1
Humanoid robots will become the largest industry or product in history.
Host:
Musk, you have a great obsession with something. We all admire “first-principles thinking”—which Huang sometimes refers to as “first-principle scaling”—and it is this kind of thinking that has allowed you to reduce battery costs from $1,000 per kilowatt-hour to below $100. Now, you are applying the same cost optimization in the robotics field with servo rotors and motors for actuators. So I want to hear how you consistently disrupt every industry with this kind of thinking.
Musk:
Actually, it’s mostly not about “disruption” but rather about “creation.” For example, SpaceX’s reusable rockets; there were no reusable rockets before, but to fundamentally change space travel, the core is reusability.
If every launch requires discarding the rocket, the cost of space exploration would be outrageous. Speaking of electric vehicles, when we first started making electric cars, there were no electric cars available on the market (as far as I know). So Tesla’s goal was to create electric cars that are both attractive and affordable.
As for humanoid robots, there are currently no truly practical products, only some gimmicks. But I believe Tesla will create the first truly useful humanoid robot, and this will be a huge revolution.
I believe everyone will want one—after all, who wouldn’t want their own personal C-3PO or R2-D2? (Yes, I remember those two characters from Star Wars). In the future, there will be a large number of robots in the industrial sector providing products and services. That’s why I say humanoid robots will become the largest industry or product in history, larger than smartphones or anything else, because everyone will want one, or even multiple.
Moreover, the demand in the industrial sector will also be enormous.
Huang:
I just want an “R2-D2 with a C-3PO shell.”
Musk:
Well said! In fact, humanoid robots will be more powerful than both R2-D2 and C-3PO combined, even ten times stronger.
People often talk about eliminating poverty. How long has this topic been discussed? There has been a lot of discussion, and many NGOs are trying, but in reality, there has been no success, and the evidence speaks for itself.
But artificial intelligence and humanoid robots will truly eliminate poverty—and Tesla will not be the only company producing humanoid robots. I believe Tesla will pioneer this field, but more companies will join afterward. The essence of making everyone wealthy is fundamentally through artificial intelligence and robotics technology.
2
Why we need AI factories
Host:
Huang, what is the next step in the development of AI factories?
Huang:
Saudi Arabia is transforming its refineries into AI factories, and this story is very beautiful. I have always said that artificial intelligence is an infrastructure, and the reason is simple: from a technological perspective, AI is disrupting every industry, and the application of digital intelligence is pervasive across various fields. In the future, every company, every industry, and every country will use it. In this sense, it is foundational and thus part of the infrastructure.
From a computer science perspective, the innovation of artificial intelligence lies in the fact that past computing was mainly “retrieval-based computing”—someone inputs a piece of text, creates a painting, or designs four versions of a digital ad, all of which are pre-made, and the system just retrieves the appropriate version for you.
Hadoop and many past frameworks and operating systems were designed to help you retrieve relevant information. But now, software will generate content in real-time—based on context, scenarios, user identity, the questions you ask, and prompts, generating unique content each time.
For example, when using Groq, the results are different each time, depending on your prompts and specific scenarios. So the computing model has shifted from “retrieval-based” to “generative,” and since it is generative and each result is different, we need AI factories around the world to generate content in real-time—this is why we need AI factories.
This computing method is unprecedented, but the benefit is that all content is not pre-set but is contextually relevant and intelligent.
3
Work will become “optional”
Host:
So with AI factories and robotics technology, last night the Crown Prince also mentioned his vision: to enhance the workforce by deploying tens of millions of robots, injecting the next wave of productivity and progress. But this also raises concerns about future employment. Musk and Huang, what are your thoughts?
Musk:
Of course, in the long run (I don’t know if long-term means 10 years or 20 years, but to me, that’s long-term), my prediction is that work will become “optional.”
Host:
That’s interesting.
Musk:
Yes, just like sports or video games. If you want to work, go ahead—just like you can go to the store to buy vegetables or grow them in your backyard.
Growing vegetables in the backyard is obviously more troublesome, but some people will still do it because they enjoy gardening. That’s how work will be in the future; it will be optional. And before we reach that state, we have a lot of work to do.
Musk:
I always recommend that people read Isaac Asimov’s “Foundation” series to get a feel for a possible positive future with artificial intelligence.
Interestingly, in those books, currency no longer exists. I guess if AI and robotics technology continue to advance (which seems likely), at some point in the future, currency will become irrelevant. Of course, basic physical resources like electricity and materials will still be constraints, but I believe currency will ultimately lose its significance.
4
AI will make you busier
Huang:
Since we’re talking about currency…
I want to say that we can look at it from different time dimensions.
First, everyone’s way of working will change—this is certain. Students’ learning methods and people’s working methods will be different because many things we currently find trivial, laborious, or difficult will become very simple in the future, so our productivity will increase significantly.
I want to say that for most people or companies, if life becomes more efficient and originally difficult things become simple, then you will have more time to pursue more ideas.
I guess Musk will become busier because of artificial intelligence, and so will I. The reason is that we both have many ideas, and there are many backlog projects in our companies that we want to push forward.
If productivity increases, we can achieve these goals faster. So in the short term, all evidence suggests that we will become more efficient, but at the same time, we will also be busier because we have too many things we want to do.
Huang:
I can give an example: I just talked to Musk about how radiology has basically been driven by artificial intelligence now.
Many excellent companies are doing this, and surprisingly, it was previously predicted that “radiologists would be among the first to lose their jobs,” but the reality is quite the opposite—now the hiring of radiologists has actually increased.
The reason is simple: the core goal of radiologists is not to “study images” but to “diagnose diseases.” Now that image analysis has become very efficient, they can analyze more images, more modalities, and spend more time communicating with patients, thus being able to take on more patients.
Today, the global volume of radiological diagnosis is increasing, and disease diagnosis is becoming more accurate. This is the short-term result of artificial intelligence in productivity. As for what will happen in the long term—like when currency becomes irrelevant, remember to let me know in advance.
Musk:
You will notice it in advance, just like…
Huang:
Just like we often send messages, just send a text when the time comes. Yes, we often communicate via text.
Host:
I completely agree with both of you. Because looking at every wave of technology, every general-purpose technology ultimately brings net positive benefits to the world and humanity. I also want to share two stories.
Huang:
I think that’s the reason—innovators like Musk have so many great ideas…
5
A 500 MW AI data center will be built in the desert
Host:
I want to share the stories of two Saudi innovators whose achievements are inseparable from NVIDIA’s strong support.
The first is Professor Ouyaji, the first Saudi-American to win a Nobel Prize for his achievements in creating a new field of chemistry. He utilized NVIDIA’s AI accelerators and models like Rock to pioneer new chemistry in the field of metal-organic frameworks (MOFs)—creating a “sponge” with a pore size of only 0.33 nanometers that can capture water and carbon dioxide from the air.
The second story also involves NVIDIA-accelerated AI and models like Rock: nanorobots (sized 500 nm x 1000 nm) that can treat sickle cell anemia using CRISPR gene editing technology.
Both of these projects originated from research conducted 20 years ago, but AI has accelerated the transformation of results, allowing us to enter a new value domain. I believe that in terms of labor and productivity, humanity can always shift to new value domains with the waves of technology. But today we have some major news to announce, Musk, let’s start with your collaboration with X AI.
Musk:
We are excited to announce that we will be collaborating with Saudi Arabia to build a 500 MW (yes, 500 MW) project… Wait, if it were 500 GW, it would cost $8 trillion.
So, X AI and the Kingdom of Saudi Arabia will collaborate to build a…
Host:
A 500 MW project, starting with a first phase of 50 MW, and will collaborate with NVIDIA. Congratulations to the Humane team and the Target team for doing an excellent job. Huang, I think you also have some important news to announce this week, right?
Huang:
We have a lot of news to announce. Our collaboration with Humane is progressing very well—first, we helped this company get started, and now they have a major client: Musk.
It’s hard to imagine a startup with almost zero revenue is about to build a 500 MW data center for Musk; the scale is enormous, and this company has suddenly established itself. In addition, we are also collaborating with Amazon Web Services (AWS), as you know…
Host:
Congratulations to the Humane team for their collaboration with AWS: the first phase starts with 100 MW, aiming for 1 GW, and the scale is continuously expanding.
Huang:
So AWS will also collaborate with Humane, and we are utilizing NVIDIA’s Omniverse Digital Twins technology—artificial intelligence is not just about agents and chatbots; cognitive AI is crucial to the world, but the application of AI spans all fields: chemistry, proteins, genes, physics, fluid dynamics, particles, and of course robotics and drive systems.
We have built the Omniverse platform where robots can learn how to be excellent robots in this physics-based environment. We are working with Humane to apply Omniverse technology to various scenarios such as digital factories, robotics, and warehouses. This is another important collaboration.
We are also building a supercomputer in Saudi Arabia to simulate quantum computers and use our computers for control and error correction—quantum error correction requires massive computational power. So we are doing a lot of excellent work in this area. The collaboration with Humane is fantastic; they have made significant progress right from the start.
6
Space AI is inevitable
Host:
Is AI feasible in space?
Musk:
Feasible. If human civilization can continue to develop (which is likely), then space AI is inevitable.
Of course, I must first clarify: we cannot take the survival of civilization for granted; we need to ensure that civilization remains on an upward trajectory.
Anyone who studies history knows that the development of civilization is not always upward; in fact, civilizations have life cycles. I hope we are currently in a strong upward phase—I believe we are—but we cannot be complacent.
To understand space AI, we can think from the perspective of the Kardashev Scale: if human civilization wants to utilize even a millionth of solar energy, it must deploy solar AI satellites in deep space.
When you think from the perspective of “how much solar energy can civilization convert into useful work,” you realize that space is the key—Earth receives only about one in two billion of the total solar radiation from the sun. So if we want to obtain a million times more energy than the total power generation on Earth, we must go into space. And having a space company (referring to SpaceX) is very helpful in this regard.
Huang:
It’s also easier to cool chips in space.
Musk:
Exactly. There is no water in space, so we must use a waterless cooling method—essentially radiation cooling. My estimate is that long before the energy potential of Earth is exhausted (possibly in less than five years), the cost-effectiveness of space AI computing will far exceed that on the ground. In other words, within a maximum of five years, the cheapest way to compute AI will be solar-powered space satellites.
Huang:
Look at the supercomputer we built together: each rack weighs about 2 tons, of which 1.95 tons might be cooling equipment. Imagine what it would be like if we made these supercomputers (like the GB300 rack) into mini devices.
Musk:
And generating power has become a challenge. If you want to scale AI computing, both power generation and cooling will find that space has significant advantages.
For example, if you want to achieve 200 to 300 GW of AI computing power annually, it is almost impossible on Earth—America’s average electricity consumption is about 460 GW per year, and 300 GW is equivalent to two-thirds of America’s annual power generation, which is simply impossible to build that many power plants.
If you want to achieve 1 TW of computing power annually, that’s even more impossible; it must be achieved in space. And in space, solar energy is continuous, and batteries are not even needed (because space is always daytime), and solar panels will be cheaper (no need for glass or frames), and cooling can be done through radiation.
Huang:
That’s the dream, yes, that’s our dream.
7
What we see is not a bubble, but a fundamental shift from general computing to accelerated computing
Host:
Will we see an AI bubble?
Huang:
Okay, let me share what we see. I think to understand what is happening globally, we need to return to the fundamental principles of computer science and computing. Currently, three key trends are occurring:
First, we all know that Moore’s Law has reached its end. The gap between computing demand and the computing power that general computing can provide is widening, so the world has long begun to shift towards accelerated computing—we have been driving this change for over 20 years.
Let me give you a data point: six years ago at the Supercomputing Conference, 90% of the world’s Top500 supercomputers used CPUs; this year, that percentage is less than 15% (from 90% down to 10%), while the share of accelerated computing has risen from 10% to 90%. This marks a turning point in the high-performance computing field from general computing to accelerated computing.
Second, one of the most data-intensive and computationally demanding tasks in the cloud is data processing—every year, raw data processing alone consumes hundreds of billions of dollars in computing resources, which has nothing to do with AI; it’s just SQL processing, data frame analysis (like everyone’s name, address, gender, age, income, etc.).
Today’s world (whether in banking, credit card industry, e-commerce, or ad recommendations) relies on these data frames to operate, and the cost of processing these data frames is extremely high. This is the first impact of the end of Moore’s Law.
Third, the past 15 years have been called the “era of recommendation systems”—how to recommend information to users in social dynamics, how to recommend ads, books, or movies? The internet is so vast that without recommendation systems, our phones would not be able to find the information we need. Recommendation systems are the core engine of today’s internet, but it is shifting from traditional CPU-based models to GPU-based generative AI models.
If we only look at these two application scenarios, we will find that many internet companies will deploy a large number of GPU computers, which lays the foundation for the third opportunity—agentic AI (like Groq, OpenAI, Anthropic, Gemini, etc.).
But let’s not forget that beneath the surface of AI that everyone sees today, there is a comprehensive transformation from general computing to accelerated computing. Considering this, you will conclude that the resources supporting this revolutionary agentic AI are not only far less than you might think, but all resources are reasonable and necessary.