AI4Materials Kaiwu Ji secures 1 billion yuan in angel round funding, led by Monolith

How Does the Team’s Cross-Industry Background Help Material Large Model Industrialization?

Written by｜Xiao Jing

Edited by｜Xu Qingyang

On March 24, AI materials science company Kaiwuji announced the completion of several hundred million yuan in financing in its angel + round, led by Monolith, with follow-on investments from Guanghe Venture Capital and Jifu Asia, along with all existing shareholders including Hillhouse Capital, IDG, BlueRun Ventures, BV Baidu Ventures, and L2F Light Source Entrepreneur Fund. Light Source Capital served as the exclusive financial advisor. The funds will be used for building capabilities in material large models, industrialization of self-developed material pipelines, and team expansion.

This is the second round of financing since Kaiwuji’s establishment. All existing shareholders have significantly increased their investments, while the introduction of new lead investor Monolith reflects, to some extent, the capital market’s ongoing interest in AI-driven material research and development.

A Team Emerging from Microsoft Research

Kaiwuji was founded by Dr. Lu Ziheng. Dr. Lu was previously a principal researcher at Microsoft Research and the head of the materials team at the Science and Intelligence Center, focusing on large-scale deep learning and its application in material design, with over a decade of laboratory and industrial experience in energy materials. Co-founder Dr. Yang Mengyang also hails from Microsoft, where he served as a senior research manager, possessing a research background that spans optics, electronics, and materials. He led Microsoft’s storage and materials research projects from proof of concept to system-level deployment.

Industry CTO Dr. Ren Yu comes from the industrial side, with over 20 years of engineering and mass production experience, previously responsible for strategic product selection and the scaling of technology systems at leading materials companies like BASF. The core team also includes researchers and engineers from Microsoft Research, Google DeepMind, BASF, Cambridge University, Imperial College, Columbia University, and Tsinghua University, covering aspects such as model pre-training, Agentic AI, laboratory R&D, and industrial transformation.

Weak Priors and Strong Scaling Technical Route

In terms of technology, Kaiwuji has chosen a different path from the common industry approach of “expert experience + small models”: scaled training under weak prior constraints. The company has built a high-concurrency synthetic data and high-throughput physical experimental data environment, allowing the model to learn more generalized material and chemical space representations through large-scale pre-training, thereby reducing reliance on human chemical prior knowledge. The team’s related research has validated the effectiveness of Scaling Law in material large models, with results to be published in Nature in 2025.

Specifically, Kaiwuji has constructed a dual-engine architecture of “Prophet Prediction Engine + Creator Generation Engine.” Prophet is responsible for high-precision, broad-spectrum property prediction and screening, while Creator focuses on material generation and reverse design for cross-element systems. The company is also building a data infrastructure capable of handling millions of concurrent data, and the laboratory, automated high-throughput platform, and kilogram-level validation platform have already begun construction.

According to the company, the team has made phased progress in several directions: in the solid-state electrolyte area, model-assisted R&D has significantly compressed the R&D cycle; in thermal management materials, the team has systematically explored thermal conductivity distribution among over 640,000 inorganic crystal structures, with some key materials already validated by third parties; in recyclable PCB substrate materials, they have achieved a complete process from material synthesis to device-level products that meet real working conditions.

Currently, Kaiwuji’s exploration focuses on areas such as new energy batteries, cold storage, embodied thermal management materials, and superconducting materials. The company is advancing its self-developed material pipeline while also collaborating with industrial partners on joint research.

AI-driven material R&D is still in the early stages of industrialization. From model prediction to experimental validation, and then to ton-scale mass production, each step presents significant engineering and cost challenges.

Founder Lu Ziheng stated: “Kaiwuji seeks to build not just faster R&D tools, but a reusable, scalable intelligent infrastructure for materials, allowing the generation of material IP to gradually shift from reliance on individual experience and low-frequency serendipity to being predictable and scalable.”

Lead investor Monolith commented that Kaiwuji “has been clear from the start about how large models should be used and implemented in the materials field,” and the team’s pragmatism and long-term perspective are key factors in their decision-making.

For Kaiwuji, the next stage’s core proposition is: Can it successfully navigate the complete path from model design to mass production delivery across multiple self-developed pipelines, thereby proving the commercial viability of the full-stack model in the materials field?