Kaiwuji

Kaiwuji is Lu Ziheng’s AI-for-materials startup discussed in “你有一把能够挖出金子的铲子，肯定不会先给别人用”｜对谈开物纪陆子恒：用AI发明新材料. The episode describes it as an early, roughly 10-person company using AI to discover and validate new materials, then push promising results toward scale-up, customer testing, and commercial application.

The company’s strategic claim is that valuable AI Materials Discovery should first be used to own or control material IP rather than sold immediately as a model or software tool. Lu frames this through the “gold shovel” analogy: if the model can find valuable material, the company should use it to dig before lending it out.

Key Claims

• Kaiwuji targets both commercially valuable materials and harder frontier materials that may prove technical capability.
• The company is positioned around original material IP discovery more than ordinary process optimization.
• Its workflow combines model generation, database search, property prediction, senior expert filtering, experiment, kilogram-scale validation, customer trials, and scale-up choices.
• The episode says current costs are dominated by compute and AI talent, while early lab validation is comparatively less expensive than sustained model training.
• The company is closer to a Materials Pipeline Company than a pure SaaS, model API, or tool vendor at this stage.

Connections

• Lu Ziheng — founder and episode guest.
• AI Materials Discovery — technical domain.
• Materials Pipeline Company — business model and commercialization route.
• MatterSim and MatterGen — model examples shaping the company’s technical conviction.
• AI For Science, Frontier Model Scaling, and Domain Expert Alignment — broader themes connected to the source.