AI news

Qingxi Zhiyan raises nearly RMB 100 million in angel funding for AI materials R&D and pilot production

By Z Potentials1 viewsSource: Z Potentials
Qingxi Zhiyan AI materials research, experimental data and rare-earth magnet pilot production

Z Potentials reports that Qingxi Zhiyan is gathering real experimental data through equipment upgrades and advancing rare-earth permanent magnets toward pilot production, targeting mass production in 2027.

Editorial correction (November 13, 2026): The changes in material performance and costs are company-disclosed comparisons from small-scale laboratory testing. They do not yet represent stable performance in pilot or mass production.

Article illustration

Z Potentials recently learned exclusively that AI for Science startup Qingxi Zhiyan has completed an angel round of nearly RMB 100 million. Investors include Zhongxin Juyuan, GL Ventures, AgiBot, Fengwu Capital, Dongfang Jiafu and several industry-backed funds, spanning semiconductors, advanced manufacturing, industrial investment and international technology-innovation ecosystems. Duowei Capital serves as its long-term exclusive financial adviser. The industry's attitude toward AI-powered materials R&D has moved from watching the technology to applying it in specific scenarios.

While most AI4S companies are still telling stories about models and leaderboards, Qingxi Zhiyan has made substantive progress in AI-powered materials R&D and delivery. According to reports, within half a year of founding, it has signed orders totaling tens of millions of yuan. Customers include industrial companies and research institutions in advanced manufacturing, semiconductors and new materials, and it has signed a full-stack commissioned materials-R&D order with a leading smart-device customer. The company will also soon release the world's first expert-level intelligent characterization-analysis model covering all modalities, further productizing specialized model capabilities.

AI-native instruments and equipment are a key entry point for Qingxi Zhiyan to enable new materials to emerge.

The company began by upgrading scientific instruments and process equipment with AI, embedding specialized models into measurement, analysis and control workflows, and gradually expanding into autonomous laboratories and all-in-one AI-native platforms. This route lets models participate in real experiments and opens equipment-side data interfaces, providing real data for continued model iteration. According to the company, it has accumulated tens of millions of real experimental multimodal data records, supporting multiple in-house models in achieving state-of-the-art (SOTA) results on recognized international benchmarks.

Rarer still is its rapid ability to bring materials into industrial production. Its independently developed next-generation rare-earth permanent magnets have advanced into pilot-scale development, with mass production targeted for 2027. Meanwhile, the company is jointly developing high-performance permanent magnets with leading robotics manufacturers for applications such as hip, knee and shoulder joints and dexterous hands, while co-building a rare-earth-materials data foundation, focusing on breakthroughs in key metrics such as high torque density, light weight and high-temperature stability.

Asked about the industry's core competitiveness, Qingxi Zhiyan's judgment is: the long-term moat in model capability depends on continuous access to high-quality real scientific data, while actual orders are the most direct litmus test of model capability.

01 | A Tsinghua–Stanford dream team aiming to be the "industry practitioners" among teams of scientists

The almost "dreamlike" industrial results above came from a team that has actually worked on materials and production lines.

Founder Siming Dong holds a bachelor's degree from Tsinghua University and a materials-science PhD from Stanford University, with more than 15 years of semiconductor and industrial R&D experience. He led a major US semiconductor company's first AI-automated process-development project, which later became common industry technology. As head of its AI business, he managed a business spanning advanced processes, intelligent equipment and scientific simulation, with annual revenue of nearly $200 million, and brought multiple AI-enabled equipment projects into leading international wafer fabs. His experience combines underlying process R&D, intelligent-equipment innovation, global business management and industrial deployment.

Co-founder and COO Zhenlin Zhao also has Tsinghua and Stanford backgrounds. He was an applications scientist at a US plasma-technology company and was deeply involved in developing Tesla's battery–body integration process. In China, he led R&D and production-line construction for PVD heavy-rare-earth grain-boundary-diffusion magnetic-material technology. His experience combines frontier materials innovation, industrial engineering and global commercialization.

Co-founder and chief scientist Dr. Huan Wang graduated from Tsinghua University and has long researched multimodal large models and intelligent approaches to complex scientific systems. He studied under several academicians in China and abroad, has appeared on the global top 2% scientists list for three consecutive years, and ranks in the top 0.35% in AI. He has published more than 100 high-level papers, holds more than 30 patents and has received over 5,000 academic citations.

AI, materials and equipment capabilities meet in one founding team. From its inception, this has given Qingxi Zhiyan a foundation for moving directly from model development to industrial delivery, and a view of the "ultimate path" different from teams with purely algorithmic backgrounds.

02 | A less-traveled route: AI-native instruments and equipment build a real-data moat

Asked about AI4S's core competitiveness, Qingxi Zhiyan's judgment is that high-quality data determines a model's ceiling, particularly experimental data from the real physical world.

The company therefore starts with scientific instruments and process equipment. Developing and retrofitting AI-native instruments opens equipment data interfaces, turning measurements, process parameters and experimental records into AI-Ready data usable for model training. It has now accumulated tens of millions of real experimental data records.

This data foundation also supports iteration in model capabilities. Qingxi Zhiyan has expanded from multimodal characterization to materials generation, building a model portfolio covering mainstream characterization methods including XRD, SEM/TEM, Raman, XPS and nuclear magnetic resonance, with multiple SOTA results. For example, its in-house XRD interpretation model topped MatSciBench from the Chinese Academy of Sciences' Institute of Physics, surpassing the Nature benchmark model.

The company also plans to release soon the world's first expert-level intelligent characterization-analysis model covering all modalities, spanning more than 20 mainstream characterization and analysis methods and substantially reducing the dependence of complex materials analysis on manual operations and expert experience.

Instruments and equipment are both data entry points and the means through which models act in R&D. AI-native instruments and equipment thus form one of Qingxi Zhiyan's most differentiated AI4S paths, with three specific stages.

• Stage one: upgrade individual devices with AI for autonomous measurement. Embed specialized models into existing equipment and connect region selection, parameter adjustment, acquisition and analysis, allowing instruments to execute autonomously toward a goal and adjust measurements based on results.

• Stage two: an AI-native laboratory. Building on individual-device autonomy, integrate multiple measurement capabilities at the hardware-architecture level to obtain material composition, structure and morphology together. AI cross-validates the information, letting one device answer materials questions that previously required several instruments.

• Stage three: an all-in-one AI-native platform. An AI platform connects design, preparation, characterization and performance testing. Working around performance goals and cost constraints, AI autonomously plans experiments, schedules equipment and iterates solutions, ultimately forming an AI factory capable of continuous 24/7 operation—an “AI materials superfactory (AI SuperFactory)”

03 | Selling shovels and mining gold: the first materials IP enters pilot-scale development, targeting humanoid robots and new-energy vehicles

Commercially, Qingxi Zhiyan advances on three fronts: standardized AI-model subscriptions and AI-equipment services, commissioned materials R&D under a CRO model, and in-house materials-IP capabilities targeting strategic emerging materials. Less than half a year after founding, it has signed orders totaling tens of millions of yuan and jointly launched an intelligent advanced-memory-materials R&D platform with Peking University's School of Integrated Circuits. Customers and partners include leading industrial institutions in consumer electronics, energy, semiconductors and aerospace.

Its own materials IP offers still greater industrial possibilities. Qingxi Zhiyan has applied its AI R&D system to developing next-generation rare-earth permanent magnets for markets including new-energy-vehicle drive motors, wind power and servo motors. Global demand is projected to reach 170,000 tonnes in 2030, corresponding to a market exceeding RMB 50 billion.

According to company disclosures, some metrics improved in its small-scale laboratory tests under the selected comparison conditions, including a reported cost reduction of about 60% and improvement in heat resistance of about 30%. The specific baselines, measurement conditions and reproducibility in mass production still require further verification.

According to the account, the company has partnered with several leading robotics manufacturers around the need for "small size, high torque and high-temperature resistance" in robot joint motors. They are jointly developing high-performance permanent magnets for high-load joint motors in hips, knees, shoulders and dexterous hands, focusing on magnetic properties and high-temperature stability.

Beyond rare-earth permanent magnets, the company is also incubating R&D pipelines for high-entropy alloys for nuclear fusion and high-end equipment, and key semiconductor materials such as photoresists and new memory materials, building an AI materials-R&D system that can be replicated at scale.

04 | Conclusion

AI4S's first half competes on model metrics; its second half competes on delivery and deployment. Qingxi Zhiyan focuses on industrial R&D scenarios and is rapidly moving to the deployment frontier. After this financing, it will continue model iteration and expansion among industrial customers.

Beginning with the end in mind, Qingxi Zhiyan is currently concentrating on two things.

First, collect real physical-world data at the lowest marginal cost and fastest speed. Scientific instruments and process equipment are the most important entry points for scientific data generated by the physical world. The company has accumulated tens of millions of real experimental data records and expects to approach 100 million within a year.

Second, train generalization capabilities in selected materials fields. Magnetic materials are Qingxi Zhiyan's first chosen "prototyping" field. It is building the entire chain from AI design to pilot-scale development itself, first making "AI delivering materials" a reality, then replicating the approach across more materials systems.

AI4S cannot stop at giving scientists stronger tools. It should make materials R&D itself a continuously evolving production capability that can be replicated at scale. Whoever solves real industrial problems can win AI4S's second half.

This article comes from the "Z Potentials" WeChat account. Author: "Z Potentials."

Related topics

After reading

Join the AINRK community

Read how to join the community and connect with people working in AI.

Community information (opens in a new tab)