Alibaba Group Holding's Damo Academy has deployed an AI agent that identified four previously unknown superconducting materials, all later verified through laboratory experiments. The system screened 2.4 million crystal structures in 28 hours - a task that would take researchers years using conventional trial-and-error methods, since scientists still lack a complete theoretical framework to predict superconductivity.
How Elements Claw narrowed the field
The AI agent, called Elements Claw, was developed in collaboration with Renmin University of China and the University of Chinese Academy of Sciences. It scans scientific literature and evaluates millions of crystal structures to propose candidate materials for lab validation. According to Damo, the tool is the industry's first AI agent designed specifically for discovering superconducting materials.
Powered by a one-billion-parameter foundation model trained on 125 million molecular and crystal structures, Elements Claw processed 2.4 million stable crystal structures in 28 hours of graphics processor computing time. It surfaced roughly 68,000 candidates with superconducting potential, then narrowed those down to the most promising options for physical testing.
Why superconductor discovery is slow
Superconducting materials conduct electricity without resistance and expel magnetic fields when cooled to low temperatures - properties that could reshape power grids, quantum computing, and high-speed maglev trains. But discovering new superconductors has long depended on laborious experiments because no theoretical model can reliably predict which materials will exhibit the phenomenon.
Over decades of work, researchers have catalogued only about 2,000 known superconducting materials in the widely used SuperCon database. Elements Claw aims to accelerate that timeline dramatically by doing the computational screening up front, so laboratory researchers can focus on the most likely candidates.
What the verification step confirms
The four compounds Elements Claw proposed were tested and confirmed in physical experiments, Damo said. This validation loop - where AI predictions are checked against real-world measurements - is a concrete advance in AI for Science & Research. It suggests that trained models can surface candidates that human-led searches might miss or take far longer to find.
The system's approach combines large-scale literature mining with structural screening, reducing the search space from millions of possibilities to a focused set of testable materials. The 28-hour compute window makes it practical to run such screens repeatedly as new data and research become available.
Why this matters for science and research professionals
Elements Claw demonstrates that domain-specific AI agents can handle the discovery-phase workload that currently consumes years of researcher time. For materials scientists and condensed-matter physicists, tools like this shift the bottleneck from finding candidates to testing them - a change that could increase the pace of superconductor discovery without requiring new theoretical breakthroughs. The four verified compounds also represent real additions to a field that gains only a few dozen new superconducting materials in a productive year.
Your membership also unlocks:
