"At some point, you have to choose to believe in something."

Compiled & Edited by: TechFlow

Introduction by TechFlow: In this 63-minute deep interview, renowned tech creator Cleo Abram sidestepped the usual financial reports and stock price speculation to engage in a fundamental conversation about the evolution of underlying computation and the essence of humanity's technological future with Jensen Huang, co-founder and CEO of NVIDIA. From breaking ground with gaming graphics cards 30 years ago, to enduring over a decade of skepticism on Wall Street to build CUDA, and to astutely capturing the moment of AlexNet that ignited the deep learning revolution in 2012, Jensen Huang profoundly dissected how NVIDIA has gradually restructured the modern computing technology stack that has lasted for 65 years.

Faced with a transformative shift from generative AI to physical AI, Jensen Huang elaborated on the foundational technology of the "world model + physical simulation" constructed by Cosmos and Omniverse, depicting a vision where "everything has a robotic form" and everyone will have a lifelong companion in the form of an R2-D2 robot. At the same time, he firmly responded to the philosophical divergence regarding the specialization versus generalization of chip architecture, analyzing cutting-edge bets on digital biology and climate prediction, and offered core advice to the younger generation on embracing the art of questioning. When asked how he would like to be remembered in the future, he gave the simplest yet grand answer: "Very simply: They made an extraordinary impact."

Guest: Jensen Huang (Co-founder and CEO of NVIDIA)

Host: Cleo Abram (Renowned tech video creator, host of Huge If True channel)

Podcast Source: Huge If True (https://www.youtube.com/watch?v=7ARBJQn6QkM)

Original Title: NVIDIA CEO Jensen Huang's Vision for the Future

Broadcast Date: January 27, 2025

Summary of Key Points

• A perfect computer must balance serial and parallel computation: In most complex programs, about 10% of the core code accounts for 99% of the computational load, which is highly suitable for massive parallel acceleration by GPUs; the remaining 90% control logic is efficiently handled in serial by CPUs. The founding purpose of NVIDIA was to solve the ultra-high load problems that traditional computers could not tackle.
• The gaming industry is the best funding engine for underlying research and development: Real-time rendering of 3D virtual worlds inherently requires parallel computation, and more importantly, video games have an extremely large consumer market, capable of continuously funding billions of dollars in underlying software and hardware R&D without relying on government grants or research sponsorship.
• The essence of CUDA is to democratize parallel computing: Early researchers had to disguise physical or mathematical problems as graphics shading programs to call on GPUs. NVIDIA created CUDA to allow developers to directly invoke GPU computing power using the general-purpose C language. Although this gamble faced over a decade of Wall Street skepticism due to high R&D costs, it ultimately paved the way for the explosion of deep learning later on.
• AlexNet prompted NVIDIA to completely restructure its underlying computing stack: The AlexNet model created in 2012 by Ilya Sutskever, Alex Krizhevsky, and Geoffrey Hinton crushed traditional image classification algorithms with just two GeForce GTX 580 graphics cards. Jensen Huang acutely realized that software was shifting from "manually programmed rules" to "self-learning from data," leading NVIDIA to bet everything on AI and comprehensively restructure chips, interconnects, algorithm libraries, and system architecture.
• The core of generative AI is to become a cross-modal universal translator: This wave of AI is unprecedented because architectures like Transformers can comprehend and translate data from any modality (text, image, audio, 3D space, mechanical arm trajectories), breaking down the barrier between human knowledge and the digital world.
• Physical AI must be driven by both "world model + physical simulation": Large language models can hallucinate, requiring retrieval and context anchoring for real answers; in the physical world, robots must possess common sense of gravity, inertia, friction, and spatial constancy. NVIDIA launched Cosmos as the foundational model for the physical world, using the Omniverse simulation engine that adheres to Newton's physical laws to generate realistic synthetic data, allowing robots to complete millions of extreme condition drills before entering the physical world.
• In the next decade, everything will possess a robotic form: Whether autonomous vehicles, smart buildings, robotic lawnmowers, or humanoid robots, all moving objects in the future will exhibit robotic capabilities. Everyone will have their own lifelong digital companion (R2-D2), which will continue to evolve alongside the user, taking forms across glasses, phones, computers, cars, and home entities.
• Refuse the trap of specialized ASICs and uphold computational flexibility: In response to the idea of completely hardening Transformers into dedicated chips (ASICs), Jensen Huang believes innovation will never stop at any specific algorithm architecture. Solidifying computers for a single purpose is akin to turning them into microwaves. NVIDIA has always insisted on building flexible, programmable acceleration architectures that support academia's continued exploration of newer, more efficient attention mechanisms like FlashAttention and hierarchical attention.
• From a gaming company to the eye of the storm in human frontier science: In the RTX 50 series, in a 4K image of 8 million pixels, only 500,000 pixels were computed through traditional pipeline ray tracing; the remaining 7.5 million pixels were all accurately predicted and filled in by AI. This paradigm allows scientists and engineers to focus their precious energy on core key points, amplified by AI into superhuman productivity. From digital biology, protein language analysis, and microclimate prediction to future robots, NVIDIA's ultimate historical vision is to "create an extraordinary impact."

Notable Insights

• On technology belief and forward-looking layout: "At some point, you have to choose to believe in something. If you build it, people may not come; but if you don't build it, people will never have the chance to come. Our strategy is built on hope, but it is a rational hope that has been deeply thought out."
• On the division of labor between serial and parallel computing: "Perhaps only 10% of the code occupies 99% of the computational processing, and that 99% of tasks can be done in parallel; but the remaining 90% of the code must still be executed in order, serially. A truly perfect computer is not either-or, but excels at both serial and parallel processing capabilities simultaneously."
• On the driving force behind the birth of CUDA: "Many great innovations often emerge from the same complex mélange: part comes from lofty ambitions and inspirations, and the other part purely from urgent self-rescue in a desperate situation. On one hand, we saw researchers at Massachusetts General Hospital using gaming graphics cards to reconstruct CT images, and on the other, we faced immense challenges trying to render flowing water, explosions, and particle collisions in games, and these intertwined factors led to the birth of CUDA."
• On the specialization debate of computing architectures: "Will Transformer be the last AI algorithm architecture discovered by human researchers? Absolutely not. Looking back at the history of computer science, no algorithm has ever dominated everything forever. If you had completely solidified the algorithms of ten years ago into chip hardware, the computer would have become a microwave, and all subsequent innovations would have been impossible to discuss."
• On the essential relationship between AI and human work: "Out of the approximately 8 million pixels on a 4K display, we only ray traced 500,000 pixels; the remaining 7.5 million pixels were entirely predicted and perfectly filled in by AI. This is the vivid metaphor that AI will make humanity superhuman: humans can concentrate their precious energy and attention on the most critical core parts, leaving the rest of the cumbersome and lengthy processes to AI, thereby achieving a hundredfold leap in productivity."
• On the survival guide for the younger generation: "If I were still a student today, the first thing I would do is to learn AI. Learning how to interact with AI is very much like becoming someone who is exceptionally good at asking questions. You can't just casually ask questions; to make AI your assistant, you need to master the skills and art of prompting. Whether you study law, medicine, chemistry, or biology, you must continuously ask yourself: How can I use AI to do my work better?"
• On the endpoint of life and historical legacy: "Very simply: they created an extraordinary impact. I hope the next generation knows that NVIDIA is the company that grew up with them playing games, but also sees that at the center of the storm of changes in digital biology, materials science, robots, and autonomous driving stands this company that upholds its core beliefs."

Main Text

The Birth of Heterogeneous Computing: Why Does 10% of Code Need a Dedicated GPU?

Cleo Abram: In today's talk, I want to steer clear of the quarterly financial reports and stock price numbers often seen in financial news and focus on more essential questions: What exactly are you building? How have computers evolved over the past few decades to reach today's state? And how will future technology shape the lives of ordinary people?

Jensen Huang: That's great, I’m very much looking forward to this discussion. In fact, at some point, you have to choose to believe in something. What we are doing now is essentially a complete reconstruction of the modern computing architecture that has lasted for decades.

Cleo Abram: To understand this AI era that the whole world is focusing on, we must first go back to the 1990s. At that time, game developers wanted to create more realistic and immersive 3D visuals, but the traditional hardware at that time simply could not handle the massive pixel calculations in real time. Why couldn't computers at that time handle graphics? How did GPUs come to exist?

Jensen Huang: This stems from a core feature of computer programs: in a segment of software code, typically, only about 10% of the critical instructions account for a full 99% of the computational load. This 99% of intensive calculations tends to have a very high degree of repetitiveness, making it ideal for breaking down into thousands of tiny tasks that can be executed in parallel; the remaining 90% of the code primarily involves system control, branching, and scheduling logic, which must be processed in order.

At that time, traditional computers were completely dominated by the central processing unit (CPU), which was designed to be extremely proficient at processing serial logic at high speeds, but when faced with the massive demand for rendering vast numbers of pixels simultaneously, it was entirely inadequate. We keenly observed that a truly ideal computer could not solely possess serial processing capabilities, nor could it possess only parallel processing capabilities; it must excel in both.

Thus, we founded NVIDIA with the goal of creating a dedicated accelerator processor that could solve problems unattainable by traditional computers. This is the origin of the graphics processing unit (GPU).

Cleo Abram: This reminds me of a very classic educational video NVIDIA released over a decade ago: the Mythbusters team created a robot that fired paintballs one at a time from a single nozzle to demonstrate how a CPU handles tasks; then they presented a massive mechanical array where thousands of nozzles fired paintballs simultaneously in a millisecond, instantly creating a complete image of the Mona Lisa, vividly showcasing the power of GPU parallel acceleration.

Jensen Huang: That demonstration very intuitively showcased the essential differences between the two. Gaming is the first field to have an extreme demand for such massive parallel computational power.

Breaking Ground in Gaming: Using the Consumer Market to Feed Hardcore R&D

Cleo Abram: Why did NVIDIA choose to first enter the video game space and not directly tackle research computing?

Jensen Huang: First and foremost, it’s because we have a deep love for this application. 3D video games are a digital simulation of the real or fantastical world; who wouldn't want to enter a realistic, splendid virtual world?

But from a business strategy perspective, there was a deeper decisive consideration: we foresaw that video games would become one of the largest consumer entertainment markets for all of humanity. Developing cutting-edge semiconductor chips requires unimaginable amounts of R&D funding; if you only serve niche high-performance scientific computing laboratories, you have to rely on government research grants or expensive custom contracts, which cannot sustain continuous massive technological iterations.

In contrast, video games provide an incredibly vast consumer market. Millions of players are willing to pay for smoother, more brilliant graphics, which allows us to become the world's largest manufacturer of parallel processors. The huge volume of shipments grants us a continuous and self-replenishing source of funds for R&D, enabling us to push this parallel architecture to extreme physical limits over the ensuing decades.

Betting the House on CUDA: From Medical Imaging Inspiration to a Universal Parallel Computing Platform

Cleo Abram: By the early 2000s, some sharp scholars and developers began to realize that GPUs could not only be used to render game textures but also to run physical collision simulations and were even used by external scientists for complex scientific calculations. However, my understanding is that doing this at the time was extremely painful; scientists had to disguise matrix operations as triangular pixels or texture maps to deceive the graphics cards into computing. Then you made an extremely bold decision: to launch the CUDA platform. Can you talk about the background at that time?

Jensen Huang: You have done your homework; that's very accurate. Many great innovations often arise from the same complex mélange: part comes from lofty ambitions and inspiration, while another part stems purely from desperate self-rescue out of necessity.

One of the earliest external cases using our GPUs for general acceleration came from Massachusetts General Hospital. Researchers there used our gaming graphics cards for CT image reconstruction, which provided us with tremendous inspiration.

Simultaneously, NVIDIA internally faced bottlenecks: we wanted the water flow in the gaming world to move like real water, the thousands of fragments created by explosions to conform to aerodynamic principles, and the smoke to diffuse naturally. This required the graphics cards not only to process graphics but also to run particle physics and fluid mechanics simulations simultaneously. If the graphics card hardware pipeline were rigidly bound to traditional computer graphics workflows, none of this would be achievable.

Thus, we resolved to completely break the constraints of the dedicated graphics pipeline and transform it into a universal parallel computing architecture, creating the Unified Computing Architecture (CUDA). CUDA allows engineers and scientists all over the world to directly write code using the C language they already know, freely invoking thousands of computational cores within the GPU.

Cleo Abram: Looking back from today’s perspective, creating CUDA was a visionary gamble. At that time, NVIDIA bet all its assets on CUDA, forcibly incorporating general computing logic into every gaming graphics card sold, leading to significant increases in chip area, high costs, pressing profit margins, and facing harsh criticism and skepticism from Wall Street for over a decade.

Jensen Huang: When you create a completely new disruptive technology, critics will always say: "If you build it, people may not come." But from an optimist’s perspective, our belief is: "But if you don’t build it, people will never have the chance to come."

We are acutely aware of the immense physical potential this technology contains, and we strongly believe that through the vast market for consumer-grade GeForce graphics cards, we can lay this advanced parallel computing architecture on the desks of every geek and researcher around the world. It was this rational hope that sustained us through those long and lonely years.

The Enlightenment of AlexNet in 2012: Restructuring a 65-Year-Old Modern Computational System

Cleo Abram: Fast forward to 2012, a landmark battle in computer science arrived: the ImageNet image recognition competition. Alex Krizhevsky, Ilya Sutskever, and their mentor Geoffrey Hinton submitted a deep neural network model named AlexNet, which crushed all traditional computer vision algorithms significantly reducing error rates. They trained this groundbreaking model using two consumer-grade GeForce GTX 580 graphics cards. What was your first reaction at NVIDIA upon seeing this result?

Jensen Huang: That was a moment that completely changed the course of history. At that time, a computer vision R&D team within NVIDIA was trying to solve the image recognition problem using traditional algorithm libraries and experienced significant barriers and frustrations. When AlexNet burst onto the scene, we felt not only shock but also a profound resonance: this new algorithm completely abandoned the old path of manually crafted rules, relying entirely on vast amounts of data to self-train weights, achieving an unprecedented leap in capability.

We immediately stepped back and asked ourselves a fundamental question: "If this deep learning architecture can demonstrate such terrifying scalability in computer vision, what is its limit? If it can be infinitely scaled, what does that mean for the entire computing industry and the underlying software and hardware architecture?"

We made a crucial judgment at the time: the paradigm for software development was undergoing an essential shift. Since IBM System/360 established the instruction set architecture for modern general-purpose computers in 1964, for 65 years, humans had been telling computers what to do by meticulously programming instructions step by step; the rise of deep learning marked a transition where computers began to learn rules autonomously from vast data.

Since the essence of the software changed, the underlying hardware and system stack supporting the software must all be rebuilt from scratch. We immediately made the decision to completely bet on AI, redesigning the entire computing stack comprehensively from single chip architecture, interconnect buses, network communications, algorithm libraries, to distributed systems. The later DGX supercomputer system originated from this epiphany.

Why Now: A Universal Translator Across All Data Modalities

Cleo Abram: Why has the general public only recently felt the explosion of AI? What is the essential technological leap behind this?

Jensen Huang: Because in the past decade, the hard challenges in the field of computer science have been tackled one after another. Initially, computer vision was completely solved, followed by speech recognition reaching superhuman levels, and subsequently, natural language understanding achieved qualitative breakthroughs.

However, the real trigger for a comprehensive qualitative change was the maturity of the Transformer architecture and generative AI. Previous AI models were mostly single-task classifiers, such as determining whether a photo contains a cat; whereas today's generative AI is a universal "cross-modal translation system."

It can understand text and translate it into another text, for example, summarizing paragraphs or translating across languages; it can also translate text into images, images into 3D assets, and sensor data into mechanical arm angles. It can move smoothly between various representations in the digital and physical worlds, which is why today's AI feels so omnipotent.

Physical AI and World Models: Cosmos Provides Common Sense, Omniverse Anchors Physical Laws

Cleo Abram: Next, I want to delve into a significant transformation happening now: moving from the digital world of text and images to the physical world of embodied intelligence and robotics. Previously, training robots required physically placing them in real workshops for step-by-step movement or getting humans dressed in motion capture suits to collect extremely limited data, which could easily wear down the machines or result in breakage, making data collection very expensive and slow. Now we are starting to train robots in virtual worlds. NVIDIA launched Cosmos and Omniverse; how are you addressing the robot learning challenges in the virtual world?

Jensen Huang: The best reference for understanding physical AI is the development path of language models like ChatGPT.

When ChatGPT first appeared, although it performed impressively, it had a critical weakness: hallucinations. If it rambled on for long or discussed untrained blind spots, it would spout nonsense because its output was not anchored in factual truth (Ground Truth). Subsequently, the industry introduced retrieval-augmented generation (RAG); when you input an authoritative PDF or real-time internet search into the model, that PDF and search results become the model's reference facts, stabilizing the model's reasoning.

Now, we project the same logic onto the physical world. For robots to act freely in the real world, they first need a foundational model that understands physical common sense. What is physical common sense? They must understand that the Earth has gravity, objects have inertia and friction; they must have geometric and spatial perception capabilities; they must grasp object permanence, meaning when I turn my head and turn back, the cup on the table is still there; and they must understand causal logic; if I knock over the cup, the water will spill out.

To imbue AI with this common sense, we built Cosmos, a foundational language model for the physical world that encodes the universal laws of physics.

But just having Cosmos is not enough; we must provide it with an "absolute physical truth anchor," just like feeding authoritative PDFs to ChatGPT. This anchor is Omniverse. Omniverse is not a typical video game engine, but a physical simulation system that strictly follows first principles, based on rigorous mathematical models of Newtonian mechanics, fluid mechanics, and electromagnetism solvers.

By combining the Cosmos world model with the Omniverse physical simulation engine, we can generate endless virtual environments and sudden conditions in the digital space for robots that absolutely comply with physical laws: lighting changes, wet surfaces, obstructions, and mechanical failures. Robots can complete extreme training within hours that would otherwise take decades in the real world before deploying their models into actual robots.

Vision for the Next Decade: Everything Will Have a Robotic Form, Every Person Will Have Their Own R2-D2

Cleo Abram: Following this logic, what will the physical world look like a decade from now? How will ordinary people interact with these physical AIs?

Jensen Huang: In the future, all moving objects will possess robotic capabilities, and this day will arrive faster than most people imagine.

In the future, people will no longer have to push lawnmowers across their yards under the blazing sun, which seems a bit absurd today. Every car in the future will be a fully autonomous robot; humanoid robots will also enter factories and homes. All these robots will complete their training in the digital twin world constructed by Omniverse and Cosmos, then transition into physical reality.

What excites me even more is that everyone will have their own lifelong digital companion, just like R2-D2 from Star Wars. It might not necessarily be a round metallic bucket; it could exist in your smart glasses, your smartphone, your office computer, or in your car until you arrive home, where it will greet you in some form of a physical robot.

This dedicated R2-D2 will accompany you throughout your life, evolving alongside your experiences, deeply understanding your joys, sorrows, habits, and work styles, becoming the strongest extension of your perception and transformation of the world.

Debate on Architectural Philosophy: Why Can't Transformers Be Directly Hardened into Chips?

Cleo Abram: While preparing for this interview, I discussed with many engineering friends who insisted I ask you this question: As the Transformer architecture has become the foundation for most large models today, many chip companies are attempting to manufacture specialized chips (ASICs) completely tailored for Transformers. Specialization can indeed bring extreme efficiency for specific tasks but loses general flexibility. How does NVIDIA balance specialized optimization with general architecture? How do you make such heavy technological bets worth billions?

Jensen Huang: This touches on our core belief: do you believe the Transformer is the last AI algorithm architecture that human researchers will discover?

If your answer is yes, then you can indeed weld all the mathematical operations of Transformers directly onto silicon chips; however, our core belief is that the Transformer is not the endpoint; it is merely a stepping stone on the ladder towards more advanced AI evolution. Future next-generation cutting-edge architectures may not bear any resemblance to the Transformer.

Looking back at the history of computer algorithms and innovations, is there any brilliant idea that can forever dominate everything? The answer is no. The essential charm of computers lies in their ability to execute wondrous programs today that no one could even dream of a decade ago. If you had solidified the algorithms of ten years ago into single-function hardware back then, the computer would have regressed into a microwave, and the prosperity of the entire application ecosystem would have abruptly halted.

Take the core attention mechanism of the Transformer as an example; its original intention was to calculate the relevance between each word and all the other words in the context. When the text only has 10 words, the computation is effortless; but when the context window expands to tens of thousands of words and millions of tokens, the total computation cost will explode quadratically, making it impossible to be implemented on physical hardware. To solve this problem, new mathematical algorithms like FlashAttention, hierarchical attention, and waveform attention have emerged in the past few years.

Computer science has no end, AI researchers have not given up their exploration, and NVIDIA will not give up either. Our most important mission is to always build an acceleration computing system with a high degree of programming flexibility, allowing the smartest algorithm scientists in the world to navigate freely in this sea of computing power, exploring the next generation of unknown, great algorithms for humanity.

Limits of Chip Manufacturing and Physical Challenges: From Wafer Packaging to Extreme Energy Efficiency

Cleo Abram: Today's top chips are approaching the fundamental limits of physics. You design chips, and partners like TSMC are responsible for lithographically manufacturing them at a few nanometers of physical scale. How does NVIDIA approach design when facing the slowdown of Moore's Law and power consumption bottlenecks?

Jensen Huang: Traditional Moore's Law and Dennard Scaling are already practically reaching their end in a physical sense. The era of solely relying on shrinking transistor sizes for multiplication of computing power and reduction of power consumption is gone.

This is why we must shift comprehensively from "chip thinking" to "data center thinking." The optical limit dimension (Reticle Limit) of a single wafer is fixed; to break through the physical limits of a single chip, we employ CoWoS (Chip on Wafer on Substrate), ultra-high bandwidth interconnect links (NVLink), and ultra-high density on-chip HBM (high bandwidth memory).

More importantly, parallel accelerated computing follows entirely different physical laws. Assigning a complex task to thousands of efficient cores to process in parallel can greatly enhance energy efficiency by dozens or even hundreds of times compared to letting a few high-frequency general-purpose CPU cores handle it. Through collaborative optimization at algorithm, chip, network, and system levels, we have opened up a new path for driving exponential increases in computing power in the post-Moore's Law era.

Next Major Bet: Physical AI, Digital Biology, and Microclimate Simulation

Cleo Abram: Looking back, NVIDIA's major bets on GPUs, CUDA, deep learning, and autonomous driving have all proven correct with time. Besides the robots mentioned earlier, what major bets do you have in the pipeline?

Jensen Huang: In addition to the physical AI and humanoid robot hardware and software toolchain, one area I am extremely excited about is digital biology.

When computers understood codes made up of 0s and 1s, today’s AI comprehends human natural language; whereas the essence of biology lies in the complex interaction language of molecules, amino acids, proteins, and living cells. Just as we use physical models to understand the real world, if we can employ AI to deeply analyze and predict the interactive language between biological macromolecules, we can build digital twins of the human body. This means drug discovery, gene therapy, and disease mechanism research will shift from previous trial-and-error methods to high-precision scientific simulations that can rapidly iterate computationally.

Another development I take great pride in is climate science. With digital twin technology, we are constructing high-resolution regional climate and extreme weather prediction systems capable of accurately simulating micro-meteorological evolution within a kilometer above your head.

Scientific tools are fundamentally "time machines" invented by humanity. Scientists strive desperately to predict the future to clearly see the evolution trajectory in advance, allowing them to make optimal designs and interventions in the present to create the best possible future version.

Neural Rendering and Personal Supercomputers: Productivity Leap from RTX 50 and DGX

Cleo Abram: At the recent Consumer Electronics Show, you showcased the latest GeForce RTX 50 series graphics cards and the next generation of DGX systems on display. What do these new hardware innovations mean for ordinary creators and developers?

Jensen Huang: Behind this lies an incredibly fascinating technological loop: originally, GeForce gaming graphics cards gave birth to modern AI, helping Alex Krizhevsky and Geoffrey Hinton train AlexNet; today, powerful mature AI has returned to empower GeForce graphics cards, completely overturning the underlying logic of graphic rendering.

In the images shown in the latest RTX 50 series graphics cards at 4K resolution (approximately 8 million pixels in total), our physical ray tracing pipeline actually intensively computed only 500,000 of those pixels, while the remaining 7.5 million pixels were all accurately predicted and perfectly filled in by a deep neural network based on the 500,000 high-quality references in real-time.

This is the ultimate metaphor of AI empowering human productivity: human energy, focus, and physical effort are extremely precious and limited. If we have to grind through every detail manually, the output will inevitably be extremely limited; but if we leverage AI’s generalization and reasoning capabilities, humans can focus all their energy on the most critical and valuable "500,000 pixels," and AI can help fill in the rest, unleashing superhuman creativity across the board.

At the same time, we personally delivered the world’s first DGX supercomputer to OpenAI in 2016, which was received directly by Elon Musk. At that time, the supercomputer was massive and cost $250,000, only a few top laboratories could own it; now we have scaled it down to about $3,000 for personal AI super workstations that can sit right on the desk of students or engineers. This means that every ordinary student and independent developer now has their exclusive supercomputing center to train their own AI models and R2-D2 robots in their dorms.

Advice for the Younger Generation and Legacy of Life: The Art of Prompting and Extraordinary Impact of the Era

Cleo Abram: For students, the younger generation, and anyone looking to make an impact in this era who are watching this interview, what crucial advice do you think I haven't asked yet?

Jensen Huang: I think one of the most important pieces of advice is: if I were a student now, the first thing I would do is to fully learn about AI.

How to learn to interact with ChatGPT? How to learn to interact with Gemini Pro? And how to learn to interact with Grok? Learning to interact with AI is essentially about becoming exceptionally skilled at asking questions. For instance, you are very good at asking questions, and writing prompts for AI is very similar. You can't just casually ask random questions; to make AI a helpful assistant, you need to master the skills and art of prompting.

So if I were a student today, regardless of studying math, science, chemistry, or biology, no matter which scientific field I plan to enter or what profession I will pursue, I will continuously ask myself: how can I leverage AI to do my job better?

If I want to be a lawyer, how can AI make me a better lawyer? If I want to be a doctor, how can AI make me a better doctor? If I want to be a chemist, how can AI make me a better chemist? If I want to be a biologist, how can AI make me a better biologist? This question should permeate everyone’s thinking.

Just like in the era I grew up in, we were the first generation to confront this question: how to use computers to do our jobs better. The previous generation had no computers at all; remember, I entered the industry even before Windows 95, which was in 1984, when there weren't even computers in the office. Soon after, computers became widespread, and we had to ask ourselves how to use computers effectively. The next generation no longer needs to ask this question, but clearly, they must face the next proposition: how can I use AI to do my job better? This is the starting point and the endpoint of everything.

AI has completely lowered the barriers to understanding, knowledge, and obtaining intelligence. In the past, putting a computer in front of someone who had never used one before would have meant they could not learn to use it in a day without someone teaching them step by step; but now with ChatGPT, if you don't know how to use it, you simply input: "I don't know how to use ChatGPT, please teach me," and it will immediately respond and give examples. The most wonderful thing about intelligence is that it will assist you along the way, making you more powerful throughout the process.

Cleo Abram: That’s great. If you still have time, I have one last question. This isn't the one I originally planned to ask, but on my way here, I actually have a bit of a fear of flying, which isn't very rational. On the flight here, it was a bit bumpy, quite bumpy. Sitting in the violently shaking cabin, I couldn't help but think about what people would say at my funeral.

Jensen Huang: "She always asked great questions!" That’s definitely what would be written on your tombstone.

Cleo Abram: I hope so! Yes. Besides "loving her husband, friends, and family," I hope people can discuss my "optimism" and understand the goals I strive for. I'm very curious about you; you've been deeply involved in this industry for so many years, and the vision you've outlined is so grand. What theme would you hope that people use to evaluate the career you have worked hard to accomplish?

Jensen Huang: Very simply: "They made an extraordinary impact."

I believe we are very fortunate because we established some core beliefs long ago and have consistently upheld these beliefs, continually building upon them, making us one of the most important and influential technology companies in the world, if not of all time.

Therefore, we take this responsibility very seriously. We strive to ensure that the capabilities we create are not only for large enterprises but are equally accessible to independent researchers and developers across various scientific fields, regardless of profitability, size, or fame. It is because we deeply understand the profound significance of the work we engage in and its potential far-reaching impact on countless lives that we aim to disseminate this capability to every corner possible.

I firmly believe that when we look back a few years from now, I sincerely hope the next generation will see: first, they will recognize us for the gaming technology we created; but I am certain that looking back, the field of digital biology and life sciences has been completely reshaped, our understanding of materials science has been completely overturned, and robots are assisting humans in conducting dangerous and tedious tasks across various sectors. When we want to drive, we can drive ourselves; when we don't want to, we can take a break, or turn the car into a home theater, calmingly reading on our commute. By then, you might even wish to live further away just to enjoy a bit more time in the car.

When you look back and find that one company was at the storm center of all these transformations, and it happened to be the company that grew up playing games with you. I hope that’s how the next generation sees NVIDIA.

Cleo Abram: Thank you very much for your valuable time; this has been a very enjoyable conversation.

Jensen Huang: Thank you, I’ve enjoyed it too.