Deep Principle

Deep Principle (深度原理) is a Beijing-based AI-for-materials startup founded in 2024 by MIT PhDs Haojun Jia (CEO) and Chenru Duan (CTO), with a team drawn from MIT, Stanford, Harvard, Tsinghua, and industry backgrounds at Microsoft, Meta, Dow, and BASF . The company builds generative AI and first-principles chemistry tools to accelerate materials discovery, replacing traditional trial-and-error lab work with predictive pipelines .

Its technical signature is a "Diffusion + LLM" architecture: diffusion models like OA-ReactDiff and React-OT (which predicts transition-state structures in 0.4 seconds, down from days of quantum-chemistry computation) have landed as cover papers in *Nature Computational Science*, *Nature Machine Intelligence*, and *JACS* . In 2025 it launched Agent Mira, an autonomous materials-discovery agent that orchestrates molecular design, reaction prediction, and high-throughput experiment scheduling through natural-language commands .

Commercial traction has moved quickly. The company secured over RMB 10 million in commercial orders within its first year , and its funding trajectory reflects that momentum: Linear Capital led the seed round in mid-2024 , Hillhouse joined a seed+ round that October , and by July 2026 the company had closed a cumulative Series A totaling nearly RMB 1 billion, with investors including Alibaba's Gobi fund, Ant Group, Lenovo Capital, Baidu Ventures, and repeated pro-rata commitments from most earlier backers .