Artificial intelligence (AI) is at the \u00a0core\u00a0 of the ongoing technological rivalry between the \u00a0United States\u00a0 and \u00a0China\u00a0. The Trump administration took significant measures to restrict the importation of advanced AI chips designed by \u00a0Nvidia\u00a0, \u00a0AMD\u00a0, and \u00a0Intel\u00a0. These restrictions aim to \u00a0prevent\u00a0 cutting-edge AI technology from reaching \u00a0China\u00a0, governed by Xi Jinping. Conversely, China is investing heavily in its AI infrastructure, focusing on developing its own advanced \u00a0GPUs\u00a0 and AI models.<\/p>\n
<\/p>\n
The competitive landscape in the AI sector goes beyond just the \u00a0US-China\u00a0 rivalry. Over the past four years, the Chinese government has provided substantial \u00a0subsidies\u00a0, resulting in a surge of companies committed to large AI model development. While the expansive Chinese market offers immense potential, internal conflicts, particularly between \u00a0Huawei\u00a0 and \u00a0Alibaba\u00a0, have surfaced amid this competition.<\/p>\n
<\/p>\n
Ren Zhengfei, the founder and CEO of \u00a0Huawei\u00a0, made noteworthy comments in early June during an interview with a journalist from the People’s Daily. He admitted that \u00a0Huawei’s Ascend GPUs\u00a0 lag a generation behind \u00a0US\u00a0 chips. This admission raised eyebrows, as it is uncommon for a leading executive to acknowledge their company\u2019s shortcomings publicly.<\/p>\n
<\/p>\n
These statements from Zhengfei come at a crucial moment for Huawei. Just two months prior, the company announced two new AI chips, the \u00a0Ascend 910d\u00a0 and \u00a0Ascend 920\u00a0, aimed at competing with \u00a0Nvidia\u00a0 for performance supremacy in AI applications, both domestically and internationally. Notably, \u00a0Noah Ark Lab\u00a0, Huawei\u2019s research division, has successfully implemented its first large-scale model entirely on \u00a0Ascend chips\u00a0.<\/p>\n
<\/p>\n
Honestagi has published a study in Github claiming a correlation between \u00a0Pangu Pro Moe\u00a0 and \u00a0Qwen 2.5 14b\u00a0.<\/p>\n<\/p><\/div>\n<\/div>\n
For Huawei, developing a \u00a0large language model\u00a0 using its own chips rather than relying on \u00a0Nvidia\u00a0 GPUs is a major achievement. However, the launch of its latest \u00a0Pangu Pro Moe model\u00a0\u2014which utilizes a \u00a0Mixture of Experts\u00a0 configuration\u2014has not been without controversy. A research group, \u00a0Honestagi\u00a0, published a study on GitHub arguing that there is a notable correlation between Huawei\u2019s \u00a0Pangu Pro Moe\u00a0 and Alibaba\u2019s \u00a0Qwen 2.5 14b\u00a0 model. This raises questions about the originality of Huawei\u2019s model.<\/p>\n
<\/p>\n
This correlation suggests that Noah Ark Lab might have trained its AI model using \u00a0Qwen 2.5 14b\u00a0 rather than developing it independently. If true, this could imply a violation of copyright from \u00a0Alibaba\u00a0. Despite the accusations from Honestagi, \u00a0Huawei has not publicly denied\u00a0 these claims. Instead, Huawei emphasizes that \u00a0Pangu Pro Moe\u00a0 does not rely on incremental training from other manufacturers\u2019 models.<\/p>\n
<\/p>\n
Another critical point to consider is that \u00a0Huawei\u00a0 asserts its development team adhered strictly to \u00a0open-source licensing\u00a0 requirements for any third-party code used, although it has not disclosed which open-source models were utilized. Meanwhile, \u00a0Alibaba\u00a0 has yet to respond to this controversy but may offer its perspective in the coming days. The unfolding tension between two of China’s most prominent AI companies presents an intriguing narrative as the industry continues to evolve.<\/p>\n
<\/p>\n
Image | Huawei<\/a><\/p>\n