Johannes Brandstetter

Johannes Brandstetter is a machine learning researcher who builds neural models that simulate physical systems for science and industry. After studying physics in Vienna, he wrote his doctoral thesis at CERN and turned to machine learning in 2018 in Sepp Hochreiter's research group in Linz. His 2022 ICLR spotlight paper Message Passing Neural PDE Solvers, written with Daniel Worrall and Max Welling, was followed by xLSTM and Universal Physics Transformers in 2024 and by the Aurora Earth-system model in Nature in 2025. He co-founded Emmi AI in December 2024, and Mistral AI announced its acquisition of the company in May 2026. He is Vice President, AI for Science at Mistral AI and an associate professor at JKU's Institute of Machine Learning.

Now at: Johannes Kepler University of Linz and Mistral AI
Badges: Founder
Bridge

Timeline

By 2018

Physics in Vienna, a CERN doctorate

He comes from Schärding, studies physics in Vienna, and writes his doctoral thesis at CERN. His years on the CMS experiment include coauthoring papers on Higgs boson physics, and he completes the PhD in 2018.
2018

Joining Hochreiter's group in Linz

In 2018, after finishing his PhD, he turns to machine learning and joins Sepp Hochreiter's research group in Linz.
2022

Message Passing Neural PDE Solvers

Message Passing Neural PDE Solvers, written with Daniel Worrall and Max Welling, appears at ICLR 2022 as a spotlight paper. It builds a solver entirely from neural message passing, which replaces the hand-designed parts of a numerical solver, and tests it on fluid-like flow problems in 1D and 2D.
December 2023

Back at JKU in Linz

In December 2023, he is back at JKU's Institute for Machine Learning as an assistant professor, after leaving for the Amsterdam Machine Learning Lab (AMLab) and working for two years at Microsoft.
2024

xLSTM and Universal Physics Transformers

He is among the authors of xLSTM, posted in May 2024 with Sepp Hochreiter and colleagues, which scales long short-term memory networks to billions of parameters for language modeling. Also in 2024, he coauthors Universal Physics Transformers, a framework for scaling neural operators. He presents xLSTM for JKU and NXAI at Siemens' AI with Purpose Summit in Munich.
December 2024–April 2025

Co-founding Emmi AI

In December 2024, he co-founds Emmi AI with Dennis Just and Miks Mikelsons, a spin-off of JKU Linz and NXAI whose models simulate physical processes such as airflow, thermodynamics and material deformation. In April 2025, the company raises a €15 million seed round, the largest ever for an Austrian startup.
May 2025

Aurora in Nature

In May 2025, Nature publishes "A foundation model for the Earth system", which introduces Aurora, a foundation model trained on more than one million hours of geophysical data that outperforms operational forecasts for air quality, ocean waves, tropical cyclone tracks and high-resolution weather. He is one of its authors, with affiliations at Microsoft Research AI for Science in Amsterdam and Johannes Kepler University Linz, and co-architected the model during his Microsoft years.
Since May 2026

Vice President at Mistral AI

In May 2026, Mistral AI announces its acquisition of Emmi AI, and Brandstetter, the company's co-founder and chief scientist, becomes Vice President, AI for Science, responsible for Mistral's AI division for scientific and industrial applications. Linz becomes an official Mistral location. An associate professor at JKU's Institute of Machine Learning, he is also an ELLIS Scholar with the ELLIS Unit Linz.