The journey in China of the \u00a0NVIDIA GPU for artificial intelligence (AI) H20\u00a0 has been filled with ups and downs. Since this chip entered the Chinese market in mid-2024, its sales surged by \u00a050% quarter to quarter\u00a0, positioning it as \u00a0the most successful Nvidia product\u00a0 of recent years. However, this era of bonanza was short-lived. In mid-April, the \u00a0US Department of Commerce\u00a0 imposed export restrictions on the H20 GPU specifically targeting China.<\/p>\n
<\/p>\n
Following weeks of negotiations, Nvidia secured the necessary export license to sell the H20 GPU in China. Unfortunately, the joy was fleeting. The \u00a0Chinese government\u00a0 subsequently imposed a ban on the chip, citing concerns from the \u00a0Cyberspace Administration of China\u00a0, the primary internet regulatory body. They initiated a comprehensive investigation into the H20 GPU, suspecting it might incorporate a backdoor that could be exploited, which raised alarms among Chinese cybersecurity experts.<\/p>\n
<\/p>\n
In response, \u00a0Nvidia engineers\u00a0 have been diligently working on a new GPU specifically tailored for the Chinese market. Dubbed the \u00a0B30A\u00a0, this chip is crucial for the future of the company led by \u00a0Jensen Huang\u00a0 in China. The B30A must satisfy two critical requirements: it needs to be more powerful than the H20 GPU while also complying with the restrictions set forth by the \u00a0US Department of Commerce\u00a0. Failing to meet these criteria could prevent Nvidia from obtaining the export license needed to sell the B30A in China.<\/p>\n
<\/p>\n
Chinese companies involved in the development of large AI models find themselves in a precarious situation. They must navigate the \u00a0export restrictions imposed by the US government\u00a0 while grappling with their dependency on American technology. In an ideal world, these firms would halt their purchases of Nvidia chips and turn towards \u00a0Chinese alternatives\u00a0 such as those offered by \u00a0Huawei, Cambricon,\u00a0 and \u00a0Moore Threads\u00a0. These local firms have made strides in developing comparable GPUs for AI applications.<\/p>\n
<\/p>\n
Jensen Huang has just recognized that his next GPU will take to arrive in the country led by Xi Jinping<\/p>\n<\/p><\/div>\n<\/div>\n
However, according to an article in \u00a0Foreign Policy\u00a0 by analyst \u00a0Kyle Chan\u00a0, the situation is more complicated than meets the eye. Abandoning Nvidia is a daunting task for many Chinese companies\u2014including giants like \u00a0Tencent\u00a0, \u00a0Bytedance\u00a0, and \u00a0Alibaba\u00a0. They opt for Nvidia GPUs because of their superior performance, particularly for training their AI models. Moreover, \u00a0CUDA (Compute Unified Device Architecture)\u00a0 technology has ingrained itself deeply within the software development landscape, making it challenging to transition to alternatives.<\/p>\n
<\/p>\n
As most ongoing AI projects utilize CUDA, switching to a different technology could introduce significant complications. \u00a0Huawei\u00a0 has its own alternative called \u00a0Can (Compute Architecture for Neural Networks)\u00a0, but so far, CUDA remains the dominant player in the market.<\/p>\n
<\/p>\n
In this context, the B30A chip represents a vital asset for Nvidia. Reports suggest that the B30A may be half as powerful as Nvidia\u2019s most advanced GPU, the \u00a0B300 chip\u00a0. Despite this, it is expected to outperform all currently developed chips in China, particularly regarding AI model training processes.<\/p>\n
<\/p>\n
However, Jensen Huang recently acknowledged that the arrival of the B30A in China will take time\u2014not due to its readiness, but because of the lengthy approval process from the \u00a0US Trade Department\u00a0. The ongoing saga surrounding Nvidia\u2019s position in China is nothing short of dramatic.<\/p>\n
<\/p>\n
Image | Nvidia<\/a><\/p>\n More information | Reuters<\/a><\/p>\n In Xataka | The US gives Huawei a great opportunity: to get its new chip for AI with the Nvidia market in China.<\/p>\n