At the TiECon 2024 conference hosted by TiE, Jensen Huang, the co-founder and CEO of NVIDIA, was invited to attend and engaged in a fireside chat with Navin Chaddha, Managing Partner at the venture capital firm Mayfield. Huang shared his entrepreneurial journey and insights on technological advances with the many entrepreneurs present.
Huang recalled that he and two other engineers had a steadfast belief in accelerated computing. Even though they started from an unremarkable point, their collective efforts eventually changed the entire industry. He noted that over the past three decades, these ordinary things and beliefs have shaped their story.
On the topic of technological development, Huang emphasized that we are at a historic moment where we can witness the scalability of deep learning firsthand and understand the importance of learning from data. He reminded everyone that as facts change, we should also adjust our thinking in a timely manner.
Regarding the motivation to work diligently, Huang said he has the best job in the world, which is enough to inspire him to keep moving forward.
When discussing advice he would give to his younger self, Huang believes that ignorance and conviction are two superpowers. Knowing the difficulty of a problem doesn’t really help to solve it; maintaining ignorance and belief is more important.
Navin Chaddha asked Huang what prompted him to start NVIDIA. Huang responded that he didn’t start a company because of an astonishing idea or by meeting the smartest engineer in the world, but because he bonded with two friends. They wanted Huang’s support and for him to join their team, which was the starting point for NVIDIA’s entrepreneurial journey.
Huang reminisced that they would often meet at a family-style restaurant, Denny’s, to discuss business. Unlike other companies that may have an exciting story behind their creation, their beginning was very humble and unadorned.
In my view, what makes our story compelling is its simplicity and truthfulness. Our distinction comes not from extraordinary talent, but because we are ordinary people. We kept moving forward by constantly observing, thinking, and problem-solving. Though I do not wish to boast, the fact remains that our team, especially in the field of computer engineering, has gathered a group of outstanding computer scientists and founding members. They possess amazing talent and innovative ability, paving the way for a new chapter through their hard work and dedication.
In narrating this journey, I have always pondered how to make it more engaging, but unfortunately, I haven’t found a better way to express it. If possible, I would certainly love to create a perfect story chapter to captivate you. However, the truth is that we were just three ordinary engineers who, driven by a common passion and belief, started a company convinced in the power of accelerated computing. This brings to mind a profound lesson: “perseverance.”
Moreover, perseverance requires not just stable commitment and belief but also prudence to avoid potential risks and to strategically advance. After all, if a business shuts down, it cannot continue for thirty years. Therefore, the real appeal lies not at the beginning of the story, but in the entire process and our persistence in it. This is the essence of the story, and our journey is far from over.
This is just a beginning, much like an unfinished movie that’s to be continued. I believe that both my team and I are at the starting phase of our careers. Although we are fully immersed in our work, we still retain a fearless spirit, like newborn calves unafraid of tigers, tirelessly striving in our respective roles every day.
Now, let’s talk about Mayfield. Mayfield, a company with more than half a century of history, had a profound influence in the field of technology when Nvidia was founded. When we started our company, Mayfield was already investing in multiple enterprises, with graphics companies like S3 existing in the market. I believe that behind the growth of every company lies the founder’s profound dreams and yearning for success.
In dealing with market competition, our unique advantage stems from our distinctive observations and visions of the world. We faced competition from 110 to 120 graphics companies and stood out by emphasizing practical applications of our solutions. Other companies might focus on developing top-notch computer graphics chips or systems, but we always focused on how to make applications more enriching and enjoyable.
The concept of “application acceleration computing” that we proposed, especially the acceleration computing aspect, is not just about 3D graphics chips but also includes the underlying architecture and the libraries we have created for various specific fields. We developed CDNN to support AI, RAPIDS for data processing, and KQuantum for quantum simulation. Over time, we have built a range of algorithm libraries for different sectors.
We serve not just the technology itself but the entire application industry. Our team consists of graphics chip designers and experts who work closely with the application industry. They constantly push our partners to use our technology and collaboratively optimize and rebuild applications, allowing games and other applications to utilize our developed technologies or algorithms for better results.
We have always been at the intersection of application demands and independent architecture, pioneering the scientific computing industry and promoting the concept of “co-design,” which requires the simultaneous development of applications, algorithms, libraries, systems, and architecture. This global perspective has been our core philosophy since our founding in 1993.
This different perspective makes our organizational structure, market strategy, and technological development approach stand out from the rest. When we showcase our unique insights into the world and maintain an open mind, allowing others to delve deeply into this viewpoint, the reason Nvidia stands out from the competition becomes clear.
I remember when Nvidia first embarked on its journey, I was working at Microsoft, which had acquired my first startup company, which specialized in internet video streaming technology. From Bill Gates, I learned that a company’s influence is limited if it operates solo. To achieve success, it is more important to create an ecosystem that can drive the entire industry forward, rather than working alone.
In the era when numerous proprietary Unix systems shone like stars, Jobs once again proved the value of specific platforms by introducing terminal devices such as the iPhone and iPad, and I can’t help but see the commonalities between them. When it comes to end-to-end platforms, one cannot omit the remarkable capabilities your ecosystem demonstrates. Your company not only earns its own profits but also ensures that partners can be successful within this ecosystem. Isn’t this a piercing summary of the profound impact of your company’s work? It’s as if making a larger cake together and then letting everyone enjoy a slice.
Jensen Huang: Speaking precisely, our foundation is built on a strong accelerated computing platform, which is vibrant due to its wide application in practice. Without an active developer community, this platform would lose its value. Therefore, the developer ecosystem always remains at the core of our corporate spirit. Also, we do not pursue the so-called “general-purpose accelerator,” as experience has taught us that no single accelerator can suit all application areas.
Our goal is not just to create these libraries but also to ensure the entire ecosystem can use them efficiently. Once these libraries are widely adopted, all users will benefit from the technologies we develop, and these benefits will directly affect their work and life. People originally chose NVIDIA for accelerated computing because it offered unprecedented capabilities.
Now, with the surge in computing demand, almost every task requires accelerated processing to cope. The scalability of Central Processing Units (CPUs) is nearing its limits, but the demand for computing continues to double each year. If CPUs cannot double their performance at the same power consumption or cost, computational inflation is inevitable. This phenomenon is already manifesting globally, with more people feeling the pressure of computational inflation.
It’s worth mentioning that in the past decade, through accelerated computing, we have nearly reduced the cost of deep learning by a million times. This tremendous reduction has made people realize that using computers to search for patterns and relations in data is no longer unattainable.
With the continuous accumulation of intelligence in digital experiences, we have witnessed the rise of generative artificial intelligence. If I were in your position, this phenomenon would certainly not surprise me. Imagine if this gathering was entirely about startups and entrepreneurs assembling, utilizing the great technological innovation we have jointly witnessed.
This is undoubtedly an unparalleled time, and our contribution is to significantly reduce the cost of computing, revealing a new way of computing. This is the first discovery in nearly sixty years. Looking back to 1964, IBM brought the concept of the central processing unit, laying the cornerstone of the computing field we discuss today. Since then, these core concepts have not substantially changed. However, people have finally recognized that accelerated computing is the key to leading the future.
So, how do we make an extraordinary choice?
Navin Chaddha: As an entrepreneur, I’ve fought alongside countless peers. Among the various choices faced in life, many are good options, but to achieve excellence is especially challenging. You are undoubtedly a master of the art of refusal and are decisive when it’s necessary to terminate existing projects. How do you develop this intuition?
Jen-Hsun Huang: I advance my thinking based on first principles—a fundamental philosophical and scientific methodology that inspires us to deduce novel insights or solutions by analyzing the basic components and underlying principles of a problem or a matter. In the field of deep learning, I was fortunate to be involved early and attempted to understand this domain. However, what I saw was not different from others, except for shocking results like AlexNet and memorable achievements, and the high scalability of the deep learning architecture itself.
Indeed, this is the first time in history we had a clear schematic of software design, not limited to software, but also able to outline concepts such as GOP and GOS, whereas in the past we could never describe software functionality so explicitly through diagrams. This indicates the high degree of structure in deep learning. In exploring its scalability, I focused on whether software could scale like chips: with increased computing power or more data provided, could it maintain its ideal characteristics? Is this akin to adding more transistors, deepening pipelines, or more threads for greater hardware scalability? This was the first time I saw such potential in the software domain.
What’s crucial is, what can we learn from data? About 15 years ago, after careful consideration, we concluded: this will revolutionize the way software is developed. That was my bet at the time. Once you’ve established a core belief, you test your hypothesis every day, as I did. If facts change, your views should also adjust. Each day, I test this belief in multiple ways, becoming more convinced that this is the direction of the future. You know, whatever you do, this is always the essential path.
Navin Chaddha: Indeed, this is extremely critical for the audience. Establishing a personal view, digging deeply and driving development.
On our journey to success, we may encounter barriers, but that doesn’t mean we should give up. The key is to continuously shift our thinking and look for new solutions. The road to success may be long and filled with challenges, and it requires us to use an array of skills and strategies. In fact, we don’t necessarily have to bet everything on an idea that may not provide immediate returns. Instead, we can gain profits by applying our technologies in practical scenarios.
In building enterprises and creating a bright future, we don’t need anyone to make sacrifices. Remain firm in your beliefs, give your all, but don’t take blind risks.
With the advancement of technology, we are getting closer to achieving the goal where machines can understand human natural language. At present, although computers do not understand language, we can transform language into numbers to make machines comprehend our intentions. Through machine learning, computers can learn languages such as English, Japanese, Chinese, and French, and even process biologically complex information.
In recent years, with the help of big data, we have made significant progress in multimodal and unsupervised learning. With these technologies, we can teach machines to understand cross-domain content, for example, how to convert text into images, recognize speech from acoustic patterns, and more. Furthermore, if we have a deep understanding of how artificial intelligence works and its potential applications in various industries, we will be able to see clearly how to use these technologies to drive innovation and transformation in the industry.
Facing the application and development of technology, entrepreneurs need to be forward-thinking, not blindly imitate others, but think about how to solve existing and future problems. Ultimately, we should think deeply based on first principles and try to intuitively grasp the essence of the problem.
In the process of continuous exploration and analysis of phenomena, we sometimes come up with unexpected insights and indicate the direction for the future. The changes brought about by the new era are not only filled with anticipation but also come with a series of response measures and development strategies.
About the Fourth Industrial Revolution and its four “cognitive pipelines”:
We are witnessing an important shift where server-side and data centers are gradually moving from decentralization to centralization. Mega companies are buying critical semiconductor system infrastructure from all over the world. However, in tandem with this is the rise of artificial intelligence, with an expected hundreds of billions of IoT devices interconnected, triggering widespread concern over data sovereignty. In this rapidly changing context, how should we view future development, and as an entrepreneur, how should we participate in and promote this process, especially on the issue of data sovereignty, and deal with the needs and challenges from different countries around the world?
Data centers have powerful data processing capabilities and financial support, but the real era of intelligence means intelligence is ubiquitous. Entrepreneurs should take on the responsibility of democratizing technology. Questions about where to start and how we can provide support are critical.
Further thoughts on data sovereignty:
Why do we regard a nation’s social data as its natural resources? In the past, natural resources typically referred to mineral resources hidden underground. However, in the digital future, data becomes an important asset for countries. It is unreasonable to export data for other countries to process and then import it back at an additional cost. Countries should fully utilize their data resources, localize training data models, and provide services, allowing data to flow and add value domestically. This is a crucial strategy for many countries.
Examining the New Dynamics of the Industrial Revolution:
The first Industrial Revolution brought the rise of steam power and mechanized manufacturing. Subsequently, the second Industrial Revolution realized the mass production of electronic products and alternating current, which at the time was a miracle for most people. We are currently in the midst of the third Industrial Revolution, which has introduced unprecedented technological achievements, still difficult for many to fully understand.
The value of these intangible assets has been overlooked by the public for a long time. Initially, value was attributed to tangible hardware, systems, and physical devices. As companies like Microsoft offered integrated solutions through technologies like Cuda, the value of software gradually became recognized. Hence, the last industrial revolution witnessed a surge in software value, giving rise to the profession of software engineers, as well as comprehensive software development methodologies, tools, and ecosystems.
As we stand at the forefront of the fourth Industrial Revolution, we face the challenge of producing cutting-edge industrial products that are still elusive to the public. However, we should not worry, as the rise of the new era of factories predicts the possibility of intelligent mass production. Through smart production, manufacturing, improvement, and marketing, these technologies are becoming increasingly feasible.
It is worth noting that artificial intelligence factories are leading a revolution that will completely transform the traditional data center model to support the so-called “Tokens”—the symbolic floating-point numbers. The accelerated computing technologies we previously discussed remain key here, but must be enhanced in innovative ways for these factories.
Let’s envision how the new industrial revolution will implement the production of these Tokens. Will every corner of the globe become a production base? Countries with rich energy resources, especially renewable energy, will have a significant advantage in this transformation. This means intelligent production bases can be far from the point of use, bringing up the issue of “edge computing.”
To take artificial intelligence to the edge, there are two feasible methods. First is remote sensing technology, which helps us capture and analyze events in the physical world. The second method, more complex but equally valuable, is called physical artificial intelligence. This kind of AI must understand not just English or other natural languages, but also physical properties. In the physical world, any misunderstanding could lead to harmful, anti-physical environment behavior, so we must ensure that AI systems act safely and rationally in the physical world, adhering strictly to the laws of physics.
The AI systems of the future will deeply understand the physical laws such as the conservation of energy, mass conservation, and invariance, ensuring that their actions are feasible in the physical world:
Next, as we delve into the concept of “intelligence as a service”, although the term was originally proposed half-jokingly, it is now getting serious attention in the industry. Perhaps, we should call it “Cognitive as a Service” (CAS), because this concept is similar to the existing ones like “Infrastructure as a Service” (IaaS) and “Software as a Service” (SaaS).
The value of CAS far exceeds the commonly seen surface applications. It involves building applications centered around artificial intelligence, digital assistants, and digital twin technologies. The success of these technologies depends on the underlying architecture and technology. These underlying architectures hold immense potential and possibility, encompassing the foundational layers of system training reasoning, model layers, data layers, as well as middleware and tool layers, which together constitute a multi-level “cognitive pipeline” architecture, carrying the important cornerstones for the future development of cognitive technologies.
In recent years, the fields of artificial intelligence and accelerated computing have made leaps and bounds. They have expanded beyond data centers, not only paving the way for the integration of shared resources and microservices but also shaping a new paradigm of hyperscale computing. Looking forward, we can anticipate the emergence of a system known as the artificial intelligence factory, which will leverage the computational resources of entire data centers to provide a solid foundation for the construction of various applications, digital twins, and digital comrades.
NVIDIA has transformed into a data center company, focusing on the unique needs of data centers to create scalable solutions that integrate into mainstream cloud platforms such as Azure, Google Cloud, and AWS. Our goal is to ensure that our platform is ubiquitous, whether it’s in the cloud, on-premises, or on the PC, and compatible with various architectures and software. We provide flexible toolkits to support developers in applying their software to any required environment.
In terms of leadership and organizational structure, NVIDIA’s founder, Jensen Huang, has his unique insights. He pointed out that traditional organizational structures are largely derived from military needs, imperfect information systems, and leaders not wishing to be questioned. However, at NVIDIA, the organization is designed with first principles in mind, taking into account the company’s purpose, the nature of its products, and its environment. Huang believes that in an era of rapidly developing and ever-changing technology, the flow of information within the company must be highly fluid, to ensure flexibility and rapid response. He emphasizes that team members who report directly to the CEO must be top experts in their respective fields, as this structure helps the company understand and execute decisions more quickly, keeping pace with the developments of the era.
This group of outstanding senior managers can achieve impressive results with minimal intervention. As such, the management methods of the CEO naturally need to evolve. We cannot simply apply the management guidelines for new graduates to guide these senior leaders. Therefore, based on first principles, I have created for NVIDIA a unique organizational structure, personnel configuration, corporate culture, and business processes. This all endows NVIDIA with a unique position in specific fields, although we are not the best in every aspect. But this precisely reflects our uniqueness.
In my career, the most difficult decision pertains to the choice of chip architecture, which almost led to the company’s downfall. However, for many CEOs, such a problem might be easier to answer. But I have spent 31 years at NVIDIA, during which I had to make countless decisions, many of them critical to life and death, and many I made myself. Therefore, it is very difficult to pick out the single most challenging one. However, I would still like to talk about the “forward texture mapping” architecture we chose a long time ago. This architecture abandoned the use of the Z-buffer and triangles and instead utilized surfaces and turns. In hindsight, the right direction should have been the complete opposite—namely, “reverse texture mapping,” which is based on triangles and makes use of the Z-buffer. It is evident that our choice was off the mark.
How serious was this mistake? In simple terms, we chose an entirely wrong path from several options, much like getting a zero on the SAT exam. The reason lies in the fact that we smart engineers, when facing the rapidly changing world, made seemingly rational decisions based on the situation at the time. Even if the technical details are not the most attractive part, they are closely related to DRAM prices, floating point calculations, the state of technology at the time, and our competitors.
However, there was a severe disagreement internally about the future direction of this technology. You can imagine the pressure that comes with having invested heavily in this technology. I personally was very fond of this technology and had high hopes for its success. But when it was definitively proven to be going in the wrong direction, I was extremely reluctant to accept this reality.
But this experience also brought a valuable lesson to the company: we realized that the company is not just about technology choices, and these choices do not define us entirely. If you can separate your choices from the purpose of your existence, and realize the reason you exist and your ultimate goals, you can scrutinize and critique your choices more objectively without letting your ego get too involved. For both me personally and the company, this was an extremely important lesson.
It taught me how to view strategic decisions, technology decisions, and any decisions made by the CEO more rationally, rather than being tied to the company’s short-term goals. As facts change and the world changes, your thoughts should also change. In fact, the world around us has already changed dramatically, and a rational person would draw reasonable conclusions based on these changes. Perhaps, given the situation at the time, we already made the best decision possible. But as the world changes, you should adjust accordingly.
Therefore, it was a tough decision because we were still young at the time, and the company was in its infancy, we were all vulnerable. We were all unsure and lacked confidence. But I am grateful that we were able to get back to first principles, relying on the confidence and wisdom of those present. We led the company away from those wrong decisions, then moved forward, and eventually achieved remarkable accomplishments.
If you had the chance to start over, what advice would you give to your younger self? In these 31 years at NVIDIA, I indeed learned a lot. First of all, what I want to say is, I would not tell my younger self all of the experiences and lessons in one go. I want to retain these experiences and share them step by step on the journey forward. The reason is that there is a kind of superpower in ignorance, which is not knowing the difficulty of something, and this ignorance is actually a driving force. Not believing that something is impossible is also a superpower.
And as we know more, the more history and stories we accumulate, we become more aware of the difficulty of things. So, I like the feeling of “how hard can this be?”, it motivates us to keep moving forward. My first reaction to all the challenges we face is: How hard can this be? In fact, it is always much more difficult than one can imagine and always takes a longer time to overcome. But I believe knowing this in advance doesn’t really help. It’s like running a marathon; I’ve never run a marathon, but how hard can it be? Only by trying it will one know.
Building a company is truly a marathon, not a sprint. In building a company, how can we maintain perseverance? It is by moving forward step by step, experiencing the challenges and difficulties. You know, if someone can do it, then we can too. So, how hard can it be, really? It may be one of the superpowers of a startup.
