【By Observer Group, Qi Qian】

The U.S. government had previously announced that it would allow NVIDIA to export H200 artificial intelligence (AI) chips to China, but with a 25% fee.

On December 12 local time, David Sacks, the White House AI chief, said this move was aimed at encouraging U.S. companies to challenge Chinese tech giants like Huawei in the Chinese market. However, he also said he was unsure whether this strategy would work.

"They are refusing our chips," Sacks said in an interview with Bloomberg, stating that China had rejected the U.S. H200 chips, "obviously they don't want them, and I think the reason is that they want to achieve semiconductor independence."

The report said that Sacks was one of the main forces pushing the U.S. government to allow exports of H200 chips to China.

He explained that the H200 chip was already "outdated" technology, no longer the most advanced chip.

He then said: "What you're seeing now is that China is not accepting them because they want to support Huawei. This was part of our calculation, which was to sell 'not the best but lagging' chips to China to capture Huawei's market share. But I think China has seen through this, which is why they are not paying for it."

White House official David Sacks, video screenshot

Bloomberg said that Sacks' comments have raised doubts about whether NVIDIA can recover its revenue from the Chinese market.

NVIDIA had completely excluded the Chinese market revenue from its forecasts. According to Bloomberg Intelligence analysts, the annual revenue from the H200 chip in China could reach $10 billion, but only if the Chinese side accepts the chip and pays for it.

A NVIDIA spokesperson said in a statement that the company will continue to cooperate with the U.S. government to provide H200 licenses to reviewed customers.

The company said: "Although we have not yet had results to report, it is clear that three years of excessive export controls have helped foreign competitors develop and cost American taxpayers billions of dollars."

Graphic of NVIDIA H200 chip, NVIDIA official website

According to the introduction, the H200 chip has more high-bandwidth memory than its predecessor, the H100, enabling it to process data faster. According to a report by the U.S.-based think tank the Progressive Policy Institute, the performance of the H200 is nearly six times that of the H20. The report also states that exporting this chip would enable Chinese AI laboratories to build supercomputers with performance close to the top U.S. AI supercomputers, albeit at a higher cost.

On December 8, President Trump announced that he would allow NVIDIA to export its H200 chips to China. Trump said this decision came with conditions, including safeguarding what the U.S. calls "national security," and the U.S. government would receive a 25% cut.

He said the U.S. Department of Commerce is finalizing the agreement details, and the same approach would apply to other AI chip companies such as AMD and Intel.

Foreign Ministry Spokesperson Guo Jiancong said on the 9th: "We have noted the relevant reports. China has always advocated mutual benefit and win-win cooperation between the U.S. and China."

U.S. media said this decision was a major victory for NVIDIA CEO Huang Renxun, and also a policy shift for Trump, as the White House had initially pledged to restrict the sale of AI chips to China.

From the Biden administration to the Trump administration, the United States has implemented a series of measures to restrict chip exports to contain and suppress the development of Chinese technology, and these restrictions have been tightened continuously, with NVIDIA's AI chips being a "top priority." To this end, NVIDIA sought to sell a "downgraded special edition" H20 chip to China.

"But the question now is, will China still pay for NVIDIA chips?" The New York Times said that the H20 chip is a "special edition阉割 version" designed specifically for the Chinese market, and is currently the most advanced model approved by Washington for export, but more and more Chinese buyers are unwilling to pay for it. Huang Renxun recently also admitted that he was unsure whether China would accept the H200.

Earlier this year, the U.S. government had banned NVIDIA from exporting the H20 chip to China until NVIDIA agreed to pay 15% of its sales in China to the U.S. government, after which the export license for the chip was restored.

However, NVIDIA faces issues of security vulnerabilities and monopoly risks in China. In July, Chinese authorities interviewed NVIDIA and requested that it explain and submit relevant proof materials regarding the security risks of backdoors in the H20 computing chips sold to China. In September, Chinese regulatory authorities issued a statement saying that NVIDIA violated the anti-monopoly law and decided to conduct further investigations accordingly.

Chinese enterprises are striving to develop domestic AI chips that can replace NVIDIA, seizing the market share that NVIDIA once dominated. For example, Huawei announced in September its product iteration roadmap for the Ascend AI chip over the next three years. Internet giants such as Alibaba, Tencent, Baidu, and ByteDance have also increased their investment in chip R&D and design to gain greater control over the supply chain.

At the regular press conference on December 10, a foreign reporter asked, according to a report by the Financial Times, although President Trump has decided to allow the export of NVIDIA's H200 chips to China, the Chinese side is considering limiting access to these chips. What is China's comment on this?

Guo Jiancong responded that we have already made a response to the relevant issues yesterday. For specific situations, it is recommended to consult the relevant Chinese authorities.

This article is exclusive to Observer Group, and any reproduction without permission is prohibited.

Original: toutiao.com/article/7583171981609026098/

Statement: This article represents the personal views of the author.