Recently, a journalist from Wired interviewed Nvidia CEO Jensen Huang.

The reporter stated that a conversation with Jensen Huang should come with a warning label, because this Nvidia CEO is so invested in the direction of artificial intelligence that after nearly 90 minutes of intense conversation, I (referring to the journalist of this interview, same below) am convinced that the future will be a rebirth of neural networks. I can also see all of this: the robot renaissance, medical godsends, self-driving cars, and chatbots with memory. The buildings in the company's Santa Clara campus do not help at all. No matter where I look, I see triangles within triangles, a shape that helped Nvidia make its first fortune.

Huang has been a person of the year in the past and may even be the person of the decade in the future. Because technology companies really love Nvidia's supercomputing GPUs. This is not the old Nvidia, it is the supplier of X-generation video game graphics cards, which makes images come to life by effectively rendering countless triangles. This is Nvidia, whose hardware has created a world where we talk to computers, computers talk to us, and ultimately, depending on which technology expert you talk to, they surpass us.

Advertisement

In our meeting, Huang, who is 61 years old, was wearing his iconic leather jacket and minimalist black sneakers. He told me on that Monday morning that he hates Monday mornings because he works all day on Sunday and is already tired when he starts the official workweek.

Huang has established a pattern that keeps Nvidia at the forefront of every major tech trend. In 2012, a small group of researchers released a breakthrough image recognition system called AlexNet, which used GPUs (not CPUs) to process code, opening a new era of deep learning. So, Huang immediately directed the company to pursue artificial intelligence with all its might. In 2017, when Google released a new neural network architecture called Transformer (the T in ChatGPT) and sparked the current artificial intelligence gold rush, Nvidia was in the perfect position to start selling its AI-focused GPUs to hungry technology companies.

Nvidia currently holds more than 70% of the AI chip market in sales, with a valuation just over 2 trillion dollars. The revenue for the last quarter of 2023 was 22 billion dollars, a 265% increase from the previous year. Its stock price rose by 231% last year. Huang is either incredibly good at what he does, extremely lucky, or both - everyone wants to know how he did it.

But no one can rule forever. He has now become the focus of the Sino-American tech war and is subject to the whims of regulatory agencies. Some of Huang's challengers in the field of artificial intelligence chips are household names - Google, Amazon, Meta, and Microsoft - and have the most financial resources in the tech field. At the end of December, semiconductor company AMD launched a large processor for artificial intelligence computing, aiming to compete with Nvidia. Startups are also aiming at this goal. Data from research firm Pitchbook shows that venture capitalists invested more than 800 million dollars in artificial intelligence chips in just the third quarter of last year.

So, how does Huang view these? Let's take a look at the original interview:

Huang: You and I are both graduates of Stanford University.Lauren Goode: Yes. Well, I studied journalism, and you did not.

Jensen Huang: I wish I had.

Lauren Goode: Why is that?

Jensen Huang: Well, the person I truly admire as a leader and as an individual is Adobe CEO Shantanu Narayen. He said he always wanted to be a journalist because he loves telling stories.

Lauren Goode: Being able to tell the story of a business effectively seems to be an important part of building a business.

Jensen Huang: Yes. Strategy formulation is storytelling. Building culture is storytelling.

Lauren Goode: You've said many times that you don't sell Nvidia based on the hype.

Jensen Huang: That's correct. It's actually about storytelling.

Lauren Goode: So I'd like to start with something another tech executive told me. He pointed out that Nvidia was a year earlier than Amazon, but in many ways, Nvidia has a more "day one" approach than Amazon. How do you maintain this perspective?

Jensen Huang: Frankly, it's really a good word. I wake up every morning as if it's the first day, and the reason is that we're always doing something we've never done before. It also has a vulnerable side. We are very likely to fail. Just now, I was in a meeting, and we were doing something completely new for our company, but we didn't know how to do it right.Lauren Goode: What's new?

Jensen Huang: We are building a new kind of data center. We call it the AI factory. According to the way data centers are built today, many people share a set of computers and put their files in this large data center. The AI factory is more like a generator. It's quite unique. We've been building it for the past few years, but now we have to turn it into a product.

Lauren Goode: What are you going to call it?

Jensen Huang: We haven't named it yet. But it will be everywhere. Cloud service providers will build them, and we will build them. Every biotech company will have it. Every retail company, every logistics company. Every future car company will have a factory for making cars (actual goods, atoms) and a factory for making AI for cars (electrons). In fact, as we speak, you see Elon ·Musk is doing this. He is far ahead of most people in thinking about what industrial companies will look like in the future.

Lauren Goode: You once said that you run a flat organization with 30 to 40 executives reporting directly to you because you want to be in the information flow. What has recently aroused your interest and made you think, "I may finally need to bet on Nvidia in this matter?"

Jensen Huang: Information does not have to flow from the top to the bottom of the organization like in the Neanderthal era, when we did not have email, text messages, and all these things. Today, information flows more quickly. Therefore, there is no need for a hierarchical tree to explain the information from top to bottom. A flat network allows us to adapt more quickly, which is what we need because our technology is developing so fast.

If you look at the way Nvidia technology has developed, Moore's Law doubles every few years. Well, in the past 10 years, we have advanced AI by about a million times. This is many times that of Moore's Law. If you live in an exponential world, you don't want information to spread from top to bottom in one layer.

Lauren Goode: But I ask you, what is your Roman Empire? This is a meme. What is the version of today's transformer paper? What is happening now that you think will change everything?

Jensen Huang: There are a few things. One of them does not really have a name, but it is some of the work we have done in the field of basic robotics. If you can generate text, if you can generate images, can you also generate motion? The answer may be yes. Then, if you can generate motion, you can understand intentions and generate a general version of clarity. Therefore, humanoid robot technology should be around the corner.

I think the work around state-space models (SSM: state-space models) may be the next transformer, which allows you to learn extremely long patterns and sequences without a quadratic increase in computation.Lauren Goode: What does this bring? What are the real-life examples?

Jensen Huang: You can have a conversation with a computer that lasts a very long time, but the context is never forgotten. You can even temporarily change the subject and return to the previous topic, and the context can be preserved. You might be able to understand extremely long chains of sequences, such as the human genome. Just by looking at the genetic code, you can understand its meaning.

Lauren Goode: How far are we from this goal?

Jensen Huang: From when we had AlexNet to the superhuman AlexNet, it only took about five years. The foundation models for robots may be imminent—I will announce it at some point next year. From then on, five years later, you will see some very amazing things.

Lauren Goode: Which industry will benefit the most from the widely trained robot behavior models?

Jensen Huang: Well, heavy industry represents the largest industry in the world. Moving electrons is not easy, but moving atoms is extremely difficult. Transportation, logistics, moving heavy objects from one place to another, discovering the next drug—all of these require an understanding of atoms, molecules, and proteins. These are huge and incredible industries that artificial intelligence has not yet impacted.

Lauren Goode: You mentioned Moore's Law. Is this law now irrelevant?

Jensen Huang: Moore's Law is now more of a system issue rather than a chip issue. It's more about the interconnectivity of multiple chips. About 10, 15 years ago, we started the journey of decomposing the computer so that you can connect multiple chips together.

Lauren Goode: This was the original intention behind your acquisition of the Israeli company Mellanox in 2019. Nvidia said at the time that modern computing posed huge demands on data centers, and Mellanox's networking technology would make accelerated computing more efficient.Jensen Huang: Yes, absolutely correct. We acquired Mellanox so that we can scale our chips, turning the entire data center into a super chip, thereby creating a modern AI supercomputer. This is actually about recognizing that Moore's Law is over. If we want to continue scaling computing, we have to do it at the data center scale. We studied how Moore's Law was formulated, and then said, "Don't be limited by it. Moore's Law is not the limit of computing." We must abandon Moore's Law so that we can consider new ways of scaling.

Lauren Goode: Mellanox is now considered a very wise acquisition for Nvidia. Two years ago, you tried to acquire one of the world's most important chip IP companies, Arm, but were thwarted by regulatory agencies.

Jensen Huang: That would've been wonderful!

Lauren Goode: I'm not sure if the U.S. government agrees, but yes, let's establish that. When you consider acquisitions now, what specific areas are you focusing on?

Jensen Huang: The operating systems for these large systems are extremely complex. How do you create an operating system in the computing stack to coordinate tens of millions, hundreds of millions, or even billions of microprocessors in GPUs? It's a very difficult problem. If there are teams outside our company that do this, we can collaborate with them, or we can do more.

Lauren Goode: Does your implication mean that for Nvidia, having an operating system and building it into a platform is crucial?

Jensen Huang: We are already a platform company.

Lauren Goode: The more you become a platform, the more problems you face. People tend to take more responsibility for the output of the platform. How autonomous vehicles behave, the margin of error for medical devices, and whether AI systems have biases. How do you address this issue?

Jensen Huang: However, we are not an application company. This may be the simplest way to think about it. We will do our best, but serve an industry as little as possible. So, in terms of healthcare, drug discovery is not our expertise, computing is. Manufacturing cars is not our expertise, but making extremely good AI computers for car manufacturing is our expertise. Frankly, it's difficult for a company to be good at all these things, but we can be very good at the AI computing part.

Lauren Goode: Last year, there were reports that some of your customers had to wait for months for your AI GPUs. What is the situation now?Lauren Goode: What is the current wait time?

Jensen Huang: I'm not sure what the delivery times are now. But, you know, this year is also the beginning of a new generation for us.

Lauren Goode: Are you referring to Blackwell, your rumored new GPU?

Jensen Huang: Yes, the new generation of GPUs is coming, and Blackwell's performance is beyond imagination. It will be incredible.

Lauren Goode: Does this mean customers will need fewer GPUs?

Jensen Huang: That's the goal. The goal is to greatly reduce the cost of training models. Then people can scale up the models they want to train.

Lauren Goode: NVIDIA has invested in a lot of AI startups. There were reports last year that you invested in more than 30 startups. Will these startups be in a long line to buy your hardware?

Jensen Huang: They face the same supply constraints as everyone else because most of them use public clouds, so they have to negotiate with public cloud service providers themselves. However, what they do get is our AI technology, which means they can use our engineering capabilities and our special technology for optimizing their AI models. We make them more efficient. If your throughput increases by five times, you actually get the equivalent of five more GPUs. That's what they get from us.

Lauren Goode: Do you see yourself as a kingmaker in this regard?Jensen Huang: No. We invest in these companies because their work is incredible. Being able to invest in them is our honor, not the other way around. They are some of the smartest people in the world. They don't need us to support their credibility.

Lauren Goode: What happens when machine learning shifts more towards inference rather than training (basically, if the computational intensity of AI work decreases)? Will this reduce the demand for GPUs?

Jensen Huang: We love inference. In fact, I would say that if I guess, Nvidia's business today is probably 70% inference, 30% training. This is a good thing because then you will realize that artificial intelligence has finally succeeded. If Nvidia's business is 90% training and 10% inference, you might say that artificial intelligence is still in the research stage. It was like that seven or eight years ago. But today, every time you input a prompt in the cloud, it generates some content - it can be a video, an image, 2D, 3D, text, graphics - it is most likely that there is an Nvidia GPU behind it.

Lauren Goode: Do you think the demand for AI GPUs will weaken at any time?

Jensen Huang: I think we are at the beginning of the generative AI revolution. Most of the computing done in the world today is still based on retrieval. Retrieval means you touch something on your phone, and it sends a signal to the cloud to retrieve a piece of information. It may compose a response with something different and present it on your phone's beautiful screen using Java. In the future, computing will be more based on RAG (Retrieval-augmented generation: a framework that allows large language models to extract data outside their usual parameters), its retrieval part will be less, and the personalized generation part will be much higher.

That generation will be completed by GPUs. So I think we are at the beginning of this retrieval-augmented generation computing revolution, and generative AI will become an indispensable part of almost everything.

Lauren Goode: The latest news is that you have been working with the US government to develop chips that can be shipped to China in compliance with sanctions. My understanding is that these are not the most advanced chips. How closely are you cooperating with the government to ensure that you can still do business in China?

Jensen Huang: Well, this is actually export control, not sanctions. The United States has determined that Nvidia's technology and artificial intelligence computing infrastructure are of strategic importance to the country and will implement export controls on it. We first complied with export control in August 2022.Jensen Huang: Yes. And the United States has added more clauses to export controls in 2023, which has forced us to redesign our products once again. So we did just that. We are developing a new set of products that comply with today's export control regulations. We are working closely with the government to ensure that our proposed solutions align with their thoughts.

Lauren Goode: How concerned are you that these restrictions will stimulate China to launch competitive artificial intelligence chips?

Jensen Huang: China has some competitive things.

Lauren Goode: Correct. It's not yet at the data center scale, but the Huawei Mate 60 smartphone launched last year has garnered some attention due to its self-developed chips.

Jensen Huang: Really, really good company. They are limited by the semiconductor processing technology they possess, but they have still been able to build very large systems by aggregating many chips together.

Lauren Goode: Are you generally concerned about China's ability to catch up with the United States in the field of generative artificial intelligence?

Jensen Huang: The regulations will limit China's ability to access the most advanced technology, which means that the Western world, that is, countries not subject to export control restrictions, will be able to access better technology and develop at a quite fast pace. So I think these restrictions have imposed a significant cost burden on China. Technically, you can always aggregate more chips to make systems to get the job done. But this will only increase the unit cost of these products. This might be the simplest way to think about it.

Lauren Goode: The fact that you are manufacturing compliant chips to continue selling in China, does this affect your relationship with TSMC (the pride and joy of Taiwan's semiconductor industry)?

Jensen Huang: No. The regulations are specific. It's no different than speed limits.

Lauren Goode: You have said many times that among the 35,000 components in your supercomputer, 8 come from TSMC. When I heard this, I thought it must be a very small part. Are you downplaying your dependence on TSMC?Jensen Huang: Not at all. Not in the slightest.

Lauren Goode: So what point are you trying to make?

Jensen Huang: I'm just emphasizing that to build an AI supercomputer, there are many other components involved. In fact, in our AI supercomputers, almost the entire semiconductor industry is working with us. We have been working closely with Samsung, SK Hynix, Intel, AMD, Broadcom, Marvell, and others. When we succeed in our AI supercomputers, a bunch of companies will also succeed with us, and we are happy about that.

Lauren Goode: How often do you speak with TSMC's Morris Chang or Mark Liu?

Jensen Huang: Every moment. Constantly. Yes. Constantly.

Lauren Goode: What are your conversations like?

Jensen Huang: These days we talk about advanced packaging, planning for capacity in the coming years, advanced computing power. CoWoS [TSMC's proprietary method of stuffing chip chips and memory modules into a single package] requires new factories, new production lines, and new equipment. So their support is really, really important.

Lauren Goode: I recently spoke with a CEO who focuses on generative AI. I asked who Nvidia's competitors might be, and this person suggested Google's TPU. Others mentioned AMD. I don't think this is a binary issue for you, but who do you think is your biggest competitor? Who keeps you up at night?

Jensen Huang: Lauren, they all do it. The TPU team is excellent. Most importantly, the TPU team is really great, the AWS Trainium team and the AWS Inferentia team are really outstanding, very good. Microsoft is developing an internal ASIC called Maia. Every cloud service provider in China is building internal chips, and there are a bunch of startups as well as existing semiconductor companies building great chips. Everyone is building chips.

This should not keep me up at night—because I should make sure I am already exhausted from work to the point that no one can keep me up at night. This is indeed the only thing I can control.What surely wakes me up in the morning is that we must continue to honor our commitment, which is to say, we are the only full-stack company in the world where anyone can collaborate to build a data center-scale artificial intelligence supercomputer.

Lauren Goode: I have some personal questions I'd like to ask you.

Jensen Huang: [Huang says to the PR representative] She's done her homework. Not to mention, I'm just enjoying this conversation.

Lauren Goode: I'm glad. I am too. I do want to—

Jensen Huang: By the way, when TSMC or someone I've known for a long time asks me to be the interviewer, the reason is that I won't sit there and interview them by asking questions. I'm just having a conversation with them. You have to have empathy for the audience and what they might want to hear.

Lauren Goode: So I asked ChatGPT a question about you. I wanted to know if you have any tattoos because I was thinking of proposing a tattoo for you at the next party.

Jensen Huang: If you get a tattoo, I'll get one too.

Lauren Goode: I already have one, but I've been looking to expand.

Jensen Huang: I have one too.

Lauren Goode: Yes. This is what I learned from ChatGPT. It's said that Jensen Huang got the company logo tattooed when the stock price hit $100. Then it says, "However, Huang indicated that he is unlikely to get another tattoo, noting that the pain was more intense than he anticipated." It's said that you cried. Did you cry?Jensen Huang: Just a little bit. My suggestion is, before you do this, you should have a shot of whiskey. Or take an Advil. I also think women can take more pain because my daughter has a rather large tattoo.

Lauren Goode: So, if you were to get a tattoo, I think a triangle might be nice, because who doesn't like triangles? They are perfect geometric shapes.

Jensen Huang: Or the silhouette of the Nvidia building! It's made up of triangles.

Lauren Goode: That's a commitment. I'm curious, how frequently do you personally use things like ChatGPT or Bard?

Jensen Huang: I've been using Bard, and I also like ChatGPT. I use both almost every day.

Lauren Goode: For what?

Jensen Huang: Research. For example, computer-aided drug discovery. Maybe you want to understand the latest advancements in computer-aided drug discovery. So you want to construct the entire subject so that you can have a framework, and from that framework, you can ask more and more specific questions. I really enjoy these large language models.

Lauren Goode: I heard you used to lift weights. Do you still do that?

Jensen Huang: No, I try to do 40 push-ups every day. It doesn't take more than a few minutes. I'm a lazy exerciser. I do squats while brushing my teeth.

Lauren Goode: Recently, you commented on the Acquired podcast, which has quickly become popular. The hosts asked what you would start if you were 30 years old today and considering starting a company. You also said that you wouldn't start a company at all. Do you have any modifications to that?Jensen Huang: This question can be answered in two ways, and here's how I would answer it: If I had known all the things I know now back then, I would have been too scared to do it. I would have been too afraid. I wouldn't have done it.

Lauren Goode: You have to have a certain level of delusion to start a business.

Jensen Huang: That's the advantage of ignorance. You don't know how hard it's going to be, you don't know how much pain and suffering it will involve. These days, when I meet entrepreneurs who tell me how easy it will be, I am very supportive of them, and I actually don't try to burst their bubble. But deep down, I know, "Oh my God, things are not going to be as they think."

Lauren Goode: What do you think has been the biggest sacrifice you've had to make in running Nvidia?

Jensen Huang: Other entrepreneurs make the same sacrifices. You work really, really hard. For a long time, no one thinks you will succeed. Only you believe you will succeed. Insecurity, vulnerability, and sometimes even humiliation, these are all real. No one talks about it, but it's all true. CEOs and entrepreneurs are human just like everyone else. When they publicly fail, it's embarrassing.

So when someone says, "Jensen, with everything you have today, wouldn't you start it?" It's like, "No, no, no, of course not."

In fact, if I knew Nvidia would become what it is today, and you ask me would I start this company? Are you kidding? I would sacrifice everything to do this.