Jensen Huang's drunken remarks

On the evening of February 3rd, U.S. time, a top-tier technology dialogue that should have been serious ultimately turned into a "confession session after five drinks."

Just back from a trip to China and still adjusting to the time difference, NVIDIA CEO Jensen Huang sat across from Cisco CEO Chuck Robbins.

After a few drinks, Huang's voice began to grow hoarse, but his words became sharper. Fueled by alcohol, Huang not only "smashed" programmers' jobs, "confronted" management dogmas, but also took a swipe at several world-class giants:

About programmers: "Programming? That's just typing. Typing is no longer valuable."

About control issues: "If you want to control innovation, you should see a psychologist."

About Moore's Law: "Computing power has increased a million times in ten years; at this speed, the old Moore's Law seems as slow as a snail crawling."

About traditional giants: "I love Disney, but I'm sure they want to be more like Netflix; I love Mercedes, but I'm certain they want to be more like Tesla."

About AI evolution: "Why should people adapt to tools? Let AI learn to use tools, and we can create a true 'digital workforce.'"

Huang used these "explosive remarks" to remind all leaders: in the face of exponential evolution, the experience you once prided yourself on will inevitably be ruthlessly eliminated by the times.

"Uncontrolled" Blooming: Your First Lesson is Not ROI, but "Letting Go"

When Robbins asked what the first step for companies moving towards AI should be, Huang's answer bypassed all conventional business rhetoric. "I am often asked about return on investment, but I won't talk about that first," he said bluntly. In his view, during the dawn of a technological explosion, using spreadsheets to define value is futile, even dangerous, as it only stifles exploratory efforts.

He cited NVIDIA's internal practices as an example: allowing "blooming." Within NVIDIA, AI projects are so numerous that they are almost out of control. "Notice what I just said: out of control, but fantastic," Huang emphasized. His explanation was filled with philosophical meaning: "Innovation is not always under control. If you want to control everything, you should first consult a psychologist; secondly, that's an illusion, you can't control it at all."

His management logic is surprisingly simple, treating the team's curiosity like that of children. "When one of our teams says they want to try a certain AI, my first response is 'yes,' and then I ask 'why.' I don't ask why first and then agree; I agree first and then ask why." He compared this to how we never require proof at home, but do so at work, which makes no logical sense to him.

The roadmap Huang described is clear yet counterintuitive: the first step is not to formulate a rigorous pilot plan, but to allow and even encourage "safe trial and error." Let every team with an idea engage with and try various AI tools, whether it's Anthropic, Codex, or Gemini The purpose is not to generate immediate benefits, but to cultivate the organization's "AI sense." Only after a sufficiently long and somewhat chaotic period of flourishing can leaders, through intuition and observation, know when to start "pruning the garden" and concentrate resources on truly important directions. "But you can't concentrate your efforts too early, or you'll choose the wrong arrows," he warned.

From "Screwdriver" to "Creating Labor": The Essence of AI Factory is Value Transfer

So, when companies begin to explore, where should they look? Jensen Huang used the concept of "AI factory" to depict a vision far grander than merely improving efficiency. He believes we are undergoing a fundamental shift from "manufacturing tools" to "creating labor."

"The industry that Chuck and I are in has always been about making tools, always in the business of screwdrivers and hammers." For decades, tech companies have produced software, chips, and networking equipment, all of which are tools, extensions of efficiency in the atomic world.

But AI, especially physical AI that can understand the physical world and possesses causal reasoning abilities, will change the game. "For the first time in history, we are going to create what people call 'labor' or 'augmented labor.'" He cited the example of autonomous vehicles, which are essentially digital drivers, and the economic value of this digital driver over its lifecycle will far exceed that of the car itself as hardware.

This is the deeper meaning of the "AI factory": it is not a server room storing servers, but a new type of value creation center that continuously produces "digital labor." This labor can be an ever-tireless customer service representative, a dispatcher optimizing the supply chain in real-time, or an engineering assistant capable of collaborative design.

Jensen Huang provided a shocking numerical comparison: the global IT industry is about $1 trillion, while the total global economy is $100 trillion. "For the first time, we are facing a potential market size that has expanded a hundredfold." This means that the greatest opportunity brought by AI is not to divide the existing IT budget but to penetrate and reshape the remaining $99 trillion of the real economy.

Every industry has the opportunity to reshape itself into a tech company by injecting this digital labor. "I believe Disney would rather be Netflix, Mercedes would rather be Tesla, and Walmart would rather be Amazon." He pointedly noted that all of you are like this.

"Infinite Speed" and "Zero Gravity": Redefining Problems with AI Thinking

How can we truly seize this trillion-dollar opportunity? Jensen Huang proposed a disruptive thinking model: think everything through the lens of the assumption of "Abundance."

He sarcastically remarked that in the AI era, the old Moore's Law seems to crawl like a snail. Now, we need to establish new cognitive benchmarks. "In the past 10 years, how much has our computing power increased? 1 million times in 10 years." Under this premise of exponential abundance, the thinking of business leaders must upgrade "Now, when I imagine an engineering problem, I assume my technology, my tools, and my spacecraft are infinitely fast. How long does it take to get to New York? One second." He heuristically asked, if you could reach New York in one second, what different things would you do? If something that used to take a year can now be done in real-time, what different things would you do? If something very heavy in the past now has no gravity, how would you handle it?

He asked managers to apply this assumption of infinite speed and zero gravity to the company's core and most challenging problems. For example, when faced with a complex network analysis with trillions of connections, the past approach was to break it down. "Now it's: give me the whole graph, no matter how big, I don't care." This logic is being applied everywhere. If you don't apply this logic, you're doing it wrong.

This is no longer incremental optimization, but rather redefining the boundaries of the problem itself with the possibilities of technological abundance. He warned that if your competitors or a startup challenge you with this way of thinking, they will fundamentally change the rules of the game.

Your "questions" are worth more than answers: Sovereign AI and the future core of companies

After envisioning a ubiquitous digital workforce, Jensen Huang brought the topic back to a more realistic and hidden concern: data sovereignty and core intellectual property.

Regarding whether companies should fully rely on public clouds or build their own AI capabilities, his advice is akin to teaching a child to ride a bike. "Build one yourself. Even though PCs are everywhere, go ahead and build one, and figure out why these components exist." He believes that companies must have "a personal grasp of technology."

More importantly, he raised a sharp point: the most valuable intellectual property of a company may not be the answers stored in databases, but the "questions" generated during employee interactions with AI. "I am not comfortable putting all of Nvidia's conversations in the cloud because, for me, core intellectual property is not the answers, but my questions."

Jensen Huang explained: "My questions are my most valuable IP. What I am thinking reflects that. Answers are cheap. If I know what to ask, I have locked in the focus. I don't want others to know what I think is important."

Therefore, he believes that conversations involving strategic thinking must take place in a controlled local environment. He described that in the future, every employee will have many AI assistants that continuously learn from the employee's decisions and questions, and ultimately, these evolved AIs will become the unique intelligent assets accumulated by the company. "This is the future of the company; it will capture our life experiences."

Five-layer cake and the end of the "pre-recorded" era: A fundamental revolution from retrieval to generation

When discussing specific implementation paths, Jensen Huang traced back a cognitive revolution that has lasted for 15 years.

He traced back to the first contact moment represented by AlexNet and concluded that most of the world's problems do not have precise physical laws to follow, and the answers often depend on specific contexts. This type of context-dependent problem is precisely the area where AI can shine The real turning point is the breakthrough in self-supervised learning, which has caused parameters to explode from hundreds of millions to trillions. He asserted: “We will reshape computing from the ground up. Computing will shift from explicit programming to a whole new paradigm, which is learning software through models.”

Immediately following, Jensen Huang used a clever metaphor: “We are moving from the ‘pre-recorded’ era to the ‘generative’ era.”

“The reason past software was ‘pre-recorded’ is that it was stored on CD-ROMs.” In the old paradigm, software was like a burned disc, and user interaction was essentially retrieval. Future software will be highly contextualized. “Every scenario is different, every user, every prompt, and every background is different. Every instance of software is unique.”

Future applications will dynamically generate unique responses, interfaces, and even functionalities based on real-time context, intent, and background. This is the essence of generative; the construction logic of traditional hardware, frameworks, and model layers has changed.

The Implicit Programming Revolution: When “typing” becomes a universal skill, your expertise is the ace up your sleeve

In this profound transformation, the value of industry knowledge is skyrocketing. “From explicit programming to implicit programming, you only need to tell the computer what you want, and the computer will write the code.”

He pointed out that the computing paradigm centered on writing precise code, which has lasted for 60 years, is coming to an end. “Because it turns out that writing code is just typing, and typing has become commonplace.”

This means that the threshold for technical ability will be greatly lowered. In contrast, those who are well-versed in business but do not understand technology will stand at the forefront of the wave. “Fresh graduates in computer science may be great at coding, but they don’t know what the customers want. You do. The coding part is simple; let AI handle it. Understanding customers and understanding the domain expertise of the problem is your superpower.”

This conversation concluded with gratitude towards Cisco, but the core message left behind is crystal clear: the AI revolution is not an upgrade for the IT department, but a “reset” of business logic.

As night fell, the conversation ended with jokes about skewers and chips. But the truths that Jensen Huang revealed in his tipsy state, like the digital workforce he predicted, have already begun to silently permeate, ready to reshape everything we know.

【The following is the full transcript】

Robbins: Hey, hey, hey. Yes, everyone stay there and don’t move.

Jensen Huang: I feel like I’m in a state of drinking while working.

Robbins: When we brought the drinks up just now, Jensen Huang reminded me. He said, “You realize you’re live streaming this, right?” Hey, who cares, it’s already late.

Jensen Huang: The first principle is: do no harm and be aware of how lucky you are.

Robbins: First of all, thank you all for spending a long day here. We started early this morning, with speakers one after another, and after about two and a half hours of break, everyone is back to see you Jensen Huang: I woke up at 1 AM today.

Robbins: So, this guy is at the end of a two-week journey across four or five cities in Asia.

Jensen Huang: Just a day ago, I was in Taiwan, last night in Houston, and now I'm here.

Robbins: He has been out for two weeks, and now we are interrupting his way back to his warm bed at home; he has had enough of hotels. So, we’ll have some fun and then let him go. Although you don’t need much introduction, thank you for being here, man. We really appreciate it.

Jensen Huang: Thank you for our partnership, and I’m proud of you all.

Robbins: Let’s start by talking about the partnership. We’ve always collaborated, and you introduced the whole concept of the "AI factory." We are working together to advance it, although it may not be as fast as we thought in the enterprise space. Can we first talk about what "AI factory" means to you?

Jensen Huang: First, remember that we are in the process of reshaping computing for the first time in 60 years. In the past, it was "explicit programming," right? We wrote programs and variables, passing them through APIs, all very explicit; now we are shifting to "implicit programming."

You now tell the computer your intent, and it figures out how to solve your problem. From explicit to implicit, from general computing (essentially computation) to artificial intelligence, the entire computing stack has been reshaped. Nowadays, when people discuss computing, they focus on the processing layer, which is where we are. But remember the components of computing. Besides computation and processing, there are storage, networking, and security. All of this is being reshaped right now. The first part is that we need to elevate AI to a level— we will discuss this later— that is useful to humans. Until now, the kind of chatbots where you give a prompt and they find a way to tell you the answer, while interesting and curious, are not truly useful.

Robbins: Sometimes it helps me finish crossword puzzles.

Jensen Huang: Yes, but only limited to what it remembers and generalizes. If you look back, it was just three years ago when ChatGPT first appeared, and we were amazed at how it could generate so many phrases and create Shakespearean works. But that was all based on its memory and generalization. We know that the core of intelligence is problem-solving. The problem-solving part involves knowing what you don’t know and reasoning how to solve problems you’ve never seen before. Breaking down problems into known and easily solvable elements, combining them to solve unseen problems, and formulating strategies (which we call "planning") to execute tasks, seek help, use tools, conduct research, and so on.

These are all fundamental aspects of the current term "agentic AI," including tool use, research, retrieval-augmented generation (RAG), memory, etc. You will hear these when discussing agentic AI. But the most important point is to evolve from this "explicit programming" of general computing (the era we wrote in Fortran, C, C++) Robbins: Right.

Jensen Huang: That's a good thing, Robbins.

Robbins: That's my safety net job.

Jensen Huang: That's a very good skill, and those skills are still valuable.

Robbins: I know.

Jensen Huang: They are still valuable.

Robbins: I've received a lot of acceptance letters.

Jensen Huang: "Dinosaurs" are always valuable. We just confirmed that you are older than me. I know I am from that "prehistoric era," even though I don't look like it, but it's true. Well, this is quite interesting. I might be the oldest person in this room.

Robbins: Let's talk about this. Jensen Huang, when you think about the future.

Jensen Huang: We're right here. I went to Robbins and said, "Hey, listen, we need to reshape computing, and Cisco has to be part of it." We have a whole new computing stack about to launch, called Vera Rubin. Cisco will synchronize with us to go to market. There's also the network layer, where Cisco will integrate our AI networking technology and put it into Cisco's Nexus control plane. From your perspective, you'll get all the performance of AI while retaining Cisco's controllability, security, and manageability. We'll do the same in the security domain. Every pillar must be reshaped for enterprise computing to benefit. We'll talk about this later: why enterprise AI wasn't ready three years ago and why now you have no choice but to get involved as soon as possible. Don't fall behind. You don't have to be the first company to use AI, but don't be the last.

Robbins: Exactly. So if you are a business owner today, what are the first, second, and third steps you would recommend they take to start preparing?

Jensen Huang: I often get asked about return on investment (ROI) questions, and I won't talk about that first. The reason is that in the early stages of any technology deployment, it's hard to calculate the ROI of new tools and technologies using Excel. What I would do is figure out what the essence of my company is. What is the most impactful work our company does? Don't waste time on those marginal, secondary things. In our company, we let "a thousand flowers bloom." We have so many different AI projects internally that it's almost out of control, and that's great. Notice what I just said: out of control and great. Innovation is not always under control. If you want to control everything, you should first consult a psychologist. Secondly, that's an illusion; you can't control it. If you want your company to succeed, you can't control it; you can only influence it.

I hear too many companies wanting clear, specific, and provable ROI, but it's very difficult to demonstrate the value of something in the early stages. I suggest letting a thousand flowers bloom, letting people experiment, and letting people experiment safely. Our company experiments with all sorts of things; we use Anthropic, we use Codex, we use Gemini, we use everything. When one of our teams says they want to try a certain AI, my first response is "yes," and then I ask "why." I don't ask why first and then agree; I agree first and then ask why The reason is that my expectations for the company are the same as for my children: to explore life. When they say they want to try something, the answer is "yes," and then you ask why. You wouldn't say, "Prove to me first that doing this will lead to financial success or future happiness; otherwise, I won't let you do it." We never do that at home, but we do it at work.

Robbins: Do you understand what I mean?

Jensen Huang: Yes. It makes no sense to me. So the way we treat AI, just like the previous internet or cloud technology, is to let a hundred flowers bloom. Then at some point, you need to use your judgment to decide when to start "tidying the garden." Because letting a hundred flowers bloom can make the garden messy. But at some point, you have to start organizing and find the best methods or platforms to focus your efforts on doing big things (put all your wood behind one arrow). But you can't concentrate too early.

Robbins: Otherwise, you might choose wrong.

Jensen Huang: Exactly. So first, let a hundred flowers bloom, and then tidy up at some point. Based on the current progress, I haven't started tidying up yet; I'm still letting flowers bloom everywhere. But I certainly know what is most important for our company. I ensure that there is a lot of expertise and capability focused on leveraging AI to revolutionize those core tasks, which in our case are chip design, software engineering, and systems engineering. You may have noticed that we are collaborating with Synopsys, Cadence, Siemens, and today Dassault Systèmes, so we can inject our technology. Whatever they want, I will provide it to revolutionize the tools we use to design products.

We are using Synopsys, Cadence, Siemens, and Dassault everywhere. I will ensure they receive 1000% support, so I have the tools needed to create the next generation of products. This reflects my attitude towards reshaping our work. Think about what AI has done: it has reduced the cost of intelligence by orders of magnitude, or created vast amounts of intelligence. In other words, what used to take one unit of time to work on now might only take a day; what used to take a year might now take an hour, or even be done in real-time.

In a world of "abundance," Moore's Law is moving at a snail's pace. Remember, Moore's Law doubles every 18 months, tenfold in five years, and a hundredfold in ten years. But where are we now? A hundred million times in ten years. We have pushed AI so far in the past decade that engineers are saying, "Hey, guess what? Let's just train an AI model using all the data in the world." They are not talking about collecting data from my hard drive, but pulling data from all over the world. That is the definition of "abundance." Its definition is: when you see a huge problem, you say, "Forget it, I want to solve it all." I want to cure every disease in every field, not just cancer. We are directly addressing all human suffering. That is "abundance." Now, when I imagine an engineering problem, I assume my technology, my tools, and my spacecraft are infinitely fast. How long does it take to get to New York? One second. If you could get to New York in one second, what different things would you do? If something that used to take a year can now be done in real-time, what different things would you do? If something that used to be very heavy now has no gravity, how would you handle it? When you approach everything with this mindset, you are applying "AI logic." Does that make sense? For example, many of the companies we collaborate with have dependencies, nodes, and edges in their graphs that number in the trillions. In the past, you would process the graph piece by piece; now it's: "Give me the whole graph, no matter how big, I don't care." This logic is being applied everywhere. If you don't apply this logic, you're doing it wrong. Is speed important? Not at all, because you are at the speed of light. Is quality important? Zero weight, zero gravity. If you haven't applied this logic to think about the hardest problems in your company, you haven't done it right. This is the mindset of all of them. If you don't think this way, just imagine your competitors thinking this way. Imagine a startup thinking this way; it would change everything. So, go find the most impactful work in your company, apply the concepts of "infinity," "zero," and "light speed" to it, and then ask Robbins how to achieve it.

Robbins: Just give it to me.

Jensen Huang: We will achieve it together.

Robbins: You have an analogy of a "five-layer cake" because everyone is talking about infrastructure, models, and applications. What should I do? Talk about this.

Jensen Huang: One thing successful people do is reason about what is happening. About 15 years ago, an algorithm allowed two engineers to solve a computer vision problem. Vision is the first part of intelligence: perception. Intelligence consists of perception, reasoning, and planning. Perception is: "What am I? What happened? What is my context?" Reasoning is: "How do I reason based on my goals?" The third is to make a plan to achieve the goals. Just like the fighter jet problem, perceive, locate, and then act. Without perception, there is no second and third part. If you don't understand the context, you don't know what to do. And context is highly multimodal; sometimes it's a PDF, sometimes it's a spreadsheet. Sometimes it's sensory and smells, like where we are, what we are doing, who the audience is, how to read the atmosphere in the room, and so on.

About 13 or 14 years ago, we made a huge leap in the field of computer vision. AlexNet was the first breakthrough we saw; it was like "First Contact," I love that movie. That was our "first contact" with AI. So we thought: what does this mean? How could two engineers defeat algorithms that we had all been studying for 30 years? I just talked with Ilya Sutskever yesterday, and Alex Krizhevsky. How could two young people with a few GPUs solve this problem? What does this mean? Ten years ago, I broke this down and reasoned that the conclusion is: most valuable problems in the world actually do not have fundamental physical algorithms. There is no F=ma, no Maxwell's equations, no Schrödinger equation or Ohm's law, no thermodynamic laws; it's not that precise. Those valuable things we call "intuition" and "wisdom," like the problems you and I encounter, the answer is often: "It depends." If the answer were always 3.14, that would be great, but valuable problems in life often depend on context and environment.

Since visual problems have been solved, we reasoned that through deep learning, this is not only scalable, but you can also make the models larger and larger. The only problem we needed to solve at that time was how to train the models, and the huge breakthrough was "self-supervised learning." AI began to learn on its own. Note that today we are no longer limited by manually labeled data, completely unrestricted. This opened the floodgates, allowing models to scale from hundreds of millions of parameters to hundreds of billions, trillions. The knowledge we can encode and the skills we can learn have exploded. We will reshape computing from the ground up. Computing will shift from "explicit programming" to a whole new paradigm, which is learning software through models. What does this mean for the computing stack? What does it mean for software development? What does it mean for engineering organizations, product marketing, QA teams? What will these products look like in the future? How do we deploy? How do we keep them updated? How do we patch the software? I asked thousands of questions about the future of computing and concluded: this will change everything. So we turned the entire company around. Simply put, we came from a world where everything was "pre-recorded."

Robbins: Indeed, it's amazing stuff.

Jensen Huang: It's been running for a long time. I solemnly declare that those were indeed described in Hebrew (humorously hinting at the old technology's antiquity).

Robbins: Indeed. That's another skill.

Jensen Huang: After all, you are probably the only one in this room who is proficient in both Hebrew and COBOL at the same time. Anyway, that was pre-recorded. We write algorithms, describe thoughts, and then pair them with data. Everything is pre-recorded. The reason past software was pre-recorded is that it was stored on CD-ROMs, right?

What is modern software? It is contextualized; every context is different, every user, every prompt, every background is different. Every instance of software is unique. The past pre-recorded software was called "retrieval-based." When you tap on your phone, it retrieves some software, files, or images and presents them to you. In the future, everything will be "generative," just like it is now. This conversation has never happened before. The concepts existed, the context existed, but the order of every sentence is brand new. The reason this is happening is obvious because we've had four drinks.

Robbins: And cold brew coffee. Cobol, Hebrew, thank goodness this isn't on campus, nor is it live Jensen Huang: Hmm.

Robins: Does everyone understand what you're saying?

Jensen Huang: Did you all get it? Robins only fed me four drinks today.

Robins: To be fair, I only gave you one drink; the other three you took from the buffet.

Jensen Huang: I was eyeing that food the whole time; I was so hungry. The food was about 40 feet away from me.

Robins: That's because you were busy taking pictures.

Jensen Huang: It was really close, but I tried to go over several times and was pushed back.

Robins: Do you know what happened? Your team told us in advance that if he drinks three cups, he’s in the best state. If he drinks the fourth cup, things are going to fall apart.

Jensen Huang: That state is not ideal. Alright, listen, we need to leave some wisdom behind. Can I get another drink?

Robins: This isn’t Dave Chappelle’s stand-up.

Jensen Huang: Let’s talk about something else. Energy.

Robins: Chips.

Jensen Huang: Energy sounds good. Energy, chips, infrastructure (hardware and software), and then AI models. But the most important part of AI is the application. Every country, every company, the layers underneath are just infrastructure. What you need to do is apply this technology. Swear to God, go apply this technology. Companies using AI won’t be in danger. You won’t lose your job because of AI; you’ll lose your job because of someone using AI. So, take action; that’s the most important thing.

Robins: And call Robins as soon as possible.

Jensen Huang: You call me, I’ll call him.

Robins: We don’t have much time.

Jensen Huang: We have plenty of time. Robins charges by the hour, and I don’t even wear a watch. I won’t leave until value is delivered. If it takes all night, I’ll keep torturing all of you.

Robins: Jensen Huang, that’s why people like me need a watch. Can you talk about “physical AI”?

Jensen Huang: The idea that the software industry is declining and will be replaced by AI is the most illogical thing in the world. Let’s do the ultimate thought experiment: suppose we are the ultimate AI—a physical version of a general robot. Since you are a humanoid robot, you can solve any problem. Would you use an off-the-shelf screwdriver or invent a new one? I’d use the off-the-shelf one directly. Would you use an off-the-shelf chainsaw or recreate one? The obvious answer is to use the tool. Given that, let’s look at the digital version. If you are AGI (Artificial General Intelligence), would you use ServiceNow, SAP, Cadence, Synopsis, or reinvent a calculator? Of course, you’d use the calculator directly. Why teach AI to use tools? Because existing tools are ‘deterministic’. Many problems in the world have standard answers, like Newton's second law F=ma. You don’t need AI to give you a fuzzy answer that is probabilistically close to ma; ma is ma. Another example is Ohm's law V=IR, which is absolute in science, not ‘statistically IR’. So, we want AI to pick up these precise tools directly like humans do, rather than guessing on things that already have standard answers We hope AGI uses tools, which is a big logic. The next generation of physical AI will understand the physical world and causal relationships. If I knock this over, it will knock everything down. They understand the concept of "dominoes." Every child can understand what it means to knock it over; this combination of causality, contact, gravity, and mass is very profound. Large language models currently do not have this concept, so we need to create physical AI.

Where is the opportunity? So far, the industry that Robbins and I are in has been about making tools. We have been in the business of "screwdrivers and hammers." For the first time in history, we are going to create what people call "labor force," or "augmented labor force." For example, what is an autonomous vehicle? It is a digital driver. The value of a digital driver far exceeds that of the car itself. For the first time, we are facing a potential total addressable market (TAM) that is a hundred times larger. The global IT industry is about one trillion dollars, while the global economy is one quadrillion. All of you have the opportunity to become a technology company by applying this technology. I believe Disney would rather be Netflix, Mercedes would rather be Tesla, and Walmart would rather be Amazon. Do you agree? Did I get these three examples right?

Robbins: Yes. You all are like that.

Jensen Huang: We have the opportunity to help every company transform into a "tech-first" company. Technology is your superpower, and your industry domain is the application scenario. Why tech-first? Because that way you are dealing with "electrons" rather than "atoms." The value of atoms is limited by mass, while the value of electrons exploded a thousand times the moment CD-ROM turned into electronic streams. You need to become a technology company. Even Robbins, who only knows Hebrew programming, has this talent.

Robbins: This programming direction is written from right to left, quite clever.

Jensen Huang: Smart people do smart things. The wonderful thing is that your company's advantages are knowledge, intuition, and domain expertise. Now for the first time, you can explain to the computer what you want in your own language. From explicit programming to implicit programming, you just need to tell the computer what you want, and the computer will write the code. Because it turns out that writing code is just typing, and typing has become commoditized. This is your huge opportunity. You can all break free from the constraints at the physical (atomic) level and achieve a qualitative leap. We are no longer limited by the lack of software engineers because typing is cheap, and you have something extremely valuable—understanding customers and domain expertise in understanding problems.

A freshly graduated computer whiz may be great at writing code, but they don’t know what the customer wants or what problem to solve. You know. The coding part is simple; let AI do it. That is your superpower. This summary was made after I had five drinks.

Robbins: This is simply a miracle.

Jensen Huang: It is a great honor to collaborate with all of you. Cisco has profound expertise in the two pillars of computing invention—networking and security. Without Cisco, there is no modern computing. In the world of AI, both of these pillars have been reshaped. The computing part we excel at is commoditized in many ways, while what Cisco masters is extremely valuable Earlier, someone asked me: should I rent cloud services or build my own computer? My advice, which is the same I would give to my children, is to build one yourself. Even though PCs are everywhere, take the time to build one and understand why these components exist. If you're in the transportation industry, don't just use Uber; open the hood, change the oil, and understand it. This technology is too important for the future; you must have a tactile understanding of it. You might find that you have a great talent for it. You may discover that the world is not just about "renting" or "buying"; you need a part of it to be on-premises. For example, when it comes to sovereignty and private information, you wouldn't want to share all your "issues" with everyone. For instance, when you visit a psychologist, you wouldn't want your questions posted online.

Robbins: A hypothetical example, right?

Jensen Huang: Yes, hypothetical. So I believe many conversations and uncertainties should remain private. The same goes for companies. I'm not comfortable putting all of NVIDIA's conversations in the cloud, so we built a super AI system on-premises. Because for me, the most valuable core asset is not the answers, but my questions. Do you understand? My questions are my most valuable IP. What I'm thinking reflects in my questions. Answers are cheap. If I know what to ask, I've locked in the focus. I don't want others to know what I think is important. So I want to create my own AI in my little room, on-premises.

One last thought, it's already 11 o'clock. There used to be a viewpoint called "human in the loop," which is completely wrong. It should be "AI in the loop." Our goal is to make the company better, more valuable, and more knowledgeable every day. We don't want to regress or stand still. This means that if AI is in the loop, it will capture our life experiences. In the future, every employee will have many AIs in the loop, and these AIs will become the company's intellectual property. That's the future of companies. So, I think the wise thing to do is to call Robbins right away.

Robbins: I called Jensen Huang. After a two-week journey, Jensen flew here and spent the last night with us; it's the first time in a long while that he could sleep in his own bed. We are forever grateful.

Jensen Huang: Thank you very much. Also, I've been eyeing those skewers out of the corner of my eye.

Robbins: I hope they're still there.

Jensen Huang: Where's that bag of chips you promised me?

Robbins: Let's go eat. Thank you, everyone!

Risk Warning and Disclaimer

The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investing based on this is at your own risk