AI for Science: The Underlying Secret of China's AI Explosion
When AI becomes the "infrastructure" for basic research and the transformation of scientific research paradigms touches the underlying logic of technological development, the great explosion of Chinese technology will no longer be a distant dream.
While application - layer concepts such as AI agents and generative AI are causing waves of hype in the capital market, a concept with more fundamental transformative significance is quietly reshaping the development logic of Chinese technology - AI for Science (Scientific Intelligence, abbreviated as AI4S). It is not simply a technological application but uses artificial intelligence to empower basic scientific research, driving a disruptive transformation of the scientific research paradigm from "trial - and - error driven" to "data + model driven". While the capital market is still chasing short - term application dividends, Chinese enterprises have been deeply involved in AI4S in three core industries: new materials, biomedicine, and chips, attempting to break through the efficiency bottleneck of basic research. This in - depth integration of basic research and AI may be the key for Chinese technology to break through the "chokepoint" dilemma and achieve a great explosion.
The core of AI for Science is to make artificial intelligence the "super assistant" of scientists. Traditional scientific research is often plagued by long cycles, high costs, and difficult data processing. For example, the research and development of new materials may require tens of thousands of experimental iterations, the average time for new drug research and development is more than 10 years, and chip design faces the computing power challenge of hundreds of billions of parameters. AI4S can extract patterns from massive data, build prediction models, and even generate scientific hypotheses in reverse through technologies such as machine learning, big data analysis, and quantum computing simulation. It can not only compress the scientific research cycle several times but also address high - dimensional and complex problems that are difficult for human intuition to reach. This is also the core value that differentiates it from ordinary AI applications - the former focuses on industrial implementation, while the latter is rooted in basic innovation, which is the "source of water" for technological development.
In China, AI4S is not just a theoretical concept but has already formed implementation cases in three core industries. From the atomic - level design of new materials to drug discovery in biomedicine and then to the computing power breakthrough of chips, listed companies are becoming the main force in technological exploration, paving a development path for AI4S with Chinese characteristics.
In the field of new materials, the cooperation between Fangda Carbon New Material Co., Ltd. (600516.SH) and Jingtai Technology has become a typical example of traditional manufacturing enterprises embracing AI4S. Carbon new materials are the core foundation for new energy and high - end manufacturing. However, under the traditional R & D model, optimizing the formula and debugging the process of high - performance carbon - based materials often take several years, and it is difficult to break through the performance bottleneck.
In 2025, Fangda Carbon signed a strategic cooperation agreement with Jingtai Technology. Relying on Jingtai Technology's AI algorithms and quantum chemistry computing capabilities and combining its own industrial experience in the carbon material field, they created a "AI + robot" super - intelligent body for new material R & D. By building a vertical large - scale model, they achieved atomic - level design of high - end materials such as silicon - carbon composites and graphene, compressing the material R & D cycle from the traditional 2 - 3 years to 3 - 6 months. At the same time, they used digital twin technology to optimize the production process, increasing the yield rate of high - end carbon products by more than 15%. Fangda Carbon plans to invest 1 billion yuan in innovation funds over three years and jointly establish a special talent fund with Jingtai Technology to cultivate AI material R & D talents. The core value of this model lies in closing the loop of "AI algorithms + industrial data + experimental verification", enabling basic research to directly target industrial needs and avoiding the problem of the disconnection between scientific research and industry. In the future, with the deepening of cooperation, the two sides are expected to achieve technological breakthroughs in fields such as anode materials for lithium - ion batteries and carbon - based materials for semiconductors, providing underlying support for the high - end transformation of China's new material industry.
In the field of biomedicine, Medicilon Inc. (688202.SH) has reconstructed the entire chain of AI - driven drug R & D, becoming a benchmark for AI4S in the field of pharmaceutical R & D. Traditional drug R & D is known as a "money - burning and time - consuming" industry, with bottlenecks in every link such as target discovery, molecular design, and pre - clinical research. The average cost of developing a new drug exceeds 2 billion US dollars. Medicilon was the first to achieve in - depth integration of "AI + CRO", integrating open - source technologies such as Google's AlphaFold3 and NVIDIA's BioNeMo, and building an AI drug discovery platform covering target screening, molecular design, and pre - clinical research.
In the target screening stage, its self - developed algorithm, combined with AlphaFold3's atomic - level protein structure prediction ability, can complete 5000 automated iterations of virtual compound libraries per week, increasing the accuracy of toxicity prediction to 92%. In the molecular design stage, the generative model developed based on BioNeMo can explore a potential chemical space of the order of 10^60, designing innovative molecules that are difficult to reach by traditional methods. In the pre - clinical research stage, the deployment of NVIDIA's DGX SuperPOD computing power cluster has significantly improved the accuracy of the drug metabolism prediction model and reduced the dependence on animal experiments by 30%.
A typical case is the ISM3412 project in cooperation with Insilico Medicine. Through full - process AI empowerment, the pre - clinical R & D cycle was compressed by 40%, and the IND application was quickly completed. In 2024, the proportion of Medicilon's AI - related revenue reached 18%, and it is expected to increase to 45% in 2027. This exploration not only reduces the cost of drug R & D but also brings the efficiency of Chinese pharmaceutical R & D to the global first - tier, providing a new path for meeting unmet clinical needs such as rare diseases and tumors.
In the field of chips, Dowstone Technologies Co., Ltd. (300409.SZ) has targeted the computing power bottleneck of AI4S and built a dual computing power base for new material and chip R & D by deploying atomic - level scientific computing chips. The implementation of AI4S relies on powerful computing power support. Especially in scenarios such as molecular simulation and atomic - level material design, extremely high requirements are placed on the parallel computing ability of chips, and the traditional CPU/GPU architecture is difficult to meet the demand.
Dowstone Technologies entered the AI4S - specific chip track by participating in the equity of Xinpeisen. The APU chip launched by Xinpeisen is specially designed for atomic - level scientific computing, solving the computing power bottleneck of traditional chips in scientific computing. At the same time, Dowstone Technologies built the Hexi Atomic Computing Center, combining the APU chip with its own new material R & D needs, achieving a two - way cycle of "chip R & D empowering material design, and material needs feeding back to chip optimization".
In the R & D of lithium - battery materials, the Hexi Computing Center can accurately predict the electrochemical performance of electrode materials through atomic - level simulation, increasing the efficiency of new material formula screening by more than 10 times. The problems of material heat dissipation and performance optimization encountered in the chip R & D process can be solved with the help of Dowstone Technologies' accumulation in the new material field. This collaborative innovation model of "chips + materials" not only fills the gap in domestic AI4S - specific computing power chips but also constructs a technology - closed loop with independent controllability, providing underlying computing power support for basic research in fields such as chips and new materials.
The reason why Chinese enterprises' exploration in the field of AI4S can achieve good implementation essentially lies in leveraging the triple dividends of "industrial demand traction + technological independent innovation + continuous policy support". Different from overseas, which focuses more on basic theoretical research, China's AI4S has targeted industrial pain points from the beginning. The explorations of Fangda Carbon, Medicilon, and Dowstone Technologies are all carried out around the enterprises' own industrial needs, avoiding the disconnection between technology and the market. At the technical level, the breakthroughs in domestic computing power chips and AI algorithms have provided guarantees for the independent and controllable development of AI4S. The computing power bases of enterprises such as Loongson Technology and Hygon Information have formed a synergistic effect with application - layer enterprises. At the policy level, AI4S has been incorporated into the core layout of the national scientific and technological innovation system, and various regions have introduced support policies such as super - computing centers and scientific research data sharing, providing fertile ground for technology implementation.
Of course, the development of AI4S still faces many challenges: problems such as the scarcity of high - quality scientific data, the lack of interdisciplinary talents, and the insufficient interpretability of models are still bottlenecks restricting its large - scale implementation. However, it cannot be denied that AI4S provides an unprecedented opportunity for Chinese technology to break through the short - board of basic research. Basic research is the "root" of technological development. In the past, the rapid development of China's technology industry mainly relied on innovation at the application level and optimization of business models. AI4S gives China the opportunity to achieve "overtaking on a new track" in basic fields such as new materials, biomedicine, and chips.
When AI becomes the "infrastructure" for basic research and the transformation of scientific research paradigms touches the underlying logic of technological development, the great explosion of Chinese technology will no longer be a distant dream. The explorations of Fangda Carbon, Medicilon, and Dowstone Technologies are just a microcosm of China's AI4S development. In the future, as more enterprises are deeply involved in basic research and with continuous breakthroughs in computing power, algorithms, and data, AI4S will surely become the core driving force for Chinese technology to move from "following" to "running side by side" and then to "leading", writing a new chapter in Chinese scientific and technological innovation.
This article is from the WeChat official account "Competition and Cooperation in Artificial Intelligence". Author: Competition and Cooperation. Republished by 36Kr with permission.