Huang Renxun, who was severely "trapped" by Trump, appeared in China at the speed of light. The sudden ban on sales of the H20 chip, known as the "special version for China," has once again cast a shadow on Nvidia's prospects in the Chinese market. Against this background, Huang Renxun’s recent visit to China has two key purposes. The first is to express that NVIDIA will not give up the Chinese market. After the H20 sales ban, the company will still find ways to launch products that adapt to the new policies.


The second is "learning from the experience". If we don’t give up on the Chinese market, H20 can no longer be sold, and Huang Renxun needs new products. How to design new products requires dialogue with the demand side.

The Financial Times reported that Huang met with Nvidia customers, including DeepSeek founder Liang Wenfeng, "to discuss new chip designs for Chinese customers."

At present, neither Huang Renxun nor Liang Wenfeng have responded to the news.

The rise of DeepSeek, Nvidia’s “stability maintenance”, tariff war and H20 ban, it is hard to believe that this is what happened in just three months. In the past three months, Nvidia has appeared in DeepSeek's technical reports and in almost every piece of content analyzed by the outside world about DeepSeek.

Nvidia cannot avoid DeepSeek. Huang Renxun actively or passively talked about this company at financial conferences and GTC conferences. DeepSeek not only caused problems for Nvidia and caused the outside world to question the company's soul, but also brought new vitality to Nvidia in the Chinese market.

Now, the two founders behind it, Huang Renxun and Liang Wenfeng, have finally met. In such a chaotic background, it is somewhat regrettable. But trying to break through the influence of geopolitics and continue the "win-win" attitude is enough to become a good story in the business world.

However, even if Huang Renxun has locked in the person who can best help him at the moment, Liang Wenfeng, no one can say what Nvidia will experience in the Chinese market in the future. If NVIDIA's chips sold in China are repeatedly castrated and domestic chip manufacturers catch up, will AI companies including DeepSeek continue to favor NVIDIA chips?

01. Jen-Hsun Huang’s special trip to Beijing

Three months ago, it was the Chinese New Year. As the new U.S. President Trump takes office, Nvidia CEO Jensen Huang is busy visiting China for a week. The outside world calls it Huang Renxun's "longest road show" in China, and calls Huang Renxun the "hardest sales person."

At that time, Huang Renxun could not have imagined that he would visit China again in less than a hundred days, taking off his leather jacket and putting on a suit. The seriousness of the situation is evident.


Source: CCTV financial video screenshot

After Trump took office, he began to exert greater pressure on China. The tariff war started, and the pressure was quickly transmitted to the chip industry. Nvidia's stock price fluctuated, and its market value once evaporated by trillions of dollars overnight.

About a week ago, good news came out of Nvidia. According to the National Public Radio NRP, Huang Renxun personally participated in Trump’s dinner at Mar-a-Lago, and then the White House temporarily gave up its plan to ban the sale of H20.

H20 is a special product of Nvidia under the Biden administration's "chip ban". It is specially designed for the Chinese market. Compared with the flagship chip, it has been "castrated" to a certain extent to comply with export regulations to China. The increase in tariffs, coupled with rumors that Trump may even ban the sale of H20, has become the biggest concern for Nvidia.

The news of Mar-a-Lago's "successful lobbying" caused Nvidia's stock price to rise by 18% in a stock market filled with disasters.

However, a shocking reversal soon took place. Just less than five days later, on Monday, the Trump administration notified Nvidia that H20 chips were included in export controls. If exported to five countries including China, a license is required. This requirement will take effect indefinitely in the future.

This unexpected bad news caught Huang Renxun off guard. In a document submitted to the U.S. Securities and Exchange Commission on Tuesday, Nvidia stated that this move will cause Nvidia to lose US$5.5 billion in the first fiscal quarter. "(This loss) is related to H20 product line inventory, purchase commitments and related reserves."

It should be noted that NVIDIA's first fiscal quarter corresponds to the natural date, which ends on April 27.

In other words, after signing the order with the customer and not yet delivering it, a lot of products have already been released, and now they are all in the hands. Unsurprisingly, Nvidia's stock price fell another 6% as soon as the news came out.

Within 48 hours, Huang Renxun arrived in Beijing. He made it clear that Nvidia has grown in China over the past 30 years and China is very important to Nvidia. Nvidia and the Chinese market have achieved a win-win situation and will continue to serve the Chinese market.

Whether it is the timing of his visit to Beijing or Huang's unabashedly clear stance, it can be seen that Huang is not prepared to bow to the Trump administration's "concerns." Business will continue and we will have to work harder than before.

02.Why did you find Liang Wenfeng?

Not giving up on the Chinese market means that Nvidia needs to continue to make "China special edition" AI chips. It is necessary to comply with the new ban requirements and meet the needs of Chinese customers, both of which are indispensable.

As for how to meet the needs of Chinese customers, no one is more suitable than Liang Wenfeng to discuss this issue.

Regardless of whether the news is true or not, Huang Renxun does need to "align" with Liang Wenfeng.

When Huang Renxun visited China during the Spring Festival, a major event was happening in the AI ​​field, and that was the rapid rise of DeepSeek.

This is a very dramatic event: In China, which is restricted by the United States from obtaining high-end chips, a start-up company emerged. They used Nvidia's "castrated" version of the chip to train an inference model R1 with excellent performance at a cost of millions of dollars.

Although there are different voices in Silicon Valley, and even accusations from the CEOs of OpenAI and Anthropic (Altman suggested that DeepSeek "plagiarized", and Amoudi questioned DeepSeek for "lying" about the chips used and costs), DeepSeek used "Open Source Week" to prove its technical strength. Almost every open source project is a "squeeze magic" for Nvidia chips.

Before the Sino-US tariff war and the US ban on the sale of H20, NVIDIA's biggest crisis and biggest challenge was caused by DeepSeek. And this crisis hits Nvidia's soul directly. This chip giant iterates at a high speed, breaks Moore's Law, and continuously launches the most advanced AI chips. DeepSeek makes a question naturally arise: Does Nvidia really need to do this? In other words, does the industry really need Nvidia to do this?


In the NVIDIA GTC speech in March, Huang Renxun spent a long time explaining to the world why the scaling law in the inference era has not died and why NVIDIA is still very important. Moreover, Huang Renxun also announced the chip architecture plan of the next generation and the next generation in advance.

At the same time, orders from major customers such as Meta and OpenAI for NVIDIA's new generation of high-end Blackwell architecture chips have also given Huang Renxun's argument a factual basis. OpenAI even launched GPT-4.5, the most expensive model in history, which talks about the same principle.

If this continues, Nvidia's biggest challenge will be to deliver Blackwell, let sales figures speak for itself. As for the Chinese market, apart from DeepSeek’s “soul torture” of Nvidia, it cannot be ignored that it has greatly helped Nvidia’s performance in the Chinese market. DeepSeek has proven the "scent" of H20, and many Chinese giants spent US$16 billion to purchase the "castrated version" of H20.

In other words, from a global market perspective, DeepSeek forces Nvidia to explain that high-end chips are still important, but in the Chinese market where it can only sell the special version of H20, DeepSeek's role is positive.

Until the ban on H20.

Now that NVIDIA has to design new chips specially supplied to China, it is imperative to visit this country's "NVIDIA chip crushing magician".

03.Liang Wenfeng’s choice

No matter which angle you look at it, imagining the scene where the two meet is enough to make people emotional.

As a Chinese born in the 1960s, Huang Renxun followed the elite path of the previous generation: going to the United States to study, work, and start a business. Liang Wenfeng, who was born in the 1980s, followed the path of the new generation: studying in Chinese higher education institutions, staying in China to work, and starting a business. In the end, the two people met at the top amidst the geopolitical chaos of 2025 and sought common development.

It’s just that time has changed, and there is still a question mark as to whether Liang Wenfeng will still need NVIDIA as he did in the past, or whether this will continue to be the case in the future.

Regarding the Chinese market, Huang Renxun is determined not to give up, but will the downstream computing power demand side be forced to accelerate the migration to domestic alternatives? This was an unavoidable question before him.

Just on April 10, Huawei announced a breakthrough new development in its AI infrastructure architecture, launching the CloudMatrix 384 super-node cluster based on a new high-speed bus architecture, and announced that it has been launched on a large scale in the Wuhu data center.

In the field of chips and AI computing, "super nodes" are data processing units with ultra-large computing power and high bandwidth built through high-density integrated computing resources and high-speed interconnection technology. They emerged to meet the needs of large-scale AI model training and inference. NVIDIA's NVL72 is a representative product of super node technology. It consists of 72 GPUs and uses NVLink and NVSwich technology to achieve high-speed communication between GPUs.

Zhang Ping'an, Managing Director of Huawei and CEO of Huawei Cloud Computing, said at the Huawei Cloud Ecosystem Conference 2025: "Huawei does not release large model applications for C-sides. We are more concerned about how to enable industry customers to truly implement AI."

Its competitive relationship with Nvidia in China is self-evident.

Huawei is not the only company in China that is working on AI chip technology, but the tension between Huawei and Nvidia is a typical example: just like DeepSeek "suppressed" its "squeezing magic" despite being restricted, Chinese companies cannot sit still and wait for the United States to "unlock" Nvidia chips. In a sense, the further ban on sales of Nvidia H20 in the United States has injected impetus into China's chip industry.

Huang Renxun has long been aware of the competitive pressure of Nvidia in China. Nvidia has included Huawei on the list of competitors for two consecutive years. In an interview with the Financial Times, Huang Renxun once praised it as "China's most powerful technology company, conquering every market they enter."

Under the layers of castration, Nvidia's chips sold in compliance with regulations in China will almost inevitably be further compressed in terms of cost performance and attractiveness. At the same time, local Chinese companies are planning to make moves.

The ban on the sale of H20 in China may not be enough to deal a fatal blow to Nvidia's future in the Chinese market, but if DeepSeek and Liang Wenfeng abandon or reduce their reliance on Nvidia, that may be the real nightmare for Nvidia in China.