AI Photo:VCG

China should establish a national team to address the country's supercomputing bottleneck, a political advisor told the Global Times on Tuesday. He said he believed this initiative is crucial for China to achieve sovereign Artificial Intelligence (AI) development.

Zhang Yunquan, a member of the Chinese People's Political Consultative Conference (CPPCC) National Committee, also a research fellow from Institute of Computing Technology, Chinese Academy of Sciences, said China's innovation and supply of domestically produced intelligent computing chips for large model training is currently lagging behind, compared to that of the US. 

Zhang wrote in his proposal that domestic AI chip manufacturers in China are limited in their ability to produce cutting-edge chips due to constraints on semiconductor producing and being placed on the US entity list.

Domestic chip manufacturers have to either start from scratch or be forced to be compatible with Nvidia's Compute Unified Device Architecture (CUDA). In the long run, China's chip development would be still subject to restrictions and legal risks, Zhang said, bringing up the challenges facing domestic chip manufacturers.

Zhang proposed establishing special supercomputing research projects, including chips project, with the national strength.

China is able to fully utilize the talent and technological advantages accumulated in the field of supercomputing, so as to quickly make breakthroughs, and narrow the gap between China and the US in the field of sovereignty-based large model training as soon as possible, Zhang noted.

Sovereign AI capabilities can affect the entire country's technological competitiveness, including military, political, economic, and other aspects, in the international community. If we fall behind in this regard, we may face the risk of being sanctioned and suppressed, Zhang noted. 

Sovereign AI refers to a nation's capabilities to produce artificial intelligence using its own infrastructure, data, workforce and business networks, according to Nvidia. 

To achieve a rapid breakthrough in sovereign AI capabilities, Zhang stressed that the issue must be addressed at a national level, by fully coordinating various forces and forming a national team.

The special national team can be composed of leading authoritative experts in the industry and major AI chip manufacturers, and can work to develop domestically produced chips, software, and systems for large-scale model training that can compete with high-end training GPUs from Nvidia. 

With their efforts, national standards and the best chip development technology route used for sovereign AI training can be determined, Zhang noted.