Permanent magnets are inside almost every electric motor, wind turbine generator, hard drive, and speaker you own. The most powerful ones — the kind that make electric vehicles practical and offshore wind turbines economically viable — require rare earth elements: neodymium, praseodymium, dysprosium. Over 80% of the global supply of these materials is controlled and processed in China. That supply chain exposure has been a recognized strategic vulnerability for years.
Now, Prashant Singh at Ames National Laboratory has published a framework in Advanced Functional Materials that uses physics-informed artificial intelligence to systematically search for rare-earth-free alternatives — before any candidate material is synthesized in a lab.
What the AI does differently
This is not a pattern-matching system trained on historical data. Singh's approach encodes the actual physics of magnetism — how a material's atomic structure and electronic behavior determine its magnetization strength, energy storage capacity, demagnetization resistance, and thermal stability — directly into the model. This means the AI can evaluate entirely hypothetical compounds it has never seen by reasoning from first principles, rather than interpolating from prior examples. It can search, as Singh put it, "arbitrary material space."
The framework draws on earlier work developing DuctGPT, an agentic AI tool built to predict ductility in extreme-environment alloys for fusion reactors and aerospace systems. DuctGPT demonstrated that embedding physical constraints directly into an AI dramatically expands its useful prediction range — the same principle now applied to the permanent magnet discovery problem. The system also factors in practical constraints like elemental abundance and production costs, not just theoretical magnetic performance.
Why Ames is the right lab for this
Computational power is only as useful as the data behind it. Ames National Laboratory has decades of accumulated experimental magnet data — measurements and synthesis records that took years to generate and that most institutions simply don't have. As Singh noted, Ames's strength "comes from its deep expertise and a long history of data in the magnet space that no other institution has." That dataset gives the AI a grounding that a model trained only on published literature would lack.
The work is part of the DOE Genesis Mission, a Department of Energy program that unites national labs, industry, and academia to apply AI to scientific breakthroughs — with securing America's critical mineral supply listed explicitly as a priority.
How close is an actual replacement?
Honest answer: not close yet. No rare-earth-free magnet matching the performance of neodymium-based magnets has been synthesized from this work. What this research establishes is a faster, smarter path to finding one — a method that can evaluate huge material spaces computationally, flag the most promising candidates, and dramatically reduce the number of failed synthesis attempts researchers need to run. Given that rare earth dependency is now framed as a national security issue rather than just a supply chain inconvenience, the urgency driving this work is real. Whether a viable candidate emerges in three years or fifteen is the open question — but the search is now considerably more intelligent.