International Expert Gao Zhikai: If This Is the Case, Nvidia Should Pay China a 15% Security Deposit

Deep News
Aug 25

On August 22, foreign media reports citing informed sources indicated that Nvidia has notified suppliers including Advanced Semiconductor Engineering and Samsung to suspend H20-related production. Previously, on July 31, China's Cyberspace Administration summoned Nvidia to address potential security vulnerabilities and backdoor risks in H20 AI chips sold to China, requiring the company to provide explanations and submit relevant documentation to protect Chinese users' network and data security.

Nvidia's production halt has undoubtedly increased external concerns about H20 security. To address this issue, we connected with renowned international affairs expert Gao Zhikai to seek his recommendations for China's risk prevention regarding imported AI chips.

**Expert Interview Analysis**

Regarding the recent H20 chip "backdoor" controversy, investigations have revealed that the United States has systematically considered installing "backdoors" in AI chips. The US specifically mentioned that if companies cooperate with the US government in installing "backdoors," the government could exclude them from export controls, including relaxing exports to "low-risk Chinese customers."

Gao Zhikai explained that recent US positions on Chinese chip issues have been inconsistent, requiring careful analysis of their true intentions. Initially, when Nvidia sold chips to China, the industry thrived significantly. Nvidia's chips had numerous users in China, and the company profited substantially, providing ample resources for subsequent research and capacity expansion. However, in recent years, the US government launched a chip war, attempting to restrict China in mid-to-high-end chip sectors and limit US chip exports to China, creating a new conflict between the two nations.

During Yale University's anniversary celebration in October 2023, Gao had a brief conversation with US Commerce Secretary Gina Raimondo. He mentioned to Raimondo that China previously spent over $400 billion annually importing chips from the US and other regions. If a chip war were launched against China, China might ultimately stop purchasing chips overseas entirely. As US Commerce Secretary, was she concerned about China's chip supply becoming completely independent and self-sufficient?

Raimondo expressed genuine concern about such a scenario but indicated it depended on China's response. Gao has been contemplating the implications behind Raimondo's remarks.

Currently, we see that Nvidia AI chips, previously strictly restricted and prohibited from sale to China by the US government, have been given an opening. Why did the US government change course? Gao believes several factors are involved:

First, Nvidia CEO Jensen Huang reportedly approached the White House to plead that without allowing Nvidia chip sales to China, the company's revenue would suffer, making it impossible to maintain high R&D investment levels.

Second, the US government realized that Chinese companies like Huawei have achieved considerable technical standards in AI chip production. The US government also worries that Chinese chip manufacturers will not only rapidly dominate the domestic market but may also develop competitive advantages in international markets.

Therefore, the US has changed its initial strategy of completely blocking AI chip sales to China, allowing Nvidia to sell its modified H20 chips designed specifically for the Chinese market. If these China-targeted chips contain no backdoors or suspicious elements, some Chinese manufacturers might find H20 sufficient for their business objectives and purchase them without concern.

However, the question remains: has the US forced Nvidia to install backdoors in chips specifically exported to China? US senators and representatives have consistently demanded that chips exported to other countries, especially to what they call "hostile nations" or countries that "don't align with the US," must contain backdoors.

This means three things: First, the US wants to determine who the final users are and whether they violate purchase regulations or misuse the chips. Second, they want to monitor what these chips are actually doing. Third, the US wants capability to shut down these chips on command - "when we tell you to stop, you must stop." These are contents the US is putting on the table through legislation.

Does H20 currently have other backdoors? Nvidia claims there are no backdoors, citing a 1990s example where attempts to manipulate chips at software and hardware levels crashed entire systems. They believe installing backdoors in chips is inadvisable and could affect overall network security.

However, this is only Nvidia's word. Over a decade ago, the US experienced the major "PRISM Gate" incident involving Edward Snowden, who accused the US government of monitoring and intervening in many online activities through "Regin" software. US intelligence agencies have far-reaching capabilities.

So the question becomes: when Nvidia says "no backdoors," is this actually true?

**Legal Framework Recommendations**

Gao suggests that while China's cyberspace authority's summoning of Nvidia for explanations is necessary and timely, China needs more legally binding agreements with foreign manufacturers. For AI chips, regardless of manufacturer claims, China should force Nvidia and other foreign manufacturers to make clear commitments:

If backdoors or other functions are discovered in chips or products sold in mainland China that harm Chinese users, chip manufacturers must unconditionally accept legal responsibility. Upon discovery, these manufacturers should unconditionally accept penalties. Only companies signing such agreements should be allowed to sell AI chips in China.

Additionally, if chips and products sold in China have vulnerabilities allowing entities like the CIA or FBI to easily breach security and threaten Chinese users, Nvidia or other manufacturers should bear corresponding legal responsibility and provide compensation.

Gao noted that Trump recently introduced a new rule requiring Nvidia, AMD, and other companies to pay 15% of their chip sales revenue in China to the US government in exchange for export licenses. The US government is essentially collecting a "franchise tax" through companies like Nvidia, intervening in their overseas chip sales.

He recommends that the Chinese government sign formal agreements with companies like Nvidia, requiring them not only to guarantee no backdoors or related functions but also to pay 15% of their revenue to the Chinese government as security deposits.

Companies like Nvidia and AMD should deposit 15% of their AI chip sales revenue in China at designated locations as collateral guaranteeing their products contain no backdoors or illegal functions. After a period confirming no backdoors exist, these deposits could be returned.

This is crucial because relying solely on manufacturer statements or paper signatures without substantial legal constraints could have unimaginable consequences if problems arise. US agencies might use elements in Nvidia's software or servers to spy on Chinese users' online activities, generate analytical materials, or even restrict certain training or program operations.

**Institutional Safeguards**

Gao recommends learning from the US by introducing "class action" mechanisms - establishing specialized collective litigation systems. An estimated 1 million H20 chips will enter the Chinese market in 2025, potentially affecting tens of thousands of users. If hidden functions or third-party abuse are discovered in H20 chips, making Nvidia an "accomplice," individual lawsuits would be impractical.

China should allow several users to represent all 1 million chip users in collective action against manufacturers, enabling minority users to defend majority rights if serious violations are proven.

He also suggests China should explicitly prohibit certain entities from using imported AI chips, especially military, police, security departments, and highly sensitive research institutions and technological universities. Rather than gambling on the absence of backdoors, it's better to firmly close the door.

Using domestic chips would be safer, as important information theft or surveillance through foreign AI chips could have unimaginable consequences.

Gao recommends designating a specialized court to handle disputes and litigation involving imported AI chips, staffed with professionals knowledgeable about rapidly developing AI technology. The nation should recruit more AI litigation lawyers and professionals to work with specialized AI courts and collective litigation institutions, forming comprehensive legal protection systems.

Such institutional arrangements would protect legitimate Chinese user and societal interests while deterring overseas entities with ulterior motives from using illegal means to achieve undisclosed objectives.

These recommendations are based on Gao's experience in equity investment funds and corporate management, reflecting on information protection issues. As a lawyer and current chairman of the Yale Law School China Society, he closely monitors US Congressional discussions on AI legislation, hoping China can stay informed about US legal developments to prevent problems before they occur.

Disclaimer: Investing carries risk. This is not financial advice. The above content should not be regarded as an offer, recommendation, or solicitation on acquiring or disposing of any financial products, any associated discussions, comments, or posts by author or other users should not be considered as such either. It is solely for general information purpose only, which does not consider your own investment objectives, financial situations or needs. TTM assumes no responsibility or warranty for the accuracy and completeness of the information, investors should do their own research and may seek professional advice before investing.

Most Discussed

  1. 1
     
     
     
     
  2. 2
     
     
     
     
  3. 3
     
     
     
     
  4. 4
     
     
     
     
  5. 5
     
     
     
     
  6. 6
     
     
     
     
  7. 7
     
     
     
     
  8. 8
     
     
     
     
  9. 9
     
     
     
     
  10. 10