Introduction
The global demand for high-performance magnetic materials continues to rise sharply due to the growth of clean energy technologies, electrified transportation, robotics, medical devices, advanced space systems, and semiconductor industry. Permanent magnets based on rare-earth elements, particularly Neodymium Iron Boron (NdFeB) and Samarium Cobalt (SmCo), currently define the upper limits of magnetic performance thanks to their high coercivity, high residual flux density, and excellent energy product.
Nevertheless, the reliance on rare-earth elements introduces concerns associated with geopolitical supply-chain fragility, environmental impact, and cost volatility. In addition, there is increasing industrial interest in materials capable of achieving ultra-high saturation magnetization, with the ambition of reaching values in the range of 30–50 kG, although such performance remains beyond the capabilities of current commercially viable materials. To address these scientific, economic, and sustainability challenges, Artificial Intelligence (AI) has emerged as a transformative tool capable of accelerating the discovery, optimization, and industrial translation of new magnetic materials.
New Magnetic Materials Development
The adoption of AI for magnetic materials research stems from its ability to explore compositional and structural design spaces far larger than what traditional experimental or intuition-driven approaches can manage. Machine-learning models, informed by vast datasets from experimental measurements and first principles calculations, can learn complex relationships between chemical composition, microstructure, processing conditions, and magnetic properties such as saturation magnetization, coercivity, magnetocrystalline anisotropy, and Curie temperature. AI systems can then rapidly propose novel materials or optimized compositions, drastically shortening development cycles that once required years or decades. This acceleration has been demonstrated by recent advances reported in the industrial and academic sectors.
MagNex Project
A prominent example is the AI-assisted discovery of MagNex, a rare-earth-free permanent magnet developed by Materials Nexus. Using an AI platform capable of analyzing more than one hundred million possible compositions, researchers produced a workable material in roughly three months, reducing the development time by a factor of two hundred relative to traditional methods and achieving a cost reduction of approximately eighty percent along with a significant reduction in carbon emissions during production (Materials Nexus, 2024). News analyses have highlighted the importance of this breakthrough, noting that AI may now enable companies to diversify away from the constrained supply of Neodymium or Dysprosium while maintaining magnet performance at industrially relevant levels.
ARPA-E Project
Parallel efforts are exploring whether AI can accelerate the discovery of alloys with extremely high saturation magnetization. Although reaching the multi-tesla targets desired by some advanced electromagnetic applications remains extremely challenging, AI-accelerated high-throughput computational screening has yielded promising leads. Work funded by the U.S. ARPA-E program and conducted by Chelikowsky, Wang and collaborators (2025) has demonstrated the potential of machine-learning models integrated with ab initio electronic-structure methods to identify stable or low-energy Fe–Co–X ternary compounds with saturation magnetizations approaching Js = 1.4 T while simultaneously maintaining substantial magnetocrystalline anisotropy. The significance of this achievement lies not only in the performance of the predicted candidates but also in the viability of the workflow, which can be scaled to explore increasingly complex multicomponent spaces.
MIT Project
On the other hand, R. Okabe and colleagues provides a concise overview of how AI and Machine Learning (ML) are increasingly being used to accelerate the discovery and design of new magnetic materials. Magnetic materials are essential in a wide range of technologies, yet their traditional discovery process is slow and resource intensive. The authors argue that materials informatics the application of data science to materials research, offers a powerful framework to overcome these limitations. R. Okabe and colleagues present how modern AI and ML techniques can support different stages of magnetic materials research. The paper reviews several ML approaches relevant to magnetism, including random forests, neural networks, and decision-tree methods. To help researchers choose an appropriate technique, the authors propose a flowchart that connects specific problems (such as classification or regression and the amount of available data) to suitable ML methods. They then highlight various magnetic-materials domains where informatics is already proving useful. For example, in topological materials, ML can help predict or classify topological magnetic phases; in Heusler alloys, data-driven screening accelerates the exploration of their vast compositional space; at interfaces, informatics can guide the selection of promising structures in multilayer systems; and for permanent magnets, ML models can predict critical magnetic properties such as anisotropy, saturation magnetization, and Curie temperature. While emphasizing the promise of materials informatics, the authors also discuss current challenges. These include the need for high-quality, standardized datasets for magnetic properties, the importance of model interpretability, and the necessity of integrating computational predictions with experimental validation. They envision a future in which autonomous, closed-loop discovery systems combine ML models with automated synthesis and testing, greatly speeding up the development of next-generation magnetic materials. Overall, the paper serves as a concise roadmap for how AI and ML can support magnetism research and offers practical guidance for researchers aiming to adopt informatics tools in their own work.
UNH Project
Researchers at the University of New Hampshire (UNH) have developed an AI-based approach to significantly accelerate the discovery of new magnetic materials. They created a database of 67,573 magnetic compounds, using machine learning to extract and predict magnetic properties from scientific literature. Among these, they identified 25 previously unrecognized materials that remain magnetic at high temperatures. The goal of this work is to reduce reliance on rare-earth elements, which are expensive, difficult to obtain, and crucial in many current magnetic technologies.
Their AI system works by reading scientific papers and extracting experimental data, which is then used to train models that predict if a material is magnetic and estimate its Curie temperature (i.e., the temperature at which it loses its magnetism). The results are organized in a searchable database called the Northeast Materials Database, enabling other researchers to more easily explore promising magnetic materials. The researchers believe that this tool could help discover sustainable and high-performance magnets that don’t rely on rare earths, which would lower costs and support green technologies.
They also suggest that their modeling strategy, especially the Large Language Model (LLM) behind their data extraction, could have broader uses, such as digitizing older scientific literature (e.g., converting images from old papers into modern, machine-readable text).
Sandia National Laboratories Project
Solid-state transformers are emerging as a promising technology for strengthening the resilience of the electrical grid, especially in DC grid systems (de Leon, Francisco paper). Compared with today’s large, foreign-manufactured grid transformers, solid-state transformers can be replaced more rapidly and offer greater resistance to physical and cyber threats. However, to fully exploit their advantages, particularly the high switching speeds enabled by ultrawide-bandgap power electronics, new magnetic materials are needed. Current inductor materials suffer from excessive hysteresis losses at high frequencies, limiting performance.
The Sandia National Lab project described in this document aims to overcome this materials bottleneck by using AI to accelerate the discovery, synthesis, and testing of high-frequency ferrites suitable for next-generation power systems. Instead of relying on the traditional trial-and-error approach, which can take decades to yield new magnetic materials, the team seeks to compress the entire discovery-to-application cycle into less than three years. The approach integrates AI at every stage. First, candidate magnetic materials are identified using AI models trained on density functional theory (DFT) calculations. Although DFT provides highly accurate predictions of magnetic properties, it is too slow for broad exploration of unknown ferrite compositions. By training AI on a smaller set of DFT data, researchers can rapidly screen large numbers of potential materials and then validate promising candidates with full DFT simulations.
The project also uses AI to guide the synthesis of these materials, which can be as challenging as predicting their existence. The team simulates the expected X-ray diffraction (XRD) pattern for each target ferrite and then collects real-time XRD data during synthesis. AI compares the evolving pattern with the desired one and dynamically adjusts reaction parameters, such as time, temperature, and precursor concentrations, to steer the system toward the correct crystal phase. Over time, the system learns from experience, improving its ability to produce complex, previously unknown ferrite phases.
All materials are synthesized as nanoparticles to enable rapid reactions and continuous XRD monitoring. Once successfully produced, these nanoparticles are consolidated through warm compaction into toroidal inductors. After winding the inductors with wire, the team tests them directly in relevant power-electronics circuits to evaluate performance in realistic operating environments.
This AI-driven methodology is expected to produce magnetic materials with improved high-frequency behavior within the duration of the project, a timeline far shorter than normal materials-development cycles. The resulting materials will support the advancement of new electrical architectures for a more resilient grid, including solid-state transformers. Beyond this specific application, the project represents a new generalizable paradigm for materials research, one that replaces intuition-driven experimentation with data-guided discovery and adaptive synthesis. Its broader impacts could extend to defense, fusion, non-proliferation, sensing, and other technology areas requiring advanced functional materials.
Conclusions
In practice, AI accelerates magnetic materials development through integrated workflows that combine data aggregation, machine learning, physics-based modeling, optimization algorithms, and experimental feedback. Researchers typically assemble datasets containing both measured and computed values for magnetic properties, formation energies, crystal structures, and thermodynamic stability. ML regression models such as gradient-boosting ensembles, neural networks, and Gaussian-process surrogates are then trained to predict continuous magnetic properties, while classification models can screen for phase stability or synthesizability. Optimization methods, including particle swarm optimization, genetic algorithms, and Bayesian optimization, guide the exploration of compositional and structural spaces by balancing competing objectives such as high saturation magnetization, cost, thermal stability, environmental impact, and manufacturability. Physics-informed modeling, typically through density functional theory, supplies precise electronic and magnetic property calculations while improving the reliability of ML predictions. Closed-loop experimental feedback systems, sometimes involving robotic synthesis platforms, validate AI-generated predictions and provide new data that enhances model accuracy over successive iterations. These workflows have begun to appear in autonomous materials-discovery platforms for magnetic intermetallic and L1₀-ordered alloys (Iwasaki et al., 2024), demonstrating that AI can increasingly operate at the interface of computation and laboratory experimentation.
To achieve industrial impact, these workflows require collaboration across a broad range of disciplines. Materials Scientists and Metallurgists design and interpret alloy systems; crystallographers characterize structures, defects, and phase transformations; computational physicists conduct first principles simulations; data scientists develop and refine prediction models; manufacturing and process engineers translate laboratory compositions into scalable production routes; and sustainability experts evaluate the economic and environmental implications of proposed materials. Such interdisciplinary cooperation ensures that AI-generated discoveries are not merely theoretical curiosities but realistic, manufacturable, and industrially relevant products.
The impact of AI on the magnetic and electromagnetic industries is expected to be substantial. By reducing the dependence on critical rare-earth elements, AI-designed materials can stabilize supply chains and decrease production costs. By accelerating research cycles, AI reduces time-to-market and allows companies to more rapidly iterate on magnet grades tailored to specific applications such as high-temperature motors, miniaturized actuators, or heavy-duty alternators. By incorporating environmental and economic considerations directly into the materials-design process, AI enables the development of magnets with dramatically lower carbon footprints, such as those demonstrated by the MagNex project. More broadly, AI-assisted materials discovery may facilitate future breakthroughs in high-saturation alloys capable of supporting next-generation electromagnetic devices with unprecedented power density and efficiency.
Although the potential of AI-enabled magnetic-materials discovery is remarkable, several challenges remain. Machine-learning predictions depend on the quality and diversity of the input data and can struggle to extrapolate reliably into unexplored chemical spaces. Synthesizing AI-predicted materials may reveal unforeseen kinetic barriers or microstructural complications that were not captured in computational models. Scaling laboratory discoveries to industrial production requires careful control of purity, reproducibility, and cost. Despite these risks, the rapid progress observed in the past several years demonstrates that the synergistic combination of AI, materials science, and advanced characterization is already reshaping the landscape of magnetic materials development.
Finally, AI has become a powerful accelerator for discovering and optimizing magnetic materials, from rare-earth-free permanent magnets to improved NdFeB and SmCo compositions and emerging high-saturation alloys. Its ability to integrate machine learning, ab initio modeling, experimental feedback, and sustainability assessment offers unprecedented opportunities for industrial innovation. As interdisciplinary research teams continue to refine AI-driven methodologies and expand databases, algorithms, and autonomous platforms, the magnetic materials industry is likely to benefit from faster development cycles, reduced costs, improved performance, and greater resilience in global supply chains. This convergence of AI and materials science represents a critical step toward a new era of magnetic technologies that better meet the technological, environmental, and economic challenges of the coming decades.
References
[1] Chelikowsky, J. R., Wang, C.-Z., et al. (2025). AI and quantum simulations for the discovery of rare-earth-free magnets. ARPA-E Publication.
[2] Materials Nexus. (2024). Materials Nexus discovers new rare-earth-free magnet using AI algorithm. MD ONE News.
[3] BGR Science. (2025). AI Developed A New Rare-Earth-Free Magnet 200 Times Faster Than Humans. BGR Science Report.
[4] TechXplore. (2025). Magnetic materials discovered by AI could reduce rare earth dependence. TechXplore Technology News.
[5] R. Okabe et al., "Materials Informatics for the Development and Discovery of Future Magnetic Materials," IEEE Magnetics Letters, vol. 14, pp. 1-5, 2023.
[6] Iwasaki, Y., Ogawa, D., Kotsugi, M., & Takahashi, Y. K. (2024). Autonomous materials search using machine learning and ab initio calculations for L1₀-FePt-based quaternary alloys.
[7] UNH researchers harness AI to discover magnetic materials | EurekAlert!
[8] Itani, S., Zhang, Y. & Zang, J. The northeast materials database for magnetic materials. Nat Commun 16, 9415 (2025)
[10] F. de León, "The future belongs to dc: Edison will beat Tesla after all," in IEEE Power and Energy Magazine, vol. 21, no. 2, pp. 78-80, March-April 2023.