CuspAI's Welling pitches AI hunt for not-yet-existing materials

CuspAI co-founder and chief technology officer Max Welling used a 50-minute Dutch-language interview on AIToday Live to describe the Cambridge startup's platform as "a search engine for materials that don't yet exist": generative models propose millions of candidate compounds, downstream algorithms filter them for stability and practical applicability, and lab results feed back for iterative refinement.

The concrete example he cites is perovskite solar cells "progressing toward 30% efficiency" through that cycle. Welling is also a machine learning professor at the University of Amsterdam and co-inventor of the variational autoencoder, and he co-founded CuspAI in 2024 with Chad Edwards; the company has since raised over $220M from investors including NEA, Temasek, Samsung and Hyundai and built out an AI Materials Foundry of over 45 partners spanning NVIDIA and Meta.

Much of the episode dwells on the scaffolding around the models rather than the models themselves. Data infrastructure has to exist before the AI becomes useful, Welling argues; regulatory and organisational processes tend to be the real bottleneck; and verifying a result in the lab matters more than being able to interpret why the model chose it.