xAI's Final Cofounder Is Gone. All 11 Have Now Left Elon Musk's AI Company
Ross Nordeen and Manuel Kroiss, the last two of xAI's original eleven cofounders, departed in mid-March 2026, completing a full exodus that leaves Elon Musk rebuilding the company from scratch while Grok continues to lag competitors in coding benchmarks.
All eleven of xAI’s original cofounders have now left the company. Ross Nordeen, the operations lead described by Business Insider as Musk’s “right-hand operator,” departed in the final week of March. Manuel Kroiss, who headed the pretraining team, left earlier in the same month. That completes the roster.
The departures were not sudden. The exodus began in early 2025 with Christian Szegedy, who came from Google, and accelerated dramatically through February and March 2026. Tony Wu, who led reasoning research, left February 10. Jimmy Ba — co-author of the Adam optimization paper that underlies most modern neural network training — followed within 24 hours. Igor Babuschkin, Greg Yang, Toby Pohlen, Zihang Dai, Guodong Zhang, and Kyle Kosic all departed by mid-March. Then Kroiss. Then Nordeen.
Elon Musk’s own words frame this accurately. On March 13, he posted that “xAI was not built right first time around, so is being rebuilt from the foundations up.” At the Abundance Conference, he acknowledged that “Grok is currently behind in coding.” Both statements confirmed what researchers had apparently been saying internally for months.
The consequence of the leadership collapse is visible in the product. Grok 4 — built on Colossus, a supercomputer with over 200,000 NVIDIA H100 GPUs — trails Cursor, Claude, and GPT-4.1 on coding benchmarks. The hardware exists. The research team that would have built something competitive with it has entirely dispersed. In March, xAI attempted to fill the gap by poaching Andrew Milich and Jason Ginsberg from Cursor, two senior engineers who built its core assistant. Whether two hires can substitute for eleven departing research leads is the question xAI’s rebuilding effort will answer.
What makes this situation structurally unusual: xAI still has enormous assets. The Colossus supercomputer is real, and it represents a genuine infrastructure advantage. Grok has distribution through X’s user base — hundreds of millions of accounts. The company has capital. What it does not have is the organizational knowledge and research culture that the founding team built and then took with them.
Musk has rebuilt organizations before. Tesla had multiple near-death moments. SpaceX failed its first three Falcon 1 launches. But those companies faced engineering and manufacturing problems; xAI faces a research talent problem in the most competitive AI hiring market in history. The people who left built their careers on publishable research and technical credibility. Replacing them requires offering similar conditions — which becomes difficult when the signal being sent to the research community is that the previous team found the environment untenable.
The SpaceX IPO complicates the picture further. Investors looking at xAI as part of Musk’s broader corporate ecosystem will price in the leadership vacuum. A company rebuilding its foundational team from scratch while simultaneously trying to close the gap with OpenAI and Anthropic is a harder story to tell to institutional money.
What happens next at xAI matters for the entire competitive landscape. If Musk succeeds in rebuilding a credible research organization, there will be four serious frontier labs rather than three. If he does not, xAI becomes a well-funded distributor of Grok — useful, but not a frontier player.