The leadership landscape at OpenAI is shifting again. Liam Fedus, the company’s vice president of research, is leaving to found a materials science AI startup, marking the latest in a series of departures of senior researchers.
OpenAI has confirmed it will be an investor in and collaborate with Fedus’s new company, which aims to use AI to accelerate scientific discovery.
Fedus joined OpenAI in 2022 from Google and was promoted to VP last fall. His background in physics is driving his focus on the application of AI in materials science—an area OpenAI considers “strategically important.”
The move marks yet another change in the upper ranks of OpenAI. Fedus, who led post-training operations, is striking out on his own to use AI to discover new materials. A company spokesperson confirmed his departure to The Information.
Fedus’s promotion to VP came amidst other significant executive changes. This included the departure of his predecessor, Luke Metz, who joined former OpenAI Chief Technology Officer Mira Murati at her new venture, Thinking Machines Lab.
“My undergrad was in physics, and I’m keen to apply this technology there,” Fedus said in a statement posted on X. His new company will be entering a competitive field already drawing significant investments from Google DeepMind, Microsoft, and other players. In 2023, DeepMind unveiled GNoME, a system they claim can identify new crystals for materials development, while Microsoft introduced MatterGen and MatterSim, tools designed for materials discovery.
Despite the departure, OpenAI appears to be maintaining close ties with Fedus.
“Because AI for science is one of the most strategically important areas to OpenAI and achieving [artificial superintelligence], OpenAI is planning to invest in and partner with my new company,” Fedus noted in his statement.
The exit comes during what sources describe as a “renewed surge of defections” from OpenAI. This talent exodus has seemingly prompted action: According to reports, investor Josh Kushner held a presentation for employees last month, arguing they stood to lose money by leaving—a clear indicator of the competitive talent landscape in advanced AI.
As VP of post-training, Fedus oversaw the critical phase where AI models are refined after their initial training. This role placed him squarely at the heart of OpenAI’s product development pipeline, making his departure especially noteworthy.
Fedus’s new venture will focus on an area where AI’s ability to accelerate scientific discovery has shown early promise but has yet to achieve widespread commercial applications. By leveraging machine learning to predict material properties and potential new compounds, these AI systems aim to significantly reduce the time and resources generally required for materials research and development.
The ongoing partnership between OpenAI and Fedus suggests both entities see strategic value in maintaining close ties, even as leadership continues to evolve at one of AI’s most prominent organizations.
