How is AI being used to discover new materials that don't exist in nature?

The answer

Google DeepMind's GNoME (Graph Networks for Materials Exploration) system, which initially predicted 2.2 million potential new stable materials in late 2024, saw major real-world validation in 2025 as hundreds of these AI-predicted materials were successfully synthesized in laboratories worldwide. The materials include new types of superconductors, battery components, and solar cell materials that don't exist in nature. Traditional materials discovery is painstakingly slow - testing one new compound at a time through trial and error. AI accelerates this by predicting which combinations of elements will form stable, useful materials before any physical experiments begin. The breakthrough is expected to accelerate development of next-generation batteries, more efficient solar cells, and novel electronic components - potentially compressing decades of materials science into years.