The weather is good today The weather is good today.
Author: Bu Shuqing

Regarding the future direction of AI development, $NVIDIA (NVDA.US)$ CEO Jensen Huang has made a new determination. He mentioned two AI trends at the earlier Nvidia Japan summit:

The scalability of AI and multimodal capabilities. This means that AI can process and understand various types of data, such as text, speech, images, and videos, and correlate these data types to function across multiple applications.

The "Cambrian explosion" of AI applications. The explosive growth of AI applications will create many new industries and companies. Jensen Huang compared this period to the "Cambrian explosion," a geological period marked by a dramatic increase in biodiversity.

He emphasized that although chips are one of the core components of AI systems, the real value and potential of AI lie in the comprehensive capabilities of the entire system and its wide range of application prospects.

Jensen Huang believes that there will be two types of AI that will be very popular in the future: digital AI workers (AI agents) and physical AI (robotics). Digital AI workers can perform various tasks, such as marketing and customer support, operating like digital employees. Physical AI is embodied in mechanical systems, such as autonomous vehicles and industrial robots, which can perform complex tasks in the real world.

Jensen Huang views Nvidia as a simulation technology company focused on simulating physics, virtual worlds, and intelligence, helping to predict the future through simulation, more like creating a time machine. He said that Nvidia's future strategy focuses on AI technology and ecosystem collaboration, accelerating computing and strengthening CPU with GPU, while providing rich software and library support for AI development.

He also mentioned that a new industry—artificial intelligence manufacturing—is emerging, where every company will become an AI manufacturer.

Highlights of Jensen Huang's speech:

• Essentially, Nvidia is a simulation technology company. We simulate physics, virtual worlds, and intelligence. Through our simulation technology, we help you predict the future. Therefore, in many ways, Nvidia is like a company that has built a time machine.

• Nvidia invented accelerated computing but did not replace CPUs. In fact, we are almost the only company in the computing field that does not want to replace CPUs. Our goal is to release the capabilities of CPUs by offloading compute-intensive workloads to GPUs.

• In the past decade, we have scaled up artificial intelligence and machine learning by a million times. By increasing the scale of machine learning by a million times, we achieved a huge breakthrough, and it is this breakthrough that led to the emergence of today's ChatGPT—the arrival of artificial intelligence.

• Software 1.0 is the code written to run on the CPU. Now we have entered the era of Software 2.0, because computer speeds are very fast, and you can provide it with a large amount of sample data for it to learn and predict functions on its own. We call this Software 2.0.

• The current software is no longer hand-written code, but neural networks running on GPUs. These neural networks running on GPUs are forming a new operating system, a new way of using computers, which is the operating system of modern computers, especially large language models.

• This machine learning method has proven to have incredible scalability. You can use it for a variety of things, including digitizing text, sound, speech, images, and videos. It can be multimodal. You can teach it amino acid sequences and enable it to understand anything with a large amount of observational data. Now, AI applications are exploding at a Cambrian explosion pace, and we have just begun.

• AI is not just a chip problem. (GPU) systems cannot work alone. Even the most advanced computers in the world cannot work for AI alone. Sometimes it must work with thousands of other computers, functioning together like a single computer. Sometimes they must work separately because they are responding to different customers and queries, sometimes it is individually, and sometimes it is as a whole.

• I believe there are two types of AI that will be very popular. One is digitized, which we call AI agents that you can use in the office to collaborate with your employees. The second is physical AI system robots. These physical AIs will be the products built by companies. Therefore, companies will use AI to enhance employee productivity, and we will use AI to drive and empower the products we sell. In the future, auto manufacturers will have two factories, one for making autos and the other for producing AI that operates within those autos.

• Now we have a new industry (AI manufacturing) that has never existed before—AI is at the top of the computer industry, but it is utilized and created by every industry. Every industry, every company, and every country must produce its own AI; this is a new industrial revolution.

• To achieve robotics, we need to build three computers. The first computer trains the AI, just as all the examples given to you before. The first is to simulate AI. You need to provide the AI with a place to practice, a place to learn, a retreat area to receive synthetic data that it can learn from. The Omniverse platform enables you to create AI. Ultimately, what you want is AI. Ultimately, the AI you expect will see a world where it can recognize videos, the surrounding environment, and your needs, and generate corresponding actions.

The full text of Huang Renxun's speech (AI translation) is as follows:

Welcome to the nvidia ai summit. Everything you just saw was simulated. There were no animations.

Essentially, nvidia is a simulation technology company. We simulate physics, virtual worlds, and intelligence. Through our simulation technology, we help you predict the future. Therefore, in many ways, nvidia is like a time machine.

Today, we will share some of our latest breakthroughs with you. But most importantly, this is an event about the japan ecosystem. We have many partners here, including 350 startups, 250,000 developers, and hundreds of companies. We have been here for a long time.

Since the company's inception, the japan market has been very important to us. In japan, we have made many 'firsts'. The first game developer to collaborate with us was yu suzuki from sega, who is a well-known 3D game developer, and he first worked with us to port sega's amazing 3D games to nvidia's graphics processors. Tokyo institute of technology was the first to use nvidia CUDA to build the supercomputer subamer 1.2, allowing us to leverage our graphics processors to drive scientific computing. Japan has been first in many ways. This also marked the first time we were able to create mobile processors, which gave birth to one of our very important projects — the nintendo switch. So many 'firsts'.

We are now at the beginning of a new era, the ai revolution, a new industry, and extraordinary technological shifts. This is a very exciting moment, but also a very critical one. Therefore, we are here to collaborate with outstanding companies in the japan ecosystem to bring ai to japan so that we can fully capitalize on this extraordinary opportunity before us.

Today, we have many partners here, and I want to thank platinum sponsors such as gmo internet group, hp inc, microsoft azure, and mitsui group. I would like to thank all of you.

There are also 56 other sponsors. Thank you all for being here, and thank you for your support. nvidia invented accelerated computing but did not replace the cpu. In fact, we are almost the only company in the computing field that does not want to replace the cpu. Our goal is to free up cpu capacity by offloading compute-intensive workloads to gpus.

This is the gpu, and this is the cpu. By combining the two, we can leverage the best capabilities of both processors: cpus are very good at sequential processing, while graphics processors are very good at parallel processing. I will discuss this in more detail later, but this is what accelerated computing is about — it’s not just parallel computing, but rather the collaboration between cpus and gpus.

这种计算模型对世界来说是全新的。事实上，CPU自1964年开始就已经存在了，也就是我出生的第二年，至今已有60年。我们今天在计算机上看到的绝大多数运行在CPU上的软件。但现在有一个新的变化，计算模型正在发生根本性变化。然而，为了实现这一点，你不能仅仅将顺序运行的CPU软件放到GPU上并行运行。我们必须创建一大堆新的算法，就像OpenGL使计算机图形应用程序能够通过图形处理器连接到加速一样，我们必须为许多不同的应用程序创建许多特定领域的库。这些是我们公司拥有的350个不同库中的一些，非常重要的库。

Kulethos是一个用来加速计算光刻的工具，这是芯片制造过程中的一个步骤。计算光刻是一个复杂的过程，通常需要数周时间来计算许多层的图案。但使用Kulethos后，这个时间可以缩短到几个小时。

当然，我们能够缩短芯片制造的周期，但同样重要的是，我们能够让光刻技术的算法变得更加复杂，这意味着我们可以推动半导体技术达到更高的精度，比如2纳米、1纳米甚至更小的尺度。因此，计算光刻的过程将通过Kulethos和Spark Solver的DSSAI Ariel技术得到加速。我今天将会详细讨论这个话题。

这个新开发的库非常了不起，它使计算机能够运行5G无线电的技术栈。简单来说，就是可以在Nvidia的CUDA加速器上实时操作一个无线电。此外，CUDA也被用于量子模拟，比如模拟量子电路。还有用于基因测序的配对技术，以及KUV技术，这是一种用于存储向量数据的技术，也用于索引和查询向量数据库，这些数据库在人工智能领域特别有用。

NumPy是一个用于数值计算的库。它是世界上最流行的数值处理库之一，大约有五百万不同的开发者在使用。这个库非常受欢迎，仅在上个月就达到了30 million次的下载量，这是一个惊人的数字。现在，NumPy已经完全支持在多个GPU和多个计算节点上进行加速计算，这使得它在处理大规模数据时更加强大。建议你去了解一下这个库，它的强大功能确实令人难以置信。

QDF是一个用于处理数据帧和结构化数据的库，它支持像SQL、Pandas、Polars等数据处理技术，以及解决复杂的旅行商问题（TSP，一种组合优化问题）。这个库极大地加速了这些问题的解决，速度提升了数百倍。

接下来提到的是KUDNN，这是我们创建的最重要的库之一，全称是深度神经网络的qdnn。这个库负责处理深度学习模型中不同层次的数据。通过创建qdnn并推动深度学习的普及和加速，在过去的十年里，我们让人工智能和机器学习的规模提高了 1 million倍。通过把机器学习的规模提升1 million倍，我们实现了一个巨大的突破，也正是这个突破，催生了如今的 ChatGPT——人工智能的到来。简而言之，KUDNN库对于推动人工智能的发展起到了关键作用。

Qdnn做了一些特别的事情，它改变了我们编写和使用软件的方式。软件 1.0 就是编写在 CPU 上运行的代码。现在我们进入了软件2.0时代，因为计算机速度已经非常快，你可以给它提供大量的样本数据，让它自己学习并预测函数。我们称之为软件 2.0。这样它就能自我学习并预测函数是什么，这就是机器学习。

So, the current software is no longer handwritten code, but neural networks running on GPUs. These neural networks running on GPUs are forming a new operating system, a new way of using computers, which is the operating system of modern computers, especially large language models.

This machine learning method has proven to have incredible scalability. It can be used for various tasks, including digitizing text, sound, speech, images, and videos. It can be multimodal. You can teach it amino acid sequences and help it understand anything with a large amount of observed data.

The first step to understanding the meaning of data is through studying a vast amount of text on the internet, enabling the understanding of words, vocabulary, grammar, and even the meanings of words by finding patterns and relationships. Using the same approach, we can now understand the meanings of different data types connected to different modalities, for example, the relationship between words and images (e.g., the image of the word 'cat' is now connected to images of cats). By learning multimodal understanding, we can even translate and generate various intelligent information now.

If you observe all those amazing emerging companies and the applications they create, you will find that they can be categorized into two types, displayed on a slide from one side to the other. The first type is text-to-text applications, including text summarization, question-answering systems, text generation, and storytelling. The second type is video-to-text applications, such as generating subtitles for videos. There are also image-to-text applications, such as image recognition, and text-to-image applications, like image generation services such as Mid Journey. Moreover, there is text-to-video creation, like platforms such as Runway ML.

All these different combinations are truly groundbreaking. You can even send text messages about proteins explaining their functions or about chemicals describing the characteristics of a chemical that could potentially become a successful drug. For drug discovery, one can even have applications from video and text to robotics. Each of these combinations represents a new industry, new companies, new application cases. AI applications are now mushrooming at a Cambrian explosion speed, and we're just getting started.

Of course, one feature of machine learning is that the larger the brain, the more data it can be taught, and it becomes smarter. We refer to this as the law of scaling. There is ample evidence that as we scale up the model sizes, the quantity, effectiveness, quality of training data, and the intellectual performance improve year by year. The industry is expanding the size of models approximately twofold, requiring correspondingly twice the amount of data. Hence, we need four times the computational capacity. The amount of computing resources needed to push AI to the next level is extraordinary. We refer to this as the scaling law, the training scaling law. Pre-training is part of it, while post-training involves reinforcement learning, human feedback, reinforcement learning, and AI feedback. Now there are many different methods using synthetic data generation in the post-training phase. Therefore, training, pre-training, and post-training are enjoying significant scaling, and we continue to see excellent results.

Well, when 'Strawberry' or 'OpenAI S01,1' was announced, it demonstrated to the world a new form of reasoning that when interacting with AI is like ChatGPT. But ChatGPT is one-off. You ask a question, and you have it do something for you. Whatever your question is, regardless of the prompt provided in a single shot, it can provide an answer. However, we know that thinking is often not just one-off. Thinking requires us to make multiple plans, consider multiple potential answers, and choose the best one. Just as we think, we may reflect on an answer before giving it. Reflection allows us to break a question down into a step-by-step reasoning chain. We have invented many different techniques, and as we apply more and more computing power, the performance of reasoning improves.

Now we have a second scaling law, the reasoning scaling law, which is not just about generating the next word but thinking, reflecting, and planning. These two coexisting scaling laws will require us to drive computing at ultra-fast speeds. Every time we deliver a new generation and architecture, we enhance performance by a factor of X, but we also reduce power by the same factor of X. We reduce costs by the same factor of X. Therefore, increasing performance is identical to reducing costs. Improving performance is the same as reducing energy consumption. Hence, as the world continues to absorb and embrace AI, our mission is that we have the responsibility to continuously enhance performance as fast as possible. In the process, expanding the reach of AI, improving its effectiveness, reducing costs, and lowering power consumption is why we choose a yearly cycle.

However, AI is not just a matter of chips. These AI systems are massive. This is the Blackwell system, named after a GPU, but it is also the name of the whole system. The GPU itself is extraordinary. There are two Blackwell chips. Each Blackwell chip is the largest chip in the world, with 104 billion transistors, manufactured by taiwan semiconductor at the advanced 4-nanometer node. These two Blackwell chips are interconnected through a low-energy link of 10TB per second. In the middle, that line, that seam, consists of thousands of interconnections between the two chips, at 10TB per second. It is connected by eight HBM3E memories, which together operate at 8TB per second. These two GPUs connect to another low-power, very energy-efficient CPU in the city, at 1TB per second. Each GPU is connected via MVLink at a speed of 1.8TB per second. That is a lot of TB per second. The reason is that this system cannot work independently. Even the world's most advanced computers cannot operate AI on their own. Sometimes it has to work with thousands of other computers, nodes like this working together as a single computer. At times, they must work separately because they are responding to different customers and queries; sometimes they operate individually, sometimes as a whole.

To enable MV and GPUs to work together, we of course have network 2 CX Sevens connecting this GPU with thousands of other GPUs. However, we still need this MV link, allowing us to connect several GPUs to a rack behind me. A rack connects to this MV at 5.8 TB per second. The bandwidth is 35 times that of the highest bandwidth networks in the world, enabling us to connect all these GPUs to this MV link switch.

There are nine MV link switches in a rack. Each rack has 72 such computers. This connects through a spine. This is the MV link spine. These are cables, copper, weighing 50 pounds, driven directly by this incredible Ceres, which we call the MV link. They are connected in this way to the MV link of the computer, this switch connects all these computers together. Therefore, the result is that 72 of these computers connect to form a large GPU, a very large GPU. From a software perspective, it is just a giant ship. These racks, these MBs connect 72 systems; this rack weighs 3000 pounds. It is impossible to get on this stage. Otherwise, I show you, it is 3000 pounds at 120 kilowatts.

That is, I have my friends here; that is the power of many Nintendo Switches. It is not portable, but it is very powerful. This is the black wall system. We designed it so that it can be configured like this as a Superpod, or a giant data center, with tens of thousands, hopefully tens of thousands, linked by switches to it. Some of these are quantum infinite bandwidth switches. If you want a dedicated AI factory or Spectrum X, nvidia's Spectrum X revolutionary Ethernet system, you can integrate it into your existing Ethernet environment. We can build AI supercomputers with these. We can integrate them into corporate datacenters, hyper-scale servers, or configure them for the edge.

The Blackwall system is not only powerful but also highly adaptable, capable of fitting into every corner of the world's computing infrastructure.

Of course, there are computers, but most importantly, this computer cannot operate without all the software running on it. When you see these computers, all the liquid cooling, all the wires, your brain would explode. How do you write such an incredible computer? This is where the nvidia software stack comes in, it is all the effort we put into Cuda Nickel, all our Megatron cores, all the software we have created, Tensor, RTLM, Triton, all the software integrated over the years into the system, making it possible for everyone to deploy AI supercomputers around the world. Of course, most importantly, we have AI software that allows people to easily build AI. So what is AI?

We talk about AI in many different ways, but I believe there are two types of artificial intelligence that will be very popular. I think there are two models that are very helpful.

This is very helpful to me. First, digital AI workers. These AI workers can understand, they can plan, they can take action. Sometimes, digital AI workers are asked to execute marketing campaigns, support customers, plan the manufacturing supply chain, optimize chips, help us write software, maybe as research assistants, laboratory assistants in the drug discovery industry. Perhaps this agent, you know, serves as a mentor to the CEO. Maybe all our employees have a mentor, the AI, these digital AI workers we call AI agents, are essentially like digital employees. Like digital employees, they must be trained, special data must be created to welcome them into the company, teaching them to understand the company. You train them in specific skills, depending on the functions you want them to have. You evaluate them after training to ensure they have learned what they should have. You protect them, ensuring they do the work they are asked to do rather than what they have not been asked to do. Of course, you operate them. You deploy them, giving them energy from Blackwell, providing them AI tokens from Blackwell, they interact with other agents, solving problems as a team.

You will see a variety of agents; we have created something that makes it easier for companies to establish AI agents in the ecosystem. Nvidia does not engage in service business; we do not create, provide end products, or solutions. However, we do provide enabling technology that allows the ecosystem to create AI, deliver AI, and continuously improve AI. The AIAgent lifecycle library, the lifecycle platform is called Nemo. Nemo has libraries for every stage I mentioned, from data management to training to fine-tuning to synthetic data generation to evaluation and to protective railings. There, these libraries are integrated into workflows and frameworks around the world.

We are collaborating with AI startups, service providers like Accenture and Deloitte, and companies worldwide to bring this to all major corporations. We are also partnering with ISVs like Service Now so they can create agents using Service Now. Today, you use services through licensed platforms; your employees interact with the service platform for assistance. In the future, Service Now will also provide a multitude of AI agents that you can rent, essentially digital employees that can help you solve problems. We are currently using services. We are working with SAP, Cadence, ANSYS, companies worldwide, and snowflake companies globally so we can all build agents that help improve your company's productivity.

Now these agents can understand reasoning, plan, and take action. These agents, our collection of AI models or systems, are not just a single AI model but a system of AI models. Nemo helps us build these. We also created pre-trained AI models that we package in what is called NIM. So these NIMs are microservices. They are basically AI from the old era packaged, like software in a box, with a CD-ROM. Today, AI is packaged in a microservice, and the software inside is intelligent. You can talk to the software; you can converse with the software because it understands what you mean, and you can connect software with other software. You can link this AI with other AIs, creating an agent, an AI agent. So that’s the first point.

Let me give you an example of these agents. Agent AI is using complex reasoning and iterative planning to solve complicated multi-step problems, thus transforming every business. AI agents help marketing campaigns launch faster through real-time insights, optimize supply chain operations, save hundreds of millions in costs, and reduce software security processes from days to seconds by assisting analysts in categorizing vulnerabilities. What makes AI so powerful is its ability to transform data into knowledge and knowledge into action. The digital agent in this example can provide insights to individuals using a set of information-intensive research papers. It is built using the Nvidia AI blueprint. These reference workflows include Nvidia accelerated libraries, SDKs, and Nim microservices, which help you quickly build and deploy AI applications. The multimodal PDF data extraction blueprint helps establish data ingestion pipelines, while the digital person blueprint provides seamless human-like interactions. Hi, I’m James, a digital agent ingesting PDF research papers, including complex data like images, charts, and tables, and generating advanced summaries through an interactive digital human-machine interface.

Weather forecasting has made exciting breakthroughs. The development of the new generative model cordiff is an important step in accurately predicting weather patterns. By combining unit regression models with diffusion models, the condensed court.

James can also answer questions based on papers or generate new content. Nvidia AI enables businesses to automate processes, leverage real-time insights, and improve workflow efficiency.

AI agents consist of three parts, the Nemo names and blueprints. These are all references. They are provided to you in source code form so you can use them as you wish and build your AI agent workforce. None of these agents can complete 100% of any person's tasks, any job. No agent can do 100%. However, every agent can do half of your work. This is a great achievement. Instead of thinking of AI as replacing 50% of human jobs, one should think of AI as doing half of the work for 100% of people. Thinking this way, you realize AI will help improve company productivity and enhance your productivity. You know, people ask me, does AI take your job? I always say, and it’s true, AI will not take your job. The AI used by others will take away your work. So be sure to activate AI as soon as possible. The first is digital AI agents; these are digital artificial intelligences. The second application is physical AI. The same basic technology is now embodied in a mechanical system.

Of course, robotics will become one of the most important industries in the world. So far, robotics has been limited, and the reasons are clear. In fact, in Japan, 50% of the world's manufacturing robots are manufactured. Kawasaki, Fanuc, Yaskawa, and Mitsubishi are the four leaders that manufacture half of the robot systems in the world. Just like robotics boosted manufacturing productivity, it has been challenging to scale up. The robotics industry has largely remained stagnant for a long time because it is too specific and not flexible enough to adapt to different scenarios, conditions, and jobs. We need more flexible AI that can adapt and learn on its own.

Please note the technology we have described so far, agent AI, which you should be able to interact with regardless of who you are. It can respond to you. Of course, sometimes the response may not be as good as what you could generate, but in fact, many responses might even be better than what we can produce. Therefore, we can now apply this general AI technology to the world of embodied AI, physical AI, or what is referred to as robotics.

To achieve robotics technology, we need to build three computers. The first computer trains AI, just like all the examples we provided earlier. The first is to simulate AI. You need to give AI a place to practice, a place to learn, a retreat where it can receive synthetic data to learn from. We call it the omniverse, which is our virtual world digital twin library suite used to create physical AIS robots. The omniverse. Then, after validation, training, and evaluation, you can put the model into a physical robot. In this, we have processors specifically designed for robotics technology. We call it Jetson. Thor is a robot processor designed for humanoid robotics.

To achieve robotics technology, we need to build three computers. The first computer trains AI, just like all the examples we provided earlier. The first is to simulate AI. You need to give AI a place to practice, a place to learn, a retreat where it can receive synthetic data to learn from.

This cycle continues indefinitely. Just as there is a NemoAI agent lifecycle platform, the Omniverse platform enables you to create AI. Ultimately, what you want is AI. Ultimately, the AI you expect will see a world where it can recognize videos, its surroundings, and your needs, generating corresponding actions. You tell it what you want, and this AI will produce joint movements. Just as we can extract text, we can generate videos, we can extract text and generate chemicals. For pharmaceuticals, we can extract text and generate joint movements. Well, this concept is very similar to generative AI. This is why we believe that now we have the necessary technology between the Omniverse and all the computers we are building, these three computers, combined with the latest generative AI technology, the era of human or robotics technology has arrived.

Now, why is humanoid robotics technology so difficult? Well, clearly, developing software for humanoid robots is very challenging. However, the benefits are incredible. Only two robotic systems can be easily deployed around the world. The first robotic system is autonomous vehicles because we have created a world adapted to cars. The second is humanoids or robots. These two robotic systems can be deployed anywhere in the world in brownfield areas because we created our world for it. It is a very challenging technology, and the time is ripe, but the impact could be tremendous.

Last week at the robotics learning conference, we announced a very important new framework. It is called the Isaac Lab, which is a reinforcement learning virtual simulation system that allows you to teach humans or robots how to be human or robot. Most importantly, we created several workflows. The first workflow is the Groot Mimic Group.

Mimic is a framework for showing robots how to perform tasks. You use human demonstrations, then simulate that environment using domain randomization to generate hundreds of other examples similar to your demonstration, so that the robot can learn to generalize. Otherwise, it can only perform very specific tasks using mimic, and we can generalize its learning.

The second is the grouping Gen assembly. Using generative AI technology in the Omniverse, we can create a large number of I-random domain environments and random examples of the actions we want the robots to perform. So we are generating a large number of tests, evaluation systems, and assessment scenarios that the robots can try to execute and improve themselves, learning how to become a good robot.

The third one is group control, a model distillation framework that allows us to refine all the skills we have learned into a unified model, enabling the robot to perform kinematic skills. The robots will not only be autonomous, but remember that future factories will also be robots. Thus, these factories will become robotic factories, orchestrating robots and building robotic machinery systems. Let me show you. Of course.

Physical AI embodies robots, such as self-driving cars that safely navigate the real world, robotic arms performing complex industrial tasks, and humanoid robots working alongside us. Factories will be embodied by physical AI, capable of monitoring and adjusting their operations or conversing with us. Nvidia has manufactured three computers that enable developers to create physical AI. Models are first trained on a DGX, and then finely tuned and tested using reinforcement learning and physical feedback in Omniverse. The trained AI runs on Nvidia's Jetson AGX robotic computers. Nvidia Omniverse is a physics-based operating system for physical AI simulations. Robots learn and fine-tune their skills in the Isaac Lab.

The robot gym built on Omniverse features group workflows such as Group Gym, generating diverse learning environments and layouts. Group Mimic generates large-scale synthetic motion datasets based on a small amount of real-world capture and neural whole-body control group control. This is just one robot.

Future factories will coordinate teams of robots and monitor entire operations with thousands of sensors. For the digital twins of the factories, they use a versatile blueprint called Mega. With Mega, the digital twin of the factory is filled with virtual robots whose AI simulates the brain of the robots. The robots perform tasks by perceiving their environment, reasoning, planning the next action, and ultimately converting it into action. The World Simulator in Omniverse simulates these actions in the environment, and the robotic brain simulates perception results through Omniverse sensors. Based on sensor simulations, the robotic brain determines the next move, continuing the cycle, while Mega accurately tracks the status and position of everything in the factory.

Digital twin. This software-in-the-loop testing brings software-defined processes into the physical space and implementation cases, allowing industrial enterprises to simulate and validate comprehensive digital twin changes before deployment to the physical world, thus saving tremendous risks and costs. The era of physical AI has arrived, transforming the heavy industry and robotics of the world.

An incredible era. So we have two robotic systems, one being digital, called AI agents, which can be used in the office to collaborate with your employees. The second is the physical AI system robots. These physical AIs will be the products built by companies. Hence, companies will use AI to enhance employees' productivity, and we will use AI to drive and empower the products we sell. In the future, automotive companies will have two factories: one for manufacturing autos and another for producing the AI that operates within autos.

Well, here comes the robotics revolution. There is so much activity happening all across the globe. It is hard to imagine any country leading the robotics AI revolution better than Japan. The reason is, as you know, this country loves robots. You love robots. You have created some of the best robots in the world. These are the robots we grew up with together. These are the robots we've loved all our lives. I haven't even shown some of my favorites. Majin Gazi, I hope Japan can leverage the latest breakthroughs in AI and combine them with your expertise in large electronics. No other country in the world has higher skills than Japan in super electronic integration. This is an extraordinary opportunity that you must seize. So, I hope we can work together to make this dream a reality.

Nvidia is doing very well with AI in Japan. We have many partners here. We have partners building large language models. Tokyo Institute of Technology, Rakuten. Self-service banks, Intuition, NTT, Fujitsu NEC, Nagoya University, Kota Bar Technologies. If you go to the top right corner, we also have AI cloud, along with SoftBank, Sakura Internet, Transgenic Internet Group. Hi, Rezzo KDDI Rutilia, building AI cloud here to let the ecosystem flourish in Japan.

Therefore, many robotics companies are beginning to understand the capabilities that AI provides to take advantage of this opportunity. Yaskawa, Toyota, Kawasaki, Repute, Reputa, medical imaging systems, Canon, Fujifilm, and Olympus are all utilizing AI. Because in the future, these medical instruments will become even more autonomous. It's almost like having a nurse AI inside the medical instruments that helps nurses guide diagnosis. There are many different ways that AI is used in the drug discovery industry.

So, there is happiness about the progress here, and there is hope to leverage the AI revolution more swiftly. Well, the industry is changing; as I mentioned earlier, the computer industry has fundamentally shifted from coding running on CPUs to now running machine learning on GPUs. We have moved from an industry that produced software to one that manufactures AI. AI is produced in factories. They are running 247. When licensing software, it is installed on computers. The manufacturing and distribution of that software have been completed. However, intelligence is never complete. There is interaction with all AIs, whether they are AI agents or AI robots.

Token, intelligence is represented by Tokens; a Token is a unit of intelligence. This is a number. These numbers are constituted, these symbols are formed in an intelligent and linguistic manner. Intelligence and steering wheel, the intelligence of autonomous vehicles, intelligent electric machines for expressing humanoid robots, intelligence of proteins and chemicals, as well as in drug discovery.

All these Tokens are produced in these factories. This infrastructure, these factories never existed before. This is something entirely new, which is why we see so much development worldwide. We have for the first time a new industry, a new factory that produces what we call AI. These factories will be built by companies. They will be constructed, and each company will become an AI manufacturer. Of course, no company can afford not to manufacture or produce AI. How can a country afford not to produce intelligence? You don’t have to produce chips. You don’t have to produce software, but you must produce intelligence. This is crucial. This is your core. This is our core.

So we have new industrial AI factories, which is why I call it the new industrial revolution. The last time this happened was 300 years ago, when electricity was discovered, generating and distributing electricity, a new type of factory was created. That new factory was not a power plant. Then a new industry was created called energy. Hundreds of years ago, there was no energy industry. It happened during the industrial revolution. Now we have a new industry that has never existed before—AI is at the top of the computer industry, but it is utilized and created by every industry. You must create your own AI. The pharmaceutical industry creates its own AI. The automotive industry creates its own AI. The robotics industry creates its own AI. Every industry, every company, every country must produce its own AI; this is a new industrial revolution.

Today, there is a very important announcement. It is announced that there will be a collaboration with SoftBank to bring, build, and create AI infrastructure in Japan. Together, we will build the largest AI factory in Japan, which will be constructed using nvidia technology. When completed, it will have a 25 AIx flip. Note that recently the world's largest supercomputer is a 1x flip. This is an AI factory producing 25 times the AI. However, to distribute AI, SoftBank will integrate nvidia's Ariel, which is the engine I previously mentioned that runs 5G radios on Cuda. By doing so, we are able to unify and integrate radio computing, baseband, and AI computing running on 5G. We can now develop and transform the telecom network into AIRAM. It will be able to carry voice, data, and video, but in the future, we will also carry AI, a new form of information intelligence. This will be distributed across 0.2 million sites for 55 million customers served by SoftBank. The AI factory and area will produce an AIAI distribution network to distribute AI performance.

Therefore, we will build these applications on top of the nvidia AI enterprise I previously mentioned and shown to you. There will also be a new store that will enable everyone to access AI. This is just a grand development. The result will be an AI grid spanning across Japan.

Now this AI grid will become part of the infrastructure and also one of the most important infrastructures. Remember that you need factories and roads as part of the infrastructure so that you can manufacture and distribute commodities. Infrastructure needs energy and communication parts. Every time you create something entirely new for infrastructure, new industries and new companies will be created, along with new economic opportunities and new prosperity. Without roads and factories, how would we have an industrial revolution? Without energy and communication, how would we have an IT revolution? Each of these new infrastructures opens up new opportunities.

Thus, collaborating with softbank to achieve this goal in japan is very exciting. Mia Kawasan's team, they should be in the audience. Working with you is incredible, and we are very happy to do this. This is completely revolutionary. This is the first time converting telecom networks and communication networks into ai networks.

Okay, let me show you what you can do. You can do some amazing things. For example, I am standing under a base station, a radio tower, where cars have video, and the car's video is streamed to the radio tower, where the radio tower has ai. This radio tower has video intelligence. It has visual intelligence. So it can see what the cars see and understand what the cars see. That ai model might be heavy in the car but won't be too heavy in the base station. Using the video streamed to the base station, it can understand the cars and anything happening around them. Okay, this is just one example of using ai on the edge to keep people safe, maybe this is air traffic control, essentially for autonomous vehicles. The applications are endless.

We can also turn the entire factory into ai using this basic concept. Here is a factory with many cameras. The cameras are streamed to the base station. Amazingly, now this factory has become ai because of all the cameras and ai models in ai. You can talk to the factory, asking what is happening in the factory. Ask the factory, has there been an accident? Is there anything abnormal happening? Did someone get injured today? Give you a daily report. You only need to ask the factory because the factory has now turned into ai. The ai model does not have to run in the factory. This ai model can run in the softbank broadcasting.

Okay, here is another example. But there are countless examples, you can basically turn every physical object into ai, stadiums, roads, factories, warehouses, offices, buildings... they can all become ai, and you can talk to it just like you chat with gpt. Okay, what is the condition of the aisle, is there any blockage or overflow? You are just talking to the factory. The factory observes everything, understands what it sees, it can reason, it can plan actions, or just talk to you. Here it says no, there are no obstacles, overflow, or dangers in the aisle of the warehouse. The conditions of the aisle in the video look orderly, clean, with no obstacles or dangers.

Okay, you are talking to the factory. It’s incredible. You are talking to the warehouse, you are talking to the cars, because all of these have now become intelligent.

Editor / jayden