Jensen Huang, CEO of NVIDIA, the leading stock in the AI ​​boom, said on Wednesday that NVIDIA's products have now become the most sought-after commodities in the tech industry, and customers are competing for limited supply, especially for AI chips. The limited growth of Blackwell supplied by suppliers has frustrated some customers. He also hinted that if necessary, NVIDIA would reduce its reliance on Taiwan Semiconductor and turn to other chip manufacturers.

He told the audience at a technology conference hosted by Goldman Sachs in San Francisco:

"Our product demand is so high, everyone wants to be the first to get it and get the most share. We may have more emotional customers today, and that's understandable. The relationship is very tense, but we are trying our best."

Huang Renxun introduced to the audience that the company's latest generation of AI chips, Blackwell, is facing strong demand. Currently, NVIDIA outsources the production of Blackwell, and he said that NVIDIA's suppliers are trying their best to keep up with demand and making progress.

However, the majority of NVIDIA's revenue relies on a few customers, such as Microsoft and Meta Platforms Inc., data center operators. When asked if the massive AI spending has brought investment returns to the customers, Huang Renxun said that companies have no choice and can only accept "accelerated computing." He explained that NVIDIA's technology can not only accelerate traditional workloads such as data processing, but also handle AI tasks that old technologies cannot cope with.

Huang Renxun also stated that Nvidia heavily relies on Taiwan Semiconductor for chip production, as Taiwan Semiconductor is far ahead in the chip manufacturing sector.

However, he also stated that NVIDIA has developed most of the technology internally, which allows the company to transfer orders to other suppliers. However, he said that such changes may lead to a decrease in the quality of their chips.

"The agility of Taiwan Semiconductor and their ability to respond to our needs is truly incredible. That's why we chose them because they are outstanding, but if necessary, of course, we can turn to other suppliers."

In addition, it is reported that the US government is considering allowing NVIDIA to export advanced chips to Saudi Arabia, which could help the country train and run the most powerful AI models. Some individuals working for the Saudi Data and AI Management Agency said Saudi Arabia is making efforts to comply with US security requirements to accelerate the process of obtaining these chips.

After the interview content was released, Nvidia's stock price turned from a drop to a rise during the day, closing up more than 8% at $116.91, which also led the Nasdaq to turn from a 1.6% drop to a 2.17% rise. This year, Nvidia's stock price has more than doubled, and it has risen 239% in 2023.

The following is an excerpt from Huang Renxun's interview:

1. First, let's talk about your thoughts when you founded the company 31 years ago. Since then, you have transformed the company from a GPU company focusing on gaming to a company that provides a wide range of hardware and software for the datacenter industry. Could you first talk about this journey? What were you thinking when you started? How did it evolve? What are your key priorities for the future, and how do you see the future world?

Huang Renxun: I would like to say that one thing we did right was that we foresaw that there would be another form of computing in the future, one that could enhance general computation and solve problems that general-purpose tools could never solve. This processor would initially do something that was extremely difficult for CPUs, which is computer graphics processing.

But we would gradually expand into other domains. The first domain we chose, of course, was image processing, which is complementary to computer graphics processing. We expanded it to physical simulation because in the video game domain we chose, you not only want it to be visually appealing but also dynamic, capable of creating virtual worlds. We expanded step by step and introduced it to scientific computing. One of the first applications was molecular dynamics simulation, and another was seismic processing, which is basically inverse physics. Seismic processing is very similar to CT reconstruction and is another form of inverse physics. So we solved problems step by step, expanded into adjacent industries, and eventually solved these problems.

The core concept we have always adhered to is that accelerating computation can solve interesting problems. Our architecture remains consistent, meaning software developed today can run on the large installed base you leave behind, and software developed in the past can be accelerated by new technologies. This way of thinking about architecture compatibility, creating large installed bases, and developing with the ecosystem started in 1993 and continues to this day. This is why NVIDIA's CUDA has such a large installed base, because we have been protecting it. Protecting the investments of software developers has always been our top priority.

Protecting the investments of software developers has always been our top priority. Looking to the future, some of the problems we have solved along the way include learning how to become founders, how to become CEOs, how to run a business, and how to build a company, all of which require new skills. This is somewhat like inventing the modern computer gaming industry. People may not know, but NVIDIA has the largest installed base of video game architecture in the world. GeForce has around 0.3 billion players and is still growing rapidly and very active. So, I believe that every time we enter a new market, we need to learn new algorithms, market dynamics, and create new ecosystems.

The reason we need to do this is that unlike general-purpose computers, once a general-purpose computer is built with a processor, everything will eventually run. But we are accelerating computers, which means you need to ask yourself, what do you want to accelerate? There is no such thing as a universal accelerator.

2. Let's talk more about the differences between general-purpose and accelerated computing?

Huang Renxun: If you look at modern software now, the software you write includes a lot of file input and output, parts where data structures are set up, and some magical core algorithms. These algorithms vary depending on whether they are used for computer graphics processing, image processing, or something else. It could be related to fluid, particles, inverse physics, or image domains. So these different algorithms are all different. If you create a processor that specializes in these algorithms and complement the CPU processing tasks it is good at, theoretically, you can greatly accelerate the running of applications. The reason is that usually 5% to 10% of the code accounts for 99.99% of the running time.

Therefore, if you offload that 5% of the code to our accelerator, you can technically increase the speed of the application by 100 times. This is not rare. We often can accelerate image processing by 500 times. Now what we are doing is data processing. Data processing is one of my favorite applications because almost everything related to machine learning is evolving. It could be SQL data processing, Spark-type data processing, or vector database type processing, dealing with unstructured or structured data, these are all data frames.

We greatly accelerate these, but to do this, you need to create a top-level library. In the field of computer graphics processing, we were fortunate to have Silicon Graphics' OpenGL and Microsoft's DirectX, but beyond these, there are no truly existing libraries. So, for example, our most famous library is a library similar to SQL. SQL is a storage computing library, and we created a library that is the world's first neural network computing library.

We have cuDNN (a library for neural network computing), cuOpt (a library for combinatorial optimization), cuQuantum (a library for quantum simulation and emulation), and many other libraries, such as cuDF for data frame processing functions similar to SQL. Therefore, all these different libraries need to be invented, they can rearrange the algorithms in the application to allow our accelerator to run. If you use these libraries, you can achieve 100 times acceleration, gain more speed, which is quite amazing.

So, the concept is very simple and meaningful. But the question is, how do you invent these algorithms and make the video game industry use them, write these algorithms and make the entire seismic processing and energy industry use them, write new algorithms and make the entire AI industry use them. Do you understand what I mean? Therefore, all these libraries, each library, first we must complete the research of computer science, and then we must go through the development process of the ecosystem.

We have to persuade everyone to use these libraries, and then consider which types of computers they run on, each computer is different. So, we step by step into one field after another. We have created a very rich library for autonomous driving cars, a very impressive library for robot development, and an incredible library for virtual screening, whether it's physics-based or neural network-based virtual screening, and an amazing library for climate technology.

So, we have to make friends and create markets. It turns out that what NVIDIA is really good at is creating new markets. We have been doing this for so long now that NVIDIA's accelerated computing seems to be everywhere, but we really have to complete it step by step, develop markets one industry at a time.

3. Many investors on the scene are very interested in the datacenter market. Can you share your views on medium and long-term opportunities? Obviously, your industry is driving what you call the "next industrial revolution." How do you see the current situation of the datacenter market and the challenges in the future?

Huang Renxun: There are two things happening simultaneously, they are often confused and discussed separately is helpful for understanding. First, let's assume that AI does not exist. In a world without AI, general computing has already stagnated. As we all know, some principles in semiconductor physics, such as Moore's Law and Denard scaling, have come to an end. We no longer see the doubling of CPU performance every year. We have been lucky to see performance double in ten years. Moore's Law used to mean a tenfold increase in performance in five years, and a hundredfold increase in ten years.

But now these have come to an end, so we must accelerate everything that can be accelerated. If you're doing SQL processing, accelerate it; if you're doing any data processing, accelerate it; if you're creating an internet company and have a recommendation system, it must be accelerated. The largest recommendation system engines are now all accelerated. A few years ago, these were running on CPUs, and now they are all accelerated. So, the first dynamic is that the global datacenter, worth trillions of dollars, will be modernized and transformed into an accelerated computing datacenter. This is inevitable.

In addition, because NVIDIA's accelerated computing has brought such a huge cost reduction, over the past decade, computing power has grown not at a rate of 100 times, but at a rate of 1 million times. So, the question is, if your airplane could go a million times faster, what would you do differently?

So, people suddenly realize, "Why don't we let computers write software instead of us imagining these features or designing algorithms ourselves?" We just need to give all the data, all the predictive data to the computer and let it find the algorithm - this is machine learning, generative AI. So, we have applied it on a large scale in many different data fields, where the computer not only knows how to process the data, but also understands the meaning of the data. Because it understands multiple data patterns at the same time, it can perform data translation.

Therefore, we can convert from English to images, from images to English, from English to proteins, and from proteins to chemical substances. Because it understands all the data, it can perform all these translation processes, which we call generative AI. It can convert a large amount of text into a small amount of text, or expand a small amount of text into a large amount of text, and so on. We are now in the era of this computer revolution.

And now, what is amazing is that the first batch of data centers worth trillions of dollars will be accelerated, and we have also invented this new type of software called generative AI. Generative AI is not just a tool, it is a skill. And because of this, a new industry is being created.

Why is that? If you look at the entire IT industry until now, we have been creating tools and instruments that people use. And for the first time, we are creating skills that can enhance human capabilities. Therefore, people believe that AI will surpass data centers worth trillions of dollars and the IT industry, and enter the world of skills.

So, what are these skills? For example, digital currency is a skill, autonomous driving cars are a skill, digitized assembly line workers, robots, digitized customer service, chatbots, digitized employees planning the supply chain for NVIDIA. This can be a digital agent for SAP. Our company uses ServiceNow extensively, and now we have digital employee services. So, we now have these digitized humans, and this is the AI wave we are currently in.

4. There is an ongoing debate in the financial markets about whether the return on investment is sufficient as we continue to build AI infrastructure. How do you evaluate the return on investment that customers get in this cycle? If you look back at history, at PC and cloud computing, how did their ROI compare in similar adoption cycles? What is different now?

Huang Renxun: That's a great question. Let's take a look. Before cloud computing, the biggest trend was virtualization, if you remember. Virtualization basically meant that we virtualized all the hardware in the data center into virtual data centers, and then we could move workloads across data centers without being directly tied to a specific computer. The result was increased utilization of data centers, and we saw a reduction in data center costs by two to two and a half times, almost overnight.

Then, we put these virtual machines into the cloud, and as a result, not only one company but many companies could share the same resources, costs decreased again, and utilization increased again.

All the progress in recent years has overshadowed the underlying fundamental change, which is the end of Moore's Law. We have achieved a two-fold, or even greater, cost reduction from increased utilization, but it has also encountered the limits of transistors and CPU performance.

Furthermore, all these improvements in utilization rates have reached their limits, which is why we now see data centers and computational inflation. Therefore, the first thing that is happening is accelerated computation. So, when you are dealing with data, such as using Spark - one of the most widely used data processing engines in the world today - if you use Spark and accelerate it with NVIDIA accelerators, you can achieve a 20-fold acceleration. This means you will save 10 times the cost.

Of course, your computation cost will increase a bit because you need to pay for NVIDIA GPUs. The computation cost may double, but you will reduce the computation time by 20 times. Therefore, you ultimately save 10 times the cost. And such return on investment is not uncommon for accelerated computation. So, I suggest accelerating anything that can be accelerated and using GPUs for acceleration, so you can immediately get your investment return.

In addition, the discussion of generative AI is the first wave of AI at the moment. Infrastructure players, such as ourselves and all cloud service providers, are putting their infrastructure in the cloud for developers to use these machines for training models, fine-tuning models, securing models, and so on. Because the demand is so high, for every $1 spent on our side, cloud service providers can earn $5 in rental income. This situation is happening globally, and everything is in short supply. Therefore, the demand for this kind of demand is very high.

We have seen some applications that are well-known, including OpenAI's ChatGPT, GitHub's Copilot, or the shared generator we use internally, and the productivity improvement is incredible. Every software engineer in our company now uses the shared generator, whether it's the one we created for CUDA, or the one for USD (another language we use), or the generators for Verilog, C, and C++.

Therefore, I believe the days when every line of code is written by software engineers have come to an end. In the future, every software engineer will have a digital engineer by their side, assisting their work 24/7. That is the future. So, when I look at NVIDIA, we have 32,000 employees, but there will be many more digital engineers around these employees, possibly 100 times more digital engineers.

5. Many industries are embracing these changes. Which use cases and industries are you most excited about?

Huang Renxun: In our company, we use AI in computer graphics. Without artificial intelligence, we cannot continue in computer graphics. We calculate only one pixel and then extrapolate the other 32 pixels. In other words, we 'imagine' the remaining 32 pixels to some extent, and they are visually stable and appear photo-realistic. The image quality and performance are excellent.

Calculating one pixel requires a lot of energy, while predicting the other 32 pixels requires very little energy and can be done very quickly. Therefore, AI is not just about training models, that is only the first step. What is more important is how to use the models. When you use the models, you save a lot of energy and time.

Without AI, we would not be able to provide services to the autonomous driving industry. Without AI, our work in the fields of robotics and digital biology would also be impossible. Almost every technology and life science company is now centered around NVIDIA, using our data processing tools to generate new proteins, small molecule synthesis, virtual screening, and other areas that will be completely reshaped by artificial intelligence.

6. Let's talk about competition and your competitive barriers. There are currently many public and private companies that want to challenge your leadership position. How do you view your competitive barriers?

NVIDIA: First, I think there are a few things that make us different. The first point to remember is that AI is not just about chips. AI is about the entire infrastructure. Today's computers are not about manufacturing a chip and people buying it and putting it into a computer. That model belongs to the 1990s. Today's computers are developed under the names of super computing clusters, infrastructure, or supercomputers. It's not just a chip, it's not entirely a computer.

So, in fact, we are building the entire data center. If you take a look at one of our super computing clusters, you will find that the software required to manage this system is very complex. There is no 'Microsoft Windows' that can be directly used for these systems. This customized software is developed by us for these super clusters. Therefore, the companies designing chips, building supercomputers, and developing this complex software are naturally the same company, ensuring optimization, performance, and efficiency.

Second, AI is fundamentally an algorithm. We are very good at understanding how algorithms work, and how computing stacks distribute computation and run on millions of processors for days, maintaining computer stability, energy efficiency, and the ability to complete tasks quickly. We are very good at this.

Lastly, the key to AI computing is the installed base. Having a unified architecture across all cloud computing platforms and on-premise deployments is very important. Whether you are building super computing clusters in the cloud or running AI models on a device, there should be the same architecture to run all the same software. This is called the installed base. And this consistency in architecture since 1993 is one of the key reasons why we have achieved what we have today.

Therefore, if you want to start an AI company today, the most obvious choice is to use NVIDIA's architecture because we are present on all cloud platforms. Regardless of which device you choose, as long as it has the NVIDIA logo, you can directly run the same software.

7. Blackwell is 4 times faster in training and 30 times faster in inference than its predecessor product, Hopper. Your innovation speed is so fast, can you maintain this pace? Can your partners keep up with your pace of innovation?

Huang Renxun: Our basic innovation approach is to ensure that we continuously drive architectural innovation. The innovation cycle of each chip is about two years, in the best case scenario, it is two years. We also perform midterm upgrades on them every year, but the overall architectural innovation is about once every two years, which is already very fast.

We have seven different chips that work together for the entire system. We can launch new AI supercomputing clusters every year that are more powerful than the previous generation. This is because we have multiple parts that can be optimized. Therefore, we can deliver higher performance very quickly, and these performance improvements directly translate into a decrease in Total Cost of Ownership (TCO).

The improvement in performance with Blackwell means that customers with 1 gigawatt of electrical utilities can generate three times the revenue. Performance directly translates into throughput, and throughput translates into revenue. If you have 1 gigawatt of electrical utilities available, you can generate three times the revenue.

Therefore, the return on investment for this performance improvement is unparalleled, and the 3x revenue gap cannot be compensated by reducing chip costs.

8. How do you view the dependence on the Asian supply chain?

Huang Renxun: The Asian supply chain is very complex and highly interconnected. NVIDIA's GPU is not just a chip, it is a complex system composed of thousands of components, similar to the construction of an electric vehicle. Therefore, the Asian supply chain network is very extensive and complex. We strive to design diversity and redundancy in every link, to ensure that even if there are problems, we can quickly shift production to other places. In general, even if the supply chain is disrupted, we have the ability to make adjustments to ensure the continuity of supply.

Currently, we manufacture at Taiwan Semiconductor because it is the best in the world, not just a little bit better, but much better. We have a long history of cooperation with them, and their flexibility and scale capabilities are impressive.

Last year, our revenue saw significant growth, thanks to the rapid response of the supply chain. Taiwan Semiconductor's agility and their ability to meet our needs are remarkable. In less than a year, we have greatly increased our production capacity, and we will continue to expand next year and further expand the year after. Therefore, their agility and capabilities are excellent. However, if needed, we can certainly turn to other suppliers.

Your company is in a very advantageous market position. We have discussed many very good topics. What are you most worried about?

Huang Renxun: Our company currently collaborates with every AI company in the world, as well as every data center. I don't know of any cloud computing service provider or computer manufacturer that we do not collaborate with. Therefore, with such scale expansion, we shoulder a great responsibility. Our customers are very emotional because our products directly affect their income and competitiveness. The demand is enormous, and there is also great pressure to meet these demands.

We are currently in full production of Blackwell and plan to start shipping and further expand in the fourth quarter. The demand is so high and everyone wants to get the product as soon as possible and get the largest share. This tension and intense atmosphere are unprecedented.

While it is very exciting to create the next generation of computer technology and see the innovation of various applications, we bear a great responsibility and feel a lot of pressure. But we are trying our best to do a good job. We have adapted to this intensity and will continue to work hard.

Editor/Somer