Qingxi Zhiyan Raises Nearly RMB 100M Angel Round for AI-Driven Materials R&D
Qingxi Zhiyan, a Chinese AI-for-science startup, has raised an angel round of nearly RMB 100 million (about $14 million). The company announced it on November 9, roughly four months after a first round in the tens of millions of yuan.
The investor list is unusually heavy for an angel round. It includes SMIC's investment arm, Hillhouse's venture unit and the humanoid robot maker AgiBot. That mix says a lot about where the company is headed: chips, robots and the materials both depend on.
From lab instrument to "super factory"
Plenty of AI-for-science startups train models on public datasets and publish strong benchmark scores. Qingxi Zhiyan is betting that the edge lies in the lab itself. It starts with scientific instruments and process equipment, adds AI so they can measure and adjust on their own, and turns every reading and process setting into training data.
The roadmap has three steps. First, single devices that run themselves.
Next, AI-native labs where one machine handles several kinds of measurement and AI checks the results against each other. Last, what the company calls an "AI SuperFactory": a platform that runs around the clock and plans, schedules and repeats experiments until a material hits its performance and cost targets.
CEO Dong Siming studied at Tsinghua and earned a materials science PhD at Stanford. He spent more than 15 years in semiconductor and industrial R&D, including running an AI business at a leading US semiconductor company. Co-founder Zhao Zhenlin has worked on battery integration at Tesla and on magnet production lines in China. Chief scientist Wang Huan is a Tsinghua PhD with more than 100 papers.
Strategic money, not just financial
The round brought in SMIC-affiliated fund Zhongxin Juyuan, Hillhouse Ventures, AgiBot, Fengwu Capital, Oriental Jiafu and several industry-backed funds. No lead was named. The first round, in July, was led by Wuyue Feng Technology.
Each backer brings a potential customer as well as a cheque. A chipmaker's fund points to semiconductor materials such as photoresists and memory compounds, both on Qingxi Zhiyan's list. A robotics company points to something more concrete: the magnets inside robot joints.
Five months instead of five years
The company's flagship result is a new-generation rare-earth permanent magnet. It says AI cut development time from the usual three to five years down to five months. It also reports that, compared with the previous generation, the magnet costs about 60% less and handles heat about 30% better. The magnet is moving into pilot production, with mass production targeted for 2027.
The company is working with leading robot makers on magnets for hip, knee and shoulder motors and for dexterous hands. Further out are high-entropy alloys for fusion and high-end equipment, photoresists, and new memory materials.
The timing works for Qingxi Zhiyan. Humanoid robots, electric vehicles and wind turbines all need more and better magnets, and the article cites global demand of about 170,000 tonnes by 2030. As the company puts it in its announcement: "The first half of AI4S competed on model metrics; the second half competes on delivery."
What still needs proving
For now, almost everything comes from the company's own account. The cost and heat figures are from small-scale tests, and no third party has published them. Signed orders total tens of millions of yuan in the first six months, including a full-stack R&D contract from a leading smart-device maker. Revenue isn't disclosed.
Scaling up is the classic trap in materials. Plenty of compounds look excellent in a lab and then struggle to reach consistent quality and yield at production volume. The 2027 target for magnet mass production is the date to hold the company to.
There's also a data question. Qingxi Zhiyan says it holds tens of millions of real experimental data points and expects close to 100 million within a year. That data is its moat only if it stays proprietary and keeps growing faster than rivals can collect their own. Its XRD model topped a Chinese Academy of Sciences materials benchmark, but leads on benchmarks don't tend to last.
What to watch
Next the company plans to iterate its models, sign more customers and open a pilot base in Beijing. It also plans to release a characterization model covering more than 20 methods, aimed at cutting labs' reliance on scarce human experts.
The milestone that matters most is the magnet. If it moves from pilot to mass production in 2027 with the promised cost and heat gains, Qingxi Zhiyan will have shown that AI can deliver a material, not just a prediction. That would be a rare outcome in AI for science, and well worth RMB 100 million at angel stage.
Sources
USD figures are approximate, at about RMB 7.1 per dollar.