.png){kind=logo}
AI models can screen millions of potential materials, reducing computational searches from decades to hours and helping researchers focus on promising candidates for laboratory testing.
Open databases such as the Materials Project provide large, curated datasets that support machine learning, material property prediction and the discovery of new compounds.
Generative AI and automated laboratories are advancing battery research, although experimental validation, manufacturing feasibility, reliable data and shared testing standards remain essential for commercial development.
Historically, materials discovery has relied on iterative, labor-intensive laboratory experimentation. Researchers systematically synthesize novel chemical formulations, evaluate their empirical properties, record outcomes, and incrementally adjust variables to achieve optimized performance characteristics. This approach gave the world steel, silicon chips, and the lithium-ion battery, yet it moves slowly when the possible materials number in the millions.
Data science offers a faster path. By learning from past results, computer models can point researchers toward the most promising options before any chemical is weighed. Battery research shows this shift most clearly, and the early numbers deserve a close look.
Battery makers want cells that store more energy, charge faster, last longer, and cost less. Each goal depends on the right mix of elements and the right crystal structure. The number of possible combinations is far larger than any lab can test by hand.
Berkeley Lab reports that experimental data exist for fewer than 1% of compounds in open scientific literature, so researchers know very little about most materials. Data science fills that gap by using calculations and learning models to estimate properties that nobody has measured yet. This lets scientists rule out weak options early and spend lab time on the strongest ones.
The Materials Project began in 2011 at Berkeley Lab and now serves more than 650,000 registered users. Its data and tools have been cited more than 32,000 times in peer-reviewed studies, and its library holds more than 200,000 materials and over 577,000 molecules. Users need no programming skills, which has widened the circle of people who can take part.
Director Kristin Persson has said that machine learning needs large amounts of well-curated data to work well. Clean and consistent records save months of preparation, so teams can spend their effort on new models and ideas instead of sorting spreadsheets.
Also Read: 10 Real-World Applications of Data Science in Business, Healthcare, Finance and Retail
Screening is where the speed gain shows most. Microsoft and Pacific Northwest National Laboratory used AI models to study more than 32 million possible inorganic materials, and found over 500,000 stable candidates. Further filtering cut the list to 150 and later to 18 for expert review.
The full computing effort took about 80 hours, and the team estimated that older methods would have needed around 20 years. Models predicted properties such as stability and ionic conductivity, then supercomputers checked the leading choices with more detailed physics. Human scientists still made the final calls on what to make in the lab.
The real test came in the laboratory. PNNL researchers made a solid electrolyte that swaps part of the lithium for sodium, using about 70% less lithium than existing lithium-ion batteries. They then built an all-solid-state battery and tested it at room temperature and at about 80 degrees Celsius.
Some reports point out that conductivity came in lower than first hoped, and the material still needs extensive evaluation before anyone can think about mass production. Even so, moving from a computer-generated list to a working cell within weeks marks a real change in how battery research can be paced.
The figures below show how large the data effort has become and how sharply the search can be narrowed. Each number comes from the Berkeley Lab report or from the Microsoft and PNNL work described above.
Screening checks known ideas, while generative models suggest fresh ones. Google DeepMind trained its GNoME models on Materials Project data to predict crystal stability, and it added nearly 400,000 new compounds to the database. Microsoft built MatterGen, a generative model for inorganic materials design, using the same resource.
These tools do not replace chemists. They hand chemists a longer and better-ordered list of ideas. A researcher can describe the properties needed and receive structures worth testing, which turns the search from a blind walk into a guided one with clear direction.
Computers can now help with the making as well as the choosing. Berkeley Lab runs the A-Lab, where robots guided by AI prepare materials, and Anubhav Jain of the Materials Project has described linking the simulation pipeline to those automated experiments. Europe is working toward a similar goal through the Battery 2030 plus initiative.
It describes an AI-guided Materials Acceleration Platform that uses data from every scale of the battery value chain. The plan also calls for shared testing methods and data standards for battery interfaces, which are the thin layers where many performance problems begin.
Progress is real, yet several problems need steady work before these methods reach everyday use. Researchers point to the following areas.
Experimental records are scarce and uneven, which limits how well models can learn.
A material predicted to be stable may still be hard to make at industrial scale.
Lab results can differ from simulations, as the lower conductivity finding showed.
Teams need shared data formats and testing rules so that results can be combined.
People who understand both chemistry and data science are in short supply.
Also Read: AI vs Data Science vs Cybersecurity: Skills, Jobs and Career Paths
Data science is changing how materials are found, and battery research is leading the way. Open databases give models the clean information they need, screening tools narrow millions of options to a handful, and automated labs turn the best ideas into real samples.
The Microsoft and PNNL result shows what is possible when these parts work together. Success will depend on better data, shared standards, and careful testing of every claim. Researchers who combine chemistry knowledge with strong data skills will shape safer and more affordable batteries.
Data science uses machine learning models to predict material properties and identify promising compounds before laboratory testing. Researchers can screen millions of candidates computationally, reducing unnecessary experiments and directing resources towards materials with greater potential.
AI helps scientists identify materials that could improve energy storage, charging speed, durability and cost. Microsoft and Pacific Northwest National Laboratory screened about 32 million candidates, narrowed the search to 18 for expert review and developed a prototype solid-state battery electrolyte.
The Materials Project is an open scientific resource developed at Berkeley Lab that provides computational materials data and research tools. Its database supports machine learning, property prediction and candidate screening, helping researchers investigate materials without beginning every search from scratch.
AI can predict promising structures and estimate their properties, yet those predictions do not establish that a material will work in a real battery. Researchers must synthesise candidates, measure performance and assess stability, safety and manufacturing feasibility before considering commercial applications.
Recent developments include Microsoft's MatterGen for generative materials design, Google DeepMind's GNoME research on crystal stability and automated laboratory workflows. These approaches help researchers propose new structures and test selected candidates, although independent experimental validation remains necessary before industrial adoption
