Fast Company Found 80 xAI Exits. Source the 69 Meta Isn't Bidding On.
Meta will outbid you for the 11 xAI co-founders. The other 69 engineers, TPMs, and infra leads are sourceable before the SpaceX June 8 roadshow.
Fast Company's late-April review counted about 80 people who left xAI in the prior year. Eleven of them are co-founders. The SpaceX investor roadshow opens the week of June 8, 2026, which means for roughly ten days every senior recruiter on the planet will be staring at Jimmy Ba, Igor Babuschkin, and Tony Wu while Meta waves $300M retention packages at them. That is the wrong fight.
The right fight is the other 69. They have identical pedigree on paper, no LinkedIn inbox fire, and no Zuckerberg dartboard with their face on it. Here is who they are, where they sit, and how to reach them in the two-week window before the IPO swallows the news cycle.
The headcount math nobody is doing
Business Insider pegged xAI at roughly 1,200 staff in March 2025. Fast Company's tally of ~80 exits is about 7% of the company, and the publication was explicit that it only counted the publicly searchable ones. The real number is higher. The named portion of that exodus, the part everyone will fight over, is 11 co-founders plus a handful of senior researchers whose departures were individually covered in TechCrunch and Business Insider.
That leaves at least 69 unbranded exits with public profiles, plus an unknown long tail. These are RL trainers, eval engineers, pretraining infrastructure ICs, multimodal researchers, and senior TPMs who shipped Grok, Grok Imagine, and Colossus. They are the most operationally experienced frontier-lab ICs available right now, and almost no one is sourcing them by name.
Why the co-founders are a trap, not a target
Tony Wu left February 10, 2026. Jimmy Ba resigned within 24 hours. Toby Pohlen, Zihang Dai, and Guodong Zhang followed in late February and early March. Manuel Kroiss and Ross Nordeen, the last two, were out by March 27 and 28. Every name on that list is in a comp conversation that starts at eight figures and tops out near Meta's reported $300M-over-four-years ceiling.
You are not going to win those. Neither is your portfolio company. The expected value of an outreach to Jimmy Ba in June 2026 is approximately zero, and the opportunity cost is the hour you didn't spend on someone who will actually take a meeting.
The June 8 window
SpaceX prices June 11. The expected Nasdaq debut is June 12 under ticker SPCX, targeting a $1.75T to $2T valuation and a roughly $75B raise. From the week of June 8 through about June 19, recruiter and investor attention is concentrated on Musk's orbit. SPCX news will eat every AI-talent newsletter. Every executive recruiter with a Musk-adjacent retained search will be in airports.
This is a sourcing window, not an obstacle. Outreach to ex-xAI ICs landing in inboxes between June 8 and June 19 competes with almost nothing. The marquee names are getting hammered. The unbranded ones are getting silence.
The marquee names are getting hammered. The unbranded ones are getting silence.
Who the 69 actually are
The research note from our index surfaces a handful of named mid-level departures with public LinkedIn and X trails today. These are the templates for what you are looking for.
The RL and pretraining ICs
Juntang Zhuang worked on RL training and pre-training architecture and posted a public farewell on X. Xiuyu Li worked on infrastructure, RL programming agents, and post-training, and wrote "the die has been cast" on departure. Rad Venkataramani ran Grok's RL system. Any frontier lab building RLHF or RLAIF pipelines should have all three in a saved search by Monday.
These are not headline names. None of them appear in the Fast Company lede. All three would be top-decile hires at an Anthropic, a Mistral, a Reka, or a Series B agent startup, and none of them are currently fielding $300M offers.
The multimodal team
Hang Gao and Haotian Liu were Grok Imagine leads on video generation. Liu publicly cited burnout. If you are at Runway, Pika, Luma, Higgsfield, or any of the dozen well-funded video-gen startups, these are direct matches whose work is on shipped product. The pitch writes itself: smaller team, owned roadmap, normal on-call.
The TPMs nobody is recruiting
Fast Company specifically called out program staff, not just engineers. Senior TPMs from a frontier lab who shipped Grok, Imagine, and a 200,000-GPU training cluster are among the rarest hires in AI operations. The dominant narrative is "researcher exodus," which means almost zero recruiters are running TPM-specific boolean strings against xAI alumni right now. If your client is OpenAI, Anthropic, or any series C lab with a delivery problem, this is the lowest-competition pool in the market.
The Colossus angle that nobody is using
Anthropic announced in May 2026 that it would lease all of xAI's Colossus 1 capacity. The same week, Ross Nordeen, Musk's last co-founder and "right-hand operator," joined Anthropic to oversee compute capacity expansion. That is not a coincidence. That is the template.
Any ex-xAI infrastructure engineer who helped build or operate Colossus 1 now has a direct competitor (Anthropic) with concrete operational need for their tacit knowledge of that exact cluster. This is the cleanest warm-intro angle in AI sourcing for the next quarter. The pitch is one sentence: the company you are talking to is now running the hardware you already know.
The friction is finding them. "Worked on Colossus" is not a LinkedIn job title. People describe infrastructure work in fifteen different ways: "training cluster reliability," "GPU fleet ops," "distributed systems for foundation models," "ML platform." A boolean search loses half of them. This is the exact problem Refolk is built for: you describe the person in plain English ("infra engineer at xAI who worked on training cluster reliability, ideally 2023-2025") and get back a ranked shortlist instead of a 400-result LinkedIn page you have to grind through.
The Cursor second wave
SpaceX paid $60B in stock for an option to acquire Cursor, with a $1.5B termination fee and $8.5B in deferred services as the breakup terms. Cursor engineers are now meeting 1:1 with xAI's coding-team staff to "learn what they're working on." That is the standard pre-redundancy choreography. Expect Grok coding and agent engineers to start flipping "open to work" within 30 to 60 days of those meetings starting.
In practice, that extends the ex-xAI sourcing window through Q3 2026, with a different talent profile (coding agents, IDE integration, eval) than the June wave. Build the saved search now. Refolk surfaces ~20,941 US profiles with xAI in their history; the current-employer concentration still shows xAI itself at the top, which means most of the second wave hasn't moved yet. They will.
Where they're going (so you know who you're competing with)
Kyle Kosic went to OpenAI. Igor Babuschkin started a VC firm in August 2025, which is worth flagging as a downstream node: ex-xAI staff tend to cluster around former colleagues' funds as advisors, founders-in-residence, or first hires at portfolio companies. Ross Nordeen went to Anthropic. SignalFire's data has DeepMind-to-Anthropic flow at 11:1 in Anthropic's favor, and Anthropic just closed a $65B Series H at a $965B valuation and filed to go public.
Translation: Anthropic, OpenAI, and Babuschkin's portfolio are the three default destinations for senior ex-xAI talent. If you are not one of those three, your pitch needs to be sharper, and your sourcing needs to be faster.
The pitch the research note demands
xAI working conditions before the exits, per Business Insider: 12-to-16-hour days, 30-minute message response expected at any hour, product direction routed through a roughly 300-person X group chat instead of Slack. That tells you what the lever is, and what it is not.
It is not money. Meta has more money than you. The lever is research autonomy, stable management, normal on-call, and a Slack channel instead of a group DM with 299 other people. Every cold message to an ex-xAI IC should say that out loud in the first two lines. The candidate already knows you cannot match Meta. What they want to hear is that they will not be on call at 2am for a group chat decision.
The boolean problem
Standard LinkedIn boolean for "ex-xAI" returns the 11 co-founders, a handful of recruiters who once contracted there, and noise. It does not surface Juntang Zhuang's RL pretraining work, because his title at xAI was "Member of Technical Staff." It does not catch the TPMs, because their titles were generic. It does not catch Colossus operators, because Colossus is not a LinkedIn skill tag.
This is the structural reason ex-xAI talent sourcing is failing in June 2026: the people you want are not findable by the strings you know. Plain-English search ("xAI engineers who worked on pretraining infra and left after the Tony Wu departure") is the only practical way to rebuild that list. That is the entire reason we built Refolk, and it is the specific friction the next two weeks will reward solving.
What to do this week
- Build three searches: RL and pretraining ICs, Grok Imagine multimodal, senior TPMs. Save them.
- Pull the named examples (Zhuang, Xiuyu Li, Venkataramani, Gao, Liu) and use them as positive training examples, not as targets. They are probably already in conversation. Use them to find the next 50 who look like them.
- Write three outreach templates, one per cohort. Lead with autonomy and management stability. Do not lead with comp.
- Schedule the sends for June 8 through June 19, when the SPCX roadshow will be absorbing every competing recruiter's attention.
- Build a second saved search for Cursor-adjacent Grok coding engineers. Check it weekly through Q3.
The 11 co-founders will be a Bloomberg headline by the end of June. The other 69 will be at someone's company by the end of August. The only question is whose.
FAQ
How do I find ex-xAI engineers who aren't the 11 co-founders?
LinkedIn boolean is the wrong tool because xAI titles are generic ("Member of Technical Staff") and the work people actually did, RL training, Colossus operations, Grok Imagine multimodal, is not in their headline. Use plain-English search across LinkedIn, GitHub, and public X farewell posts. Refolk's index currently returns about 20,941 US profiles with xAI history, and the ranking surfaces recency of exit and public signal first.
Is the SpaceX IPO roadshow a sourcing problem or a sourcing window?
It is a window. From roughly June 8 through June 19, 2026, every senior tech recruiter is chasing co-founder names and watching SPCX pricing. Outreach to mid-level ex-xAI ICs in that window competes against almost nothing. The headline names will be unreachable; the unbranded 69 will be unusually responsive because nobody else is in their inbox.
Should I try to recruit any of the 11 co-founders?
Probably not. Meta is reportedly offering up to $300M over four years for top AI researchers, and Anthropic is the most aggressive secondary bidder coming off a $65B Series H. Unless you are one of those, the expected value of outreach to Jimmy Ba, Igor Babuschkin, or Tony Wu is close to zero. The opportunity cost is high because those hours come out of the time you would otherwise spend on the 69 unbranded exits with the same pedigree.
What about the Cursor integration, does that change the timing?
Yes. SpaceX's $60B Cursor option and the ongoing 1:1s between Cursor and xAI coding-team engineers will likely produce a second wave of departures from Grok's coding and agent teams within 30 to 60 days. That extends the ex-xAI sourcing window through Q3 2026 with a different talent profile (coding agents, IDE tooling, eval) than the June pretraining and RL wave. Build the saved search now and check it weekly.