In summary: Meta has hired five founding members of Thinking Machines Lab, the AI startup built by former OpenAI CTO Mira Murati, after she rejected a reported $1 billion acquisition offer. The most expensive individual hire, co-founder Andrew Tulloch, reportedly received a $1.5 billion package over six years. The talent raid is part of a broader AI restructuring that includes the $14.3 billion Scale AI investment, Alexandr Wang’s appointment as chief AI officer, Yann LeCun’s departure after 12 years, 600 FAIR research cuts, and the creation of Meta Superintelligence Labs, whose first closed-source model, Muse Spark, launched on 8 April.
Meta has now hired five of the founding members of Thinking Machines Lab, the AI startup that Mira Murati built after leaving her role as OpenAI’s chief technology officer. The most recent departure, founding engineer Joshua Gross, joined Meta Superintelligence Labs in March after building Tinker, the startup’s core API product. His exit follows that of co-founder Andrew Tulloch, who left for Meta in October with a compensation package reportedly worth $1.5 billion over six years, a figure that, if accurate, would make it the most expensive individual talent acquisition in the history of the technology industry.
The pattern is not subtle. After Mark Zuckerberg reportedly offered roughly $1 billion to acquire Thinking Machines Lab outright and was rejected, Meta pivoted to recruiting the founding team one by one. Multiple outlets have described the strategy as a “full-scale raid.” It has been effective. Of the startup’s original founding group, five have gone to Meta, three have returned to OpenAI, and one has joined Elon Musk’s xAI. Murati’s company, which raised $2 billion at a $12 billion valuation in a seed round led by Andreessen Horowitz in July 2025 and was reportedly in talks for a new round at a $50 billion valuation by November, has lost the majority of the team it was built around.
Who left and where they went
Tulloch and Gross are the two departures whose destinations and roles are most clearly documented. Tulloch, an AI researcher who had previously worked at OpenAI, joined Meta Superintelligence Labs and is now working under Alexandr Wang, the 28-year-old former Scale AI chief executive whom Meta installed as its first chief AI officer in June 2025. Gross, who had previously worked at both OpenAI and Meta, now leads engineering teams within the same division.
The 💜 of EU tech
The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!
The other defections from Thinking Machines followed different paths. Barret Zoph and Luke Metz returned to OpenAI in January 2026, along with Sam Schoenholz. Zoph was reportedly fired by Murati for what the company described as “unethical conduct” before immediately rejoining OpenAI. Devendra Chaplot left for xAI in March. The departures have left Murati with a significantly reconstituted leadership team: she remains as chief executive, Soumith Chintala, the creator of PyTorch who joined from Meta’s own FAIR lab, serves as chief technology officer, and John Schulman continues as chief scientist.
Meta’s new AI hierarchy
The talent acquisitions are part of a broader restructuring that has transformed Meta’s AI organisation over the past year. In June 2025, Meta paid $14.3 billion for a 49% non-voting stake in Scale AI and brought Wang in to lead a new division called Meta Superintelligence Labs, alongside Nat Friedman, the former GitHub chief executive. Zuckerberg called Wang “the most impressive founder of his generation” in an internal memo.
The restructuring has not been smooth. Yann LeCun, Meta’s chief AI scientist for 12 years and one of the most influential figures in deep learning, departed in November 2025 after being asked to report to Wang. “You don’t tell a researcher what to do. You certainly don’t tell a researcher like me what to do,” LeCun told the Financial Times in January. He called Wang “young and inexperienced” and warned that “a lot of people have left, a lot of people who haven’t yet left will leave.” LeCun subsequently raised $1 billion to found AMI Labs in Paris, drawing the founding team “almost entirely from Meta’s AI research organisation.”
By August 2025, Meta Superintelligence Labs had been split into four groups: the TBD Lab for large language models, led by Wang; FAIR for fundamental research; a products and applied research division led by Friedman; and an infrastructure unit led by Aparna Ramani. The AGI Foundations team, which had been responsible for the Llama model family, was dissolved after Llama 4’s lukewarm reception. LeCun publicly stated that the AI team had “fudged” some of the results. Approximately 600 roles were cut from FAIR and AI infrastructure units in October 2025.
The economics of the talent war
The compensation figures circulating in the AI talent market have reached a scale that distorts normal recruitment dynamics. OpenAI’s chief executive, Sam Altman, has acknowledged that signing bonuses of up to $100 million have been offered to lure top researchers. OpenAI’s chief scientist, Mark Chen, described Meta’s poaching as “akin to someone breaking into our home.” Altman’s counter was that Meta “had to go quite far down their list,” a characterisation that five founding-team departures from a single startup would seem to contradict.
The competition extends beyond Meta and OpenAI. Anthropic is winning what Fortune described as a “one-sided talent war” against both OpenAI, which retains 67% of its researchers, and Google DeepMind, which retains 78%. DeepMind has responded by enforcing six- to twelve-month non-compete clauses with full salary. The talent market for frontier AI researchers has become a zero-sum contest in which every hire by one lab is a direct loss for another, and the compensation required to move individuals has escalated from millions to hundreds of millions to, in Tulloch’s case, potentially billions.
Meta can afford the escalation. The company reported $201 billion in revenue for 2025, up 22% year over year, with $43.6 billion in free cash flow. It is spending $115 to $135 billion in capital expenditure this year on AI infrastructure, including a $27 billion joint venture with Nebius for a gigawatt-scale data centre. The 8,000 layoffs beginning on 20 May are explicitly framed as a reallocation: shedding roles in Reality Labs, recruiting, sales, and global operations to fund the AI pivot that Wang’s division represents.
What Meta has to show for it
The first output from Meta Superintelligence Labs arrived on 8 April with the release of Muse Spark, a natively multimodal reasoning model that Meta described as the first step toward “personal superintelligence.” It now powers Meta AI across Facebook, Instagram, WhatsApp, Messenger, and the Ray-Ban Meta AI glasses. The model is closed-source, breaking with Meta’s open-source Llama tradition and signalling that the intellectual property produced by the researchers Meta is hiring at extraordinary cost will not be shared freely.
A second model, internally codenamed “Avocado,” is reportedly in development under tighter control within the TBD Lab. Its progress has been uneven: internal benchmarks showed it falling short of Google’s Gemini in some evaluations, which contributed to the dissolution of the AGI Foundations team and the consolidation of model development under Wang.
The question for Meta is whether the talent it has assembled, at a cost that includes $14.3 billion for Scale AI, a reported $1.5 billion for a single engineer, the departure of one of the field’s most respected scientists, and the systematic dismantling of another company’s founding team, will produce AI capabilities that justify the investment. Thinking Machines Lab, for its part, has survived the exodus with new leadership, a $12 billion valuation, and the distinction of having produced a team so valuable that the world’s largest social media company was willing to spend more to acquire its members individually than it offered to buy the entire company.
