Artificial intelligence is no longer merely the driving force behind global semiconductor demand, but is also actively reinventing how chips are designed and built, industry leaders and scientists said at a major science forum in Shanghai on Friday.

Chen Yunji, co-founder of AI chipmaker Cambricon Technologies, predicted that the integration between AI and chip design would eventually automate the chip design process entirely.

“In the [next] 10 years, the most important development … is the RSI (recursive self-improvement) techniques for artificial intelligence, where AI models improve AI models, and AI chips improve AI chips,” Chen said.

“With the potential of RSI, it is possible that one day AI systems will autonomously design better chips, which leads to better AI systems in a self-recursive loop.”The transformation is already reaching the factory floor, according to the speakers. Lu Guangquan, chairman of Chinese semiconductor equipment maker Piotech, said industrial AI had evolved from generating text to executing physical fabrication processes.

With front-end manufacturing tools accounting for over 80 per cent of fabrication costs, Chinese foundries were increasingly deploying AI-powered digital twins and predictive models to improve chip equipment engineering and operations, and eventually boost wafer yields, Lu said.

The remarks, delivered at the 2026 World Laureates Forum, underscored how China’s chip sector is embracing AI to overcome domestic manufacturing bottlenecks and drive automation across the entire supply chain.

The shift comes as AI shapes the architecture of scientific computing. Jack Dongarra, an American computing pioneer and Turing Award winner who also addressed the forum, said AI was forcing those in the industry to rethink processor design.

“AI does not simply add another workload to high-performance computing; it also changes the hardware that scientific computing will have,” Dongarra said, noting that AI workloads depended less on 64-bit FP64 calculations and placed greater strain on memory bandwidth and energy across systems.

FP64 is a 64-bit double-precision floating point, the mathematical format that most scientific simulations and supercomputing benchmarks rely on.

Beyond the production line, the technology was reshaping how China trained its next generation of chip engineers, experts said.

Wu Hanming, director of the Information Science Division at Zhejiang University, said the university’s integrated circuits college was using AI systems such as FabGPT alongside simulation engines. The set-up created virtual fabs where students could experience CMOS production lines digitally, he said.

But even as AI automated code and semiconductor layouts, it remained dependent on hardware to perceive the real world, according to Liu Sheng, dean of the School of Integrated Circuits at Wuhan University.

Liu highlighted surging demand for smart sensors and micro-electromechanical systems (MEMS) in humanoid robotics and autonomous driving.

“Sensors are the foundation … acting as the link between the physical and information worlds,” Liu said, adding that developing advanced packaging and testing platforms was critical to keeping complex MEMS chips reliable and cost-effective.