Chinese companies dedicated to developing large \u00a0artificial intelligence\u00a0 (AI) models face a profound challenge. They are caught in a web of \u00a0export restrictions\u00a0 imposed by the U.S. government on graphics processing units (GPUs), which are essential for AI training. Adding to their struggles is the reality of their dependence on \u00a0American technology\u00a0. Ideally, a solution would involve ceasing purchases from Nvidia and similar U.S. firms, instead turning to “comparable” GPUs offered by Chinese companies like Huawei and Moore Threads.<\/p>\n
<\/p>\n
However, as noted in an article by Foreign Policy<\/a>, the situation is not straightforward. American analyst Kyle Chan emphasizes that abandoning Nvidia is more complicated than it may appear. Companies such as Tencent, Bytedance, and Alibaba prefer Nvidia GPUs due to their superior performance, particularly in training AI models. Their allegiance to Nvidia is also largely driven by the \u00a0CUDA\u00a0 (Compute Unified Device Architecture) platform.<\/p>\n <\/p>\n The majority of currently developed \u00a0AI projects\u00a0 utilize CUDA, the ecosystem that integrates compilers and development tools required for software engineering on Nvidia GPUs. Transitioning to an alternative platform for ongoing projects presents significant challenges. Although Huawei is making strides with its \u00a0Cann\u00a0 (Compute Architecture for Neural Networks) initiative, CUDA still dominates the landscape.<\/p>\n <\/p>\n “China must develop an alternative system to achieve self-sufficiency in AI”<\/p>\n<\/p><\/div>\n<\/div>\n This statement by \u00a0Li Guojie\u00a0, a prominent computer scientist at the Chinese Academy of Sciences, illuminates the critical need for developing alternative systems to achieve lasting \u00a0self-sufficiency in AI\u00a0. Li highlights that while China’s \u00a0Deepseek\u00a0 has made inroads into the CUDA ecosystem, complete dominance has not yet been realized due to existing barriers. He stresses the need for a comprehensive set of software tools that can surpass CUDA functionalities.<\/p>\n <\/p>\n Addressing these challenges is essential for China’s aspirations in the AI domain, and the Cann initiative may be key. Over the past five months, Huawei has introduced two competitive GPUs and is gearing up to position Cann as an \u00a0open-source toolkit\u00a0. This shift aims to “accelerate innovation among developers and simplify the usability of the Asce Family chips,” according to \u00a0Eric Xu Zhijun\u00a0, Huawei’s Deputy Chairman.<\/p>\n <\/p>\n While Xu does not explicitly state it, his strategic vision seems to focus on bolstering the \u00a0Huawei ecosystem\u00a0 by targeting Nvidia’s core strengths. Discussions are underway with major players in China’s AI industry, including business partners, universities, and research institutions, to build this open-source ecosystem. If this initiative flourishes, it could mark a crucial breakthrough toward \u00a0China’s technological independence\u00a0 in the realm of AI.<\/p>\n <\/p>\n As these companies navigate the complexities of their technological environment, the implications extend beyond mere business competition. The AI race between the U.S. and China reflects broader themes of \u00a0geopolitical strategy\u00a0 and economic self-sufficiency. With rising endeavors to cultivate indigenous technologies and reduce reliance on foreign entities, the stakes could not be higher. <\/p>\n In conclusion, the landscape of AI development in China is marked by strategic challenges and significant potential. As companies begin to foster self-reliance while contending with restrictive trade measures, their efforts could reshape the future of AI technology globally.<\/p>\n Image | Nvidia<\/a> | Huawei<\/a><\/p>\nCUDA: Nvidia’s Key Advantage in AI Hardware<\/h2>\n
<img alt=\"Stars and the mysteries of the universe\" width=\"375\" height=\"142\" src=\"https:\/\/i.blogs.es\/32b524\/estrellas-ap\/375_142.jpeg\"\/><\/code><\/pre>\n<\/div>\n<\/div><\/div>\n<\/div>\n