Manuel Kroiss, a co-founder of xAI and former DeepMind researcher, has left Elon Musk's AI lab, becoming the third co-founder to depart in Q1 2026. Kroiss brought deep expertise in reinforcement learning and optimization, having previously conducted research at DeepMind on training algorithms that improve the efficiency and capability of large neural networks.
His work was considered foundational to the training methodology behind Grok's early iterations. The departure continues a troubling trend for xAI, which despite raising billions in funding and building one of the world's largest GPU clusters, has struggled to retain the elite researchers who formed its founding team.
Kroiss's background in optimization was particularly valuable given xAI's emphasis on training efficiency — the company has publicly claimed to achieve competitive model performance with fewer training resources than rivals. His exit, combined with the departures of other co-founders from DeepMind and Google Brain backgrounds, suggests a cultural mismatch between the academic research mindset of xAI's founding scientists and the fast-moving, product-driven environment Musk has cultivated.
