Talent Outflow from Meta: how startup Thinking Machines Lab poaches PyTorch creators and leading AI researchers

The recent move of Weiyao Wang, an engineer with eight years of experience at Meta, to the startup Thinking Machines Lab (TML) has become another twist in the large-scale battle for specialists in the field of artificial intelligence. Wang, who was involved in creating multimodal perception systems and open-world segmentation projects, including SAM3D, at the corporation, left his first post-university job last week. He was joined at the startup by Harvard University PhD holder Kenneth Lee. This talent migration is clearly bidirectional: it was previously reported that Meta poached seven TML founders.

The scale of the corporation's personnel losses becomes particularly evident when examining the list of transferred specialists who have, over recent years, effectively shaped the modern landscape of machine learning. TML's most notable acquisition is Soumith Chintala, who worked at Meta for eleven years and co-authored PyTorch – an open-source deep learning framework on which most of the world's AI research is based today. According to the source, Chintala left the corporation at the end of 2025 and took on the role of the startup's CTO at the beginning of this year. Petr Dollar, former Meta research director with eleven years of experience and co-author of the influential Segment Anything model, also joined TML's staff.

Thinking Machines Lab's recruiting campaign is not limited exclusively to former Meta employees: the startup is actively building its team by gathering expertise from many leading technology companies, as a result of which the total team size has already reached approximately 140 people. At the beginning of the year, Neil Wu joined the project – a three-time International Olympiad in Informatics gold medalist and one of the founders of the popular AI startup for code writing, Cognition. TML also succeeded in attracting Jeffrey Tao, who has experience working at Waymo, Windsurf, and OpenAI. The research potential was strengthened by Muhammad Maaz, who previously worked at Anthropic, and Erik Wijmans, who moved from Apple.

This massive talent reinforcement is occurring amidst a powerful infrastructural leap, guaranteeing engineers access to advanced computational power for creating next-generation foundational models. At the recent Google Cloud Next conference, it was officially announced that Thinking Machines Lab signed a multi-billion dollar cloud contract with Google. This agreement provides the startup with access to the latest NVIDIA GB300 chips, making it one of the first developing companies to gain the opportunity to work on such high-class equipment. The deal with Google followed a previously established direct partnership with NVIDIA and effectively brought TML to the same infrastructure level as industry giants like Anthropic and Meta.

The financial motivation behind this mass exodus deserves separate attention, as it illustrates the specific pragmatism of the current stage of artificial intelligence development. It is known that Meta offers its leading specialists guaranteed seven-figure compensation packages without any hidden conditions. However, for researchers evaluating their career prospects, the mathematics of the venture market prove more attractive: Thinking Machines Lab is currently valued at 12 billion dollars.