(Yicai) March 4 -- Two executives at Chinese technology companies have put forward proposals about the layout of artificial intelligence inference computing power and the AI industry commercialization at the ongoing Two Sessions, the annual meetings of the National People's Congress and the National Committee of the Chinese People's Political Consultative Conference.

The Chinese government should introduce policies to guide the layout of computing power required for AI inference, establishing a 'national coordination + regional refinement' system for the distribution of AI inference computing power, Zhou Hongyi, member of the National Committee of the CPPCC and founder of Chinese cybersecurity giant 360 Security Technology, said in his proposal to the Two Sessions today.

Zhou advocated for the construction of low-latency, high-density AI inference computing clusters in key industrial agglomeration areas and the promotion of dynamic allocation of computing resources across different levels and regions.

Moreover, he called for encouraging the domestic development of dedicated AI inference chips, focusing on breakthroughs in high-precision, low-latency, and multimodal chip technologies to build an independent and controllable industrial chain and support the in-depth application of intelligent agent technologies.

Demand for AI inference computing power is growing exponentially in the "era of billions of intelligent agents," Zhou noted. The gap between clusters dedicated to AI inference needs to be addressed, the supply and demand relationship among regions should be optimized, and breakthroughs in AI inference chip technology must be achieved, he added.

China should accelerate the transition of the AI industry from 'computing power accumulation' to 'business closed loop,' Guo Yufeng, member of the National Committee of the CPPCC and deputy general manager of Chinese central processing unit maker Phytium Technology, said in its proposal submitted to the Two Sessions. He also emphasized the need for the simultaneous advancement of security access and intelligent upgrades within industrial control systems.

China's domestic localization rate of core software and hardware for high-end industrial control systems is less than 40 percent, Guo noted, adding that the industry is facing challenges, such as fragmented replacement and lagging intelligence levels. Simple import substitution can no longer meet the dual demands of 'safety + intelligence,' he explained.

The replacement process must incorporate new capabilities, including AI, digital twins, and predictive maintenance, Guo believes. He suggested establishing a risk compensation mechanism to promote the transition of domestic industrial control systems from just 'usable' to 'user-friendly and confidently usable.'

Moreover, Guo advocated for abandoning fragmented upgrades and establishing an integrated adaptation principle of 'chip - operating system - industrial application,' and conducting comprehensive replacement pilot projects in sectors such as electric power and chemical engineering.

Guo also called for the improvement of the legal framework for AI governance, addressing legal gaps for issues including liability determination in autonomous driving, copyright ownership of AI-generated content, and the behavioral boundaries of intelligent agents. He highlighted the need to prevent systemic risks and safeguard enterprises' innovative large-scale application of AI.

Editor: Futura Costaglione