AI and New Materials

Every object around you — the phone in your pocket, the window glass, the cables carrying electricity, the fabric of your clothes — is made of materials that someone had to discover or invent. For most of human history, new materials were found by accident or through exhausting trial-and-error: mix these elements, heat to this temperature, observe what happens. The pace of materials discovery has always been a bottleneck on technological progress. The Bronze Age, the Iron Age, the Silicon Age — each era was defined by a material that unlocked new possibilities. Today AI is opening the possibility of compressing that timeline dramatically, enabling scientists to search spaces of possible materials so vast that no human laboratory could explore them physically.

Why Materials Discovery Is So Hard

The space of possible materials is almost incomprehensibly large. Consider just inorganic crystal structures: there are roughly 94 stable elements in the periodic table, and they can combine in countless ratios and arrangements. The number of conceivable compounds runs into the trillions. Of those, some fraction will be stable enough to actually form a crystal. Of those, a much smaller fraction will have a desired property — high electrical conductivity, unusual magnetism, room-temperature superconductivity. Traditionally, discovering a useful new material required years of laboratory synthesis, characterization, and property measurement. Each candidate had to be physically made and tested. Negative results — materials that did not have the desired property — told you something, but slowly. Computational approaches began helping in the 1990s, using quantum mechanical equations (density functional theory) to predict properties from first principles. But these simulations are computationally expensive: each candidate material might take hours of supercomputer time to evaluate. AI changes this dramatically.

How AI Searches Materials Space

AI approaches to materials discovery work in two broad modes: screening and generation. Screening uses a trained AI model as a fast predictor. You describe a candidate material — its chemical composition, crystal structure — and the AI predicts its properties based on patterns it learned from a database of known materials. Instead of simulating every candidate with expensive quantum mechanics, you run the cheap AI predictor on millions of candidates, flagging the most promising few hundred for detailed follow-up. Generation uses AI to propose entirely new materials. Generative models — similar in concept to the models that produce text or images — can be trained to output valid crystal structures with specified target properties. You tell the model you want a material that is stable, conducts heat poorly, and has a specific electronic band gap, and the model proposes candidate structures you have never seen before. In 2023, Google DeepMind released GNoME (Graph Networks for Materials Exploration), an AI system that predicted 2.2 million new stable crystal structures — roughly ten times more than all previously known stable inorganic materials combined. Of these, 380,000 were deemed promising candidates for synthesis, and some have already been verified experimentally.

Real-World Targets: Batteries, Superconductors, and More

AI-accelerated materials discovery is being aimed at several high-priority targets. Better battery materials are perhaps the most economically significant: a lithium-ion battery with twice the energy density and half the cost would transform electric vehicles, grid storage, and portable electronics. AI is screening thousands of candidate solid electrolytes and anode materials to find safer, more energy-dense options. Room-temperature superconductors are the holy grail of materials science: a material that conducts electricity with zero resistance at everyday temperatures would revolutionize power transmission, magnetic levitation, and computing. Current superconductors require extreme cold. AI is being used to identify candidate structures that might achieve this at more practical temperatures — a problem that has frustrated physicists for decades. Beyond energy, AI materials discovery is targeting catalysts for industrial chemistry (to make ammonia production — essential for fertilizer — far less energy-intensive), structural materials for lighter aircraft and spacecraft, and photovoltaic materials for more efficient solar cells.