Huang Renxun's latest interview: If not going all out, NVIDIA may go bankrupt within 30 days

Key Points

①Huang Renxun believes that every company is essentially in a precarious state, and if it cannot give its all, NVIDIA may also go bankrupt within 30 days.

②NVIDIA's investment has three guiding principles: Is the problem challenging? Can NVIDIA provide unique contributions? Will it have far-reaching impacts?

③Huang Renxun stated that NVIDIA's vision is to help every future car achieve autonomous driving to ensure they meet the highest safety standards possible.

④Huang Renxun foresees that generative AI will disrupt the field of software writing and processing, helping develop new types of software and solve new problems.

On July 2nd, according to foreign media reports, at the 2024 Download Day held in Donington National Park, Leicestershire, UK, NVIDIA CEO Huang Renxun had a fireside chat with Chris Gibson, co-founder and CEO of the American biotechnology company Recursion. In the interview, Huang Renxun admitted that if they do not give their all, NVIDIA may also go bankrupt within 30 days.

NVIDIA is a major investor in Recursion. Last July, the chip giant announced a $50 million investment in Recursion to accelerate the development of breakthrough foundational models for AI in drug discovery, further indicating NVIDIA's entry into the new field of AI pharmaceuticals. This cooperation is a win-win situation, as NVIDIA's investment can help Recursion utilize AI technology to accelerate the development of new drugs, optimize and expand Recursion's foundational models. Recursion can also help train NVIDIA's models with its vast data to better identify and design new disease therapies.

The following is a transcript of the fireside chat between Huang Renxun and Gibson:

"Design Rules" + "Methodology"

Gibson: Thank you very much for flying over to participate in this event, it means a lot to us. Over the years, we have been following your team, including NVIDIA's Vice President of Healthcare Kimberly Powell, Global Head of Life Sciences Business Development Rory Kelleher, and the entire elite team, their outstanding performance is admirable. Your personal insights into the field of biology are of great value to usHuang Renxun: This is also my team, probably one of NVIDIA's largest external investments!

Gibson: In our last conversation, you mentioned how in the early days of your career, the silicon chip industry transitioned from being based on laboratory experiments and evidence to almost entirely relying on computer simulations. What experiences from the field of biology can we learn from this transition? Are there any similarities between the two? Could you please briefly discuss this?

Huang Renxun: There are indeed many similarities between the two. My career began 41 years ago, at a time when Computer-Aided Design (CAD) was emerging in chip design. While computer-aided design had been mentioned before, it wasn't until then that algorithms, fast enough computers, and real-time knowledge combined to truly introduce the term "methodology" into modern chip design. Prior to this, this term was not commonly used, and it was introduced into modern chip design by me and Professor Lynn Conway, author of "Introduction to VLSI Systems".

I don't know if you have read the book "Introduction to VLSI Systems", which describes how we used a simple methodology based on first principles to create and simplify the methodology for chip design, enabling the creation of huge chips. This book is about Very Large Scale Integration (VLSI) systems, a concept that made silicon chips large and complex enough to accommodate entire systems. The book details the methodology of design, transistor layout, simulation, and scaling. This work has truly inspired generations of chip designers.

These three core elements - algorithms, computing power, and expertise - are now flourishing in your industry. In the field of chip design, although expertise is not core to the amount of data required in your industry, these three elements play a crucial role when we delve into the essence of Recursion. The emergence of new algorithms or families of algorithms like deep learning, and the supercomputing power you utilize, are the results of our collaborative efforts. Of course, there are also the professional skills of systematically generating and collecting data from robot labs, as well as the expertise in extracting biological significance from this data, all deeply rooted in the mysteries of life. The fusion of all these elements in the field of biology is demonstrating its immense potential and value.

I was fortunate to witness a similar process in chip design 40 years ago. It is amazing that chip designers at that time, including myself, gradually moved out of the laboratory and were able to work on designs outside the lab. Nowadays, chip designers hardly need to enter the lab, unless it is to celebrate the successful operation of a chip. Imagine thousands of engineers working together for three to four years, consolidating their wisdom and efforts into a small chip. This chip is then embedded in a huge system, collaborating with thousands of such chips (many of them different types). When we start up this system, it begins to work normally, which to me is not a miracle, but a completely expected resultIn fact, this is just another ordinary day in the life cycle of a chip. The reason is that this chip has long existed in silicon, and it has been doing what it should be doing. All of this is just an iteration and evolution based on the chips we manufactured before. Therefore, we have this cycle and iteration: chips continuously create and evolve, providing us with the algorithms and tools needed to design the next generation of chips. This process is almost like a recursion, but it is a true reflection of what my generation experienced in the field of chip design.

Gibson: Did you feel at the time that this progress was inevitable? Is such development also inevitable for others?

Huang Renxun: At that time, most people might tell you that this approach wouldn't work. They believed that due to the complexity of boundary conditions, the long-tail effects of problems, various difficulties and challenges in the laboratory, and the frequent failures of chips, they could not believe that this was achievable. However, I believe that the evolution of every industry follows a similar trajectory. Early pioneers went through countless pains and setbacks, to the point where when things started to run smoothly, they couldn't even believe it would be so simple. Of course, it's not simple, but we have integrated these experiences into our tools.

For us, we have the ability to reshape our transistors, which is one of the difficulties you face and the reason why we spend a lot of time. We can change the structure of transistors until they can be designed according to our expectations. But for you, you have to accept biological transistors - the organisms themselves, they are what they are and cannot be changed. Whereas we shape our transistors to make their behavior conform to our expectations or simulation results. If we cannot predict the performance of transistors or chips under extreme conditions, we will not attempt to manufacture them. We simply adjust the design rules, that's what we do.

That's why we have these things called "design rules." Unfortunately, biology follows these rules, and so does evolution. We have the opportunity to shape our transistors and chips until they become very small, to the point where they show differences statistically. For example, if one transistor points in one direction and another points in another direction, their performance will be different. To solve this problem, we make all transistors point in the same direction. This way, our chip design operates in the way we understand until we reach the limits of technology.

We have these things called design rules and methodologies, and everything runs within this framework. And for you, the challenge is much greater. You have to learn the behavior of biology, understand their meanings, behaviors, and characteristics, just as they naturally exist. But the good news is, you finally have the technology needed to achieve this goal. I firmly believe that with your innovation in robotics labs, data processing capabilities, system data collection, machine learning, and the supercomputer we have jointly built, you are just one step away from truly understanding the meaning of life

NVIDIA Investment Follows Three Principles

Gibson: This is indeed a challenging task, but we have been working tirelessly. NVIDIA, in addressing these issues, seems to have positioned itself at the forefront of the industry, almost touching the pinnacle of success. In order for investors to have a clearer understanding of your vision, NVIDIA, as one of the pioneers in cross-industry work, your stance in the healthcare sector is particularly noteworthy. I heard that you have three guiding principles: Is the problem challenging? Can NVIDIA provide a unique contribution? And will this have a profound impact? Clearly, life sciences in the healthcare sector is undoubtedly a huge challenge, with its impact being self-evident. So, what is NVIDIA's unique contribution in the healthcare sector? What is your overall vision for the healthcare sector?

Huang Renxun: In addition to our collaboration with Recursion, from a more macro perspective, our other choice is to do better what others have already done well. We understand that the pursuit of quick investment returns and victories is the norm in the business world, but this is not our ultimate goal. We aspire to do things that others have never done, things that if we don't do, others won't do either. When we choose this path, we are well aware of the difficulties and challenges involved, but it is these challenges that make our lives more meaningful and our contributions more unique. This is NVIDIA, this is how we deal with opportunities, threats, and challenges.

Of course, understanding the mystery of life, using computers for drug discovery, is undoubtedly an extremely challenging task. However, I believe that in our generation's lifetime, we have the ability to make substantial breakthroughs in this field. As we have emphasized before, algorithms, computing power, and methodology are the three core elements. In large-scale scenarios, this is more like a specialized knowledge domain in a specific field. It is worth emphasizing that we can provide profound insights and contributions in two of these key elements. Given that we do not possess, nor do we aspire to possess, this specific domain expertise, we prefer to be a trusted partner. Our vision is to help every future car achieve autonomous driving to ensure they meet the highest safety standards possible. However, we have no intention of becoming an automotive company.

We are eager to drive significant advancements in artificial intelligence in terms of safety, speed, and efficiency. However, at the same time, we are not aiming to operate or provide large language model services. Please note that in many fields and industries, as you have mentioned, we are not seeking to be industry leaders. On the contrary, we look forward to industry leaders emerging so that we can focus on our unique value contribution. Therefore, I believe that at the intersection of the three core pillars we mentioned, we can play a truly outstanding role. With your profound domain expertise, passion for methodology, and pioneering spirit, you are dedicated to achieving all of this. Therefore, I greatly appreciate people like you, your efforts, and who you are. I think this is extremely outstanding.

Hope to Simulate Weather Around the Clock in the Future

Gibson: When it comes to collaboration, our team once worked day and night, even sleeping on the floor of the data center, just to complete the relevant settings in a short three weeks. And your team has also shown the same effort. I often wonder, if we don't sleep and continue working, can we complete this task in an even shorter time? Now, our achievement - the fastest supercomputer in Biofarma, has been born. Are you surprised that this supercomputer is built and operated by a small company like us rather than a large biopharmaceutical company?

Huang Renxun: Is it really here?

Gibson: No, we once considered hanging the supercomputer from the ceiling, but in the end, for practical reasons, we placed it in a data center several kilometers away in the south.

Huang Renxun: When I walked in, I realized that this is where our supercomputer is located. NVIDIA is the first chip company to manufacture supercomputers for itself, and it has proven to be a wise decision. Similarly, Tesla, as a car company venturing into the supercomputer field, is also a case worth learning from. There are many other similar cases. In short, do we believe that we can discover knowledge from certain things through principled simulations alone? In the past, we thought that having enough supercomputing power could simulate the human body, but now we have basically given up on this idea. Our perspective has changed, hasn't it? I even once harbored a dream: to have a powerful enough supercomputer that can simulate the weather in every region of the world by the minute, accurate to kilometers or even hundreds of meters. However, now we realize that even with the advanced accelerated computing system of NVIDIA, achieving that scale still requires a billion times more computing power than we can imagine, which may take decades to achieve. Nevertheless, I still hold hope, hoping to witness the birth of this miracle in my lifetime.

In the process of solving problems, for example, in the field of computer graphics, we have adopted a technique called ray tracing. We used to think that it would take a long thirty years to achieve ray tracing. But now, we have more advanced path tracing technology. Our problem-solving method is not only efficient but also remarkably effective. We start with simulating one pixel and then use artificial intelligence to predict the situation of the remaining 64 pixels. When dealing with climate issues, we also use a similar strategy. We understand the importance of physical principles, so there is no need to brute force simulate every protein and cell's physical processes.

Obviously, we don't need to simulate every physical process one by one. We understand the physical principles of the weather, without simulating the specific physical state of every square kilometer. We can teach artificial intelligence to predict these physical phenomena and let intelligent algorithms complete this task. You see, our goal is not to deeply understand the causality of the weather, we already understand these relationships, what we truly care about is the weather conditions this afternoon. Do you understand what I mean? This represents a fundamental shift, a huge leap in thinking. We understand the causality of things and are eager to know how these causes affect people's lives, the development of diseases, and how to cure certain diseasesWe may have mastered the underlying biological principles behind it. However, the complexity of multi-omics makes comprehensive understanding exceptionally difficult. Despite understanding the basic principles of biology, it becomes extremely difficult to comprehend at a broader level and in more complex internal relationships. At this point, we need to rely on another type of algorithm, namely artificial intelligence, to lend us a helping hand.

If you firmly believe that your company is fundamentally committed to the creation of intelligence, then your company is not only producing drugs, but also nurturing intelligence, which will ultimately promote drug development. If the amplification of intelligence is such a process - being discovered, enhanced, and applied, then how can your toolbox lack intelligent tools? When you think from first principles, those who can think like this are often able to take a crucial step forward. Even if these ideas have never been attempted before, if they are true and you fundamentally believe in their value, it does not mean you should stand still. On the contrary, this is exactly why you should forge ahead courageously. This is how we understand and drive the development of things. I believe that most pioneers, leaders, and innovators think in this way.

"A Leap of Faith is Needed"

Gibson: Looking back at NVIDIA's founding 30 years ago, we faced many analysts, investors, advocates, and many potential founders, even within our own company, who doubted the future. Nevertheless, you always adhere to the guidance of first principles. As founders, early employees, visionary leaders, and pioneers, how did you continue to build the company for decades under heavy doubt?

Huang Renxun: First of all, we always strive to think and judge things based on first principles. But sometimes, first principles may not be so obvious, because we may not fully understand the mysteries within them, so a leap of faith is needed to believe that certain things will inevitably be so. These beliefs or reasoning do not come out of thin air, they stem from our external information, the knowledge we have, and our understanding and application of that knowledge. If the facts themselves have not changed, why do we need to change our beliefs? If the facts remain the same, just like we are all universal function approximators in this huge deep learning model, how can the output change as long as the input remains unchanged? Therefore, I always hold on to this belief, constantly challenging and examining my reasoning, in fact, I do this almost every day.

Next, I carefully examine many of the assumptions we make, because you know, these assumptions are related to the hard work and interests of many of our employees. Therefore, we must reason these assumptions carefully, but at the same time, we must constantly test and retest them. If the facts have not changed, we need to understand why we would stop believing in things we once believed in wholeheartedly. So, while holding on to our beliefs, we also seek to understand as comprehensively as possible, reason as rationally as possible, and always return to first principlesAt the end of the day, all great achievements are made possible by that leap of faith. If an idea seems obvious to everyone, it has already been realized. It is those who dare to take that step forward that drive the progress of the world. Therefore, we firmly believe in first principles, that universal computation, as a universal tool, cannot be applied to all types of calculations. Take the CPU for example, the Arithmetic Logic Unit (ALU) only occupies a small part of the computing resources, just like in a company where only 3% of the engineers are actually working, while the remaining 97% are management overhead. This reveals the essence of universal computation, and our first principles tell us that it is not the most rational choice.

We are not trying to replace universal computation, as CPUs have their own significance. We have designed many CPUs ourselves, which are very useful in their respective fields. But why can't we add more functions on top of this? We call it "accelerated computing," not "parallel computing." Because parallel computing is usually in contrast to sequential computing, involving the execution of single-threaded and multi-threaded code. Based on first principles reasoning, if we can find a way to make the essence of computers and all computers have these characteristics, and design a programming model that can handle this complexity properly, then we can truly change the future of computing. From the beginning, we have firmly believed in this. The next question is how to find these applications, and we explore them one by one. Computer graphics was the first reasonable application we found, followed by image processing, molecular dynamics, particle physics, fluid dynamics, and so on. We continue to search for the direction to move forward. Until one day, we discovered deep learning.

Will face bankruptcy threat in the next 30 days

Gibson: I remember about 18 or 24 months ago when we first met, you mentioned that every company is essentially in a "D" (danger) natural state, meaning that every company is constantly facing the risk of bankruptcy. This view undoubtedly resonates strongly with many founders of startups, but considering the brilliant achievements NVIDIA has made, hearing such a statement from you may still be surprising. How was this philosophy formed in your company's current stage? What can we, as companies in the early stages, learn from this philosophy?

Huang Renxun: NVIDIA has always been a confident company because we know deeply that only confidence can drive us to achieve great things. However, confidence is by no means the same as complacency. Between confidence and complacency, we have found our position. We have always believed that we can create remarkable results. But once we stop innovating, stop achieving amazing things, we will lose this ability. Therefore, every time I wake up in the morning, including today, the first thing that comes to my mind is the unresolved problems, especially those that were not solved last night. I face these problems head-on, greet them with a "good morning," because they are the driving force that keeps me moving forward. These problems are all challenging, and I always see them as fundamental issues that determine the life and death of the company, because once you ignore them, the company will fall into trouble and eventually declineIn 1993, I had such an understanding, and to this day, I still hold the same vigilance. I believe that it is crucial to exercise the sense of responsibility of adults in the right way.

On the one hand, you must be confident that you can accomplish things that have never been done before, extremely difficult things; on the other hand, you must have a sense of urgency, you must continue to work hard, never be satisfied, and never be complacent. I deeply understand the importance of these emotions, so I always strive to ensure that they are not buried or diluted. Long ago, I used to habitually share our financial situation with employees. Today, this information is already transparent, but I still insist on sharing these key data with team members. Over the years, I have reviewed our financial statements with employees every Friday. I would bring my laptop, directly show spreadsheets, and analyze the financial situation together with everyone.

When someone pointed out that our cash flow might turn negative in June, I calmly admitted that if we could not make a profit before that, the company would face the risk of bankruptcy. Subsequently, engineers would ask what this meant. I explained that it simply meant that if we could not increase revenue in the next thirty days, we would face the dilemma of running out of funds. Faced with such a challenge, someone asked how we should respond. I replied, "Do not let yourself fall into a cash shortage situation." So, we faced two choices: one is to work hard to make money, and the other is to actively raise funds. Frankly, we were not good at either at that time. So I thought our company might go bankrupt. But things would turn around. Over time, I want to remind everyone that if we do not give our all, we may face the threat of bankruptcy in the next 30 days in the future.

At that time, every employee present conveyed a clear message to their colleagues: we must maintain a high level of focus, because any slackness could lead the company to the brink of bankruptcy within 30 days. Even at a company meeting where we had already successfully gone public, when someone mentioned rumors about the company possibly going bankrupt within 30 days, I still responded, "Listen carefully, if we do not give our all in the next 30 days, then the company indeed may go bankrupt on some future day. Yes, it may even be in the next 30 days."

This news shocked and unsettled new employees. However, I believe that this mindset is necessary for a startup company. Let me tell you, for a startup company, every day is crucial. I can tell you without hesitation that every decision you make, every effort you put in, every recruitment, every achievement, and the company culture you shape will determine the fate of the company in the next 30 days - whether it will be vibrant and successful or lead to bankruptcy. I fully believe in this, and I am confident in myself and the team.

Gibson: In our industry, we know the value of time. Because we are eager to bring hope to patients waiting for treatment, every day carries extraordinary meaning and expectations.

Huang Renxun: Indeed, every day is crucial. Look at us, everyone here is a volunteer, 100% volunteers. You are not required or forced to work here, you can choose other ways of life or career pathsBut you chose to be here out of genuine passion and belief. Your drive comes from a firm belief in the mission, love for the team, enjoyment of interacting with others, and recognition of the importance of what you do. Therefore, you are able to wholeheartedly immerse yourselves in it every day. If my words truly reflect the reality, then you can and should strive to do your best every day. This is not to say that we carry obligations or pressure every day, but rather that we have the ability and willingness to show our best selves every day. I believe that you are in an extraordinary period in your industry, experiencing unprecedented opportunities. You must realize that the tools in your hands have not undergone such a thorough innovation or reshaping in the past 60 years.

In the year I was born, IBM introduced the groundbreaking System 360, which laid the foundation for general computing architecture and still roughly maintains the essence of this architecture. Today, we are ushering in a new era - accelerated computing, this breakthrough progress has given rise to a new computing model - generative artificial intelligence, which is reshaping everything in unprecedented ways. From the way software is written and processed, to the types of software we can create, to the types of problems we can solve, everything is undergoing a thorough revolution. This revolution is sweeping through every industry, with profound and far-reaching implications.

As a member of NVIDIA, being able to stand at the forefront of this industrial and computing revolution, we feel immensely honored. And for you, it is a once-in-a-lifetime opportunity to use our technology to innovate one of the most important industries for humanity. Such an opportunity is rare for each of you in a lifetime. This is a unique company, at such a unique historical moment. If you can efficiently compress two days of work into one day, I will definitely fully support it. Right now, it is the most interesting and exciting moment for you. Standing here, I just want to say to you, I really envy you. This is really great, I am proud of you. (Translated by JinLu)

Tencent Technology, original title: "Huang Renxun's Latest Interview: If Not Going All Out, NVIDIA Could Go Bankrupt Within 30 Days"