Thinking Machines benefits as Meta faces setbacks

Weiyao Wang spent eight years at Meta — beginning with his first role after college — where he worked on multimodal perception systems and contributed to open-world segmentation initiatives, including SAM3D. His tenure at Meta concluded last week, and he has now joined Thinking Machines Lab (TML).

His transition comes at a time when Thinking Machines Lab is scaling rapidly across several areas. The company recently secured a multibillion-dollar cloud agreement with Google, granting it access to Nvidia’s latest GB300 chips and positioning it among the first startups to operate on that hardware.

The deal, unveiled earlier this week during the Google Cloud Next event, builds on a prior collaboration with Nvidia and places TML in the same infrastructure category as companies like Anthropic and Meta. Reports indicate that Meta had previously explored acquiring Thinking Machines around the same period last year and has more recently recruited several of TML’s founding team members.

The movement of talent between the two companies remains dynamic. Alongside Wang, Kenneth Li — a Harvard PhD who spent about 10 months at Meta — also joined TML this month. These hires reflect a broader exchange of talent flowing in both directions. According to earlier reporting, Meta has recruited seven of Thinking Machines Lab’s founding members, while TML has been actively hiring from Meta. A review of LinkedIn activity suggests that Meta is currently the largest single source of new hires for TML.

Among the most notable figures is Soumith Chintala, the company’s chief technology officer. He spent 11 years at Meta and co-founded PyTorch, an open-source deep learning platform widely used in AI research. Chintala departed Meta in late 2025 and took on the CTO role at TML earlier this year. Piotr Dollár, another longtime Meta researcher who served as a director and co-authored the Segment Anything model, has also joined TML’s technical team.

Additional hires from Meta include Andrea Madotto, who worked in Meta’s FAIR division on multimodal language models and joined TML in December, and James Sun, who spent nearly 9 years at Meta focusing on pre- and post-training large language models.

Thinking Machines Lab has also attracted talent from outside Meta. Neal Wu, a three-time gold medalist at the International Olympiad in Informatics and a founding member of Cognition, joined earlier this year. Jeffrey Tao arrived after experience at Waymo, Windsurf, and OpenAI. Muhammad Maaz previously held a research fellowship at Anthropic, while Erik Wijmans joined from Apple. Liliang Ren, who spent two and a half years on Microsoft’s AI Superintelligence team, working on pretraining models for code, joined TML in March.

The company’s workforce has now grown to roughly 140 employees.

Meta’s compensation packages — often reaching seven figures without long-term restrictions — are widely recognised across the industry. For researchers evaluating opportunities, however, the appeal of Thinking Machines Lab may lie in its growth potential. The startup is currently valued at around $12 billion, a figure that would have seemed extraordinary for a company at this stage in earlier technology cycles, particularly given that it has released only one product so far. Still, compared with the record-setting valuations achieved by firms like OpenAI and Anthropic, there remains considerable upside.

A spokesperson for Thinking Machines Lab declined to comment when contacted on Friday morning.