Alibaba (09988.HK) subsidiary DAMO Academy has achieved a major breakthrough in materials science. On July 3, DAMO Academy announced that, in collaboration with Renmin University of China and the University of Chinese Academy of Sciences, it successfully developed “ElementsClaw,” the industry’s first AI agent focused on discovering superconducting materials. The system not only autonomously predicted 68,000 potential superconducting materials but also synthesized and experimentally verified four entirely new ones, confirming the immense potential of artificial intelligence in cutting-edge materials exploration.
Superconductors are viewed as highly promising new materials. When cooled below a critical temperature, their electrical resistance abruptly vanishes, and they completely expel external magnetic fields, holding significant application value in areas such as lossless power transmission and magnetic levitation. However, for nearly a century, because the underlying physical mechanisms remain incompletely understood, researchers have largely relied on “Edisonian trial-and-error” to discover elemental combinations that yield superconducting materials—a process that is not only time-consuming and labor-intensive but also has an extremely low success rate.
Currently, the mainstream international superconductivity database, SuperCon, contains only about 2,000 materials accumulated over decades. To overcome this bottleneck, the DAMO Academy team leveraged its technical strengths in AI for Science to develop ElementsClaw. This system can not only predict whether a material is superconductive but also, like a human materials scientist, review literature, assess synthesis feasibility, and design experimental protocols, dramatically improving the efficiency and success rate of material discovery.
Market sources indicate that ElementsClaw employs a “specialized-generalist fusion” architecture. At the specialized model level, the team pre-trained a 1-billion-parameter atomic foundation model called “Elements” based on a database containing 125 million molecular and crystal structures. This model achieves an AUC of 0.996 in determining whether a material is superconductive, and its average error in predicting superconducting critical temperature is controlled within 1K. At the generalist agent framework level, ElementsClaw implements an entire automated material screening pipeline, including tool creation, workflow orchestration, and literature verification, and can even “self-evolve” after mining new clues from the literature.
In practical operation, this AI system screened 2.4 million crystal structures in just 28 GPU hours, predicting 68,000 superconducting candidates from them. The research team has selected and experimentally synthesized four entirely new materials, confirming their superconductivity, with the highest critical temperature reaching 6.5K. These four materials are: Hf21Re25, a “missed discovery” retrieved from existing databases; Zr4VRe7, successfully “vindicated” after correcting a configuration error in the database; HfZrRe4, designed from scratch by the AI; and Zr3ScRe8, derived by drawing inferences from similar structures.
Rong Yu, Head of Scientific Intelligence at DAMO Academy, stated: “These are the first batch of superconducting materials discovered and verified by an AI agent, preliminarily validating the potential of the AI agent framework in the field of material discovery.” With a vast number of candidate materials still awaiting exploration, DAMO Academy has decided to open-source the entire database of 2.4 million stable crystals predicted by ElementsClaw, making it freely available to academia and researchers to further mine potential new materials.
Huang Wenbing, Associate Professor at Renmin University of China’s Gaoling School of Artificial Intelligence, noted that the application prospects of this AI agent extend beyond superconducting materials. In the future, it is expected to expand into the discovery of new materials such as solid-state battery electrolytes, multiphase catalysts, and thermoelectric materials, injecting powerful momentum into the Materials Genome Initiative.
This case of AI autonomously discovering and experimentally verifying superconducting materials not only demonstrates artificial intelligence’s role as an accelerator in fundamental scientific research but also opens up greater imaginative space for technological innovation in future industries such as energy and electronic components.