Key Takeaways AI models, particularly graph neural networks (GNNs) and generative approaches, predict material properties with >90% accuracy in targeted domains like thermoelectrics, slashing virtual screening time from weeks to hours. DeepMind’s GNoME project identified 380,000 new stable materials, demonstrating how large-scale ML on databases like Materials Project accelerates discovery by 10–100×. Autonomous laboratories combining AI planning with robotics achieve ~71% sy