Jensen Huang’s Fireside Chat with Wang Jian at the Chain Expo, Hu Xinjin's comment on H20 ban lift

On July 17, during the second day of the Chain Expo, Jensen Huang sat down for a fireside chat with Wang Jian, Director of Zhejiang Lab and founder of Alibaba Cloud.

As longtime friends who have known each other for over a decade, the two opened with warm greetings and reflections on the development of AI over the years. “Over the past twelve years or so, AI has progressed at an incredible pace, with a major breakthrough roughly every three to five years,” Huang said.

In Huang’s view, AI has moved past the stage of perception intelligence and generative AI, and is now in the phase of reasoning intelligence, edging ever closer to Artificial General Intelligence (AGI). “We’re now approaching a stage where AI can solve cognitive tasks and outperform most humans on most tests. That’s what we call AGI,” he explained.

During the exchange, Huang asked Wang which technology had excited him most in recent years. Wang replied, “For me, what’s truly exciting is actually the foundation of everything. What we call AI is all built on computing, and computing is changing everything.”

When talking about the rapid progress of AI, Huang emphasized that while it’s driven by NVIDIA’s computing innovations, it’s also based on the global openness of research. He noted that Chinese researchers have played an extremely active role in pushing global AI forward, even calling Hangzhou "China’s Silicon Valley."

"The vast majority of AI research is done openly," Huang said. “I saw a statistic that Chinese researchers now publish the most papers on arXiv in the world.”

When asked whether computing power is fundamentally based on silicon, Huang gave a firm yes and emphasized that the future of silicon technology revolves around three areas: iterations in transistor structure, packaging techniques, and silicon photonics. “We still have at least twenty years of work ahead in this field. I can say that because NVIDIA’s technology roadmap already extends nearly a decade into the future.”

(Original dialogue transcript below, adjusted slightly for clarity without altering the meaning)

Wang Jian: Good morning everyone. Jensen, long time no see—it’s really great to see you again. Welcome to the Chain Expo. I’ve written my questions down on my phone.

Jensen Huang: I think the first time we met was a long time ago in Beijing.

Wang Jian: It was around 2012 or 2013—about a decade ago. I still clearly remember visiting you in Silicon Valley. You personally introduced NVIDIA’s technologies to me and gave me a tour. That moment left a strong impression on me. I saw how passionate you were about what you were doing, and I understood how important a founder is to a company.

Jensen Huang: Back then we were talking about computer graphics and mobile devices, right? That must’ve been around 2012. The video earlier mentioned that I came to China in 2007 to introduce CUDA. [In Chinese] That was a long time ago.

Wang Jian: It’s been quite a journey.

Jensen Huang: [To the audience, in Chinese] Can they hear us?

Wang Jian: I remember you were working in LA at the time, and we met at SIGGRAPH, the graphics conference. You launched the GPU and transformed the field of computer graphics. It’s been an amazing journey. My first question is about technology: AI is the hottest topic now, but people have very different views about AI and computing. So what fundamental changes or breakthroughs have we really made in recent years?

Jensen Huang: Fundamentally, AI is a new kind of software development based on first principles. It doesn’t rely on humans writing code and describing algorithms to predict outcomes. Instead, we use a type of algorithm that learns how to predict outcomes through example data. And it turns out that using computing to learn in this way is incredibly scalable. We’ve been studying machine learning for a long time, but 2012 was a key turning point. AlexNet proved the effectiveness of deep learning. It outperformed expectations in computer vision.

Starting in 2012, over the next five years, computer vision became highly effective and eventually surpassed human-level performance. Then speech recognition followed—first it became viable, then it surpassed humans. Soon after, language understanding also made that leap.

Each modality represents a development stage. The first wave was perception intelligence. The second wave was generative AI. Now we can translate between modalities—from English to Chinese, English to images, images to English, Chinese to video. Generative AI enables that. It started about seven years ago and is still evolving quickly. Today, AI can both understand and generate information.

Now we’re in the wave of reasoning AI. This is remarkable because it can solve problems it hasn’t seen before. It breaks down complex problems, like humans do, and solves them step by step. The next wave will be physical AI, where all these capabilities get applied to machines, like robots.

In the last twelve years, the pace of AI progress has been stunning—about one major breakthrough every three to five years.

I believe we’re approaching a stage where AI will solve cognitive tasks better than most humans. That’s AGI. And that’s why everyone is now talking about superintelligence. Just like before—first we achieved viability, then we surpassed human levels. Soon, we’ll go beyond humans in problem-solving.

Wang Jian: That’s amazing. Especially this year, open-source models are changing the landscape of AI and our businesses.

Jensen Huang: Which of these technologies excites you the most?

Wang Jian: For me, it’s the foundation of it all—computing. That’s what truly excites me. AI is just the surface.

Jensen Huang: That’s amazing. And how we train models is changing too, right? The first ten years were mostly pretraining. We collected tons of data, sometimes even using AI to prepare the data. Then came reinforcement learning with human feedback. Now we’re in the post-training era. AI can think on its own, generate synthetic data, validate, and learn to reason.

Wang Jian: Yes, and the amount of computing required now is staggering. I come from a psychology background, and I don’t think AI is just mimicking human intelligence—it’s enhancing it. I see it as an extension of human creativity.

Jensen Huang: Exactly. It’s not just replacing intelligence. Like cars extended our mobility, and planes extended our reach, AI will extend our intelligence. It doesn’t work like the human brain, but it can do many things like us.

Wang Jian: Back to open source—it’s an exciting time. We have DeepSeek, Qwen from Alibaba Cloud… these are just a few examples.

Jensen Huang: And Kimi from Moonshot AI—they’re all amazing.

Wang Jian: So my question is—do you think open-source models are a disruptive force for the future of AI?

Jensen Huang: We talked about how fast AI is advancing. People say it’s because of NVIDIA’s tech—yes, that’s true. We’ve increased AI computing power by 100,000 times in the past ten years. But what’s not often mentioned is that most AI research is done openly. The number of papers on arXiv is staggering.

In fact, Chinese researchers now publish more than anyone else on arXiv. Much of this research is open, and it enables collaboration. People can read each other’s work and build upon it. That’s open science. The next step is open engineering—sharing not just ideas but implementations.

Open engineering is incredibly powerful. Instead of just depending on one company’s team, it becomes a community effort. China is doing great work in this area.

But let’s not forget—open source benefits the whole world. Models like DeepSeek, Qwen, and Kimi are helping not just China but ecosystems everywhere. They are world-class reasoning models. Whether you’re in healthcare, finance, or robotics, you can adapt them to your needs.

And open source is the safest path. Transparency invites scrutiny. Look at DeepSeek’s paper—it’s top-tier science. Publishing openly improves quality, education, and safety.

Wang Jian: By the way, DeepSeek and Qwen are from Hangzhou. I’m from Hangzhou. Very proud. I’d like to personally invite you to visit Hangzhou next time you come to China.

Jensen Huang: Hangzhou is probably China’s Silicon Valley, right?

Wang Jian: Some might not say that, but I think Hangzhou will become a global innovation hub. It’s very unique. Again, you’re very welcome.

Jensen Huang: I’ll definitely visit. Thank you.

Wang Jian: You talked about open science and open engineering. At last year’s GDC, you said this is the first time in history we could turn biology from a science into engineering. That’s mind-blowing. How do you see AI’s long-term impact on scientific discovery?

Jensen Huang: Right now we talk about human-facing AI, but science-facing AI may be even more transformative. Human-facing AI is easier—we created language.

When designing chips, we use tools to work with transistors—something we built. Biology is different. It’s natural. To engineer biology, we must first understand it.

With AI, we can now learn the structure and function of proteins, chemicals, cells—even human metabolism.

We can design drugs to extend life. We can simulate physical systems. Think of weather—it involves cloud physics, ocean physics, convection, energy transfer—all at different scales.

These simulations span seconds to years and microns to thousands of kilometers. That’s hard for traditional methods.

But maybe AI can help predict. It’s much faster than physics-based simulations.

I believe AI will help us understand and simulate nature. That will push science forward.

Wang Jian: Today’s AI still depends heavily on silicon. All that computing power comes from silicon chips. Can we keep relying on it for the next 10 or 20 years?

Jensen Huang: Absolutely. Silicon isn’t just silicon anymore. It’s evolving with new materials.

We’re seeing 3D transistors—like Gate All Around (GAA). We use nanosheets now. Later we’ll stack transistors—SFFETs. We’re also using backside power delivery.

We used to build one chip at a time. Now we’re stacking them with CoWoS packaging. We might move to panel-sized chips—chips the size of a table.

We’re also connecting chips with silicon photonics. We call this CPO—co-packaged optics.

These innovations will give us the scale and performance we need.

At NVIDIA, our roadmap stretches nearly ten years out. There’s at least 20 years of work to do.

Wang Jian: Great architecture, especially in silicon, unlocks huge possibilities. So my last question is about young people. Many are unsure about their future, but they care deeply about the world. Jensen, you’re a hero to many of them. What advice would you give?

Jensen Huang: People say, “AI writes code now, maybe we don’t need to learn it.” That’s totally wrong.

We must still think from first principles. Break down problems and build understanding from the basics.

Critical thinking is essential—whether based on physics, math, or logic. That’s how we know if an answer (from AI or anyone) makes sense.

So, learn math, reasoning, logic, programming. Even if you don’t code, understand the principles.

Second, start using AI now. It’s the new computer. It understands how we interact. If you don’t know how to use it, just ask it to teach you—it will.

AI is the greatest equalizer. It gives everyone the same starting point. Whether you’re a farmer, senior, or student, you must engage with AI now.

I envy the younger generation. They’ll grow up with an AI companion. It can remember everything, advise them, teach them.

Imagine asking your AI what you were doing at age one—and it remembers.

I wish I had that. That’s the future.

Wang Jian: I still remember the passion you showed when we first met. And the patience you’ve maintained through this journey is incredible.

For young people, passion and patience are key. Thank you for sharing your wisdom.

Well-known Chinese media personality Hu Xijin also weighed in on the issue via his personal WeChat account. In his view, the U.S. government’s decision to lift the ban on the H20 chip wasn’t because the Trump administration suddenly had a change of heart, nor was it due to Jensen Huang’s lobbying efforts. Instead, he argued, it was China’s own progress—especially the significant chip advancements made by companies like Huawei and the steady improvement of China’s overall software and hardware ecosystem—that forced Washington to rethink its chip policy toward China.

In Hu’s eyes, NVIDIA’s goal of maintaining a presence in the Chinese market is perfectly understandable from a business perspective. On the other side, China has consistently pursued a dual-track approach: developing its own technologies while remaining open to cooperation with the outside world.

Unable to Choke China's Tech, U.S. Lifts Ban to Cash In

NVIDIA’s renewed approval to sell its H20 chip to China could mark a turning point—or at least the beginning of one—in the U.S.-China chip war. Speaking in mainland China on Wednesday, Jensen Huang said he hopes to eventually sell even more advanced chips than the H20 to China. And honestly, I wouldn’t rule that out. One thing seems clear: the darkest days of the U.S. using chips as weapons to hold back China’s high-tech progress may be behind us.

This week, NVIDIA confirmed that it had received approval from the U.S. government to resume sales of the H20 to China. But let’s be clear—the reason for this isn’t that the Trump administration suddenly became generous, or that Jensen Huang delivered some incredible lobbying performance. It’s because Chinese companies like Huawei have made major breakthroughs in chip technology, and because China’s overall software and hardware ecosystem has steadily improved. That progress created pressure—enough pressure to push Washington to reassess its chip policy toward China.

Over the past few years, the U.S. treated chip sanctions as a silver bullet—an ultimate lever to suppress China’s tech rise. But instead, they ended up turbocharging China’s semiconductor development. With chips like Huawei’s Ascend 910B gaining ground, it’s become increasingly obvious that China’s progress can’t be blocked by embargoes. The country's tech industry hasn’t been strangled—it survived and is still growing. And its homegrown competitive strength is starting to take shape.

That momentum has made Washington nervous. Under pressure—and with NVIDIA lobbying hard—the U.S. is adjusting its approach. Allowing NVIDIA to resume H20 sales isn’t just about lifting a restriction. It’s a tactical move: NVIDIA wants to reclaim part of the Chinese market. That would bring the company substantial revenue, but it also serves a strategic purpose—it could slow China’s urgency to rapidly iterate on its own chips by keeping domestic companies tied to NVIDIA’s ecosystem. In short, it introduces more “choice,” and possibly even hesitation.

NVIDIA’s desire to stay in the China market is understandable. On China’s side, there’s always been a dual approach: develop homegrown capabilities while keeping doors open to external cooperation. That hasn’t changed.

Back in April, NVIDIA said it had received notice from the U.S. government of an indefinite ban on H20 chip sales to China—unless special licenses were granted. That was the environment then. But now, the situation has changed.

Today, we have a solid foundation of independent Chinese technology. That fundamentally shifts the dynamics of the chip trade. This is no longer a simple return to past cooperation. It’s a new kind of cooperation—one based on greater balance. It’s still business, but it’s no longer a one-sided relationship where the U.S. dictates terms and China passively follows. Now, it’s mutual and negotiated.

The return of the H20 chip perfectly illustrates a classic Chinese saying: “Seek cooperation through struggle, and cooperation will endure.” NVIDIA’s reentry shows exactly how that works in practice. And with China’s chip technologies bound to keep evolving, and production capacity only continuing to grow, the door is now shut for further “crippled versions” of the H20. If NVIDIA wants to stay relevant in China, it will truly have to deliver on Jensen Huang’s promise—to work toward selling China more advanced AI chips.

Sure, Huang will keep lobbying. But in the end, it’s China’s own technological progress that will continue to shape the future.

Another chip-related event is the official opening of the 5th RISC-V China Summit on July 16. Shi Huikang, Deputy Director-General of the Electronic Information Department of the Ministry of Industry and Information Technology (MIIT), attended the summit and delivered remarks. He outlined three expectations for the development of RISC-V in China:

1. Deepen collaboration and build the ecosystem together: Under the guidance of MIIT’s Electronic Information Department, the RISC-V Committee has brought together more than 80 organizations. These entities are working jointly on standards, technology, and ecosystem development to build a globally competitive platform.

2. Accelerate commercialization and expand application: Efforts should be made to bridge the gap between innovation and real-world deployment, enabling RISC-V to continue making breakthroughs in areas where it already has strengths, such as the Internet of Things, while also achieving commercial adoption in high-performance computing scenarios.

3. Stay committed to openness and lead cooperation: China encourages deeper participation in international collaboration, contributing standards and solutions to pool global resources and jointly build an open RISC-V ecosystem.

   

From the perspective of computing evolution, we've gone from standalone computers to local networks, to the internet, and then to the era of mobile connectivity—each dominated by different technology stacks: first Wintel (Windows + Intel with x86 instruction set), then Android and ARM (AA) with the ARM instruction set. As we now enter the age of the Internet of Everything, the question is: what architecture will take the lead in this next wave? Many experts and industry insiders believe it could be an open instruction set CPU paired with an open-source operating system. Personally, I agree. The fact that thousands of people are gathering at this summit is evidence of a growing consensus around the technical soundness, engineering feasibility, and commercial potential of this direction.

RISC-V has moved far beyond its academic origins and is steadily advancing toward broader industrial adoption. Since its introduction in 2010, the RISC-V instruction set architecture has seen explosive ecosystem growth, thanks to global efforts across academia and industry. According to the latest figures from the RISC-V International Foundation, over 4,500 organizations and individuals from more than 70 countries are now involved. Perhaps most impressive: in 2024 alone, RISC-V-based chip shipments reached tens of billions of units—clear proof of the competitive power of open collaboration.

Globally, the strategic importance of RISC-V is increasingly recognized. Countries are investing in it based on their own development needs. The U.S., while continuing to back x86 and ARM, is now also paying close attention to RISC-V. Europe sees open-source architectures as a way to rebuild industrial competitiveness. For China, RISC-V represents a key opportunity to achieve greater tech self-reliance. China not only has the world’s largest application scenarios and implementation base, but also shows great momentum in both technical capability and ecosystem building.

Statistics suggest that in 2024, China contributed more than half of the world’s RISC-V chip shipments, with breakthroughs in high-performance computing, AI, automotive electronics, and more. For China, RISC-V is a strategic pivot for mastering core processor technologies. It gives Chinese developers a chance to participate deeply in the design and evolution of cutting-edge instruction sets, and to gain full-stack expertise.

RISC-V is also a powerful engine for innovation. Its open-source, royalty-free, and modular nature significantly lowers the barrier and cost of chip design, opening new lanes for China to boost its global competitiveness in advanced manufacturing. More than that, it serves as a bridge for global collaboration—this is China’s first time co-creating core chip technologies with the international developer community in real time. It’s a unique chance to raise China’s voice and influence in the global semiconductor landscape.

In fact, in China’s 2025 national development plan released earlier this year, RISC-V and other emerging open architectures were specifically named as priority areas for early-stage breakthroughs. The message is clear: seize the opportunity, focus resources, and overcome fragmentation in the open-source space to chart a path for China’s innovation strategy.

At the summit, three concrete calls to action were outlined for China’s RISC-V community:

1. Deepen collaboration and build a stronger ecosystem. With the RISC-V Committee now established and 80+ members already on board, China has gathered key stakeholders across academia, industry, and application domains. Progress is already being made in standardization, common tech challenges, and ecosystem infrastructure. The call now is to keep using this platform—to jointly push forward in areas like core instruction set extensions, toolchains, operating systems, and developer support—while telling China’s story to the world and helping shape a globally competitive RISC-V platform.

2. Accelerate commercialization and drive adoption at scale. Innovation only matters if it reaches the market. RISC-V has already gained traction in IoT, industrial control, and edge computing. The goal now is to scale its use in high-value sectors like AI, data centers, smart vehicles, and high-performance computing. It’s time to take RISC-V from proof-of-concept to real commercial impact.

3. Stay open and lead globally. China must play an active and open role in the global RISC-V ecosystem. Companies, research institutes, and developers are encouraged to deeply engage with the RISC-V International Foundation, contributing to global standards, security protocols, and cross-platform interoperability. China is committed to independent innovation, but that should go hand-in-hand with global partnerships, joint research, and shared development. The vision is for China to be not only a user, but a key builder and contributor to RISC-V worldwide.