Alibaba's Damo Academy unveiled Elements Claw, a 1B-parameter AI agent that discovered 4 new superconductors. The tool screened 2.4 million crystal structures in 28 GPU hours, a task that would take humans years.
Key facts
- 4 new superconductors discovered by AI agent.
- 1B-parameter model trained on 125M molecular structures.
- 2.4M crystal structures screened in 28 GPU hours.
- 68,000 candidates identified for physical testing.
- Only 2,000 known superconductors in SuperCon database.
Alibaba Group Holding's Damo Academy has unveiled what it calls the industry's first AI agent for discovering superconducting materials, claiming the tool has already found four previously unknown compounds verified in lab experiments According to SCMP.
Superconductors conduct electricity without resistance when cooled, a property that could revolutionize power grids and computing. But discovery has been painfully slow: researchers have only cataloged about 2,000 superconducting materials in the SuperCon database over decades, relying on trial-and-error experiments because no complete theoretical framework exists.
How Elements Claw Works
The system is powered by a specialized 1-billion-parameter foundation model trained on 125 million molecular and crystal structures. In 28 hours of GPU computing time, it screened 2.4 million stable crystal structures, identifying roughly 68,000 candidates with superconducting potential. Those were then narrowed down to the most promising options for physical testing, yielding four verified new superconductors.
The agent was developed in collaboration with Renmin University of China and the University of Chinese Academy of Sciences, according to the report.
Why This Matters
This is a structural shift in materials discovery. Traditional methods are labor-intensive and slow—finding 4 new superconductors through conventional approaches could take a decade or more. Elements Claw compressed that into a weekend of compute. The 1B-parameter model is relatively small by modern LLM standards (Claude Opus 4.6 is orders of magnitude larger), but its specialized training on molecular structures makes it highly efficient for this domain.
It also highlights a growing trend: AI agents moving beyond code generation and into scientific discovery. While Anthropic's Claude Code and OpenAI's coding agents compete for developer workflows, Damo Academy is applying the same agent paradigm to materials science—a domain with massive industrial implications.
What to watch
Watch for Damo Academy's next target—likely room-temperature superconductors or battery materials—and whether the agent's hit rate improves as it incorporates feedback from lab verification. Also track Chinese AI labs' expansion into scientific discovery agents.
Source: scmp.com
