China’s Mixed Reactions to the H20 Ban Lift, and Jensen Huang's Beijing Visit

Yesterday in Beijing, Nvidia CEO Jensen Huang officially announced that the U.S. government has approved H20 shipments to China, giving the market a much-needed signal. He also revealed that Nvidia will soon release a fully compliant RTX PRO GPU for China — widely believed to be the long-rumored 6000D. On the same day, AMD confirmed it had received approval from the U.S. Commerce Department to continue exporting its MI308 chips to China.

Beijing kept its response low-key, simply restating its usual stance against politicizing and weaponizing trade and tech, and warning against disruptions to global supply chains.

At the July 15 regular press conference, Chinese Foreign Ministry spokesperson Lin Jian responded to a question regarding NVIDIA’s H20 chip exports to China:

Q: NVIDIA founder and CEO Jensen Huang said in Beijing today that the U.S. government has approved the company’s export license and NVIDIA will begin selling H20 chips in the Chinese market. Does the Foreign Ministry have any comment on this?

Lin Jian: We generally do not comment on the specific actions of individual companies. That said, China’s position on the issue is consistent and clear: we oppose the politicization, instrumentalization, and weaponization of technology and trade matters, as well as the malicious blockade and suppression targeting China. Such actions disrupt the stability of global industrial and supply chains and serve no one’s interests.

Despite it, the news quickly made waves in China’s public, particularly tech community.

Some analysts believe this shift in U.S. policy is closely tied to rare earth negotiations, especially since Treasury Secretary Bessent and Commerce Secretary Lutnick both publicly acknowledged that H20’s approval was part of broader trade talks around rare earths. Others see it as another sign of recent easing in U.S.-China tensions — and credit Huang’s behind-the-scenes lobbying for helping open up clearer communication between Nvidia and Washington.

Commentators have pointed out that chip export controls may no longer be at the top of Trump’s priority list. Compared to the Biden administration’s rigid “national security is non-negotiable” stance, the Trump team seems more flexible — with greater focus on trade deals, fentanyl, and pushing China to open its markets. That flexibility seems to have left space for Nvidia’s concerns to be heard.

From the perspective of Chinese tech companies, most see this as good news. Over the past few months, while H20 shipments were frozen under export controls, domestic GPU makers jumped on the opportunity — building out AI inference chips with different strategies. Some focused on better model compatibility, others boosted memory to support large models, and some optimized data flow to reduce memory dependence. A few leaned into power efficiency. The shared goal? Fill the gap left by H20.

That said, most industry voices still agree: when it comes to inference, H20 remains the strongest option in terms of performance and cost-effectiveness. According to Reuters, a number of Chinese firms are now rushing to place orders — and some large companies have already made substantial purchases.

As for how many H20s are actually being supplied in China, there’s a lot of speculation in the industry. Most people say it’s somewhere between 300,000 and 400,000 units, though some estimates go as high as 600,000 to 1 million. Either way, total volume is relatively limited. That’s why a lot of people in the market are now watching closely for NVIDIA’s next-gen China-compliant B-series GPUs—especially the B30.

There’s been plenty of speculation floating around online in China about the B30’s specs. The general consensus is that it's designed more for AI inference rather than training, with better efficiency and cost-performance. While it’s said to only offer around 75% of the H20’s training power, its inference performance is considered more stable. On a single card, inference might only hit 60% of H20’s level, but if you put eight cards in one server and use dynamic compression, bandwidth can reportedly reach 1.2TB/s, significantly boosting multi-card efficiency. Word is, the B30 will start shipping in Q4.

There’s also buzz that a major customer is planning to ramp up capex in Q3—very likely to prepare for B30 deployment. Rumors say that a few big tech firms have collectively placed orders for over 1 million B30 units, signaling strong demand.

That said, not everyone’s happy about all this. Under some of the news posts and WeChat blogs, readers left comments like:

“Everyone knows why Jensen Huang came to China. The U.S. Treasury Secretary just said last night—the reason they’re allowing NVIDIA to keep selling H20s is to stop Huawei from winning.”

“Now that Huawei’s broken through, the U.S. can’t really block them anymore. So instead, they let up the restrictions and find another way to profit off us.”

“Once domestic chips caught up to H20, the U.S. suddenly loosened the ban and started dumping low-end chips at cheap prices.”

“China should reject NVIDIA and go all-in on our own chips.”

“NVIDIA is just here to make money. Don’t bother with the big narratives.”

This wave of H20 re-approvals has sparked a ton of analysis around its impact on China’s chip industry. Most agree it’s a short-term negative for local AI chip makers and popular semiconductor stocks—especially companies like SMIC, Cambricon, and the Ascend ecosystem. Even though SMIC’s main business is still mature nodes like 28nm and 45nm, and it only does a few thousand AI wafers a month, in the A-share market, it’s all about expectations, not earnings. If Cambricon or Ascend take a hit, investors might start worrying about SMIC too, since they do the foundry work.

But longer-term, this doesn't really change the direction of domestic substitution. NVIDIA restrictions are just an external factor. Building a competitive local AI chip industry is going to be a long uphill climb. It’s not going to suddenly take off just because NVIDIA is banned, and it’s not going to get crushed just because H20 is back.

Meanwhile, comments from U.S. figures like Bessent, Lutnick, and Sacks are spreading on Chinese social media—and they’re not helping NVIDIA’s image. These officials basically admitted that the H20 is allowed only because it’s a downgraded product and because the U.S. wants to hold Huawei back.

There’s also an important follow-up question that’s starting to get traction: Now that H20 exports are allowed again, does that mean the original technical limits—like 1.4TB/s memory bandwidth and 300GB/s interconnect speed—are no longer valid? That matters a lot. It could affect how the B30 is spec’d, and even whether future domestic GPUs need to keep complying with the same 3A090-style export control thresholds when fabbing through TSMC.

On the morning of July 15, Jensen Huang met with Ren Hongbin, the chairman of the China Council for the Promotion of International Trade (CCPIT). According to CCPIT’s official press release, they discussed topics like strengthening U.S.-China AI industry exchanges, improving supply chain collaboration, and supporting the development of foreign businesses in China.

After the meeting, Jensen spoke to a Phoenix TV reporter and confirmed that the U.S. has lifted restrictions on the H20 chip, and shipments to China would begin soon. However, he didn’t clarify whether he had discussed tariffs with President Trump before his trip.

The following day, the official WeChat account of the CCPIT reposted a commentary from Niutanqin—a semi-official opinion blog with state media background.

The piece noted that while the Chain Expo was underway and Jensen Huang was visiting China,

“certain countries” were loudly promoting concepts like decoupling, de-risking, and friend-shoring. But in the eyes of those who actually produce goods and engage in global trade, these are little more than political slogans crafted by politicians. In a deeply globalized economy, the article argued, global industrial and supply chains are tightly intertwined—talking about cutting them off overnight is pure fantasy.

The commentary went on to point out that:

The U.S. has consistently been the largest overseas exhibitor at the Chain Expo. Elon Musk’s Tesla attended the first one; at the second Expo, Apple CEO Tim Cook made a same-day trip to visit Apple’s booth, famously saying: “There is no Apple without Chinese suppliers.”
Now, for the third Chain Expo, Jensen Huang came in person—and the number of U.S. exhibitors grew by 15% compared to last year.

This, the piece concluded, is the global business community—America included—“voting with their feet.” The trend of the times and the will of the people both point toward maintaining stable and open global supply chains.

It further emphasized that China’s door to openness is opening ever wider. Unlike the “small yards with high fences” promoted by some countries, China continues to champion international cooperation on supply and industrial chains—an anchor of certainty in an increasingly uncertain global economy.

The commentary ended with a tongue-in-cheek suggestion:

“Trump should really learn from Jensen Huang—not just learn Chinese, but also take a page from the wisdom of Chinese philosophy: prosperity comes from harmony, mutual benefit, and win-win cooperation.” (和气生财，互惠互利，合作共赢)

On July 16, Jensen attended the opening ceremony of the 3rd China Chain Expo and delivered a speech. The media quickly noticed that he wore a traditional Chinese Tang suit and opened his remarks in Mandarin.

Full Speech (Translated and Naturalized):

It’s truly an honor to be here at the China Chain Expo for the first time. China’s supply chain is nothing short of a miracle, and I’m so glad to finally witness it in person. This is also the first time I’ve had the opportunity to formally introduce NVIDIA to China’s massive supply chain community. It’s my first time attending at the invitation of Chairman Ren, and also my first time delivering remarks in Chinese. I grew up in the U.S. and have been learning Chinese—I’ll do my best.

Distinguished guests, friends, good morning!

I’m very excited to be here in China for my first Chain Expo. The scale of this event is massive, and the energy is incredible. It’s a clear reflection of China’s strong support for innovation and its commitment to building a prosperous shared future.

Let me start by telling you a little bit about NVIDIA and the industry we’ve helped shape. We started in 1993 as a small startup with a big dream—to reinvent the PC gaming experience using 3D graphics. We’re proud to have helped build what’s now one of the world’s largest tech and entertainment industries: PC gaming.

Over the years, we’ve partnered with visionary Chinese companies like Tencent, NetEase, miHoYo, and Game Science. In 1999, we invented the GPU—the first programmable graphics processor—opening the door to accelerated computing.

In 2006, we introduced CUDA, which transformed the GPU into a general-purpose computing engine and helped usher in the era of AI. A decade ago, the training of AlexNet on an NVIDIA GPU sparked the AI explosion. In the past, software was written by hand and ran on CPUs. Now, AI learns from data and runs on GPUs. This shift—from human-coded logic to machine-learned intelligence—is fundamentally reshaping the chip and computing industries.

In 2016, we launched the world’s first AI supercomputer, the DGX-1. I personally delivered the first unit to a small startup in San Francisco—OpenAI.

Since AlexNet, we’ve expanded from computing to systems, networks, and software—and we’ve increased AI computing performance by 100,000 times, which is 1,000 times faster than Moore’s Law. Today, NVIDIA is the computing platform powering the global AI ecosystem.

AI is transforming every industry—from scientific research and healthcare to energy, transportation, and logistics.

It powers China’s iconic platforms like Tencent’s WeChat and Alibaba’s Taobao. It drives Xiaomi’s autonomous driving and smart devices. It enables Baidu’s intelligent search and Meituan’s lightning-fast delivery. It even supports medical imaging systems that are improving healthcare in over 20 countries.

Behind China’s rapid innovation are the researchers, developers, and entrepreneurs—people like you. Over 1.5 million developers in China are building on the NVIDIA platform, turning their innovations into reality. Open-source models born in China—like DeepSeek, Alibaba’s Qwen, Tencent’s Hunyuan, MiniMax, and Baidu’s Ernie Bot—are helping drive global progress in AI.

China’s open-source AI has become a catalyst for global advancement, giving more countries and industries a chance to participate in this revolution. Open-source is also essential for AI safety, promoting global collaboration on technical standards, performance benchmarks, and protective measures.

Today, AI is becoming foundational infrastructure—like electricity in the past, or the internet. AI is reshaping global supply chains and transforming how we produce and move things.

Hundreds of projects in China are using NVIDIA Omniverse to simulate digital twins for factory and warehouse design and optimization. Robots are being trained in virtual environments using Omniverse so they can safely collaborate with humans in the real world.

The next wave of AI will be robotics—systems that can reason, act, and understand the physical world. Over the next decade, factories will be driven by software and AI, coordinating human-robot teams to produce intelligent, AI-designed products.

AI will sit at the core of every industry, every business, every product, and every service. It’s fueling a new industrial revolution and unlocking exciting growth opportunities for China’s world-class supply chain ecosystem. NVIDIA will continue working side by side with our long-term partners and many new friends to build a prosperous future in the AI era.

Thank you all!

On the afternoon of July 16, NVIDIA held a media roundtable at the Mandarin Oriental Qianmen in Beijing. CEO Jensen Huang took questions from reporters, offering detailed responses on the resumption of H20 chip exports to China and the company’s upcoming GPU offerings. He also addressed a wide range of topics, including U.S.-China tech competition, NVIDIA’s future cooperation with Chinese firms, and how the company is navigating both domestic and global market dynamics.

Transcript – Beijing Media Roundtable, July 16

If we zoom out and take a broader look at AI in China, it’s important to remember that AI is an infrastructure technology. It’s part of computer science—it sits on top of chips, systems, and network infrastructure. Then you have the models, and on top of that, the applications. So you can think of it as a three-layer stack: compute, models, and applications.

China is moving incredibly fast across all three layers. At the model level, you have impressive systems like DeepSeek, Alibaba’s Tongyi Qianwen, and Kimi from Moonshot AI. DeepSeek, in fact, is the world’s first open-source inference model—that’s a real breakthrough, no question about it. The pace of innovation in model development is remarkable.

But when it comes to applications, China is moving even faster. There are so many different use cases being built. Not only is China fast at inventing new technologies, it’s also incredibly quick at integrating and applying them. That’s because competition is intense, the population is massive, and so many people are hungry to build great companies or launch new ventures. We’re seeing the rise of many amazing businesses.

And the foundation of all this? It’s talent. China’s education system has produced some of the best AI researchers in the world—close to 50% of global AI researchers are based in China. You’ve got a rich ecosystem, especially in computer science, and it’s growing fast. I feel lucky to be part of it.

DeepSeek and Kimi are excellent models. They’re highly efficient for real-world applications because they’re open—you can fine-tune them for anything you want. You could build a company, a product, or a service on top of them. That makes them uniquely valuable.

Over time, I think the key question won’t be “Which model is the smartest?”—but rather, “Which model is the most useful?” It’s kind of like hiring someone—you don’t really care about their IQ test score. You care whether they get the job done.

Meta’s also going all-in on AI. They’ve built one of the world’s largest AI infrastructures, and it gives them a huge advantage. Clearly, AI is going to be a trillion-dollar industry—not billions, trillions. So it’s no surprise they’re fully committing. It’s the right move. They’re moving fast, and they’re going to be a major force.

AI is also going to make factories more flexible, reconfigurable, and productive. When product designs change, factories will adapt faster. They’ll be more efficient and output will increase—exactly what we’ve always wanted from the supply chain.

In the long run, AI and robotics could bring costs down, lower prices, and reduce inflation. And lower costs mean higher productivity. That’s a big deal.

China’s developers and entrepreneurs are fueling rapid AI innovation. Over a million developers are working in this space, with names like DeepSeek, Alibaba, MiniMax, and Baidu. They’re building world-class products that are pushing the global AI field forward.

China’s open-source AI is becoming a catalyst for global progress. It’s enabling industries and countries all over the world to participate in the AI revolution. Open-source is also key to AI safety, which is why international collaboration on technical standards is so important.

Across compute, models, and applications, China is advancing at an incredible speed. That’s due in part to strong competition and the deep talent pool. And again, that talent comes from an education system that produces world-class researchers—especially in math and computer science.

“H20 Will Do Well in China”

As for H20—yes, it’s now approved for export, and I think it’ll do very well in China. One of its advantages is its high memory bandwidth, which makes it a great fit for innovative Chinese models like DeepSeek and Kimi. It’s highly efficient and well-suited for LLMs and next-gen systems.

We just made the announcement yesterday, so I haven’t had time to meet with customers yet. It’ll take a few months to ramp up supply chain capacity, but we’re working hard to speed it up. And if we’re allowed to continue selling to China, we will—simple as that.

Export controls are beyond our control. All we can do is help governments understand the policy and its unintended consequences. But ultimately, policy decisions are up to them—not us.

“Cars Will Think Like Humans”

Right now, cars are too simple. But with inference AI, tomorrow’s cars will be able to think. You’ll have AI on the edge, in the cloud, and even for simple tasks like opening a door. It all depends on the application.

You need inference on edge devices because autonomous vehicles will inevitably face situations they’ve never seen before. That’s when the car has to reason—figure out what’s in front of it, and decide what to do.

China’s strength in hardware—like batteries and power systems—is well known. But there’s another miracle: China’s computer science is world-class. The country’s software capabilities are just as strong.

“I Want a Xiaomi Car”

I really want a Xiaomi car. Lei Jun showed me their new model—it’s beautiful. The design is sleek, the tech is incredible. I’d buy one in a heartbeat, but unfortunately, you can’t get them in the U.S. That’s our loss.

NVIDIA works closely with Xiaomi—they’re a fantastic partner. I’ve known Lei Jun since he was much younger. Back then, I was younger too—but he really was young. Even from the beginning, I could tell he was going to do something extraordinary.

And now, look at Xiaomi—it’s a phenomenal company. From amazing smartphones to high-performance vehicles to everything in between. It’s hard to believe one company can build so many great products. That’s the magic of Lei Jun.

Xiaomi, BYD, and Li Auto all make excellent cars. Li’s interiors are like living rooms on wheels. NIO and XPeng are super luxurious too.

“NVIDIA Is the Smallest Big Company in the World”

NVIDIA is one of the last semiconductor companies to be founded. I wasn’t first—I was last. But being last can be a strategy too. There’s always opportunity, as long as you stay sharp.

Our philosophy is to over-compensate employees, not under-compensate. Maybe that’s why people say we’re the smallest big company in the world—we’ve got only about 42,000 people.

For a company with such a big impact, that’s a small team. And our people stay for a long time. They’re passionate, they’re brilliant, and they’ve stuck with us. Our turnover is nearly zero—and that’s part of NVIDIA’s magic. We pay well, we take care of our people. That’s my core belief.

We don’t build self-driving cars—but we create the technology that enables them. We don’t build LLMs—but we power them.

We’re one of the few companies that can innovate across the entire stack—from chips and systems to software and algorithms. I don’t think any other company spans that full range at our speed.

We succeed because we work closely with innovators—the ones trying new things. That pushes us to evolve. And the more flexible we are, the better we can adapt to this fast-changing world.

“Never Underestimate a Competitor”

When it comes to competition, the first rule is: don’t underestimate anyone. Respect your competitors. Learn from them. Be inspired by their success and use that energy to improve.

Underestimating your rivals never helps. I have deep respect for our competitors.

As for Huawei—look, they’ve got strong chip design, systems engineering, networking, and even silicon photonics. They offer their own cloud services too. That’s powerful.

NVIDIA is a chip and systems company. Huawei goes directly to market—that’s a whole other level. We try to learn from them, and from everyone else.

Anyone who underestimates Huawei or China’s manufacturing strength is being naive. This is a serious company with real achievements.

“AGI May Be Closer Than We Think”

Do we need a breakthrough in GPU architecture to reach AGI? It depends on how you define AGI. If it means passing a suite of benchmark tests, I think we’ll beat human performance very soon—no major breakthroughs needed.

There are a lot of good ideas already in the lab. I believe we’re not far off from AGI.

On the H20 export resumption and new GPU products

Q: Can you introduce the new GPU designed specifically for the China market?
Huang: We’ve just launched a new product called RTX Pro. It’s designed for a new class of applications—digital twins. For a robot to be trained to complete tasks effectively, it needs to be simulated first as a digital robot in a digital world. At NVIDIA, we care deeply about digital factories, smart manufacturing, and robotics. And China—well, China has so much innovation happening in robotics, so many smart factories, and such a vast supply chain. I’m really excited about what we can do here.

Q: After your July 15 announcement that H20 shipments to China would resume, what kind of demand have you seen from Chinese clients?
Huang: Honestly, I haven’t had the chance to meet with our customers yet. I’m waiting for them to tell us what they need, and I look forward to serving them as soon as possible. That said, it takes time to rebuild the supply chain. From placing an order to wafer production to delivering an AI supercomputer—it can take up to nine months. So we’re going to work very hard to accelerate that process.

On working with China going forward

Q: What’s your focus for future partnerships in China?
Huang: We’ve been in China for over 30 years. It’s the second-largest tech market in the world and still growing fast, so it’s incredibly important to us. There are so many dynamic and innovative customers here. As everyone knows, to be a great company and a great supplier, you need great customers—and China has some of the most demanding and creative ones. We want to keep supporting them.

Q: Will NVIDIA continue to invest in China, or just maintain its existing footprint?
Huang: If you want to maintain your presence in any market, you have to invest. In a fast-moving and competitive space like this, nobody stays still—not even our competitors. Everyone keeps investing in order to lead globally. We have to keep improving, every single day, just to stay where we are.

On U.S.-China relations

Q: How do you balance between U.S. and Chinese trade policy?
Huang: There is no such thing as “balance.” Every country has its own security policies, and trade policy is part of that. Our job is to comply with the rules, build the best technology we can, and continue contributing to the market.

Q: Can you provide chips more advanced than the H20 to China?
Huang: I hope China can get access to more advanced chips than the H20—because technology always moves forward. The H20 is still excellent today, especially for inference and bandwidth, but a few years from now, there will be better and more powerful technologies.

On market competition

Q: Many cloud providers are now making their own chips. How do you see that competition?
Huang: Our strategy is very different. We focus massive resources on building the most advanced AI systems—designed for the entire lifecycle of AI: from pre-training and fine-tuning to reinforcement learning and inference. Our architecture works across everyone’s cloud. If you understand NVIDIA’s platform and language, you can run it anywhere—on any cloud, or on your own hardware.

Our job is to make sure the performance of our platform is always ahead of the curve. No matter where you want to run your AI workloads, we’ll support you. That’s fundamentally different from how cloud vendors think.

Q: Chinese manufacturers are already producing strong alternatives to your products. What’s your view?
Huang: If you’re building something that’s CUDA-compatible, that’s fine. In many ways, CUDA is already open.

Q: What do you think of Huawei and other Chinese companies catching up in AI chips?
Huang: Huawei is an amazing tech company. They build chips, systems, networks—they're incredibly capable. Anyone who underestimates Huawei—or China’s ability to manufacture—is being naive. I’ve been building chips for 30 years to get to where we are today. Huawei has reached comparable levels in just a short amount of time. That’s impressive.

Q: What’s the toughest challenge your competitors present?
Huang: Everyone’s working really hard. They want to win, and so do I. Competition makes all of us better.

On computing power and capacity

Q: Do you experience compute anxiety?
Huang: Absolutely—I have compute anxiety. Without enough computers, none of our chips, systems, networks, or software work. Even our own engineers have AI supercomputers and still feel that anxiety every day. They’re always asking for 10x more compute.

In this new era, hand-written code isn’t enough. The more compute we have, the more data our AI systems can learn from—and the more problems we can solve. So yeah, not just me, but every engineer will likely have compute anxiety for the next 10 years.

On humanoid robots

Q: You’re optimistic about humanoid robots. What do you think of China’s prospects in this field?
Huang: First, there’s a massive global labor shortage. If we had more people working in factories and manufacturing, global GDP would grow dramatically. More automation will make the world more productive.

Second, both humanoid robots and AI are booming. China is uniquely positioned—it has great AI capabilities, strong mechanical and electrical engineering skills, and a massive manufacturing base where robots can be deployed. So yes, I’m very bullish on robotics in China.

On future priorities

Q: At the GTC conference in June, you emphasized quantum computing. What’s next for NVIDIA?
Huang: AI is going to reshape every industry and every science. Take computer vision as an example—look at the stunning visuals in Black Myth: Wukong. That’s the power of image-based AI.

We're now developing AI for quantum error correction—helping quantum computers produce useful results. We’re also simulating complex weather models, molecular dynamics, electron bonding—you name it. Going forward, we’ll be applying AI to everything.

On the Chain Expo and open-source AI

Q: This is NVIDIA’s first time at the Chain Expo. Earlier today, you said open-source AI in China is a global catalyst. What did you mean by that?
Huang: DeepSeek is revolutionary—it’s the first open inference model, and it’s incredibly effective. Because it’s open, many countries and companies have already downloaded DeepSeek R1. It’s being used in fields like healthcare, medical imaging, robotics, and more.

Openness allows every country and industry to participate in AI.
AI isn’t just for a few companies or a few nations—everyone in the world should be able to benefit from what AI can do.