Jensen Huang to Graduates: Dare to Enter a $0 Market, Hope You Can Find Your Own GPU
TechFlow Selected TechFlow Selected
Jensen Huang to Graduates: Dare to Enter a $0 Market, Hope You Can Find Your Own GPU
The next wave of artificial intelligence is robotics, a "zero-dollar market" today that will be worth billions in the future.
Navigating Web3 tides with focused insightsThe latest commencement speech by NVIDIA CEO Jensen Huang has sparked widespread discussion.
"Believe in the unconventional. Have the courage to explore the unknown."
This was Academician Huang's encouragement to Caltech’s Class of 2024—an exhortation that also seems to mirror NVIDIA’s development history and Huang’s entrepreneurial journey.
Of course, Academician Huang also delivered a condensed history of computing from the perspective of NVIDIA.
Today, computers are the most important intellectual tools—the foundation of every scientific field and industry. When you enter this field, it's essential to understand what is happening around you.
As is well known, Jensen Huang does not hold a PhD. After immigrating to the U.S., he completed his undergraduate studies at Oregon State University and earned a master’s degree in electrical engineering from Stanford University in 1990. He founded NVIDIA in 1993 and invented the GPU for gaming in 1999.
As for why he came to Caltech to deliver a commencement address, Huang was candid:
I'm here to recruit. I’m a good boss.
Key Points from the Speech
-
Some counterintuitive lessons: facing technological and business challenges with intellectual honesty and humility, and embracing strategic retreats when necessary
-
Artificial intelligence is the only technology Huang knows of that advances on multiple exponential curves simultaneously
-
Modern computing traces back to the IBM System 360; its core ideas, architecture, and strategy still dominate today’s computer industry
-
Revolutionary work by Carver Mead at Caltech on chip design methodology and textbooks transformed IC design. It enabled our generation to design ultra-large chips and ultimately CPUs
-
Limits in Dennard scaling, transistor scaling, and instruction-level parallelism have slowed CPU performance gains. At a time when computational demand continues to grow exponentially, the widening gap between demand and computing capability—an exponentially growing chasm—if unaddressed, will eventually strangle every industry through soaring energy consumption, cost, and inflation
-
The next wave of AI is robotics—a “$0 billion market” now but destined to be worth billions, just like GPU-accelerated computing when NVIDIA first began.
……
Below is the full transcript of Jensen Huang’s speech:
Respected President Rosenbaum, esteemed faculty, honored guests, proud parents, and especially the members of Caltech’s Class of 2024:
It’s truly a joyous day for you—all of you should look a bit more excited. You're about to graduate from Caltech, an institution that nurtured great minds like Richard Feynman, Linus Pauling, and Carver Mead, who profoundly influenced our industry. This is indeed a big deal.
Today is a day of pride and joy—not just because your dreams have come true, but because of the countless sacrifices made by your parents and families. Let us take a moment to honor them with congratulations and gratitude, and let them know you love them. Don’t forget this, because you may not know how long you’ll end up living at home.
As a proud father, I loved having my kids live at home and seeing them every day. Now that they’ve moved out, I find myself a bit sad. So, I hope you’ll take time to spend with your parents.
Your journey here has demonstrated your character, determination, and willingness to sacrifice for your dreams—and you should be proud. In life, you will need this ability to endure hardship, pain, and suffering.
Two of NVIDIA’s chief scientists are Caltech alumni. One reason I stand here today is because I am looking for talent. So, let me tell you: NVIDIA is a fantastic company, I’m an exceptionally good boss, widely beloved. Join NVIDIA!
You and I share something in common: a passion for science and engineering. Though we’re about 40 years apart, we’re both at the peak of our careers. For those following NVIDIA and me, you know what I mean. But for you, there are many, many peaks ahead. I just hope today isn’t my peak.
Last year, I had the privilege of delivering a commencement address at National Taiwan University, where I shared several stories from NVIDIA’s journey and lessons that might benefit graduates. I must admit, I don’t enjoy giving advice—especially to other people’s children. So, my advice today will largely be hidden within stories I love and experiences from my life.
I believe I am the longest-serving CEO of a tech company alive today. For 31 years, I’ve neither gone bankrupt, grown bored, nor been fired. I’ve been incredibly fortunate to experience so much—from founding NVIDIA with nothing to where we are today.
I spoke about a very public canceled project with Sega and the importance of intellectual honesty. I know Richard Feynman deeply cared about and often spoke of this—intellectual honesty and humility saved our company. And knowing when and how to retreat—strategically—is one of our best strategies. All of these were counterintuitive lessons I shared at that graduation.
But I encouraged graduates to engage with artificial intelligence—the most important technology of our era. I’ll elaborate later, but you all know AI. It’s hard not to be immersed in it, surrounded by it, and drawn into endless discussions about it. Of course, I hope you’re all using it, playing with it, getting surprising results—some magical, some disappointing, some astonishing. But you must enjoy it, participate in it, because it’s advancing so fast.
This is the only technology I know that advances on multiple exponential fronts simultaneously. The pace of change is incredibly rapid. So last year, I told students at NTU: run, don’t walk—join the AI revolution. Yet, a year later, the changes are even more incredible.
So today, I want to share from my perspective some key developments happening right now as you graduate. These extraordinary shifts deserve your intuitive understanding—they matter to you and to the entire industry. I hope you seize the opportunities before you.
Accelerated Computing Has Reached a Tipping Point
The computer industry is transforming from its foundations—literally from the studs up. From nuts to bolts, everything is changing. Soon, every industry will transform too. The reason is clear: computers are today’s most important knowledge tools. They are the foundation of every scientific field and every industry. If we fundamentally change computing, it will inevitably reshape every sector.
When you enter this field, it’s vital to understand what’s happening. Modern computing traces back to the IBM System 360. That was the architecture manual I studied. You don’t need to study it anymore—there are far better documents now, better descriptions of computers and architectures.
But the IBM System 360 was profoundly important then, and its core ideas, architecture, and strategy still dominate today’s computing industry. It launched one year after I was born.
In the 1980s, I was among the first generation of VLSI engineers, learning chip design from Mead and Conway’s landmark textbook—perhaps still taught here? Likely under Introduction to VLSI Systems. Based on Carver Mead’s pioneering work at Caltech on chip design methodology and textbooks, IC design was revolutionized. It allowed our generation to design ultra-large-scale chips and ultimately CPUs.
CPU-driven computing grew exponentially. Performance and incredible technical progress—what we call Moore’s Law—fueled the information technology revolution. Our generation witnessed mass production of things never seen before: intangible, easily replicable goods—mass-producing software. This spawned a $3 trillion industry.
When I sat where you sit, the IT industry was tiny. The idea of making money selling software was fantasy. Today, it’s one of the most valuable products, technologies, and creations our industry produces.
However, limits in Dennard scaling, transistor scaling, and instruction-level parallelism have reduced CPU performance gains. As CPU improvements slow, computational demand grows exponentially. The exponentially widening gap between demand and computing capability—if left unresolved—will ultimately choke every industry due to skyrocketing energy use, costs, and inflation.
As we say, we’re already seeing clear signs of computational inflation. After two decades of developing NVIDIA CUDA, accelerated computing offers a path forward. That’s why I’m here. Because finally, after decades of unnoticed computing inflation, the industry has recognized the incredible effectiveness of accelerated computing.
By offloading compute-intensive algorithms to GPUs specialized in parallel processing, we typically achieve speedups of 10x, 100x, sometimes even 1,000x—saving money, cost, and energy. We now accelerate applications across fields: computer graphics, ray tracing, genomics, scientific computing, astronomy, quantum circuit simulation, SQL data processing, even pandas and data science.
Accelerated computing has reached a tipping point. This is our first great contribution to the computer industry, our first great societal contribution—accelerated computing. It now provides a sustainable path forward for computing.
Betting Successfully on Deep Learning
As computing demand grows, costs continue to fall. The hundredfold savings in time, cost, or energy from accelerated computing would inevitably spark new developments elsewhere—we just didn’t know where, until deep learning entered our awareness.
A whole new computing world emerged. Jeff Hinton, Alex Krizhevsky, and Ilya Sutskever trained AlexNet using NVIDIA CUDA GPUs and won the 2012 ImageNet Challenge, shocking the computer vision community. This was a pivotal moment—the big bang of deep learning, marking the dawn of the AI revolution.
The decision we made post-AlexNet was transformative. It changed our company—and possibly everything else. We saw deep learning’s potential and believed—through first-principles reasoning and our own analysis of its scalability—that this method could learn other valuable functions. Perhaps deep learning is a universal function approximator, capable of solving problems too hard or impossible to express via basic first principles.
So when we saw this, we knew it was a technology we had to commit to, limited mainly by model and data scale. But challenges remained. It was 2012, just after AlexNet. How could we explore the limits of deep learning without building massive GPU clusters?
We were still a relatively small company. Building large GPU clusters could cost hundreds of millions. But if we didn’t build them, we couldn’t prove scalability. Yet no one knew how far deep learning could scale. If we didn’t build it, we’d never know. This was one of those “if you build it, will they come?” moments. Our logic: if we don’t build it, they definitely won’t come.
So we committed based on first-principles belief and analysis. We believed it would work powerfully, and when a company believes in something, it must act. So we dove into deep learning and systematically reinvented everything over the next decade. We started with the GPU itself, re-inventing every layer of computing. The modern GPU is vastly different from the original GPUs we invented.
We continued inventing in computing, interconnects, systems, networks, and nearly every other aspect of computing—including software. We invested billions. We invested billions into the unknown. For ten years, thousands of engineers advanced and scaled deep learning, though none truly knew how far this technology could go.
We invested billions. We designed and built supercomputers to explore the limits of deep learning and AI. In 2016, we released the DGX-1, our first AI supercomputer. I delivered the first unit to a startup in San Francisco—a little-known startup, a group of friends working on AI—called OpenAI.
In 2022, ten years after AlexNet, computing power had grown by roughly a million-fold. A million times more powerful. Imagine your laptop becoming a million times faster. After that million-fold leap, OpenAI launched ChatGPT—and AI went mainstream.
Over that decade, NVIDIA evolved from a graphics company (many of you may know us first as the maker of GPUs) into an AI company building massive, data-center-scale supercomputers. We completely transformed ourselves. We also fundamentally changed computing technology. The very nature of computing today has been redefined.
The computing stack now uses GPUs to train large language models on supercomputers, rather than CPUs executing instructions written by programmers. We are now creating software humans cannot write. We are now creating software that can do things humans couldn’t imagine—even ten years ago. Computers are now intent-driven, not instruction-driven. Tell the computer what you want, and it figures out how to do it.
Like humans, AI applications will understand tasks, reason, plan, and orchestrate teams of large language models to execute them. The way future applications work and operate will closely resemble how we work: assembling expert teams, using tools, reasoning and planning, and completing tasks. What software is and what it can do has completely changed. Indeed, our industry, by transforming itself, has created another industry—one the world has never seen before.
An industry is forming before our eyes. The input and output of AI are tokens. To all engineers in the room—you know what I mean. These are floating-point numbers embedded with intelligence. Companies are now building a new type of data center that never existed—dedicated to producing intelligent tokens. Essentially, AI factories. Just as Nikola Tesla invented the AC generator in the past industrial revolution, we now have AI token generators—factories for the new industrial revolution.
There was heavy industry producing energy, electricity. Now we have a massive industry producing something invisible—software. In the near future, we will have an industry manufacturing intelligent tokens, AI generators. A new computing model has emerged. A new industry has been born—all because we formed a belief in the future from first principles and acted on it.
Robotics Is the Next Wave
The next wave of AI is robotics, where AI includes models of the physical world beyond language. We collaborate with hundreds of companies building robots, robotic vehicles, pick-and-place arms, humanoid robots, and even giant robot warehouses. But unlike our AI factory strategy—which emerged from deliberate reasoning and thoughtful action—our journey into robotics resulted from a series of setbacks.
As you know, NVIDIA invented the GPU—before we invented the AI factory. Our first great contribution to computing was reshaping computer graphics through programmable shaders. We invented the GPU and programmable shading in 2000. Wanting to integrate GPUs into every computer, we began combining GPU and motherboard chips. We launched a great integrated graphics chip for AMD CPUs.
Our chipset business immediately succeeded—growing almost overnight from zero to a billion dollars. But suddenly, AMD wanted control over all PC technologies, while we wanted to remain independent. So they acquired ATI and no longer needed us. We turned to Intel. Probably not a great idea, but we approached Intel and negotiated a license to connect to their CPUs.
Apple became very excited about our product and asked us to co-develop a new computer—the first MacBook Air. Well, Intel saw what was happening and decided they didn’t want us doing this anymore, so they terminated our agreement. Fine—we pivoted again. This time, we licensed ARM and built a low-power SoC, a mobile SoC—the world’s first SoC—essentially a complete computer, a full operating system computer. Incredible.
Our chip excited Google, who asked us to develop a new device—resulting in Android mobile devices. Well, Qualcomm decided they didn’t want us doing this either, so they refused to let us connect to their modems—and without modem connectivity, building mobile devices was difficult. With no other LTE modem makers, we had to exit the mobile device market.
Well, this happened almost on a yearly cycle: we’d build something, it would succeed wildly, generate huge excitement—and then a year later, we were kicked out of that market. Eventually, with no more markets to pivot to, we decided to build something where we were certain there were no customers—because one thing you can guarantee is that no customers means no competitors, and no one cares about you.
So we picked a market with no customers—a $0 billion market: robotics. We built the world’s first robot computer, processing deep learning algorithms no one yet understood. That was over a decade ago. Ten years later, I’m thrilled by what we’ve built and the opportunity to create the next wave of AI. More importantly, we cultivated a culture of agility and resilience.
Repeated setbacks forced us to adapt and seize the next opportunity. Each time, we gained skills and strengthened our character. We reinforced our corporate identity. Our company is truly hard to distract and hard to discourage.
Today, any setback we face is not an obstacle—it’s an opportunity. Ironically, the robot computers we build today don’t even need graphics—the very thing our journey began with. Where we stand today tells us something and teaches us something. As Richard Feynman said, the world is uncertain, perhaps unfair, dealing you bad cards. Shake them off quickly. Clearly, you’re too focused on your books. Just shake it off. Come on, that’s smart. I laughed. There’s another opportunity—or create one.
Find Your Own GPU
Let me tell you one more story. I used to spend a month each summer working at one of our international sites. When our kids were teenagers, we spent a summer in Japan. On weekends, we visited Kyoto’s Silver Pavilion Temple (Ginkaku-ji). If you haven’t been, you must go. It’s famous for its exquisite moss garden.
The day we visited was a typical Kyoto summer—hot, humid, sticky. Heat rose from the ground. The air was heavy, silent. We wandered through the meticulously maintained moss garden with other tourists. I noticed a lone gardener. Remember, this is a moss garden at Ginkaku-ji—huge, housing the largest collection of moss species in the world. And it’s maintained with extraordinary care.
I saw the solitary gardener crouched down, carefully plucking moss with bamboo tweezers into a bamboo basket. You have to do it this way—bamboo tweezers, just this gardener. The basket looked empty. For a moment, I thought he was picking imaginary moss into a pile of imagined dead moss. So I walked up and asked, “What are you doing?” He replied in English: “I’m picking dead moss. I’m tending my garden.”
I said, “But your garden is so large.” He replied, “I’ve been tending this garden for 25 years. I have all the time in the world.” Well, that was one of the most profound lessons of my life. It truly taught me something. This gardener was dedicated to his craft, engaged in lifelong work. When you do that, you have all the time.
Every morning, I start the same way—I begin with my highest-priority task. I have a very clear priority list, and I always start with the most important work. Before I even begin, my day is already successful. I’ve completed the most important task and can dedicate the rest of the day to helping others.
When people apologize for disturbing me, I always say I have enough time—and I really do. Graduates of 2024, I can hardly imagine anyone better prepared for the future. You’ve dedicated yourselves, worked hard, and earned a world-class education from one of the most prestigious schools on Earth. As you begin the next phase, please take my lessons—they may help you along the way.
I hope you believe in something—something unconventional, something unexplored—but let it be informed and reasoned. Then throw yourself fully into realizing it. You might find your GPU, you might find your CUDA, you might find your generative AI, you might find your NVIDIA.
I hope you see setbacks as new opportunities. Your pain and suffering will strengthen your character, resilience, and agility—they are ultimate superpowers. Among all the qualities I value most in myself, intelligence isn’t the top. My ability to endure pain and suffering, my capacity to persist at something over long periods, my skill in overcoming setbacks and spotting emerging opportunities—I consider these my superpowers. I hope they become yours.
I hope you find a craft. I hope you find a craft. It doesn’t matter if you decide on day one, or decide quickly—but I hope you do find a craft you dedicate your life to perfecting, honing your skills, making it your life’s work.
Finally, make your life a priority. So much is happening, so much to do—but prioritize your life, and you’ll have ample time for what matters most.
Congratulations, Class of 2024. Go get it.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
