Why the Next AI Breakthrough May Come from Physics with Max Welling

Max Welling, a theoretical physicist turned AI entrepreneur and academic, returned to the TWIML AI Podcast after several years to catch up on his work. He noted that the ideas of equivariance he helped pioneer have found a natural home in chemistry and materials science, where neural networks are now used as surrogate models to predict atomic forces — a calculation that normally requires expensive quantum mechanical approximations. "Three to four orders of magnitude acceleration" is what these models can provide, he said, and this efficiency is what led him to co-found CuspAI in 2024.

From Equivariance to Materials Discovery

The core insight behind CuspAI is that the physical world is symmetric under rotations and translations, and neural networks that respect these symmetries — so-called equivariant models — are far more data-efficient for molecular simulations. Welling's background in theoretical physics made this a natural fit. After two years as a VP at Microsoft Research in Amsterdam, he wanted the faster pace of a startup. Together with co-founder Chad Edwards, he launched CuspAI with an initial $30 million investment and has since grown the team to about 50 people across Amsterdam, Cambridge, London, Berlin, and expanding into Asia and North America. The company's advisory board includes Geoff Hinton, Yann LeCun, former ASML president and CTO Martin van den Brink,