Meta's Q4 Wave Is Signaled, Not Dated: Source the 2,743 Now

CNBC reported in May 2026 that Meta employees are openly bracing for a second layoff round in August and another later in the year. The 8,000-role May cut was the first public wave, not a single event, and if you wait for a WARN filing to start sourcing Reality Labs engineers you will be the fifth recruiter in their inbox.

The setup: three rounds already, a fourth signaled

Meta has already run three distinct 2026 reductions before the "second wave" everyone is waiting for. The May round gets the headlines because it was the largest, but it was the middle beat in a rolling restructuring that started in January.

• January 2026: 1,000 to 1,500 cuts inside Reality Labs.
• March 2026: roughly 700 cuts across five divisions, partially hitting RL.
• May 20, 2026: about 8,000 roles, roughly 10% of global headcount, spanning Reality Labs, Facebook, recruiting, sales, and non-AI product.
• Second half of 2026: signaled by employees to CNBC and, per Reuters, by executives to senior leaders. Not dated.

CFO Susan Li has publicly stated Reality Labs losses peak in 2026 before gradual reductions begin in 2027. That is the mechanical reason the reallocation continues. 2025 RL losses topped $19 billion, cumulative losses since 2020 exceed $80 billion, and Zuckerberg has committed to increasing AI spending by at least 60% in 2026, targeting $115 to $135 billion in total capital expenditure. The money has to come from somewhere inside a 78,000-person company. It is coming from Reality Labs and non-AI product.

Why Superintelligence Labs is an anti-signal, not a pool

Meta Superintelligence Labs and the Applied AI org are net importers of talent, so any "ex-Meta AI" candidate hitting the market is by definition not from the org Meta is protecting. Treat that as a probe, not a headline.

Roughly 1,000 existing Meta employees were moved into "AI builder," "AI pod lead," and "AI org lead" positions before the public May cuts, and Meta is actively hiring externally into the same titles. Hugo Barra, who led Meta VR from 2017 to 2021, is returning via the Dreamer licensing deal into Superintelligence Labs. Vishal Shah, who ran the metaverse effort for four years, was named VP of AI Products in October. That is the pattern: VR leadership is being recycled as AI leadership internally, not released to the market.

So when a candidate profile says "Meta, Applied AI, 2024 to present" and pops up as open to work, ask why. The answer is usually performance management or a team lift-out gone wrong, not a broad RIF. The sourceable arbitrage is elsewhere.

The real pool: 2,743 U.S. Reality Labs profiles

The pool that matters, and that almost nobody has mapped yet, is Reality Labs alumni and current staff outside the AI reorg. In Refolk's index there are 2,743 U.S.-based engineers with Reality Labs work histories, concentrated in Austin, the SF Bay Area, Seattle, Chicago, and NYC.

2,743

The important second number: in a 25-profile sample from that index, 13 still list Meta as their current employer. The majority of the pool has not moved yet. That is what a pre-warm pipeline looks like. Waiting for a resignation post is a competitive strategy that assumes you are faster than the other 200 recruiters watching the same feed. You are not.

Here is the comparable dataset the sourcing math rests on:

Reality Labs has already shed roughly 17 to 20% of its headcount inside a single calendar year, and December 2025 planning documents floated cuts of up to 30% of RL staff. If that trajectory holds through Q4, another 1,500 to 2,000 Reality Labs engineers land on the market before year end. Most of them will be senior systems people, not junior generalists.

The January and March cohorts are the ones you can close now

Recruiters chasing "May Meta layoffs" are six months late to the actual first cohort. The engineers cut in January and March 2026 are now deep in their next-role search, coachable, and often stuck comparing three mediocre offers.

That is a very different candidate than the freshly-severed May engineer, who is still in the "taking a month off" phase. The January and March cohorts:

1. Have completed at least one round of interview loops and know what they want.
2. Are past the initial ego bruise and open to smaller companies.
3. Often have not yet accepted, because their target list was too narrow.
4. Respond well to specific technical outreach that references what they actually built at RL (Horizon OS, Metaverse Content Group work, on-device perception).

Recruiters chasing "May Meta layoffs" are late. The January cohort is six months into a job search and coachable.

The problem is that these engineers are also the hardest to find with keyword search, because six months into unemployment they have often cleaned up their LinkedIn, removed the "Open to Work" banner, and updated their headline to something aspirational. That is the exact gap Refolk closes for this kind of pipeline: describe the person in plain English ("ex-Meta Reality Labs systems engineers in the Bay Area or Austin, cut between January and April 2026, with perception or Horizon OS work") and get a ranked shortlist rather than a keyword-brittle Boolean.

Ad-infra is smaller, harder, and worth more

The ex-Meta infra engineers most worth poaching do not self-describe as "ad infrastructure" on public profiles, which is why almost nobody has built a serious pipeline for them. Sourcing requires inference from adjacent stack signals, not keyword matches.

Refolk's index returns thousands of RL-adjacent profiles but only single-digit matches on literal "Meta ads ranking infrastructure" phrasing. That is not because the pool is small. It is because the phrasing is wrong. The engineers you want left signatures like:

• Presto and Velox query engine contributions.
• Scribe operational work.
• FBLearner Flow pipeline authorship.
• GitHub commits to facebookincubator/velox or prestodb/presto.

Any candidate with two or more of those signals plus a Meta tenure of four or more years is, functionally, an ads-infra engineer regardless of what their title says. This is where Refolk's plain-English querying compounds: ask for "engineers with Velox or Presto commits and 4+ years at Meta in the Bay Area" and skip the Boolean gymnastics.

The three shuttered studios are intact team lift-outs

Armature, Sanzaru, and Twisted Pixel are wholesale closures per an internal Meta memo, not stack-ranked reductions, which means entire orgs are hitting the market with working relationships intact. That is a rare "acquihire without the acqui" opportunity for VR studios, games shops, and simulation companies.

Meta retained Camouflaj, Glassworks, Games, BigBox, and OURO. So the sourcing map is clean:

If you are hiring for a VR or spatial-computing product, the Armature and Sanzaru senior engineers are the highest-yield six-week project you can run this quarter. Named-studio sourcing beats headcount-percentage sourcing every time, because the candidates trust each other and refer laterally. Pull one senior lead from Armature and you often get three engineers by the end of month two.

Wearables is where Meta is defending, not cutting

Optics, low-power SoC, and on-device ML engineers on the wearables side of Reality Labs are the highest-value poaching targets, and also the hardest, because Meta is actively retaining them. Meta's smart glasses sales more than tripled year on year while VR headset sales declined, and the internal reallocation is following the revenue.

Practical implication: if your target is a smart-glasses-adjacent engineer, do not wait for a layoff wave. There will not be one for that subset. Source them the way you would source a frontier-lab engineer: warm intro, specific technical hook, patient multi-quarter courtship. The California WARN filing covering 272 employees by March 20, 2026 (53 in Playa Vista, 219 in Burlingame) did not touch the wearables retain-list; those cuts skewed toward game developers, data engineers, software engineers, and AI researchers inside the Metaverse Content Group, Horizon OS, and Reality Lab Group, which is exactly the pool you can move quickly on.

For engineering leaders building this pipeline, the right query is not "ex-Meta engineers." It is "engineers who worked on Horizon OS or Metaverse Content Group between 2022 and 2026, currently in California." That returns a workable Q4 list of a few hundred names, not a 2,743-row spreadsheet you will never work through.

What to do in the next six weeks

Poaching Meta engineers ahead of the Q4 wave is a pipeline problem, not an inbound problem, so the work you do in October and November is what pays off in January.

1. Map the 2,743 now. Segment by studio (Armature, Sanzaru, Twisted Pixel), by team (Horizon OS, Metaverse Content Group, wearables), and by metro. Prioritize the roughly half who still list Meta as current employer.
2. Contact the January and March cohorts this week. They are the closable ones. Reference specific projects, not "I saw you're open to work."
3. Build an ad-infra list off Velox, Presto, and FBLearner signals, not off titles.
4. Set alerts on the three closed studios' alumni, because the referral chains inside them are dense.
5. Do not compete for Superintelligence Labs alumni. If they exist on the market, probe why.

The second wave will be dated eventually. Andrew Bosworth's next all-hands will tell you when. By then, half of your target list will already be in someone else's process. Pre-warming a pipeline against a signaled-but-undated cut is the entire game.

FAQ

When exactly is the Meta Q4 2026 layoff round?

It has not been dated publicly. CNBC's May reporting cited employees expecting a second round in August and another later in 2026, and Reuters reported executives signaling further cuts, potentially exceeding 20% of the 79,000-person staff, to senior leaders in March. Susan Li has said Reality Labs losses peak in 2026, which is the strongest structural signal that another RL-heavy reduction is coming before year end. Treat it as signaled, not dated, and source accordingly.

Are Superintelligence Labs engineers sourceable?

Generally no. Meta moved roughly 1,000 existing employees into AI builder, pod lead, and org lead roles before the May cuts and is actively hiring externally into those same titles. Superintelligence Labs is a net importer of talent, not a shedder. Any ex-Meta AI candidate on the market is worth interviewing, but do not build a pipeline strategy around a pool Meta is defending with a 60% capex increase.

What signals identify ex-Meta ad-infra engineers if their titles don't say so?

Look for contributions to Presto, Velox, Scribe, or FBLearner Flow, combined with Meta tenure of four or more years. Public GitHub commits to facebookincubator/velox or prestodb/presto are especially strong. Titles at Meta rotate frequently and rarely reflect the actual system the engineer owned, so infer from the stack, not the job title.

How does Refolk help versus a LinkedIn Recruiter Boolean?

Boolean search requires you to guess the right keywords, which fails on ad-infra engineers who don't self-describe with those terms and on RL alumni who have already cleaned up their profiles. Refolk takes a plain-English description ("ex-Meta Reality Labs engineers in Austin who worked on Horizon OS between 2022 and 2026") and returns a ranked shortlist using inferred signals rather than exact keyword matches. For a pool like the 2,743 U.S. Reality Labs profiles, that difference is roughly the difference between a workable Q4 list and a spreadsheet you never open again.