The US government has been putting extensive efforts into limiting the technological advance of Chinese companies and research organizations for more than three years. Their primary focus has been on \u00a0export restrictions\u00a0 targeting \u00a0graphics processing units (GPUs)\u00a0 from leading companies like \u00a0NVIDIA\u00a0 and \u00a0AMD\u00a0. The Department of Commerce has been the key body implementing these restrictions, serving as the enforcement arm of the administration’s technological containment strategy.<\/p>\n
<\/p>\n
Despite these substantial efforts, \u00a0cutting-edge chips for artificial intelligence (AI)\u00a0 have continued to make their way into China, albeit in smaller quantities than what organizations focused on developing large AI models actually require. Many of these chips are being acquired through \u00a0secondary markets\u00a0 and \u00a0parallel import routes\u00a0 in countries such as \u00a0India\u00a0, \u00a0Malaysia\u00a0, and \u00a0Singapore\u00a0, where US regulatory power remains limited.<\/p>\n
<\/p>\n
NVIDIA has managed to obtain an \u00a0export license\u00a0 to sell its GPUs for AI, specifically the \u00a0H20\u00a0 model. However, the Chinese government has swiftly \u00a0vetoed\u00a0 its use. The \u00a0Cyberspace Administration of China\u00a0 is reportedly conducting a comprehensive investigation into this GPU, suspecting that it contains a \u00a0backdoor\u00a0 that could be exploited by Chinese experts. This suspicion raises concerns regarding China’s ability to effectively utilize this technology without compromising security.<\/p>\n
<\/p>\n
In response to these restrictions, China has developed three notable alternatives to NVIDIA: \u00a0Cambricon Technologies\u00a0, \u00a0Huawei\u00a0, and \u00a0Moore Threads\u00a0. These Chinese chip designers have strategically positioned themselves to contend with NVIDIA’s market stronghold in China. Notably, many developers engaged in large AI projects that require \u00a0CUDA\u00a0 technology still depend on \u00a0NVIDIA chips\u00a0 due to their high performance and effectiveness. According to a report by Digitimes Asia<\/a>, a significant number of these GPUs are being sourced from the international \u00a0second-hand market\u00a0.<\/p>\n <\/p>\n Interestingly, Chinese firms are not just limited to acquiring the \u00a0H20 chips\u00a0, which have been prohibited. Instead, they are procuring more advanced GPUs, specifically the \u00a0H100\u00a0 and \u00a0A100\u00a0, both of which are \u00a0more powerful than H20\u00a0. The US faces significant challenges in regulating the second-hand market and tracking these parallel import routes. Consequently, the Department of Commerce is considering a \u00a0novel strategy\u00a0 that involves introducing technology capable of tracking these chips throughout their journey.<\/p>\n <\/p>\n Chinese companies are securing large quantities of NVIDIA A100 and H100 in the second-hand market<\/p>\n<\/div>\n<\/div>\n One proposed solution involves a form of \u00a0geolocation\u00a0 technology that would identify the location of each integrated circuit from the moment it leaves the semiconductor factory to its eventual destination. Michael Kratsios, director of the Office of Scientific and Technological Policy at the White House, has confirmed that discussions are underway to implement both hardware and software tracking solutions in the chips themselves. “We are exploring ways to improve the tracking of their location,” Kratsios stated, indicating the urgency of addressing this issue.<\/p>\n <\/p>\n The US administration remains steadfast in its commitment to maintaining leadership in the \u00a0AI hardware development\u00a0 sector as global adoption of AI technology accelerates. Nevertheless, the implementation of the proposed solutions poses \u00a0significant technical challenges\u00a0. Constant f GPS tracking is impractical for this context, according to Kratsios. Thus, one potential alternative is to embed logic within each chip to periodically confirm its location, thereby reducing reliance on external networks.<\/p>\n <\/p>\n This design, however, increases the \u00a0complexity\u00a0 of each chip and could negatively impact performance, affecting productivity in data centers. Moreover, increased chip costs and the risk of \u00a0hackers\u00a0 exploiting any vulnerabilities in the tracking logic present additional complications for this approach.<\/p>\n <\/p>\n Image | Nvidia<\/a><\/p>\n More information | Digitimes Asia<\/a><\/p>\n In Xataka | The US will not be able to contain the technological development of China. Experts from the chips industry forecast it.<\/p>\n<img alt=\"Universal quantum computers are closer thanks to this unexpected discovery\" width=\"375\" height=\"142\" src=\"https:\/\/i.blogs.es\/4c4247\/ising-ap\/375_142.jpeg\"\/><\/code><\/pre>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n