On November 29, 2023, DeepMind announced GNoME (Graph Networks for Materials Exploration), a deep-learning tool that predicts the stability of new materials, alongside a paper in Nature. GNoME discovered 2.2 million new crystal structures, of which about 380,000 were identified as the most stable and most promising for experimental synthesis.
DeepMind described the haul as “equivalent to nearly 800 years’ worth of knowledge” of stable materials. The candidates include materials relevant to future technologies such as superconductors, supercomputers and next-generation batteries - for example 528 potential lithium-ion conductors, which DeepMind reports as 25 times more than a previous study had found.
GNoME works by searching enormous spaces of candidate structures and using learned graph networks to predict which ones will be stable, focusing slow and expensive lab work on the most promising leads. DeepMind reports the approach raised the discovery hit rate from around 50 percent to 80 percent, and that external labs had already created 736 of the new structures experimentally.
GNoME extends the AI-for-science pattern from biology (AlphaFold) and weather (GraphCast) into materials science: a learned model accelerates the search through a combinatorial space that brute-force experiment could never cover. The Nature page often redirects or blocks automated access; the figures here are corroborated by DeepMind’s own announcement.