As a condition for obtaining new export licenses, Nvidia and Advanced Micro Devices (AMD) have agreed to remit 15% of their revenues from chip sales in China to the U.S. government, the Financial Times reported on Sunday. This marks an unprecedented arrangement, as no U.S. company has previously consented to revenue sharing with the government as a prerequisite for export approval.

  According to the report, Nvidia and AMD separately received U.S. licenses last week to export their H20 and MI308 artificial intelligence (AI) chips to the Chinese market, with the aforementioned revenue - sharing agreement attached. A U.S. official and other informed sources cited by the Financial Times stated that the U.S. government has not yet determined how to utilize this fund.

  In an emailed comment on Sunday, a Nvidia spokesperson neither confirmed nor denied the report. "We comply with the rules established by the U.S. government for our participation in the global market," the spokesperson said. "Although we have not exported H20 to China for several months, we hope that export control rules will enable the U.S. to maintain competitiveness in the Chinese and global markets." AMD has not yet responded to requests for comment.

  Earlier on Sunday, Reuters reported that a Chinese media article claimed Nvidia's H20 chip - specifically designed by Nvidia for the Chinese market to bypass U.S. export restrictions on advanced AI chips - poses security risks, alleging that the chip could be remotely shut down via a "backdoor".

  Nvidia denied this claim and stated in an emailed statement on Sunday, "Cybersecurity is of utmost importance to us. There are no 'backdoors' in Nvidia's chips that would allow anyone to remotely access or control them." In another blog post published last week, Nvidia emphasized that the use of "backdoors" or "kill switches" would "undermine global digital infrastructure and erode trust in U.S. technology".

  Chinese State Media Criticizes Nvidia H20 Chip as Unsafe, Outdated, and Unenvironmentally Friendly 

  After China's Cyberspace Administration of China (CAC) interviewed Nvidia, the U.S. chip giant, on July 31, Chinese state media published an article on Sunday (August 10) discussing the U.S. practice of installing backdoors in chips and criticizing Nvidia's H20 chip as unsafe, outdated, and unenvironmentally friendly.

  On July 31, the CAC interviewed Nvidia, requiring the company to explain security risks related to chip vulnerabilities or backdoors and submit relevant supporting materials.

  "Yuyuan Tantian", a new media platform under China Central Television (CCTV), pointed out in the article on Sunday that backdoors mainly fall into two categories: hardware backdoors and software backdoors. A hardware backdoor is a physical device embedded in a chip during the design or manufacturing process, primarily referring to logic circuits with backdoor functions. A software backdoor can be understood as instructions with backdoor functions implanted in software, which can damage users' systems or steal confidential information when the software is running.

  The article further stated, taking Nvidia's H20 chip as an example, "From the perspective of hardware backdoors alone, it is entirely feasible to implement functions such as remote shutdown."

  Citing security experts from the Threat Intelligence Center of Qi'anxin, "Yuyuan Tantian" noted that from a technical perspective, it is relatively easy to implement a hardware backdoor with specific denial - of - service functions during the production stage. However, this method incurs relatively high costs and risks. Installing a backdoor through software settings or a combination of software and hardware is a more flexible approach.

  Security experts from Qi'anxin's Threat Intelligence Center said that the U.S. relies on two key levers to establish its AI hegemony: hardware and software ecosystems. For other countries, it is essential not only to strive for hardware substitution but also to build an independent and controllable software ecosystem.

  The article also mentioned that to achieve these arrangements, the U.S. has systematically designed a "on - chip governance mechanism". "This mechanism states that the U.S. government needs to establish relevant departments to coordinate all links in chip design, production, and manufacturing, including coordinating enterprises and allies, to achieve control over AI chips." 

  The article quoted a report introducing the "on - chip governance mechanism" as saying that Nvidia's AI chips have already widely deployed most of the functions required for on - chip governance, although some have not yet been activated. The report also suggested that to obtain the cooperation of chip enterprises, certain "incentive" measures could be adopted, such as relaxing exports to "low - risk Chinese customers".

  The article commented, "Considering this information, when we look at the U.S. government allowing Nvidia to export H20 to China, it is hard not to feel a sense of unease. From any perspective, H20 is not a secure chip for China."

In addition to being unsafe, the article pointed out that H20 is also outdated. According to data from relevant institutions, compared with H100, the standard version of H20, H20 has an overall computing power of only about 20%. The number of GPU cores in H20 is 41% less than that in H100, and its performance is 28% lower. This makes H20 unable to meet the training needs of trillion - parameter large models.

  The article also criticized H20 for being unenvironmentally friendly, stating that according to calculations by relevant institutions, H20 has an energy efficiency ratio of approximately 0.37 TFLOPS/W, which fails to meet the energy - saving standard of 0.5 TFLOPS/W.

 "Yuyuan Tantian" wrote in the article, "When a chip is neither environmentally friendly, nor advanced, nor secure, 'as consumers, we can certainly choose not to buy it.'"

  The H20 is a chip specifically manufactured by Nvidia for the Chinese market. In April this year, the Trump administration implemented new regulations, placing H20 under export controls. Last month, U.S. President Trump lifted the ban, and Nvidia immediately submitted an export application. However, three weeks after the ban was lifted, the U.S. had not yet started issuing licenses, much to Nvidia's chagrin.

  The Financial Times quoted informed sources as saying that Nvidia CEO Jensen Huang went to the White House on Wednesday (August 6) and held another meeting with Trump. Two days after the meeting, the U.S. Department of Commerce began issuing the licenses.

  The report also mentioned that the H20 has become a focal point of debate between U.S. security officials and Nvidia. Security officials claim that the H20 may assist China in developing its military capabilities, while Nvidia argues that blocking U.S. exports will only force China to accelerate its innovation efforts.

Article reprinted from Weiku Electronic Market Network, https://www.dzsc.com