xAI Turmoil: Three Chinese Co‑Founders Exit in a Month, Halving the Original Team

In early 2026, xAI experienced a rapid succession of high‑profile departures. Greg Yang, a Harvard‑trained mathematician and former Microsoft researcher, announced he was leaving after being diagnosed with Lyme disease, attributing his illness to the "long‑term high‑intensity work" and extreme pressure at xAI. Shortly after, co‑founder Tony Wu (Wu Yuhui) posted a farewell message on X, describing the era as one where a small AI‑focused team could "move mountains". Wu, a 1995‑born PhD from the University of Toronto, previously worked at Google, DeepMind, and OpenAI on projects such as Minerva and AlphaGeometry, and was a key figure behind Grok 3’s mathematical reasoning capabilities. The following day, Jimmy Ba, Wu’s former PhD advisor and assistant professor at the University of Toronto, also announced his exit; Ba is renowned for co‑authoring the Adam optimizer with Diederik Kingma.

The attrition extends beyond the recent trio. Earlier exits include Kyle Kosic, who left for OpenAI in mid‑2024, Christian Szegedy (the Inception network pioneer) in February 2025, and Igor Babuschkin, who departed in August 2025 to launch his own venture capital firm. Altogether, six of the original twelve founders have left within three years.

Analysts cite several intertwined factors. First, Musk’s notoriously high‑pressure work culture appears to have strained staff health, as illustrated by Yang’s Lyme disease claim and Babuschkin’s remarks about a "near‑obsessive sense of urgency". Second, strategic shifts—most notably SpaceX’s acquisition of xAI announced a week before Wu’s departure—have introduced uncertainty over equity distribution, future direction, and role definitions. The integration of xAI with the X platform further distances the organization from its original mission of "understanding the true nature of the universe". Finally, product and regulatory challenges, such as allegations that Grok’s image‑generation tools can create realistic synthetic media of real people (including minors) and subsequent raids on xAI’s Paris office, add external pressure.

These departures raise concrete risks for the planned IPO. Wu and Yang were pivotal to Grok’s mathematical and architectural advances, while Ba contributed deep‑learning theory expertise. Their loss could erode xAI’s technical advantage against rivals like OpenAI and Google, prompting market skepticism about the company’s valuation—estimated at $2.5 billion for xAI within a $1 trillion SpaceX deal—and its ability to sustain growth.

Overall, the talent exodus underscores the tension between Musk’s rapid execution model and the need for stable, long‑term research leadership, a dynamic that may shape the future trajectory of xAI and its public offering prospects.