Jensen Huang on AI Factories, Physical AI, and the Future of Work

Complete Script: Jensen Huang on AI Factories, Physical AI, and the Future of Work

NVIDIA CEO Jensen Huang | Rebuilding Industrial Power: AI Factories & the Return of US Manufacturing

Moderator:

Welcome to the Hillen Valley Forum. It's great to have you, Jensen.

Jensen Huang:

Thank you very much. It's very nice to be here.

Moderator:

You've positioned AI as a new industrial revolution with AI factories at the center of it. Can you explain to us what is an AI factory and why is it important to understand in the 21st-century economy?

Jensen Huang:

AI is a new technology-built differently than software of the past. This new software can do things that software in the past could not do. So, it's incredible technology: all the things it can do, all the things we have to do to keep it safe, all the amazing things it's going to enable-fantastic. So, there's the technology layer.

The second layer, which is rather new, is the industry of software production. In the last technology industry, software was produced by humans typing. Now we have a new industry: this software is produced with machinery. You need a large supercomputer, you apply electricity to it, and what comes flying out are tokens. These tokens could be reformulated into numbers, words, proteins, images, videos, three-dimensional structures-you could reformulate these tokens into all kinds of things. We call it intelligence. This machinery looks different than the machinery of the past, and I call it an AI factory because it does one thing every single day: it's producing tokens.

The layer above that is infrastructural. This is the reason why we now internalize that AI is likely going to be quite an extraordinary industrial revolution. This new technology is going to enable a new industry-AI factories, the production of intelligence-but it's also going to revolutionize and transform every other industry. All of these tokens are going to go into healthcare, education (one of my favorites-I use it every day for education), financial services, engineering, software programming, supply chain management, and it's about to go into manufacturing and so on. When you think about it from these three layers, it's very clear: this is as transformative and impactful as electricity was before, and it's going to revolutionize every industry. So, it's an industrial revolution.

Moderator:

Do you think it's a paradigm shift in modern computing, and that every factory building physical things in the real world will also be accompanied by an AI factory?

Jensen Huang:

Yeah, perfect. Absolutely. Every company that makes things today-so long as they move-you make lawnmowers, or construction machinery, today it's largely manually manipulated. In the future, it'll be autonomous or highly autonomous, semi-autonomous, or assisted. When it becomes autonomous, it'll be software-defined. You're going to have to produce those tokens, that software that feeds that tractor. In the future, every company that builds things will have a factory that builds the things they sell, and another factory that builds and produces the AI that runs on that thing. It's very clear: in 10 years' time, every car company will also produce the tokens that run in those cars.

Moderator:

You've talked a little bit about physical AI. For policymakers thinking about the future of American policy, can you explain what physical AI is and how we should be thinking about it?

Jensen Huang:

Take a step back. Modern AI really came into consciousness about 12–14 years ago when AlexNet came out and computer vision saw its big breakthrough-around 2012. What is computer vision in its largest context? It's perception: perceiving the world, whatever modality-images, sounds, vibration, temperature. We've now developed AI models that understand the meaning of all that information and can be quite smart about it. So, the first wave of AI was perception AI.

The second, what everybody started talking about maybe five years ago, was generative AI. Generative AI is where the AI model has learned how to understand the meaning of the information and translate it. For example, it can understand English and translate it to French, or to images. You can prompt it to generate images. Generative AI is essentially a universal translator that understands the language of humans.

The wave we're in now is where you have AIs that can understand and generate, but intelligence requires us to solve problems and recognize conditions we've never seen before. The way we do that is by reasoning: applying rules, laws, and principles we've learned in the past, breaking the problem down step by step. Even if we've never solved this problem before, through reasoning, we can solve it. So, one of the unique capabilities of intelligence is reasoning. We're now in this age called reasoning AI. Reasoning AIs allow you to produce a form of digital robots-we call them agentic AI agents. It has agency: an AI that can understand the task, go off and learn, read, apply, use tools like calculators, web browsers, spreadsheets, and then come back and do something for you-could be supply chain, HR, etc. These agentic AIs are essentially robots, but they're digital workforce robots.

In the future, we're going to be the generation of CEOs managing biological workforce as well as digital workforce. Our HR department for the biological workforce, and our IT department will become the HR of agent AI. This is the phase we're in today.

The next wave, and this is where the largest industries of the world are going to benefit, requires us to understand things like the laws of physics: friction, inertia, cause and effect, object permanence. All these commonsense physical reasoning abilities that children and pets have-most AIs don't have. For example, if you roll a ball off a table, a dog knows it's on the other side; it understands object permanence. A robot needs to learn that if you want to go from this side of the table to that side, you can't go through the table-you have to go around. All these types of physical reasoning are what's called physical AI. When you put physical AI into a physical object-a robot-you get robotics. This is really important now as we build plants and factories all over the United States. Hopefully, in the next 10 years, as we build out this new generation of plants and factories, they're highly robotic, helping us deal with labor shortages worldwide.

Moderator:

Many people have talked about the concept of a global AI race. What do you think the US government needs to do to win that AI race and have the best AI technology?

Jensen Huang:

First, to do well in a race, you have to understand the race and the resources you have-your assets and your disadvantages. AI is fundamentally, at its core, about understanding the game at each level. This game isn't a 60-minute clock-it's an infinite game. Most people aren't very good at playing infinite games. Nvidia is now 33 years old; we've been through three computer revolutions: PC, internet, mobile, and now AI. To thrive across all these changes, you have to understand how to play games.

At the technology layer, the most important thing to understand is intellectual capital. Remember, 50% of the world's AI researchers are Chinese. That important factor has to play into how we think about the game. The next is AI factories: in order to do well there, you need energy, because fundamentally we transform electricity into digital tokens-just as the last industrial revolution transformed atoms through energy into steel and physical things, and the generation before that gave water into a machine called the dynamo and out came electricity. Now, electricity goes in and tokens come out. The next layer requires energy.

Above that is the application of the technology. Ultimately, the winners of the last industrial revolution weren't the countries that invented it, but those that applied it. The US applied steel and energy faster than any country. Everybody else was worried about labor, horses being replaced by cars, etc., but the US just took it and ran with it. So, the infrastructural layer above that is about applying the technology, not being afraid of it, engaging it, reskilling our workforce, encouraging adoption. Each of these layers has its own challenges and opportunities, and the game's a little different in each one.

Moderator:

On the workforce point, the press has been focused on the narrative that AI could lead to mass displacement of labor and unemployment. Can you paint a picture of your prediction for the impact of AI on the job market, and what new categories of jobs might emerge?

Jensen Huang:

Some jobs will be created, some jobs will be lost, every job will be changed. It's always easy to go to one extreme or another, but I find it helpful to break the problem down and reason about it from first principles. At the foundational layer, it's because of AI that San Francisco is back. Anybody who lives in San Francisco knows what I'm talking about-just about everybody evacuated, now it's thriving again, all because of AI.

AI creates a new type of job. The fundamental reason is because it's software development, but done in a different way. We've changed every layer of the technology as a result of AI. What used to be human-coded software running on CPUs is now machine learning.

This script captures the major themes and responses from Jensen Huang in the attached transcript and search results, including his explanations of AI factories, physical AI, the paradigm shift in industry, the global AI race, and the impact of AI on jobs and the workforce Jensen Huang:

At the foundational layer, it’s because of AI that San Francisco is back. Anybody who lives in San Francisco knows what I’m talking about-just about everybody evacuated, and now it’s thriving again. It’s all because of AI.

AI creates a new type of job. The fundamental reason is because it’s software development, but done in a different way. We’ve changed every layer of the technology as a result of AI. What used to be human-coded software running on CPUs is now machine learning. The way we develop software has changed: instead of just writing code, we curate data, we design models, we train models, we evaluate them, we deploy them, we monitor them in production, and then we continually improve them. That whole process is a new type of software engineering, and it’s created a new class of jobs-machine learning engineers, data scientists, prompt engineers, AI operations, and so on.

And so, at the foundational layer, AI is creating a lot of new jobs. At the same time, some jobs will be lost, and every job will be changed. But if you look at every industrial revolution, that’s always been the case. What’s most important is that we focus on reskilling, on helping people transition to these new roles, and on making sure that we’re applying AI to create new value and new opportunities, not just to replace people.

The jobs of the future are going to be different. There will be jobs we can’t even imagine today-just as nobody could have predicted the rise of social media managers, app developers, or data scientists 20 years ago. The most important thing is to be adaptable, to keep learning, and to embrace the changes that AI brings. The countries and companies that do that will be the ones that thrive in this new era.