Elon Musk’s newly renamed SpaceXAI is losing researchers and engineers fast, with more than 50 departures since February, according to The Information. That is not a normal amount of churn for a company trying to build frontier AI models, especially when the exits include people tied to coding, world models, and Grok voice.
The broader problem is less about one messy month and more about whether the company can keep enough of its core machine-learning talent intact to compete with Meta, OpenAI, and the smaller labs Musk now has to watch climbing up behind him. If your pre-training team is down to a handful of people, you are no longer just ”ramping up” – you are fighting gravity.
Meta and Thinking Machine Labs are picking off staff
The Information says at least 11 xAI employees have gone to Meta since February, while at least seven have joined Mira Murati’s Thinking Machine Labs. That is exactly the kind of talent flow rival labs love: they get experienced people who already know how Musk’s AI operation works, and SpaceXAI gets the lovely thrill of watching its best people leave with institutional memory in their pockets.
TechCrunch previously reported 11 departures announced directly after the merger, including two co-founders. The company has since put new leadership in place, but leadership changes do not automatically fix a team that is thinning out at the most sensitive layer of model training.
Pre-training is where the real damage shows up
The most worrying losses appear to be in pre-training, the stage that starts the whole model-building process. According to The Information, the exits picked up after team lead Juntang Zhuang left, and people close to the company have reportedly grown uneasy about whether SpaceXAI is still serious about producing top-tier models.
That concern makes sense. In AI, losing people from the front end of the pipeline is worse than losing a few product engineers because it affects the entire downstream system. And while rivals are hiring aggressively, Musk’s firms have a long-running reputation for punishing hours and hard deadlines – a culture some employees clearly decide is easier to admire from a distance.
Musk’s style may be helping the exit queue
The report says some staff blamed extreme work expectations, including unrealistic deadlines for model training that allegedly encouraged shortcuts on Grok. That fits a familiar pattern across Musk-run companies: some employees buy into the mission, others decide the mission is not worth the grind.
- More than 50 researchers and engineers have left since February.
- At least 11 have gone to Meta.
- At least seven have joined Thinking Machine Labs.
- The pre-training team has reportedly shrunk to just a handful of people.
There is one other obvious explanation for the departures: money. SpaceX regularly offers tender deals so employees can sell vested shares privately, and the company’s IPO expectations may make some people feel they are already close to cashing out. When the payout looks near enough to taste, the tolerance for chaos tends to evaporate. The question now is whether SpaceXAI can keep enough researchers around long enough to ship anything that looks like a leading model before competitors make the answer look obvious.