More than 50 researchers and engineers have left SpaceXAI since SpaceX acquired xAI in February 2026, according to a report from The Information via TechCrunch1. The departures span layoffs, firings, and resignations from a combined entity valued at roughly $1.25 trillion2, and the losses have reduced the core team responsible for training Grok models to a fraction of its former size.
The Merger and the Mass Exodus
In February 2026, SpaceX absorbed xAI in an all-stock deal that valued the combined company at approximately $1.25 trillion2, assigning SpaceX a $1 trillion valuation and xAI $250 billion. The structure folded xAI’s operations into the SpaceX corporate vehicle. Since then, the combined entity has shed more than 50 technical staff across multiple divisions, including layoffs, terminations, and resignations1.
Where the Talent Went: Meta, Thinking Machines, and Beyond
At least 11 former xAI employees have joined Meta, at least 7 have defected to Mira Murati’s Thinking Machines Lab, and Anthropic has hired at least two1, according to the same TechCrunch report. AutoGPT’s coverage3 suggests the departures include co-founders and team leads across Grok coding systems, world models, and voice features, though that outlet carries lower confidence. What is clear is that competitors are not merely hiring individual contributors; they are acquiring institutional knowledge of xAI’s training stack, data pipelines, and model architecture decisions.
[Updated June 2026] The co-founder attrition is no longer a question of confidence. By the end of March, all 11 of xAI’s original co-founders had departed6, among them Igor Babuschkin, Tony Wu, Jimmy Ba, Greg Yang, Manuel Kroiss, and Ross Nordeen. Wu and Ba left within roughly a day of each other in February; the rest trickled out through the first quarter. A company can absorb the loss of any single founder. Losing the entire founding research bench inside thirteen months is a different signal, and competitors read it as one.
Why They’re Leaving: Deadlines, Culture, and Liquidity
Sources cited by TechCrunch1 blame two overlapping factors: unrealistic deadlines for training Grok models and Musk’s “extreme work” culture. The report alleges that pressure to hit delivery dates led to cutting corners on Grok development, a claim that remains single-sourced and should be treated accordingly.
There is also a compensation structural issue. xAI’s original equity packages were tied to a startup trajectory. The all-stock merger converted those holdings into SpaceX equity governed by pre-IPO vesting schedules. SpaceX has conducted share tender offers2 that allow some employees to cash out vested equity privately. [Updated June 2026] That liquidity question has since changed shape: SpaceX went public on June 12, 2026, pricing at $135 a share for a $1.77 trillion valuation7, raising about $75 billion, and trading under the ticker SPCX. The stock closed its first day up 19 percent at roughly $160.957. For employees who stayed, the IPO converts illiquid paper into a public security, but the timing matters: most of the 50-plus departures happened while the listing was still a projection, and the standard post-IPO lockup keeps insiders from selling for months after the bell. The researchers who left in the first half of 2026 traded an uncertain SPCX lockup for the continuous liquidity of an already-public competitor like Meta. The IPO arrived after the bleeding, not before it.
The Pre-Training Team Collapse
The most consequential individual departure is Juntang Zhuang, who led the pre-training team. According to TechStartups4, the core pre-training group has shrunk to “just a handful of people.” Pre-training is not a function that tolerates attrition well. It requires months-long runs on large clusters, careful curation of data mixtures, and stable staffing to interpret scaling curves and diagnose instabilities. Losing the majority of a pre-training team mid-cycle does not just slow the next model. It risks losing the tacit operational knowledge that determines whether a training run converges or wastes tens of millions in compute.
What It Means for the SpaceX IPO and Frontier AI Retention
SpaceXAI’s talent hemorrhage is a test case for what happens when a high-growth AI lab is folded into a larger conglomerate just ahead of its public listing. The merger created a $1.25 trillion2 headline, but the absorption also reset the retention clock. Researchers who signed on for xAI’s risk-reward profile traded it for equity in a company whose culture is dictated by Musk’s operational style and whose AI roadmap is now one division inside a rocket, satellite-internet, and payments empire.
For the broader industry, the episode raises a structural question about absorbed-lab models. When a startup’s equity converts to a parent company’s pre-IPO stock, the parent assumes the burden of retention. If that parent runs an “extreme hours” culture while competitors offer immediate liquidity and saner schedules, the math is not complicated. The pre-training team collapse makes it explicit: frontier AI capability is not primarily a capital problem. It is a talent problem, and top-tier researchers can walk out the door faster than any merger agreement can lock them in.
Buying Back the Talent It Lost
The clearest evidence that SpaceXAI knows it has a staffing problem is what it did about it. On June 16, 2026, four days after the IPO, SpaceX exercised an option to acquire Anysphere, the maker of the Cursor coding tool, in an all-stock deal valued at roughly $60 billion8. SpaceX had secured that option in April, with an approximately $10 billion combined breakup-and-services fee attached if it walked away8. The transaction is expected to close in the third quarter, pending regulatory review, after which Cursor becomes a wholly owned SpaceX subsidiary.
The price tag is the largest ever paid for a venture-backed startup, and the strategic logic is hard to separate from the exodus. SpaceX is buying three things its AI division lost or never had: a working engineering organization, a high-volume stream of real coding data, and a profitable enterprise product. Cursor reached roughly $4 billion in annualized revenue8 in under four years, with about two-thirds of the Fortune 500 running it. The same coding telemetry that makes Cursor useful also feeds a training pipeline, and Cursor’s team gains access to xAI’s Colossus cluster.
Read against the founding-team collapse, the deal looks like a balance-sheet answer to an organizational problem. You cannot quickly rehire eleven co-founders or a pre-training group that scattered to Meta and Thinking Machines. You can write a ten-figure check for a team that already ships. Whether bolting an enterprise-coding outfit onto a depleted frontier-research org rebuilds pre-training capability, or simply buys revenue and a brand, is the open question. For developers, the ownership change is its own story, covered in what the Cursor and Windsurf acquisitions change for dev teams.
The Valuation Gap Behind the Retention Math
The IPO did not settle the argument over what SpaceXAI is worth, and that gap bears directly on how much the equity reset cost departing staff. SpaceX listed at a $1.77 trillion valuation and reached a market capitalization near $2.1 trillion on day one7. Independent analysts were less generous: Morningstar’s discounted-cash-flow model pegged the company at $780 billion9, less than half the offering price, a divergence examined in Morningstar’s mark against the IPO target.
That spread is not academic for a researcher deciding whether to stay. An employee holding SpaceX shares at the IPO mark is exposed to the possibility that public markets eventually price the company closer to the independent estimate than the offer sheet. A competitor offering liquid, already-public equity removes that re-rating risk entirely. When the retention pitch depends on a valuation that respected outside models cut in half, the all-stock lock-in starts to look less like golden handcuffs and more like a bet the employee did not choose to make.
Frequently Asked Questions
Are reports that all xAI co-founders left accurate?
Yes. [Updated June 2026] What was an unconfirmed aggregator claim in May is now reported by The Information5 and corroborated elsewhere: all 11 of xAI’s original co-founders have departed6, the last of them by the end of March 2026. That is on top of the 50-plus total departures spanning team leads across pre-training, coding, world models, and voice.
Has SpaceXAI leadership framed the departures as intentional?
Musk has framed the attrition as a deliberate rebuild6, saying the company was “not built right the first time around” and is being “rebuilt from the foundations up.” Whether that is spin or genuine strategy is difficult to assess, though competitors like Meta actively hiring xAI’s pre-training leads suggests they read it as an opportunity rather than a planned transition.
Is this talent drain pattern unique to SpaceXAI?
The episode is a test case for the broader absorbed-lab model. Prior AI team absorptions, Microsoft’s hiring of Inflection’s staff, Google’s deal with Character.AI, folded researchers into public companies with liquid equity. SpaceXAI’s combination of illiquid pre-IPO stock and an ‘extreme hours’ culture makes the retention math structurally harder than those precedents.
Could SpaceX’s share tender offers slow further departures?
Tender offers provide partial, episodic liquidity for vested equity, far short of the continuous liquidity available to employees at public competitors like Meta. For researchers weighing competing offers from public companies, intermittent tender offers may not compensate for an intense work culture with an uncertain IPO timeline.
