reason for that. In the last couple of years, AI has made tremendous progress because we can now reason and think and solve problems that we have never been trained before to solve. AI now has what is called three scaling laws. We teach it like a child to remember, memorize information. We also teach it called post-training, how to perform new skills, how to reason and how to think. And then during inference time, the AIs today don't just respond with memorized answers, but it responds by thinking first. The quality of the answers has improved tremendously in this last year. As a result, more people are using it because the answers are better. You can use it to solve more problems. When the answers are better, more people use it. When more people use it, the amount of computation necessary continues to grow. So the way you teach an AI, the computation is quite substantial. The way you use an AI today, the amount of computation is quite substantial. But the important thing is this. Because these tokens, these intelligence that are being produced are so good now, customers are paying for it. The tokens have become profitable tokens. This year, AI became profitable. And when something becomes profitable, you want to manufacture more of it. Just like when you're manufacturing chips and wafers and DRAM, if the manufacturing of those chips were profitable, you want to build more factories to create more chips. Now that the AI is profitable, we want to create more factories to generate more AI. We have now achieved what is called a virtual cycle. The AIs get better, more people use it, more people use it, makes more profit, create more factories, which allows us to create even better AIs, which allows more people to use it. The virtual cycle of AI has arrived. And this is the reason why you're seeing the world's CapEx going so fast. There are two fundamental things that are happening. The first is a transition, as I mentioned, from general-purpose computing to accelerated computing. Moore's Law has really run out of steam, so we need another way of doing computation. NVIDIA's accelerated computing is the foundation layer for that. On top of accelerated computing, AI has now achieved the virtual cycle, which is the reason why all of a sudden we're seeing the amount of CapEx that companies are building is growing so substantially. This is going to happen here in Korea as well. Well, the reason why I think this is such an extraordinary time for Korea is, of course, during the platform transition, it's vital to any technology industry, it's vital to any industrial nation. However, for Korea, this is an extraordinary opportunity because in very few countries in the world do you have three fundamental, essential, necessary qualities in order to succeed. The first is you need software. Software expertise is one of those skills that Korea has abundance of. Deep technical and science capability. And third, manufacturing capabilities. When you combine software and AI, technology and manufacturing, you have the opportunity to really take advantage of robotics. This is the next generation of physical AI. And you can see NVIDIA working in the robotics field all over the place. This is human robotics that we're doing with Figure. This is the work that we're doing with Caterpillar, turning themselves into a robotics industrial company. This is agility, logistics, robots. This is Johnson & Johnson, surgical robots. But the technology has now reached a level where we are starting to see robotic systems all over the place. We're also seeing entire factories become robots. This is a factory that NVIDIA is building in the United States for our own AI supercomputers. And we're working with technology companies throughout the world in order to create a robotic factory. The entire factory is a giant robot. And inside this robotic factory, it will orchestrate robots that are working together with humans, with all of us. And they're going to produce products like self-driving cars that are robots. So robots orchestrating robots, building products that are robotic. This is the future of artificial intelligence. And I think that this is an area that is just so incredible for Korea because you have all of the capabilities. You have the deep technical capabilities. You have software capability and artificial intelligence capability. And you also have such an incredible manufacturing base and industrial base. Well, this week, we're announcing several really terrific partnerships, extension of our partnerships. I've been working here for so many years. We have so many different friends. Naver, for example, was one of our earliest partners. And they partnered with us to create the world's third large language model. And that large language model was Korean. And they were one of the first customers of ours in building a supercomputer here in Korea. This week, we announced that Naver, NVIDIA will expand our GPU infrastructure here in Korea by 60,000 GPUs. We're working with Samsung in manufacturing, of course. They are the first partner NVIDIA had in building memory technology to help me invent the AI supercomputer that is now powering the world's AI. We manufacture HBMs here. We, of course, partnered with Samsung on manufacturing chips. But now we're going to advance AI together, build AI factory together, build digital twins for the factories together. Expansion of our partnership tremendously. We're going to build 50,000 GPU AI factories with Samsung. Working with SK, the SK group, of course, with HBM memory, but also in creating digital twins of their factories. And we're going to build 50,000 GPU AI factories with SK group for their AI factories. And then Honda, we partnered with them to build autonomous vehicles. But their factory will become digital twins. And inside these factories will become robotic factories, orchestrating robots, building cars that are robotic. And they're building 50,000 GPUs for their AI factory. Total, this week we announced that we will build 250,000 more GPUs here in Korea, making Korea one of the largest GPU or AI infrastructure countries in the world. And then finally, it's really vital that we build the ecosystem, not just the AI infrastructure of Korea, but the ecosystem of Korea. So that we partner with AI researchers here with Korean universities, amazing universities like KAIST, all the startups that are here, to create an environment necessary for them to benefit from AI. So that there's a rich ecosystem of AI startups here, working with the government, working with education, working with the established companies. This is an extraordinary time. I want to thank all of you for welcoming NVIDIA with open arms over the last 30 years. This is the beginning of a new journey. You were part of my journey as we were building NVIDIA. I'm honored and grateful that we're going to be part of your journey as Korea now becomes an AI nation. And you pursue your endeavors to be part of the AI frontier. Thank you very much.