Jua

Jua is a frontier AI lab building a foundation model for how physical reality behaves. Founded in Zurich and built in collaboration with ETH Zurich, Google Research, and the ETH AI Center, Jua trained EPT-2 — the world's highest-scoring AI model for atmospheric prediction, surpassing ECMWF, DeepMind's GenCast, and Microsoft's Aurora on global benchmarks. The company's core thesis is that world models — simulations of physical systems — will collapse the cost of physical iteration from weeks to minutes, unlocking redesign of jet engines, fusion reactors, materials, and power grids. Jua's first applied product, Jua for Energy, runs on five continents, pricing power markets every hour for customers including Shell, TotalEnergies, EDF, Enel, and Vitol. The same model weights that forecast weather have also predicted compressible fluid dynamics and aerodynamic fields, demonstrating generalization across physical domains.