China's AI can innovate beyond chip constraints: HK expert

Amid tightening US curbs on advanced chipmaking technology, Chinese AI researchers are turning to software and systems innovation to sustain competitiveness. While domestic AI accelerator cards cannot access cutting-edge processes, experts argue that algorithmic optimization and hardware-software co-design offer a viable way forward.

Professor Yao Xin, vice-president of Lingnan University in Hong Kong, said that leveraging low-level programming to directly tap hardware architecture could boost performance without relying on advanced fabrication. He also suggested that an integrated system of imperfect components may still deliver reliable outcomes.

Yao spoke with CGTN on July 18 while attending the World AI Conference in Shanghai.

Yet the rising cost of large language models has sparked debate. Yao warned that high-level reasoning consumes water equivalent to 1.2 million people's annual drinking needs and raises serious environmental concerns as China pursues its 2035 carbon peak goal. He advocates exploring alternative paradigms, such as spiking neural networks and analog computing, which could slash energy use by orders of magnitude.

Hong Kong, he added, offers a unique edge: free from heavy industry pressure, its universities can pursue curiosity-driven, foundational research – the kind that fosters "0-to-1" breakthroughs. With deeper Guangdong-Hong Kong-Macao Greater Bay Area integration, the city is well-positioned to complement the mainland's industrial strength with its scientific agility.

China, he concludes, has overcome technological blockades before. The same resilience, coupled with fresh thinking, may light the next AI innovation.