C12.ai (合碳智能)
Shanghai startup building a general-purpose dual-arm "robot scientist", starting with small-molecule purification in pharma labs.
1. Core Product / Service
C12.ai (founded 2022) started out selling software: its first product was an AI agent for drug-synthesis route design (retrosynthesis), built on Milvus vector search for finding similar reactions. It has since moved into physical execution. Its robot system, Talos (also shown as LabBot), is a general-purpose mobile robot with two arms, in-house sensors and a proprietary vision-language-action (VLA) model. The company says it runs existing lab instruments with no lab renovation and adjusts chromatography conditions in real time. A "Molecular AI Brain" handles retrosynthesis, property prediction and spectral analysis, which closes a design → execute → analyze loop. The first workflow is purification of small-molecule intermediates. The stated roadmap moves on to ADME assays and then GxP manufacturing and analytical methods, with flexible cosmetics production as a side exploration. The business model is per-workflow pricing, mainly lease-to-own, with rent priced by the number of robot skills. The company claims rent is 1/15–1/20 of the purchase price. No list prices or hardware specs have been published.
2. Target Users & Pain Points
- Medicinal-chemistry teams at pharma companies and CROs. The company says purification takes about half of a medicinal chemist's time and is too varied for fixed, single-purpose automation. Its goal is to have robots take over about 80% of routine purification.
- Labs that cannot rebuild around automation. The pitch is a robot that slots into existing benches and instruments.
- Broader framing: AI can propose molecules in minutes, but physical validation still takes weeks per molecule, so the bottleneck is execution rather than design.
3. Competitive Landscape
| Player | Positioning | Vs. C12.ai |
|---|---|---|
| xtalpi | Listed AI drug discovery company with in-house robotic synthesis and purification | Keeps automation in-house and sells results. A large AI-pharma player is less likely to buy a third-party purification robot |
| megarobo | China's largest custom lab-automation integrator | Builds fixed, project-based systems. C12.ai pitches the opposite: one flexible robot instead of custom lines |
| dp-technology | AI4S platform with the open-source Uni-Lab-OS | Competes for the "lab brain" software layer that C12.ai also claims |
| benyao-tech | Standardized lab-robot units | Productized, capital-light hardware. More mature as hardware, less "autonomous scientist" |
| jindu-biotech / shuye-tech | Other Chinese "physical AI / self-driving lab" startups | Same narrative, generally larger recent rounds |
| Medra (US) | "Physical AI Scientist" for biology labs, ≈US$63M raised | Closest global analog, with more funding |
| Biotage / Teledyne ISCO | Automated flash-chromatography instruments | The incumbent purification hardware Talos operates rather than replaces |
4. Unique Observations
- The headline results come only from the company. The BeOne (百济神州) Guangzhou pilot reported 50 days, 15 real intermediates, 14 purified (≈93%), 200+ steps per molecule and 3,000+ operations with "zero execution errors". The website, the investor release and media reports all trace back to the company. BeOne has not published its own account. Indirect support exists: a BeOne Guangzhou innovation-centre VP attended the Danaher signing. Fifteen molecules is a small sample, and the share of steps needing human intervention and the time per molecule have not been published.
- Danaher's release is the strongest outside validation. It is Danaher's own press release, not the company's, and calls this its first Danaher Beacon project with a Chinese embodied-AI company. The joint lab was announced in front of Cytiva's CEO and Danaher's China president. For a company this small, that gives real channel credibility. No deal value or deliverables were disclosed.
- The pivot shows where the value was. A pure-software retrosynthesis agent did not support a business: the 2023 "digital CRO" tie-up with 睿智医药 has no reported follow-up. The company found traction once it started doing physical execution. This fits a wider pattern in China's AI-for-science market: the money follows whoever closes the validation loop, not whoever designs the molecule.
- Capital is thin and the team is mostly invisible. Only the founder is publicly named, and his background is software and data, not robotics. Two rounds in about 3.5 years, the latest ¥50M, is small next to peers raising ¥100M+. Even the company's location is reported three ways (Shanghai, Beijing, Hangzhou).
5. Financials / Funding
- Revenue / orders: not disclosed. Install base and paying-customer count are also not disclosed.
- Aug 2022, first round (seed/angel): amount not disclosed. Led by Yunqi Partners (云启资本), with 靖亚资本, BioMap (百图生科) and individual investors.
- Mar 2026, new round (pre-Series A): ¥50M. Led by Yonghua Capital (涌铧资本), with 韦豪创芯 and individual investors.
- Valuation: not disclosed.
6. People & Relationships
- Founder & CEO: 陈志刚 (Jerry Chen, Ph.D., CS, University of Michigan).
- Previously SVP and first Chief Digital/Data Officer at WuXi AppTec.
- Founded Tencent's Medical Big Data Lab.
- Chief Architect at Alibaba Health.
- Tech Lead at Twitter.
- No other executives are publicly named.
- Legal entities: 上海合碳医药科技有限责任公司 (Zhangjiang, Shanghai) and 合碳(北京)智能科技有限责任公司. Offices in Shanghai and Palo Alto.
- Partners and logos:
- BeOne Medicines: Guangzhou pilot.
- Danaher: Beacon smart-lab partnership signed 2026-09-19.
- Qilu Pharmaceutical, Risun Group and Chinese Academy of Sciences institutes: logos on the website, scope unknown.
- 睿智医药: 2023 strategic agreement.
- Investors: Yunqi Partners, Yonghua Capital, BioMap, 靖亚资本, 韦豪创芯.
- Peers: xtalpi, megarobo, dp-technology, benyao-tech, jindu-biotech, shuye-tech.