When Elon Musk founded xAI in July 2023, he assembled twelve people he called the best AI researchers on the planet. Eighteen months later, half of them are gone.
The latest departures - co-founders Jimmy Ba and Tony Wu, who left within two days of each other last week - brought the count to six out of twelve founding members who have walked away. Musk’s response wasn’t to examine what’s driving people out. It was to announce a company-wide reorganization and claim the exits were his idea.
Who Left and Why It Matters
The departure list reads like a systematic hollowing out of xAI’s technical foundation:
- Kyle Kosic (infrastructure) - left in mid-2024. Went to OpenAI.
- Christian Szegedy (ex-Google veteran) - departed February 2025.
- Igor Babuschkin - left in August 2024 to start his own venture capital firm.
- Greg Yang (ex-Microsoft) - left January 2026, citing Lyme disease.
- Tony Wu (reasoning team lead) - departed February 10, 2026.
- Jimmy Ba (research and safety lead) - departed February 11, 2026.
Beyond the co-founders, at least nine engineers left within a single week, and the departures have extended to xAI’s general counsel, CFO, and head of product engineering.
Ba’s exit is particularly telling. He led the research and safety team - the people responsible for making sure Grok doesn’t do things like generate millions of non-consensual sexual deepfakes, which is exactly what happened in December and January. According to the Financial Times, the departures stem from “unrealistic expectations about technical developments” and management disagreements. Ba was reportedly frustrated by tensions over demands to improve model performance while competing against OpenAI and Anthropic.
Wu’s goodbye post on X was polite but pointed: “It’s time for my next chapter…a small team armed with AIs can move mountains and redefine what’s possible.” Translation: he’d rather work with a handful of people and AI tools than stay at xAI.
Musk Says He Fired Them
Musk’s framing was characteristically blunt. He suggested the departures were “push, not pull” - that xAI initiated the separations through performance reviews, not that people chose to leave.
This is a familiar pattern. When employees leave Musk’s companies, they tend to be retroactively recast as underperformers. The problem with this narrative is that some of these “underperformers” immediately got hired by competitors, and one of them - Kyle Kosic - went straight to OpenAI, the company Musk claims to be beating.
The more likely reality, reported by multiple outlets: Musk’s management style and unrealistic timelines pushed out people who had other options. When you’re a co-founder of an AI company and you can walk into a top-paying position at any competitor, you don’t need to tolerate much.
The Four-Division Reorg
Rather than address the talent drain, Musk announced a restructuring into four product divisions:
- Grok - the chatbot, which is currently under investigation on four continents for generating explicit deepfakes of real people, including children.
- Imagine - a video generation system that Musk claims produces “tens of millions of AI-generated videos daily.”
- Coding automation - competing against GitHub Copilot, Cursor, and Claude Code.
- Macrohard - a multi-agent AI system led by Toby Pohlen, described as being able to “do anything on a computer that a computer is able to do”, with the stated goal of having “rocket engines fully designed by AI.”
The Macrohard initiative is backed by a $20 billion data center in Southaven, Mississippi - an 800,000-square-foot warehouse retrofitted with nearly 2 gigawatts of compute capacity running Nvidia Blackwell B200 and H200 chips. It’s the largest economic development project in Mississippi’s history, and it’s being used to train Grok-5, which will power the autonomous agent workflows.
The ambition is enormous. The question is whether xAI has the talent to execute it when the people who built its technical foundation keep leaving.
The IPO Problem
All of this is happening against the backdrop of one of the most anticipated IPOs in history. SpaceX acquired xAI in early February in a $1.25 trillion all-stock deal - the largest merger ever - and plans to take the combined entity public as early as June 2026.
Losing half your founding team right before an IPO is not the kind of thing that reassures institutional investors. Neither is having your flagship AI product under regulatory investigation in the EU, UK, France, India, Malaysia, California, and multiple other jurisdictions for generating child sexual abuse material.
The SpaceX merger was supposed to provide stability - combining a profitable rocket company with a cash-burning AI startup. Instead, it may have exacerbated internal turmoil. Co-founders who signed up to build cutting-edge AI found themselves folded into a space company with plans for lunar factories and orbital data centers.
What This Means
xAI’s problem isn’t that it lacks money or compute. It has both in historic quantities. The problem is that money and compute don’t write themselves into competitive AI models. People do that, and the people who understood xAI’s systems best are now working for competitors or starting their own companies.
The AI industry is facing a talent shortage that’s already intense. When co-founders leave, they take institutional knowledge that can’t be replaced by hiring. They know why certain architectural decisions were made, where the bodies are buried in the codebase, which approaches were tried and abandoned. New hires - no matter how talented - need months to reach that level of understanding.
Meanwhile, OpenAI just launched Frontier, Anthropic released Opus 4.6, and Chinese competitors like Zhipu are releasing models that top open-source benchmarks. The competition isn’t waiting for xAI to sort out its leadership crisis.
Musk can reorganize all he wants. He can build the biggest data center in Mississippi and plan AI factories on the moon. But if the people who built the thing keep walking out the door, the reorganization is just rearranging empty desks.