AI, science and the risks in China’s reliance on imported precision equipment
The country is dependent on overseas high-end scientific instruments, crimping use of artificial intelligence, researcher says
China’s reliance on imports for the most sophisticated scientific instruments could hold back the country’s use of AI in according to a leading Chinese researcher. Advanced equipment such as mass spectrometers was essential for generating the high-quality experimental data needed to develop, validate and improve advanced scientific models, Weinan E, a professor at Peking University’s mathematical sciences school, said at the “AI for Science” conference in Shanghai last week.
“Without domestically developed precision instruments, it becomes difficult to obtain first-hand, high-quality experimental data, leaving AI ‘like cooking without rice’,” E said, according to Shanghai-based news outlet The Paper last week. E, who is also a member of the Chinese Academy of Sciences, proposed the concept of “AI for Science” in 2018 as a new approach to research.
Scientists use tools to improve scientific research, in areas ranging from improving computational modelling to experimental design. And to get the best data, researchers need the best equipment. Despite rapid market growth and , China has long relied heavily on imports for advanced scientific instruments, particularly from the United States, which leads at the high end of the industry. In 2024, China imported nearly US$17 billion in scientific equipment, and more than three-quarters of the big research instruments used in the country were from abroad, according to a report in December by Beijing-based consulting firm Puhua Policy.
And in a report earlier this year, consultancy firm LeadLeo estimated that China relied on imports for 83 per cent of its mass spectrometers and chromatographs, and 75 per cent of its spectrometers. These instruments are essential for scientific research. Mass spectrometers help identify molecules, chromatographs separate chemicals for analysis, and spectrometers use light to study the properties of materials.
China was also almost completely dependent on imports for optical instruments and biological tissue analysis equipment, the LeadLeo report said. The dependence has led to high equipment costs, long maintenance cycles, and slower after-sales support, raising concerns over China’s research efficiency and supply-chain resilience. At the same time, the US has restricted Chinese access to this kind of equipment.
By December 2020, during Donald Trump’s first term as US president, more than 42 per cent of the 4,510 China-related entries on the , according to analysis by Chinese researchers. Those efforts have continued into Trump’s second term, driven by concerns in Washington that advanced technologies could support Beijing’s military modernisation and help design new weapons through AI.
In January, the US Department of Commerce announced new export controls involving high-parameter flow cytometers and certain mass spectrometry equipment, saying that the technologies could “generate high-quality, high-content biological data, including that which is suitable for use to facilitate the development of AI and biological design tools”.
In his address to the conference in Shanghai last week, E also warned that China’s progress in AI for science could be constrained by the “significant gaps” in foundation models compared to their international counterparts, calling it a top risk that “cannot be overlooked” and “a reality that must be confronted”.
He said that simply adding scientific capabilities to existing open-source models had proved “a false premise”, as solving complex scientific problems required stronger underlying models rather than post-training modifications alone. He said the big difference was in the way the US and China equipped AI with scientific abilities.
While the US concentrated on improving general-purpose foundation models and integrating them with automated research infrastructure, China had adopted a more application-driven approach. He said China started by building scientific AI infrastructure that integrated data, software, computing resources and automated equipment, and then applied these capabilities to specific research fields and scientific tasks.
E called for a restructuring of China’s research system to adapt to the AI era, identifying three “breaks”. He said the scientific community should break down boundaries between disciplines to enable cross-field research; break the divide between theory and experimentation; and break down the barrier between academia and industry. He also suggested an overhaul of traditional research evaluation systems, calling for greater recognition of contributions beyond academic publications, such as development of data, software and research infrastructure.
Artificial Intelligence thread
- Thread starter 9dashline
- Start date
I'm on GLM Max Quarterly plan, 3 month @ $128/mo. Was on $50/mo plan but rate not enough and had to split with Kimi, then found out Zhipu is on US entity list which is an instant upgrade to max and take my money lol
For my main track GLM on Max is enough, I do exploratory work on Kimi and let my team use it, and I use DS4 Flash for small things for the speed.
I agree GLM is better than Kimi right now, Kimi never fail if I give it clear instructions on what the goal is, what to test for and what tool to use, but I can be a lot lazier with GLM and it can usually figure out my intent, IMO it questions its own work more than Kimi.
yeah agreed, I'm definitely a lot lazier with GLM and ZCode. In some ways, I think us devs using "better" models is really due to pure laziness.
I'd be curious how good the future DS4 Flash gets. That seems like the gold mine. The inference efficiency is unbelievable.
Tang Jie commenting on Zai and its progress toward AGI and self training and such. Quite interesting. Not that different than what you would hear from Dario on this matter imo
@Kalum Pupeter
The thing with the SCMP is that their authors are basically LLMs producing the most hogwash articles ever.
Ok, so China still importing more scientific instruments than any country in the world and at the same time they have export controls. Or either US export controls are useless or Europeans and Domestic companies are taking over. What is? Export controls are winning or not.
The author seems to throw LLMs in article to make look the article "AI related", because a course she does, If that is the case you don't use model directly with the instrument. You usually don't feed it the data because will overwhelm the Context Window. Maybe using agents to write some code and evaluate the results and do a report, maybe. You usually use some kind Neural Networks to analyse vast quantity of data. That alone shows the ignorance of the author.
Pretty much like semiconductor equipment, there is nothing more that domestic companies want than seeing foreign companies voluntary leave the Chinese market for them. But I don't think Germany will happy to widen the trade deficit that they already have with China.
The thing with the SCMP is that their authors are basically LLMs producing the most hogwash articles ever.
Ok, so China still importing more scientific instruments than any country in the world and at the same time they have export controls. Or either US export controls are useless or Europeans and Domestic companies are taking over. What is? Export controls are winning or not.
The author seems to throw LLMs in article to make look the article "AI related", because a course she does, If that is the case you don't use model directly with the instrument. You usually don't feed it the data because will overwhelm the Context Window. Maybe using agents to write some code and evaluate the results and do a report, maybe. You usually use some kind Neural Networks to analyse vast quantity of data. That alone shows the ignorance of the author.
Pretty much like semiconductor equipment, there is nothing more that domestic companies want than seeing foreign companies voluntary leave the Chinese market for them. But I don't think Germany will happy to widen the trade deficit that they already have with China.
lol is he falling for the American AGI pill? lol. I’m afraid he will have to wait for a longer time than 3 years without salary . Maybe never . lol. Gotta get real .
SCMP purposely post these type of articles which contain years old data and associate with random expert who actually expert in this specific field but hardly know about current market and breakthroughs.. there are many libtards in SCMP they love to bash China.
------------------------------------------------------------------------------------------------------------------------------
in Jan, 2025 US put export control on Chinese biotechnologies companies, they cannot buy equipment like high-parameter flow cytometers and certain mass spectrometry.
and guess what. China did more breakthrough and discover more drugs since Jan, 2025 and also did record breaking deals. mass spectrometry is the backbone of Biotech research also used in Ai research.
right now China is one of the leading player in mass spectrometry/chromatographs instruments globally.
Chongqing/Shanghai Drawell Instrument CO,.Ltd
clinical flow cytometers are increasingly used in AI research to automate workflows and improve diagnostic accuracy
look at the number of companies in high end scientific instruments. which are now increasingly used in Ai related research as well