Meta Poaches 7 Thinking Machines Founders | Mira Murati 2026

1. The Raid | Seven Founders, One Strategic Response to a Rejected Offer

Meta has confirmed that seven founding members of Thinking Machines Lab, the AI research startup founded by former OpenAI CTO Mira Murati, are now employees of Meta's Superintelligence Labs division. The defections, confirmed by multiple sources on Tuesday, April 21, 2026, represent the largest single talent extraction from any AI startup in the current cycle. They are also, by most accounts, the direct consequence of a strategic decision that Murati made in mid-2025: she turned down a $1 billion acquisition offer from Mark Zuckerberg.

When that offer was rejected, Meta did not walk away. It pivoted. Rather than acquiring the company, Zuckerberg began acquiring its people. The result, as of this week, is that the founding core of Thinking Machines has been effectively transplanted into the Superintelligence Labs unit led by 28-year-old Alexandr Wang , the former Scale AI founder whom Zuckerberg recruited last year to lead Meta's frontier model push.

The Talent Raid | By the Numbers

Founding members now at Meta

$1.5B

Andrew Tulloch reported 6-year package

$12B

Thinking Machines current valuation

130

Remaining employees ($92M+ per head)

The Acquisition That Wasn't:

Industry sources with knowledge of the discussions suggest that Meta's cumulative spend on individual Thinking Machines hires now exceeds what the $1 billion company acquisition would have cost. Andrew Tulloch's reported $1.5 billion package alone eclipses the original offer. Silicon Valley is calling it the most expensive rejection in startup history.

2. The Defectors | Who Left, What They Built, and Where They Landed

The individuals recruited from Thinking Machines are not peripheral contributors. They are the people who built the company's core products and set its technical direction. Each departure removes not just a researcher or engineer but an institutional founder with deep context on Murati's proprietary work.

Andrew Tulloch | Co-Founder to Meta

The most consequential defection. Tulloch co-founded Thinking Machines and is reported to have accepted a compensation package worth $1.5 billion over six years at Meta, making it the single largest publicly reported individual talent acquisition in the history of the technology industry. The structure is understood to include base salary, equity in Meta's Superintelligence Labs development outcomes, and multi-year retention vesting.

Joshua Gross | Engineering Lead to Meta

Gross built and shipped Tinker, Thinking Machines' flagship fine-tuning API that serves as the company's primary commercial product. His departure takes with it both the technical architecture and the institutional knowledge of the product roadmap Murati has been executing. Gross is now embedded in Meta's model fine-tuning infrastructure team.

Mark Jen and Yinghai Lu | Back to Meta

Both Jen and Lu are Meta alumni who joined Thinking Machines at founding. Their return to Meta follows a pattern: Zuckerberg appears to have specifically targeted individuals with prior Meta experience, offering familiar infrastructure and cultural context as part of the pitch. Both are leading reasoning and inference optimization projects within Superintelligence Labs.

Tianyi Zhang | Researcher to Meta

Zhang is a highly cited AI researcher whose published work on model optimization and architecture has been widely referenced in the field. His move to Meta this week is the most recent confirmed departure and signals that Meta's recruitment of Thinking Machines talent is ongoing rather than concluded.

The OpenAI Parallel

Meta is not operating alone in this talent market. OpenAI has also been quietly recapturing former staff. Barret Zoph, who served as Thinking Machines' first CTO, returned to OpenAI in January 2026. Founding members Jolene Parish, Luke Metz, and Sam Schoenholz followed in the months after. Thinking Machines is thus simultaneously fighting a two-front talent war against the two most heavily resourced AI organizations in the world.

3. Why Researchers Choose Big Tech Over a $12B Startup

The conventional wisdom is that startups win talent through equity upside: a bet on a smaller number that could become a much larger one. That model is under structural pressure in the current AI environment. The inputs required to do frontier AI research, namely compute at scale, are not available to startups on equivalent terms to hyperscalers.

For a researcher who wants to train at the absolute frontier, the question isn't the title or even the salary. It's how many H100s you can wake up tomorrow and use. Startups, even well-funded ones, wait in queue. The hyperscalers don't.
Senior AI Infrastructure Engineer, Anonymous, current Big Tech employee

Meta operates one of the three largest private GPU clusters in the world, alongside Google and Microsoft. Its custom MTIA chip program, extended to 2029 with Broadcom in a $2.3 billion partnership , gives it an additional advantage: proprietary silicon optimized for the exact workloads its Superintelligence Labs team is running. Thinking Machines, despite its Nvidia partnership and $12 billion valuation, is managing a fraction of that computational capacity. For researchers working at the intersection of model scale and inference efficiency, that gap is not cosmetic.

4. Murati's Counter-Move | PyTorch Creator Brought in as New CTO

Thinking Machines is not paralyzed. Murati's response to the talent exodus has been to execute what the industry is calling a "talent swap" at the highest level: while Meta takes her engineers, she is taking Meta's framework architects. Soumith Chintala, the creator of PyTorch, the dominant deep learning framework used by virtually every serious AI lab in the world including Meta itself, has joined Thinking Machines as its new CTO. Chintala spent over a decade at Meta AI Research and brings with him an understanding of Meta's internal tooling, culture, and technical priorities that no external hire could match.

The strategic logic is asymmetric but intentional. Murati is trading volume for expertise. Losing seven founders who built a product is painful. Gaining the person who built the foundational framework that the entire industry runs on is a different category of asset. Chintala's appointment signals that Thinking Machines is not pivoting away from its technical ambitions, it is restructuring around a new technical center of gravity.

The Remaining Moat

With approximately 130 employees and a $12 billion valuation, Thinking Machines sits at roughly $92 million in implied value per employee, one of the highest ratios for any private AI company. Its Nvidia partnership provides a dedicated compute allocation outside the public cloud queue. Tinker, despite losing its lead engineer, remains commercially deployed with enterprise clients and continues generating the revenue that justifies that valuation.

5. The Broader Signal | AI Researchers as Professional Athletes

The Thinking Machines raid is not an isolated event. It reflects a structural shift in how frontier AI talent is valued, allocated, and retained. The compensation packages involved, Andrew Tulloch's reported $1.5 billion over six years, are not anomalous gestures. They are the market-clearing price for individuals who have demonstrated the ability to build and ship foundational AI systems. The talent market for this cohort now operates more like professional sports than traditional technology employment.

For the broader AI ecosystem, the pattern established by the Thinking Machines case sets a precedent that will be referenced in every future startup equity negotiation. Retention packages, structure, and the geographic and computational advantages of hyperscaler employment are now explicit variables in founder and researcher decision-making. Murati built something valuable enough that Zuckerberg wanted to buy it and was willing to pay over a billion dollars when she said no. That, in an industry defined by who wins the next benchmark, is a form of validation that no talent raid can fully take away.

Sources & References

[1] Meta Recruits Seven Thinking Machines Founders — The Information confirming the defections and the $1.5B Andrew Tulloch compensation package.
[2] Bloomberg: Inside Meta's Thinking Machines Recruiting Blitz — Full account of the rejected acquisition and subsequent individual hiring campaign.
[3] Barret Zoph Returns to OpenAI as Thinking Machines CTO — The Verge reporting on Zoph's departure from Thinking Machines and return to OpenAI.
[4] Soumith Chintala Joins Thinking Machines as CTO — Official statement from Thinking Machines confirming Chintala's appointment and PyTorch legacy.
[5] WSJ: Thinking Machines at $12B Despite Founder Exodus — Wall Street Journal analysis of the startup's valuation resilience and Nvidia partnership details.