DeepPrinciple closes nearly $140M Series A, largest AI4M funding round in China
The AMW Read
Novelty 2: deepens a nascent AI4M player map with a large round but does not resolve any open debate. Significance 2: state-coordinated capital deployment signals strategic prioritization of AI science in China, segment-level impact.
DeepPrinciple closes nearly $140M Series A, largest AI4M funding round in China
DeepPrinciple (深度原理), an AI for Materials (AI4M) startup co-founded by two MIT PhDs, has closed a Series A round totaling nearly 1 billion yuan (~$140M), marking the largest single fundraising in China's domestic AI4M sector. The round was led by Fount Capital (孚腾资本) — a fund backed by Shanghai State Investment, Lingang Group, SAIC Motor, CATL, and Bilibili — and Shunxi Fund (顺禧基金), a Beijing state-owned venture platform. Industrial capital from Pharmaron (康龙化成) affiliate Kangjun Capital also participated, signaling downstream industry confidence in AI-driven materials discovery. DeepPrinciple was incubated and seed-funded by XtalPi (晶泰科技), itself a publicly listed AI drug discovery firm.
Why it matters: DeepPrinciple's Series A — backed by a rare coalition of Beijing and Shanghai state capital alongside strategic industry investors — exemplifies a structural pattern we track closely: state-coordinated capital deployment into vertical AI science. This is not a standard VC round; it is a directed bet that AI for Materials will be a strategic industry, with two Chinese provincial governments effectively co-sponsoring a single startup. The company targets materials discovery for batteries, catalysts, and consumer goods (with existing clients including L'Oréal and industry partners), placing it in the same emerging segment as CuspAI (Bezos-backed, $4B valuation) and Lila Sciences (Flagship-incubated, targeting $8.5B valuation). Unlike those Western peers, DeepPrinciple benefits from China's dense manufacturing supply chain and government willingness to fund early-stage science infrastructure.
Analyst take: DeepPrinciple's core claims — foundation models for property prediction (MPA), crystal generation (SAGA), and reaction path prediction (React-OT accelerating transition-state computation from days to 0.4 seconds) — are technically credible but unproven at scale. The real risk is execution: building a closed-loop 'AI Materials Factory' with L4 autonomy requires hardware integration, lab automation, and data feedback that few AI science firms have achieved. The company's rapid seven financing rounds in under three years suggest capital is being used to compress timelines, but materials science is governed by physical reality, not software iteration cycles. The open question is whether DeepPrinciple can convert its government-enabled capital advantage into defensible IP and revenue milestones before global competitors (DeepMind's GNoME, CuspAI, Lila Sciences) capture the most valuable problems.
