"AI competition is an infinite game; the ultimate winner is not the country that invents the technology, but the country that can apply the technology on a large scale and efficiently."

On May 3, NVIDIA CEO Jensen Huang was interviewed at The Hill & Valley Forum, where he discussed the core concept of the "AI factory," explained the development stages of "physical AI," and explored the global landscape of the AI competition, as well as how AI and digital twins empower advanced manufacturing.

Huang stated that the AI competition is an infinite game, and the ultimate winner is not the country that invents the technology, but the country that can apply the technology on a large scale and efficiently. He also predicted that AI-driven robots, due to their constrained operating environments, are expected to achieve large-scale mass production and widespread application within five years, faster than self-driving cars.

The following is a transcript of the conversation:

01 The AI factory will reshape all industries by mass-producing "smart Tokens"

Host's Question: You position AI as a new industrial revolution and place the AI factory at its core. Can you explain what an AI factory is and why it is crucial to understand it in the 21st-century economy? Additionally, do you see this as a paradigm shift in modern computing? Will every physical manufacturing plant in the real world have an accompanying AI factory in the future?

Jensen Huang:

We have had many discussions about AI over the past few years. It has multiple facets, and it is helpful to view it from the following perspectives. First, AI is undoubtedly a new technology. Its construction differs from past software, allowing it to accomplish tasks that previous software could not. Therefore, this is an extraordinary technology: it has immense potential, and we must work hard to ensure its safety, as it will bring many exciting changes. This is the technological aspect.

The second aspect is relatively new. In the past tech industry, software was produced by manually typing code. Now, a new industry has emerged where software is produced by machines. This means that large supercomputers are needed, powered by electricity, to produce Tokens. These Tokens can be reassembled into various forms such as digital, text, protein, images, videos, and three-dimensional structures—we call this intelligence. This type of machine is fundamentally different from past machines, and I refer to it as the AI factory because it focuses solely on one thing: continuously producing Tokens every day.

At a higher level, we are increasingly recognizing that AI is likely to trigger a profound industrial revolution. This new technology will not only give rise to the new industry I just mentioned—the AI factory, which is intelligent production—but will also completely transform and reshape all other industries. These Tokens will be applied in healthcare, education, financial services, and engineering. We also use AI for software programming and supply chain management in our daily lives. AI is about to enter more fields, including manufacturing. Viewed from these three levels, the transformative nature and impact of AI are comparable to that of electricity in the past; it will fundamentally revolutionize every industry. Therefore, this is a genuine industrial revolution.

Regarding the future of the AI factory, I believe it is entirely correct. Today, any company that manufactures product components, such as lawnmower manufacturers or engineering machinery manufacturers like Caterpillar—whose products currently rely mainly on manual operation—will see these devices achieve autonomous, highly autonomous, semi-autonomous, or assisted operation in the future. Once autonomy is achieved, it will be defined by software. At that point, companies will need to produce Tokens to drive these devices (like tractors), which means software. Therefore, in the future, every product manufacturing company will have, in addition to its manufacturing plant, another factory dedicated to producing the AI that drives its products. Observing the automotive industry makes this clear: today's car companies primarily manufacture cars, but in ten years, all car companies will undoubtedly also produce Tokens that operate within their cars.

02 Physical AI with reasoning capabilities is crucial for manufacturing and other physical industries

Host's Question: Over the past year, you have discussed the concept of physical AI. For decision-makers thinking about the future policies of the United States, can you explain what physical AI is and how we should view it?

Jensen Huang:

Let's take a step back. Modern AI truly entered the public eye about 12 to 14 years ago, with the advent of AlexNet and significant breakthroughs in computer vision, around 2012. From a broader perspective, computer vision at that time was essentially perception—perceiving the world, whether it be images, sounds, vibrations, or temperature, in any form of information. Today, we have developed AI models that can understand the meanings of various types of information and exhibit intelligence. Therefore, the first wave of AI was perceptual AI. About five years ago, people began to discuss the second wave: generative AI. The core of generative AI is that AI models can not only understand the meaning of information but also transform it, such as translating English into French or converting text descriptions into images—images can be generated through prompts. It can be said that generative AI is essentially a universal translator that understands human language. This is the second wave.

The current wave is that AI has the ability to understand and generate. However, true intelligence also requires solving unknown problems and recognizing new situations. We achieve this through reasoning: using learned rules, laws, and principles to gradually break down problems, and even when faced with new challenges, we can find solutions through reasoning. This is one of the core capabilities of intelligence and marks our entry into the era of reasoning AI. Reasoning AI can generate digital robots, which we call agent-based AI, meaning they possess autonomy. This type of AI can understand tasks, autonomously learn to read, use tools like calculators, browsers, and spreadsheets, and ultimately complete designated tasks, such as accessing SAP to handle supply chain matters or connecting to Workday to manage human resources. These agent-based AIs are essentially digital labor robots. In the future, our generation of CEOs will manage both biological labor and digital labor, with traditional HR departments responsible for the former, while IT departments will evolve into "human resource centers" for agent-based AI. This is the stage we are currently in.

The next wave will benefit the largest industries globally. It requires AI to understand fundamental concepts such as physical laws, friction, inertia, and causality. For example, an object that is knocked over will fall; a bottle placed on a table will not penetrate the tabletop. Most AIs lack the common-sense physical reasoning ability that children and even pets possess. If a ball is rolled off a kitchen counter and disappears, AI might think it is completely gone, but a dog knows the ball is on the other side. The dog understands the concept of object permanence and realizes the ball has not entered another universe, so it will go around the table to retrieve it.

Robots also need to learn: to get from one side of the table to the other, they cannot go straight through; they must reason out a detour. This type of physical reasoning is what we call physical AI. Embedding physical AI into a physical entity called a "robot" results in robotics. This is crucial for us today, as factories are being built vigorously across the United States. We hope to construct these facilities in a way that fully utilizes the latest technology. Therefore, we hope that the new generation of factories built in the next decade will be highly automated to help us address the widespread severe labor shortages globally.

03 The AI competition is an infinite game; the winner is the country that can apply technology on a large scale and efficiently

Host's Question: Many people believe we are in a global AI competition. How do you think the U.S. government should act to win the competition and master the most advanced AI technology?

Jensen Huang:

First, to participate in the competition and win, one must understand the competition itself: clarify one's resources, existing and lacking assets, and strengths and weaknesses. It is essential to recognize that the core of AI is foundational. Looking back at the three levels mentioned earlier, it is necessary to ensure understanding of the rules of the game at each level. This game is not limited to 60 minutes; it is an infinite game. Most people are not good at infinite games. Nvidia has been established for 33 years, going through the PC revolution, the internet revolution, the mobile revolution, and now we are welcoming the AI revolution. To continue to thrive in a constantly changing environment, one must understand how to participate in the game. Understanding the rules of the game and recognizing one's assets is crucial.

At the technological level, the key is to understand knowledge capital. It is important to note that half of the world's AI researchers come from China; we must face this significant factor and incorporate it into our game strategy. Next is the AI factory, which operates efficiently relying on energy, as it essentially converts electricity into digital Tokens. This is similar to how the last industrial revolution was driven by energy converting atoms into steel, cars, buildings, etc.; even earlier, hydropower generated electricity. Today, it is electricity input and Token output. Therefore, energy is the key at the next level.

A higher level is gradually emerging, and we must deeply understand: the ultimate winners of the last industrial revolution were not the countries that invented the technology, but those that applied it. The U.S. far surpassed other countries in the application of steel and energy. Therefore, the core at this infrastructure level is technology application—embracing it without fear, actively retraining the workforce to adapt to new technologies, and encouraging widespread adoption. Viewing AI from this perspective and framework reveals that each level has its unique challenges, opportunities, and rules of the game.

04 The AI factory will create a large demand for new skilled labor positions

Host's Question: Regarding labor issues, the media has been emphasizing the narrative that AI may lead to large-scale labor displacement and unemployment. Can you envision how AI will impact the job market? More specifically, what new job categories do you foresee emerging in the future that we currently may not even imagine?

Jensen Huang:

Some new jobs will be created, some jobs will disappear, and every job will change. It is easy for people to swing between two extremes, but I always believe that breaking down the problem and thinking from first principles will be very beneficial. In the framework I described earlier, the most fundamental situation is that you are deeply involved in venture capital and understand the dynamics of the AI field—at the foundational level, it is AI that has allowed San Francisco to recover. Almost everyone had left San Francisco, and now it is thriving again, which is entirely due to AI. AI has created new types of jobs, fundamentally because it is a completely new way of software development. The emergence of AI has changed every aspect of technology. In the past, software was manually written and ran on CPUs; now, machine learning-generated software runs on GPUs. Therefore, at every level: related tools, compilers, methodologies, the way data is collected and managed, using AI to set safety barriers, using AI for training, and using AI to ensure the safety of AI itself, all these technologies are continuously emerging and creating a large number of job opportunities.

A higher level harbors tremendous opportunities. As I mentioned earlier, we will build new factories that input electricity and output Tokens. Taking a 1-gigawatt factory as an example—we expect to build clusters of AI factories ranging from 7 to 10 gigawatts in the future. The investment for a 1-gigawatt factory can reach up to $60 billion. Currently, a 100-megawatt factory is quite common. This $60 billion investment scale is equivalent to Boeing's annual revenue. Building such factories requires financing, which will create a large number of jobs. The construction of the site and factory shell will drive significant employment in the construction industry, including carpenters, steel structure workers, and masons, all of whom are essential talents needed for factory construction. A $60 billion investment factory is extremely large. This will require mechanical engineers, electrical engineers, plumbers, as well as professionals in all subsequent low-voltage systems, IT, and networking, and finally, an operations team. The entire construction cycle will take about three years. This will give rise to a large number of new technical job categories. In the past transitions of the computer industry and computing platforms, the main bottleneck for most companies' growth was software engineers.

In the new realm of the AI factory, the most critical will be skilled workers, that is, talents with specialized skills. I think this is very good. Our country needs to recognize that specialized skills are a respected and essential job, indispensable for national development. Therefore, we should encourage the cultivation of such talents. Electricians, plumbers, carpenters, steel structure workers, and others in various fields will require a large number of such talents.

At an even higher level, we can discuss how AI Agents will change the work of doctors, financial service professionals, or customer service personnel. For example, in our company, each software engineer is currently equipped with an AI assistant as a preliminary attempt. The result is that the amount of new code generated in our company is astonishing. Our productivity has soared, allowing us to hire more talent because AI enables us to create more products that the market needs, thereby increasing revenue and hiring capacity. Therefore, I believe that higher-level applications should indeed embrace AI as early as possible. Remember, it is not AI that takes away your job, nor is it AI that destroys your company, but those companies and individuals who make good use of AI. This is something worth contemplating and accepting.

05 AI and digital twin technology are at the core of advanced manufacturing

Host's Question: Recently, there has been a heated discussion about the return of manufacturing. Many experts in the AI field have explored the concept of digital twins and how the adoption of this technology in manufacturing plants can effectively support the revival of domestic manufacturing. At the same time, the CEO of Apple recently stated that one of the main bottlenecks for the return of iPhone manufacturing to the U.S. is the lack of high-quality, high-precision robotic arm technology. Therefore, overall, it seems that AI can indeed become a key enabling technology for the development of manufacturing and the return of industries. What are your thoughts on this?

Jensen Huang:

First of all, the core of manufacturing is not low-cost labor. Today's advanced manufacturing is driven by software. The entire factory operates like a giant robot, coordinating the operations of numerous internal robots. While these advanced factories have many employees, they are essentially technology-driven. I believe the first point is that in our industry (just taking my industry as an example), achieving end-to-end manufacturing from chips to AI supercomputers in the U.S. is an excellent opportunity. I am pleased to see the government actively encouraging and supporting the industry to bring manufacturing back to the U.S. This is a high-quality, high-tech job, and conducting it domestically is a fantastic opportunity for the country. I am passionate about this, and we are staunch supporters of this trend, and we are fortunate that global partners support it. That is one.

Secondly, if we are not proficient in manufacturing, we will miss out on a huge industry that will be driven by the availability of energy in the future. Which country does not want to participate in this emerging AI industry? Why not produce AI? Why not engage in the most advanced manufacturing? It is essentially manufacturing. Its ultimate output is digital, just as the output of the last industrial revolution was electronic—at that time, most people could not understand that electricity could be created through generators. Today, we call it Nvidia AI supercomputers. But back then, the generator produced intangible electricity—something invisible and untouchable, yet it was indeed electricity, it was electronic. Now, it is a new form of "electronic"—digital. Therefore, we certainly want to engage in this emerging industry, and to achieve this, we must have domestic manufacturing.

Given the technology-intensive nature of manufacturing, we should first build in a digital twin environment, and then operate in a virtual reality environment. Nvidia has designed the most complex systems in the world. We invest about $20 billion in R&D for each generation of products, and it may be even higher now, but this $20 billion is only for producing a series of chips. We design entirely within the digital twins of these chips. Months before actual manufacturing, they exist as digital models. When the chips are released, I know they will be perfect because we have conducted thorough simulations, emulations, and rigorous testing on them. Digital factories should be the same. Especially for large factories, a complete digital twin should be created, utilizing artificial intelligence for construction and operation—achieving virtual integration, fully digitizing and integrating these grand structures, and then operating, optimizing, and planning for completely digital outputs. In the future, every factory, every car, every building, every city, and even, I hope, every person will have its digital twin version. The concept of digital twins is thus becoming a reality, thanks to artificial intelligence.

06 AI robots are expected to achieve large-scale production and application within five years

Host's Question: When do you expect AI-driven robots to become ubiquitous in our daily lives?

Jensen Huang:

First of all, self-driving cars are a type of robot. We have spent about 10 years developing to the current stage. Waymo has now entered many cities across the country and is performing excellently. It is encouraging to see Waymo's cars driving in cities like San Francisco. This took about 10 years. The time required for robots will be shorter. Because we can constrain the operating environment of robots. So robots do not need to have the same level of generality as cars. Once a car enters San Francisco, it must adapt to every street and various road conditions. For robots, we can impose more restrictions, and from prototype development, functional improvement to large-scale mass production, it will take about five years. Today, we already have powerful robots. Therefore, in about five years, we will see robots being produced in large quantities from factories. The companies that manufacture cars today will be very skilled at manufacturing robots in the future. They just need to do better in software and AI, and the relevant technologies are already quite widespread.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。