Key Takeaways
Artificial intelligence is increasingly becoming embedded in daily scientific work in China.
The country’s “AI for Science” push is backed by new policies and infrastructure investments as China aims to supercharge scientific and technological breakthroughs.
Researchers in China are moving quickly, but access to top-end chips and high-quality data remains a challenge.
Chongqing, China—Across research labs in Chinese universities and other Chinese institutions, the use of artificial intelligence is becoming more common, with models and algorithms being deployed to help sift materials candidates, design molecules, and test research hypotheses across a range of disciplines.
“It’s the time of AI, so everybody is using AI for writing and researching. Even my professor said, ‘If you don’t use AI, then people will leave you behind.’”
“AI for Science,” which describes researchers embracing AI for scientific purposes, was a notable theme at the Chinese Chemical Society (CCS) Congress held this spring in Chongqing. The meeting, held every 2 years, is considered the most influential gathering for chemistry research in the country. Chinese policymakers have been encouraging scientists in the country to integrate AI across major disciplines and core application areas, including chemistry, physics, mathematics, and astronomy.
“It’s the time of AI, so everybody is using AI for writing and researching. Even my professor said, ‘If you don’t use AI, then people will leave you behind,’ ” Bint e Zahra, a first-year master’s student from Pakistan who is pursuing a chemistry degree at Xiamen University, said during an AI-themed session at the conference.
In December, a month after the US unveiled its Genesis Mission, a major federal AI-for-science initiative, China launched an AI agent system that runs on its national supercomputing network, which comprises more than 30 supercomputing centers across the country and which has been online since 2024.
The AI agent system, accessible to both enterprise and individual researchers, is able to automatically handle the full workflow of scientific tasks with a simple prompt, shortening a day’s work into about an hour, according to Chinese media.
AI is increasingly embedded in China’s development plans
In the country’s latest 5-year plan for economic development, China’s government elevated AI to a priority role to accelerate scientific and technological breakthroughs. As early as 2023, China’s Ministry of Science and Technology, together with the National Natural Science Foundation of China, launched a national project to embed AI into frontier science research across major disciplines. Key proposals under the 2023 project include developing dedicated “AI for Science” platforms for tailored research areas, expanding computing centers, and encouraging open data for scientific research.
The government of Beijing, where many of the country’s leading AI companies and scientists are based, released a 3-year action plan in 2025 to supercharge AI-enhanced science. It proposes to develop a science-focused large language model and more than 10 scientific databases by 2027 to drive AI-powered research into at least five key areas: basic research, biomedicine and health, new materials, scientific instrumentation, and industrial intelligence.
“Acceleration [of research] is not the ultimate goal. The ultimate goal is to find new knowledge and even new functions through this data-driven approach.”
“ ‘AI for Science’ is becoming an important driving force in the new round of technological revolution,” Weinan E told the state-run People’s Daily in July 2025. The mathematician, who trained at the University of California, Los Angeles, coined the term “AI for Science” in China and heads the country’s first dedicated School of AI for Science in the southern tech city Shenzhen. Graduate students are assigned both a human mentor and an AI mentor for their academic work at the school—a first of its kind in the country.
A focus on drugs, materials, and genetic research
China’s “AI for Science” strategy has singled out drugs, new materials and chemistry, and genetic research as priorities. At the CCS Congress, researchers from institutes and universities around the country introduced items they’ve developed via AI tools and agentic systems. Feng Zhu, a professor of medicinal chemistry at Zhejiang University, said his team has developed AnnoPro, an AI tool for protein annotation, along with a database of therapeutic targets that includes data essential for drug discovery.
“I think the integration of AI technology is not a recent thing, but because now people are seeing its benefits, and with the availability of algorithms, computing, and data, you see an explosive growth.”
Zhu told C&EN that this database is now being used by researchers from both China and other countries to develop AI models for their own research. “I think the integration of AI technology is not a recent thing, but because now people are seeing its benefits, and with the availability of algorithms, computing, and data, you see an explosive growth,” said Zhu, adding that he sees AI as a way to shorten drug discovery cycles.
AI is also transforming materials discovery in China, with its use increasingly moving from experimentation to routine practice. Shan Jiang, an associate professor of molecular materials at ShanghaiTech University, told C&EN that her team uses OpenAI amid other AI models to build AI agents to assist their work. The university has purchased AI tokens that it then provides free to researchers. The researchers also work closely with computer scientists from the university to tailor general algorithms they’ve built to assist with their research and explore new materials.
“Acceleration [of research] is not the ultimate goal. The ultimate goal is to find new knowledge and even new functions through this data-driven approach,” said Jiang, who moderated an “AI for Science” session at the CCS conference. She told C&EN that her team is now working to deploy a large language model using open-source models such as DeepSeek to manage data streams produced during daily research.
“I encourage my students to use AI and build their own science agentic system,” said Jiang. “It can help you improve your work efficiency.”
Limits on microchips pose challenges
China’s “AI for Science” push is advancing quickly but not without friction. Limited access to cutting-edge chips and patchy data infrastructure continue to slow the most computing-intensive work, even as scientists increasingly build AI into everyday research.
With challenges around access to cutting-edge Nvidia chips due to US export controls, Chinese researchers have to rely on domestic chips and older Nvidia parts, with less access to advanced graphics processing units.
“On [the lack of] top-end chips, this will slow our R&D, and it does affect the computation required for large models,” said Zhu. “But it has not yet become a critical bottleneck.”
China is also pushing for domestic alternatives, including chips developed by Huawei Technologies, as it intensifies efforts toward technology self-sufficiency. In April, Sugon Information Industry, a Beijing-based supercomputer maker affiliated with the Chinese Academy of Sciences, unveiled what it says is China’s largest AI-for-science computing cluster to date, with 60,000 domestic chips built entirely with Chinese hardware. The cluster is intended to accelerate scientific research in fields such as drug discovery, materials screening, and complex computational simulations. China’s LineShine, a new supercomputer with an innovative chip design that uses only central processing units, also just topped TOP500, a list of the world’s fastest computers.
Meanwhile, practical limits on the use of AI are also visible in the everyday life of research labs, such as geolocation restrictions placed by either the Chinese government or tech companies, including Anthropic, toward users based in China.
Hassan Nawaz, who was completing a master’s degree in computational chemistry at Xiamen University, said his university lab relies on Chinese LLMs, including DeepSeek and Qwen, for work that is run on the supercomputers connected with the school internet, though he personally prefers Anthropic’s Claude.
“Usually, supercomputers are connected to the university internet or Chinese internet, so you cannot have a [virtual private network] for those supercomputers,” said Nawaz, who is from Pakistan, adding that this has created limited access to international software and data. “It’s a very secure internet. You can just access the Chinese products, and you cannot access American or other countries’ products.”
