Jensen Huang Exchange Meeting 15,000-word Transcript: Discussing China-US Chips, H20, CUDA Compatibility, Extremely High Information Content

NVIDIA CEO Jensen Huang visited China for the third time this year, making headlines across major media outlets.

At the opening ceremony of the Chain Expo on July 16, Huang appeared in a traditional Tang suit, and during the afternoon media meeting, he changed into his signature leather jacket, transforming into a "tech rock pioneer."

Although the change in style sparked considerable discussion among netizens, what attracted more attention than his appearance was the sincerity he conveyed during this visit.

At the meeting, Huang told Tencent Technology: "I originally thought I could give a complete speech in Mandarin, but my Mandarin is not good enough; it sounds a bit too stiff. I practice every time I come to China, trying to make it more natural and fluent."

Huang stated that he had two major observations during this visit: First, he was deeply impressed by China's highly developed supply chain, with its overall infrastructure being complete, ecosystem mature, technology advanced, and manufacturing scale enormous. Second, this year's exhibiting companies span the global technology ecosystem, which fully reflects the high interdependence of the global supply chain. Huang also specifically pointed out that the interconnectivity of technology ecosystems has become a key issue in industrial evolution.

Regarding the phenomenon of Chinese manufacturers developing CUDA-compatible platforms, Huang told Tencent Technology that he does not mind compatibility, as CUDA itself is quite open. Huang also joked: "If you develop products compatible with (Intel) X86 architecture, they might be unhappy. But if you develop products compatible with CUDA, I won't mind at all."

The following are the core points from Jensen Huang's media exchange:

1. China's supply chain system is complex, efficient, and has significant manufacturing advantages, forming an important foundation for global AI hardware and smart factory construction.
2. The H20 chip has been re-approved for sale, which will promote more Blackwell architecture products landing in China in the future.
3. China is globally leading in AI models, engineering talent, and industrial applications.
4. Huang gave high praise to Chinese car manufacturers such as Xiaomi Auto, NIO, and XPeng, stating that Chinese electric vehicles are a global surprise, and companies like Xiaomi are reshaping the global competitive landscape.
5. Huang commended Huawei for making significant breakthroughs in chips, network solutions, and photonic technology, calling it a model to learn from.
6. China's education system has cultivated some of the world's best AI researchers, with about 50% of AI researchers globally located in China.
7. Huang praised Lei Jun, stating that from the first day they met, he knew Lei Jun would succeed.
8. Huang mentioned that he regularly uses four AI tools: OpenAI, Gemini Pro, Claude, and Perplexity, without relying on a single tool. He employs a multi-AI comparative questioning approach to enhance the accuracy and comprehensiveness of information through cross-validation of different models' responses.
9. He encouraged young people to maintain their passion for technology and not to fear "entering the game late," emphasizing that AI is still rapidly evolving. Huang recounted his experience of graduating at 20 and founding NVIDIA, stating, "The luckiest thing is doing what you love."
10. Jensen Huang believes that according to current assessment standards, AGI may be achieved soon without fundamental technological breakthroughs, but specific progress depends on how AGI is defined.

The following is a transcript of the media conference, edited by Tencent Technology:

Jensen Huang's "Small Goals" for Coming to China

Q: This is your third visit to China this year. What is the main purpose of this visit?

Jensen Huang: I originally hoped to give a speech in Mandarin, but my Mandarin is not good enough yet. However, I will continue to set goals and work hard to improve, aiming to speak in Mandarin next time. This time, I was invited to attend the opening ceremony of the China International Supply Chain Expo, which is a very important event covering multiple aspects.

First, China's supply chain is very complex and advanced. The supply chain involves long-term cooperation and connections among numerous suppliers, capabilities, and technologies, and China's supply chain infrastructure, ecosystem, and technological level are very advanced, with a massive manufacturing scale.

Secondly, many multinational companies are attending this year, showcasing how global supply chains are interdependent. I am also very pleased to participate in the celebration of the opening ceremony.

Q: What are your plans moving forward? Will you continue to collaborate with Chinese companies?

Jensen Huang: NVIDIA has been established for over 30 years. When I first came to China, companies like Baidu, Alibaba, Tencent, and Xiaomi did not exist. I came very early, and it is my honor to collaborate with these companies, especially when NVIDIA was still young. I am very proud of this, especially of the technologies and businesses we have created together with our partners in China.

Q: Do you still plan to invest in China, or will you just maintain existing operations?

Jensen Huang: If you want to maintain growth, you must continue to invest. The market is developing so rapidly, and the competition is extremely fierce. All my competitors are not just maintaining the status quo; they are continuously investing to strive to become global leaders. Therefore, we must keep pushing forward, and we cannot afford to stop even for a day to be worthy of the business we have.

Q: Can you talk about the demand for the H20 chip from Chinese customers? You mentioned that we all face computing power anxiety. How is the market demand after the news of the lifting of restrictions was announced yesterday?

Jensen Huang: After the announcement yesterday, I have been accepting interviews and haven't had the chance to meet with customers yet! Customers should be given time to make purchasing decisions. We are fully preparing to meet the demand as quickly as possible. However, the recovery of the supply chain still takes time. NVIDIA's current supply chain cycle is as long as 9 months, meaning it takes a full 9 months from placing a wafer order to delivering the finished AI supercomputer. We are working hard to accelerate the recovery of the production capacity of the Hopper architecture products, but it indeed requires continuous effort.

Q: You have been to China three times this year. What are the different goals and gains each time?

Jensen Huang: Yes, I have been to China three times this year, all due to invitations. The first time was during the Spring Festival; if it weren't for the enthusiastic invitation from employees, I could have spent the Spring Festival in the United States. This time, I was invited to give a keynote speech at the supply chain conference. In short, as long as I receive an invitation, I am very willing to come to China again Q: Will you go to other cities in China?

Jensen Huang: After this trip, I have to go home first. I haven't been back for too long, and I miss my little dog!

The U.S. Government's Attitude Towards Jensen Huang's Visit to China and H20

Q: What was the U.S. government's reaction to your visit to China?

Jensen Huang: I was in Washington a week ago, and before coming to China, I communicated with President Trump and his cabinet officials. President Trump has always encouraged me to celebrate an important milestone for NVIDIA—our market value has surpassed $4 trillion. He is very proud of this achievement and happy for me. He told me to have fun when I go to China. I have very high expectations for China and look forward to seeing the future development of the Chinese market.

Q: Due to Trump's chip ban, NVIDIA wrote down $4 billion in assets. Can this loss be recovered? Additionally, the Trump administration has imposed many restrictions in the semiconductor sector. What do you think about the impact of these bans on the industry?

Jensen Huang: Regarding these restrictions, we may not be able to fully recover the previous inventory write-down losses, but most of the inventory can be recovered. The new customer demand can match our existing inventory products, so I can't say it will be 100% recovered, but it's not completely impossible either.

As for tariffs and trade, the world is indeed readjusting. As a company, we must adapt to changes in global trade, taxation, and tariffs. In fact, trade and tax policies existed before I joined NVIDIA and will continue to exist after I leave. So we must learn to adapt to these changes.

What is most surprising is that NVIDIA is very good at adapting to change. Our supply chain will also adjust accordingly, and we will find new solutions. In the long run, the key is to ask ourselves: Are we making products and technologies that are meaningful to the world? Are we seriously creating quality products that protect our customers' interests? Does what we do have an impact on the world? As long as these align, other issues will be resolved.

Q: Regarding the Trump administration's allowance to sell H20 chips to China, does this mean that China will relax other export controls? After all, export controls are both a pillar of national security and a governance mechanism for global trade.

Jensen Huang: I am not familiar with the specific details of the negotiations between the two countries. In the long run, this indicates that each country has unique expertise in certain areas. These core capabilities can indeed promote economic prosperity and security, which is the cornerstone of national security. But it should also be recognized that these advantageous areas are likely to become important bargaining chips in future trade negotiations. This could be a key insight.

Q: During this trip to China, what different insights do you have about your competitors? Have you been following their developments? Additionally, given the fluctuating U.S.-China trade situation, American tech companies are more or less affected. Are you worried about this?

Jensen Huang: First, to answer your question about competitors—no, I did not underestimate any competitors during this trip. We should respect all competitors, learn from them, draw inspiration, and turn that into motivation to move forward. Undermining or underestimating opponents is harmful and offers no benefits. I maintain a high level of respect for competition globally As for export controls, these are external factors that we cannot control and may indeed disrupt NVIDIA's business. Our responsibility is to explain the intent of the policies and their potential unintended impacts to the government, but ultimately, policy-making is the government's responsibility. What we can do is provide factual evidence, and the rest is beyond our control. Companies must remain sufficiently agile to better adapt to changes. We do our best, and the rest will follow naturally. Question: Have you changed your view on Trump's H20 ban?

Jensen Huang: I have not changed his position. My responsibility is to inform the president about matters I am very familiar with, namely the development of the technology industry and AI globally. This is a once-in-a-lifetime opportunity for the U.S. to maintain a lead in AI technology, and similarly, China also has the opportunity to lead in AI. If we want to be leaders, we must collaborate with global developers and interact with global markets.

I have always believed that technological leadership requires a large market, as a big market can generate revenue and profits to support subsequent investments. After all, H20 is very suitable for training large models and remains excellent. RTX Pro is specifically designed for digital twins and robotic simulations, and the two serve different purposes.

Question: How do you think to balance the relationship with the governments of both the U.S. and China in business cooperation?

Jensen Huang: There is actually no so-called secret to balance. As a global company, we need to maintain good relationships with governments around the world. The technology developed by NVIDIA has global value, which gives me the opportunity to engage deeply with governments. Through continuous communication, we help them better understand our technology, leading to more reasonable policies and regulations.

Question: Do you think the relationship between the U.S. and China in the AI field is mainly competitive or cooperative?

Jensen Huang: The AI field has a unique openness. Compared to other disciplines, the number of papers published in AI is the highest, and the sharing of foundational technologies is the most transparent. Through international conferences like ICLR, global researchers can engage in "virtual collaboration" through papers, even if they have never met. This characteristic of openness and sharing makes collaboration an essential feature of AI development.

Question: The U.S. Secretary of Commerce said H20 is the best product. Is this true?

Jensen Huang: I have many products, and I never rank them. We have many products designed for different application scenarios. One advantage of H20 is its outstanding memory bandwidth, which is very suitable for the models being developed here, such as DeepSeek and innovative architectures like Qianwen. These models are very suitable for H20. So I believe H20 will be very successful and definitely very useful.

Question: What is the future order situation for H20 chips?

Jensen Huang: There are currently several uncertainties: first, the timing of new orders from customers, as previous orders were canceled, and their needs may have changed; second, restarting the supply chain takes time and cannot be done overnight. Therefore, in the short term, we cannot predict the order volume, nor can we accurately assess supply capacity. But we will do our utmost to meet customer needs, as the "computing power anxiety" that currently exists in the industry indeed needs to be alleviated Q: Are there currently any companies placing orders for H20 chips?

Jensen Huang: Not yet, as the relevant licenses have not been approved. The U.S. government departments are in the approval process, and results are expected soon. Once approved, we will be able to officially accept orders.

Q: Given the ongoing global uncertainty and the job market being largely constrained by regulatory policy changes, how do you respond?

Jensen Huang: There is no balancing issue to think about here. First, we must recognize that each country has its own national security and trade policies, which is the right of the nation, and we need to comply with these policies. If there are reasons that lead us to believe our information is useful to them, it is our responsibility to explain how our industry operates and how the technology works. So we should respond to this just like we would to any policy.

Secondly, what we need to do is build the best technology, contribute to the market, hope that customers can see the value of our work, and continue to respect us and our existing partners, maintaining long-lasting friendships. After that, life is no longer under your control.

Jensen Huang's View on Chinese AI

Q: What do you think of the performance of Chinese AI companies in the AI chip field, such as Huawei?

Jensen Huang: There are many outstanding companies in China. For example, Huawei not only makes chips, systems, and networks but also involves itself in many fields. As a large-scale company, Huawei has a deep tradition and excellent technical strength, making it very powerful.

From the perspective of the overall development of AI in China, it is important to understand that AI is essentially an infrastructure technology that involves multiple levels such as chips, systems, and networks, while AI develops models and applications on top of this infrastructure.

Overall, this involves three levels: computing, models, and applications. China has made rapid progress in this regard, especially at the model level. Companies like DeepSeek, Alibaba, and Moonlight have made significant breakthroughs in technology.

It is worth mentioning that DeepSeek is the world's first open-source inference model, which is undoubtedly a major breakthrough. In terms of applications, China is progressing even more rapidly, with many new applications emerging. China is not only creating new technologies but also quickly integrating and applying these technologies.

Competition in China is very fierce. Many people aspire to start great companies or succeed in emerging businesses. This has led to the emergence of many excellent innovative talents. Behind all this, is China's strong education system, which has cultivated many of the world's best AI researchers. About 50% of AI researchers globally are in China. The advantages in science and mathematics here are very obvious, and computer science has its unique aspects, with the entire ecosystem being very active and developing at an astonishing pace. I am very glad to be personally involved in it.

Q: How do you view the performance of models like DeepSeek and kimi?

Jensen Huang: The models in China are all very impressive. In terms of the application scenarios they are designed for, they are highly efficient and open-source, so you can adapt these models to your needs. You can build companies, products, or businesses based on these open-source models These models have a very unique positioning globally, and each has its own advantages. I believe that over time, which model is the smartest will become less important; what matters more is which model is the most useful. Just like when you choose employees, I wouldn't give them an IQ test; I just want to know who is the most efficient and useful.

So I think that more and more questions about AI hinge on how it is used. How do you want to use AI? You want to choose what suits your needs.

Q: What do you think about the differences in AI development between the United States and China? Do you think there is room for cooperation between the two countries? Can they learn from each other?

Jensen Huang: A house should be able to accommodate at least two people. So, there is definitely room for cooperation. Fifty years ago, a hundred years ago, the world's GDP was much smaller than today, and the population was also much smaller, while the world's GDP will continue to grow in the future, and the world population will keep increasing.

I am very confident about the future of the United States and China. I believe the United States will continue to be vibrant, with all its excellent resources, and China will also continue to develop dynamically, leading more people to a better life. I am very optimistic about the future.

Q: What do you think about the future of the Chinese market?

Jensen Huang: I hope to achieve great success in China. China is the second-largest technology market in the world and is still growing rapidly; this market is very important. There are many vibrant and innovative customers here. If you want to build a company and provide quality services, you must have excellent customers and products. In China, we have many challenging customers, but they are also committed to innovation, and we want to provide them with the best service.

Q: When will Chinese customers use NVIDIA's latest GPUs? Can you describe their performance?

Jensen Huang: After the release of H20, its memory bandwidth is outstanding, especially in model inference, where H20 continues to perform excellently, particularly suitable for large language models, visual language, and action models.

In addition, we have also launched a new product, RTX Pro, specifically designed for digital twin applications. Digital twins create a virtual world, a digital robot. The reason for needing a digital robot is to train it in the digital world to teach the physical robot how to perform tasks excellently. RTX Pro was initially built to run a new application platform called Omniverse, which is crucial for digital factories, smart factories, and robotics. Therefore, RTX Pro is a unique innovative product and can be considered a world first.

I believe that in China, where robot innovation is thriving, smart factory construction is in full swing, and the supply chain system is extremely complete, RTX Pro will surely shine. I am incredibly excited about this.

Q: Recently, China's local manufacturing industry has begun to launch quite competitive alternatives. What do you think about this trend? Are you considering open-sourcing technologies like CUDA?

Jensen Huang: That's a great question. If a platform is compatible with CUDA and can support these compatible applications, then I think it's fine. In fact, CUDA itself is already quite open; you can refer to the detailed documentation of CUDA and then develop compatible versions based on that Therefore, to some extent, CUDA is already "open source." This is different from Intel's x86 architecture. If you develop products compatible with x86, they might be unhappy. But if you develop products compatible with CUDA, I wouldn't mind at all.

Question: You said, "China's open-source AI is a catalyst for global progress." What do you mean by that? How do you view China's open-source AI?

Jensen Huang: DeepSeek R1 is a revolutionary model. It is the first open-source inference model and is highly efficient. Many companies and countries have downloaded DeepSeek R1. Although not everyone uses it, 99% of users have downloaded this open-source model and adapted it according to their application needs, with usage worldwide. We see it being applied in various fields, including healthcare, energy, and more.

So openness allows every industry and every country to participate. AI should not only serve a few people or a few countries; I believe it is designed for everyone in the world, and everyone should and will benefit from it, ultimately making society better.

AI is the greatest technology; it is the biggest equalizer in anything, and we can prove this every time. The reason is that with AI, a poor artist can become a great artist, and a poor writer can become a great writer. We are enhancing everyone's capabilities. Lastly, I want to say that regarding the openness of AI, because it is open, it invites scrutiny from the global scientific community. This scrutiny and openness ensure safety.

Question: What insights does China's technological innovation provide to the world?

Jensen Huang: China has nurtured many unique innovations, such as WeChat, Xiaohongshu, and the globally popular TikTok, which have profoundly changed the global internet ecosystem. In fact, China has a unique advantage in developing new applications. In China, mobile payments have become fully popularized, and people no longer use wallets, cash, or credit cards. Meanwhile, I am still using paper money in the U.S., which makes me feel like an antique. China is indeed at the forefront of technological applications, especially in payment systems and fintech. Global companies are learning from China's innovative practices.

Optimistic about China's Electric Vehicles

Question: Are there any new products from China that have surprised you, even ones you hadn't thought of or addressed before?

Jensen Huang: The development of electric vehicles in China is truly amazing. I just met Lei Jun a few days ago, and he showed me his new car. The technology is incredible, the design is very beautiful, and the craftsmanship is exquisite. The technology inside the car is top-notch in the world.

Question: Would you buy one?

Jensen Huang: I would love to buy one, although they are not yet sold in the U.S., which is unfortunate. In fact, the configurations inside the car are fantastic, and the electric vehicles are great early models. Besides Xiaomi, there are Nio, Li Auto, and XPeng, which have very spacious cars, almost like a living room on wheels, like a New York party. What I want to say is that in the past five years, China's electric vehicles may be one of the biggest surprises in global progress. From a technological perspective, these cars are absolutely unique, and the quality is definitely top-notch Q: How is your cooperation with Xiaomi in the automotive field going?

Jensen Huang: We have close cooperation with Xiaomi in many areas. Xiaomi is a fantastic partner. The first time I met Xiaomi's founder Lei Jun in Beijing, he was still very young, and I was young too, but he was even younger. From day one, I knew he would succeed. Lei Jun has led Xiaomi to great success from smartphones to electric vehicles, which is commendable. Q: What did you talk about with Lei Jun?

Jensen Huang: We talked about AI, including the latest developments in AI in language models, autonomous driving, and robotics. As a leader with a strong technical background and excellent business acumen, Lei Jun's insights on these topics are impressive. Notably, I had the privilege of witnessing the highly anticipated new electric vehicle. Although I had learned about this car online long ago, seeing it in person was still shocking. He enthusiastically showed me the newly launched Ultra version, whose outstanding performance completely exceeded my expectations.

Q: They are also involved in air conditioning and other equipment!

Jensen Huang: That's right, it's incredible. They have many innovations under Xiaomi's system. Many people usually think that China's technology is mainly reflected in hardware, such as batteries, electric vehicles, solar panels, materials science, and even computer systems and networks.

But I believe there is an underestimated miracle: China's computer science and software capabilities are world-class. In the future, everything will be closely connected with software, whether it's hotel management systems, rice cookers, microwaves, refrigerators, or even washing machines, all of which may have software.

Even inspection equipment will have software. My jacket should also have software, like automatically cooling me down. In the future, every piece of clothing you wear will have software, and glasses will also have software. In the future, every company, regardless of whether it is a hardware company, will be involved in software and deeply integrated with cloud technology.

Xiaomi and many companies in China, like Alibaba and Tencent, have unique performances in hardware and software that are unparalleled in the world. I think more people should celebrate this. For example, applications like ByteDance may seem like consumer products, but the technology behind them is extremely advanced.

Q: In recent years, Chinese companies have made significant breakthroughs in areas such as large language models and autonomous driving. How does NVIDIA view China not only as an important market but also as a source of innovation potential? Do you have plans to jointly develop autonomous driving technology with Chinese tech companies?

Jensen Huang: While we do not directly manufacture autonomous vehicles, we develop the core technologies required for them. Similarly, we do not develop large language models, but we build the foundational technology architecture for them. By closely collaborating with innovative partners, we can drive continuous progress in our technology. Therefore, we always value and seek cooperation with innovative companies, as their needs also promote our technological innovation.

Computing power demand is far from enough

Q: Do you think computing power is overestimated? Or, if there is not enough computing power, can companies not develop AI? I guess you must have heard the term "computing power anxiety." What do you think about this? Jensen Huang: Are you saying that anxiety arises when there is too much or too little computing power? Currently, it is obviously insufficient. I have a deep understanding of this myself. At NVIDIA, without sufficient computing resources, our research and development of chips, systems, network architectures, algorithms, and software would be extremely difficult. Nowadays, we have equipped engineers with many AI supercomputers, but they are still worried about computing power every day, wishing they could have ten times more resources.

This fully illustrates that the computing revolution of the new era is no longer limited to keyboard input and manual programming, but has entered a new era of AI machine learning. When AI takes on the main learning tasks, we naturally crave more powerful and faster computers to solve more complex and challenging problems. I believe that computing power anxiety will not disappear in the next decade; it is a common challenge faced by all engineers.

Question: There are three key factors in the AI field: algorithms, computing power, and data. However, I am not sure if you think people are more focused on computing power now? Without computing power, can we even talk about AI?

Jensen Huang: That's a good question. In fact, you may not fully understand the underlying technological context; the development of AI has evolved in stages.

The first generation of AI is perceptual intelligence (Wave 1), the second generation is generative AI (Wave 2), and now we are in the third generation: reasoning AI. Reasoning is akin to the human thought process; for example, when you encounter a brand new problem, you need to analyze: Is this safe? Is it a threat or an opportunity? What action should be taken? This reasoning ability is powerful because it represents the foundation of intelligence. The next stage will be robotics. Therefore, the current development of AI is divided into four stages: perception, generation, reasoning, and robotics.

Early perceptual AI primarily learned through massive amounts of data, such as recognizing text, images, understanding story content, and even distinguishing between cats and dogs. Today's AI resembles human learning more—shifting from passively receiving information to actively thinking. Just like when we were in school, we mainly relied on listening to lectures and reading, but after graduation, we learned more through practice and reflection. Today's AI is like doing math problems at home, continuously reasoning and thinking through multi-scenario simulations, becoming increasingly intelligent. We call this learning method "post-training," while traditional data training is referred to as "pre-training." The post-training model requires high computing power but has relatively low data volume requirements.

Question: Some have suggested that as large models develop, the demand for AI computing infrastructure will decrease. What do you think about this?

Jensen Huang: DeepSeek R1 adopts a highly innovative and efficient architecture, with its uniqueness lying in the fact that the computational requirements and resources needed for each token are significantly reduced. This design innovation is inevitable—because DeepSeek R1 is essentially a reasoning engine that requires iterative calculations. This means that each generated token must be extremely efficient to ensure the smooth operation of the overall reasoning process Now, the required computational load has significantly decreased, allowing the system to think for longer periods and generate multiple high-quality answers. This has actually opened up a whole new field—Gentle AI, which is a very important direction.

Inference AI models will be widely applied in cloud and edge computing. Take autonomous driving as an example: when a vehicle encounters an unfamiliar road condition, the onboard system equipped with inference AI can actively think—what is the situation ahead? How should I respond? Rather than simply judging whether to continue driving. Current onboard systems are too simplistic, and future inference AI will give them true thinking capabilities.

Edge devices will deploy AI with thinking capabilities, while also being able to carry simple functional AI (such as automatic door opening, etc.). The specific type of AI used will depend entirely on the application scenario.

Question: Existing computing power can already train models with a certain level of intelligence. To achieve AGI, is a fundamental breakthrough in GPU architecture and model design necessary?

Jensen Huang: It depends on the definition of AGI. If AGI is measured by the current standards used to evaluate AI performance, I believe we will soon surpass human teams in these benchmark tests. Based on the technological breakthroughs we have already achieved, as well as unpublished laboratory results, I believe that a fundamental technological revolution is not necessary to achieve this goal in the near future. Of course, it depends on how AGI is defined.

Question: NVIDIA is fully committed to building AI factories. China has a huge manufacturing base; how do you think AI will help traditional factories upgrade?

Jensen Huang: AI factories will have greater agility and reconfigurability. When the products to be produced change, the factory can complete the production line reconfiguration more quickly while achieving higher production throughput. This means that factories will become more flexible, more configurable, and have higher capacity—these can meet all our expectations for future supply chains. With the combination of robotics technology and AI capabilities, production costs will significantly decrease, and lower costs will help alleviate inflationary pressures while improving overall production efficiency.

NVIDIA's Cooperation and Competition

Question: I have two questions: the first is whether you plan to restart the production of the Hopper series chips? The second question is, given that China has included the H20 chip as part of the negotiations, do you think you are overestimating the competitiveness of NVIDIA's own chips?

Jensen Huang: You cannot exaggerate the facts. In fact, anyone who underestimates Huawei or the manufacturing capabilities of China is extremely naive and ignorant. Huawei is a very powerful company, and I have seen the technologies they have developed.

It is important to understand that I have been in this field for 30 years, while they have only been doing it for a few years. And they have already been able to compete with us, which speaks volumes about their strength. No one, no one has invested their life into building something more than I have. I have poured immense passion into this... Their ability to come close to us says it all I hope to bring more advanced chips to the Chinese market, such as the H20. The reason is that technology is always advancing. Today, the H20 is still very excellent because it has reasoning capabilities and the bandwidth brought by the Hopper architecture, which is really very good. Today, Hopper is outstanding, but in a few years, we will have more and better technologies. I believe that as long as these products are allowed to be sold in China, they will continue to improve.

Question: Regarding the HBM supply chain, SK Hynix is currently the largest supplier. Is it possible to see companies like Samsung or Micron entering the supply chain in the future, especially for next-generation products?

Jensen Huang: I will not choose between SK Hynix, Micron, and Samsung; all three companies are very important partners for us. The HBM, GDDR, and LPDDR memory we use are all the most advanced technologies in the world. These three companies perform exceptionally well in these areas. The future development of AI requires a large amount of computing power and memory technology, and these companies have very good prospects. We will continue to cooperate with them.

In the past 20 years, NVIDIA has achieved tremendous growth. The development of AI technology has just begun, and it will become the infrastructure for every country and every industry. We have just started building AI infrastructure, and we have only been doing it for two years. This infrastructure is essentially AI factories—input power, output tokens (i.e., AI intelligence). These factories are very unique, equipped with computers and storage devices. Therefore, I believe that the future development prospects for SK Hynix, Samsung, and Micron are very broad. We maintain cooperative relationships with all three companies.

Question: Regarding the dysprosium capacitors used in the Blackwell chip: Since April 4, China has almost suspended the export of dysprosium and other rare earth elements. Has this affected NVIDIA?

Jensen Huang: The impact is limited. Because the dysprosium capacitors we use are among the most complex and advanced components in the world, and the usage among student groups is not large. I believe that the existing global inventory is sufficient for our normal use.

Question: You are optimistic about humanoid robots. How do you view the development of related industries in China?

Jensen Huang: Three key points:

First, the world is facing a serious labor shortage. Although the current unemployment rate is low, the labor gap in manufacturing is not in the millions but in the tens of millions. Increasing automation levels will significantly drive global GDP growth.

Second, the technological breakthroughs in humanoid robots and AI come at a timely moment, presenting rare development opportunities.

Third, China has unique advantages because you already have top-notch AI technology reserves, strong capabilities in mechatronics, and the largest application scenarios and manufacturing base in the world. These three advantages make me optimistic about the development prospects of China's robotics industry.

Question: How is the cooperation progress with robotics companies like Yushu? Jensen Huang: We are committed to supporting all approved Chinese companies, whether they are emerging technology firms or industry leaders. As long as it is within the scope of permission, we will provide full support. China has many amazing innovative companies, and I believe that companies like Xiaomi and Yushu will launch impressive robotic products in the future.

Question: What is your evaluation of Huawei's breakthroughs in system-level technology? How will this affect NVIDIA's core strategy when adjusting its strategy in China?

Jensen Huang: First of all, Huawei has indeed achieved remarkable accomplishments. They have excelled in chip design, systems engineering, network technology, silicon chips, and photonics, and they also have their own cloud services. NVIDIA focuses on chip, system, and software development and does not provide cloud services. Huawei can independently explore the market with a complete ecosystem, and this comprehensive strength is indeed admirable. We are always learning from all competitors, including Huawei. But ultimately, they are indeed doing very well.

Question: Why have the two sides failed to reach a cooperation?

Jensen Huang: I am not deliberately avoiding the topic of cooperation. In fact, NVIDIA also has unique advantages in areas of technical expertise. We are one of the few companies in the world capable of achieving full-stack innovation from algorithm architecture, system software, hardware systems, network technology to chip design. It can be said that historically, no company has been able to maintain top-notch innovation capabilities across such a wide range of technical fields as we have.

This full-stack technology integration capability is rare in the history of human technology. Typically, expertise in software algorithms, system integration, chip design, and network technology is dispersed among different companies, but at NVIDIA, these capabilities are organically integrated—from infrastructure to language model algorithms, from robotics to computer graphics, and even quantum chemistry and 5G RF technology. The convergence of such diverse professional fields within the same company is indeed astonishing. If I can say this about my company, I feel very proud.

Question: Amazon, Google, and Huawei are developing their own AI chips. How do you view the competition among cloud computing providers?

Jensen Huang: Regarding the issue of cloud service providers developing their own chips, our company strategy is different from theirs. Our primary strategy is to invest heavily in building the most advanced AI systems. These AI systems are designed to support the entire lifecycle of AI—from pre-training, post-training, reinforcement learning, to the inference process.

Our AI architecture is open, and anyone can use it. If you understand our architecture and language, you can use anyone's chips or build computers for your company. Therefore, in a sense, our system is easier to use. My job is to ensure that, first, our performance and technology are state-of-the-art; second, we ensure the usability and practicality of our system throughout the entire process from the beginning to AI applications; third, our architecture is ubiquitous. So, no matter where you want to do AI, we can play an important role.

AI will fundamentally change every industry and will also fundamentally change every scientific field. Let me give you an example: in computer graphics and video technology, GPUs have brought explosive growth in graphic computing. However, now AI is revolutionizing computer graphics technology Today's computer graphics technology would not be possible without AI.

For example, the reason modern games are so captivating is that they apply a large amount of AI technology. We are also providing AI support for quantum computing because to generate useful results from quantum computers, you must have quantum error correction technology. The preprocessing and postprocessing of quantum signals are very important, and previous technologies were insufficient to address these issues.

We also use CUDA for scientific services, helping us with physical simulations, enabling us to conduct larger-scale weather predictions, and so on. In fact, this is a necessary technology in physics. Molecular dynamics, covariant bonds, electron energy, etc., are all very complex simulations. AI can accomplish this, and it is getting better at it, as is its application in robotics. We will use these technologies. I believe quantum computing is very important, and it is changing things.

Q: What challenges have competitors brought to you?

Jensen Huang: Competitors have brought various challenges globally. I actually hope they don't work too hard, but they are giving it their all; they all want to win. Although many competitors are my friends, I also want to win. I believe competition can make each other better, can make the market better, and can make the industry better. However, my competitors are all very strong.

In my career, I have experienced the lows of a company's market value nearly reaching zero, as well as the peak feeling of a market value reaching the top in the world. NVIDIA's market value was once zero and remained at zero for a long time. But we redefined computing, which is the most important tool in human history. Moreover, we created an entirely new industry, which is AI, and it will become a foundational technology.

NVIDIA's AI Industrial Layout

Q: You have talked about the technology stack; NVIDIA originated from the American technology stack?

Jensen Huang: Every company has its unique technology stack. The United States has many top-tier technology ecosystems, and NVIDIA represents one of them. The American computer industry—including chips, systems, software, algorithms, and network computing—constitutes a world-class standard and is a national strategic asset.

Throughout history, no industry has been able to occupy such a leading position globally as the American computer industry; it is hard to match in aviation, automotive, IT, or energy sectors.

The technological system of the American computer industry is akin to China's supply chain system and electric vehicle industry, both of which are national core competitive advantages. I can't think of any other American industries that can compare. This is precisely my view on the Texas chip industry ecosystem—the construction of a complete technology stack, including systems, software, algorithms, and infrastructure, is indeed very complex.

Q: Meta recently offered top AI talents a compensation package of $100 million; what do you think about this? Does NVIDIA also pay high salaries to many top talents?

Jensen Huang: I haven't talked to Mark Zuckerberg (Meta CEO) for a while, but it is clear that he has a very grand vision for the future. For him, it may be a company centered around AI, and this is a perfect opportunity for him Because Meta has already built a very strong GPU infrastructure for their business and has also created some of the largest infrastructures in the AI field globally.

Now, AI is clearly going to become a huge industry, not just hundreds of billions, but possibly a market worth trillions of dollars. This is a tremendous opportunity. I believe Mark has decided to fully invest in the AI field, and this is undoubtedly a wise choice. Right now, his pace of action is very fast, and I believe they will definitely succeed.

Question: Has Apple sought help from NVIDIA because they are lagging behind other tech giants in AI?

Jensen Huang: I believe Apple has many outstanding talents. I have no doubt about their excellence in AI. Apple has always been advancing at its own pace, but whatever they do, I firmly believe it will attract widespread attention.

Question: In the development of AI, have we overlooked security issues?

Jensen Huang: Many security technologies have already been invented, such as human feedback reinforcement learning (RLHF) to ensure AI performs tasks that align with human values, and retrieval-augmented generation (RAG) to reduce hallucinations and improve the accuracy of generated content. Additionally, there are technologies that set operational boundaries for AI to ensure it does not do what it shouldn't. For example, engineering AI, market AI, and financial AI each have different application areas. We have also created "AI guardrails" to ensure AI behavior remains within safe limits. There are also specialized technologies like cybersecurity AI and monitoring AI to prevent potential risks. In summary, AI will become a global science, and global cooperation can ensure its safe advancement while driving innovation.

Question: Why are you launching RTX Pro in China? What is your strategy for China? In the face of domestic and international competitive pressures, what strategies does NVIDIA have in place?

Jensen Huang: This week marks the opening of the third International Supply Chain Expo, which covers complex systems such as robotics and smart factories. We launched RTX Pro specifically tailored for applications like these digital factories, digital twins, and robotics. RTX Pro is very suitable for teaching robots how to become "robots" and for teaching smart factories how to improve efficiency and quality. Therefore, this week's expo is a very appropriate time to announce this product.

Question: I have two questions to ask. First, regarding the new GPU, can you share more details about RTX Pro? How does it differ from previously announced products? The second question is, can we expect more products based on the Blackwell architecture to enter the Chinese market in the future? Is there any latest progress you can share with us?

Jensen Huang: The new RTX architecture is based on Blackwell, while the previous RTX architecture was based on Hopper. In fact, we refer to it as Beta Hopper. This new GPU is different from H20; it has computational graphics and ray tracing capabilities. Ray tracing is very important for sensor simulation, such as lidar, radar, and computer graphics. We can simulate sensors in digital factories, or sensors in autonomous vehicles, or sensors on robots RTX Pro has capabilities that H20 and other AI products do not possess; it is a product that integrates both computational graphics and AI functions, specifically designed for digital twins. In applications, we have an AI library called Omniverse, where all operations run on Omniverse. You should take a look; it is truly impressive.

Q: If you could produce a perfect updated version of a digital twin, what would you focus on first?

Jensen Huang: I think if I had a digital twin, I would let it observe me for a while first. I would let it live with me for a period, observe what I do, and understand my personality, values, and characteristics. It should learn as much as possible. I would treat my digital twin as my partner so that we can both become smarter. Q: How do you view the current supply chain landscape?

Jensen Huang: Our main production base is not located in China. But it must be acknowledged that certain products, such as batteries, solar panels, and electric vehicles, can almost only be manufactured in China. China has established an extremely complete supply chain system, especially in the field of mass consumer goods manufacturing, which has irreplaceable advantages.

Young people shouldn't be afraid to enter the AI circle late; the key is passion

Q: What do you think about the talent competition in the AI and semiconductor fields? Are most of these talents Chinese or Chinese Americans?

Jensen Huang: I work in the semiconductor industry, and among the excellent AI talents I have encountered, there are many Chinese, as well as many Americans, and even talents from the UK, Canada, and other places are also very outstanding. The excellence of technology is not related to nationality; China has a very deep foundation in mathematics and scientific culture, which provides a good soil for the development of AI. Computer science and software are crucial to AI, and this is a very good foundation. Currently, China is well-prepared in this regard, and I believe China will be very successful in the field of AI.

Q: You have many excellent talents, and the salary treatment must be high, right?

Jensen Huang: We always do our best to ensure that employees receive the compensation they deserve. This stems from our company's unique corporate philosophy. You may know that NVIDIA is the smallest big company in the world. Although the company is not large, our technological contributions, industry influence, and market position are all astonishing.

Our employees have worked at the company for many years; they are passionate and are all very talented individuals. Because they have fought alongside me for a long time, there is a deep trust and understanding among us.

In fact, our employee turnover rate is almost zero, and this situation has persisted for many years with little change. I believe this is part of NVIDIA's charm. We not only pay high salaries but also do our utmost to take care of our employees, which is also a core concept in our culture.

This is also why I spend so much time explaining these issues to you; we always maintain our focus and respect for our employees. Q: Does NVIDIA have expansion plans in Greater China or the Asia-Pacific region?

Jensen Huang: We have always maintained an expansion trend in the Chinese market, and NVIDIA has achieved rapid growth in recent years. I hope NVIDIA's development is just the beginning, not the end. As long as we continue to drive technological innovation and create value for the market, relying on our excellent engineers and professional teams, we will surely continue to progress Q: I noticed the products you are using, what do people think? They might decide whether to continue supporting based on this information, as they come here to gain this knowledge.

Jensen Huang: I use a Google Pixel phone because NVIDIA is also involved in the development of the Android system. We have our own operating systems on multiple platforms such as Linux, Windows, and Android, all designed to meet the needs of different markets. The reason I like the Pixel phone is that it has a pure Android system, without unnecessary add-ons, simple and clean, and it runs very smoothly.

Q: It's graduation season now, and many young people are very interested in the AI industry. If you were a young person graduating in 2025 with dreams, what fields would you focus on? What if you feel it's too late to enter the AI industry?

Jensen Huang: I graduated at 20 because I graduated two years early, which gave me a first-mover advantage, and it was indeed good. In fact, NVIDIA belongs to the previous generation of semiconductor companies, and now I represent the next generation. So I never feel it's too late; the key is whether you have a good strategy. If you are at the front, that's a good thing; if you are behind, as long as your strategy is correct, you can still succeed.

Opportunities always exist; the key is to maintain a sharp eye. I have always been passionate about computer technology, and I feel very lucky to be able to work in a field I love. For young people in their early 20s today, they may not choose computer science upon graduation; they might lean more towards physics or software science, but I believe the field of artificial intelligence will also attract their interest.

Q: After becoming wealthy, what other dreams do you have?

Jensen Huang: Honestly, I don't have any. Because a long time ago, I was already very wealthy. Wealth simply means you have enough money to not worry about your children's and family's lives, to send them to school, and to create a better life for your family. Then you can do other things, like founding NVIDIA, taking more risks, or perhaps helping others and contributing to society. I have probably been wealthy for 30 years; NVIDIA went public in 1999, which is almost 25 years ago. Now I just want to devote 100% of my energy to my family and building the company.

Q: What AI tools do you use daily?

Jensen Huang: I mainly use four AI tools: OpenAI, Gemini Pro, Claude, and Perplexity. I usually ask them the same questions simultaneously to gain a more comprehensive insight by comparing the different AI responses. This multi-AI collaborative working method not only improves decision quality but also constitutes a safety mechanism—multiple AI systems can verify and balance each other. I believe this represents the direction of future development.

Q: Will you come to China again this year?

Jensen Huang: I hope so, remember to invite me, and I will definitely come Risk Warning and Disclaimer

The market has risks, and investment should be cautious. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investing based on this is at one's own risk