Thinking Machines Lab has won a multibillion‑dollar Google Cloud deal that secures access to Nvidia’s GB300 accelerator chips and places the startup on the same infrastructure tier as major AI labs. The announcement comes as the company has been hiring a string of high‑profile researchers from Meta, including PyTorch co‑founder Soumith Chintala and Segment Anything co‑author Piotr Dollár.

Big cloud deal brings top-line compute

Thinking Machines Lab signed a multibillion-dollar arrangement with Google that was revealed at Google Cloud Next on Tuesday. The contract makes the startup one of the earliest customers to access Nvidia’s GB300 accelerator chips through Google’s infrastructure, according to the company announcement.

The deal follows an earlier relationship between Thinking Machines and Nvidia and moves the startup into the same infrastructure tier as major labs and firms running large generative AI workloads. Access to GB300 chips and a major cloud partner gives Thinking Machines predictable capacity as it scales model training and inference workloads.

Multibillion-dollar cloud contracts typically include multi-year commitments and support for high-volume compute, storage and networking — resources startups need to train and operate models at modern AI research scale.

The talent tug-of-war

Thinking Machines has been hiring a string of researchers and engineers who previously worked at Meta. Recent hires include:

  • Soumith Chintala — an 11‑year Meta veteran and co‑founder of the PyTorch framework, who became Thinking Machines’ chief technology officer earlier this year.
  • Piotr Dollár — long‑time Meta researcher and co‑author of the Segment Anything model, now on Thinking Machines’ technical staff.
  • Weiyao Wang — left Meta after eight years and joined Thinking Machines last week.
  • Kenneth Li — a Harvard PhD who spent about 10 months at Meta and started at Thinking Machines this month.
  • Andrea Madotto — joined in December from Meta’s FAIR division, where she worked on multimodal language models.
  • James Sun — spent nearly nine years at Meta focusing on large language model pre‑ and post‑training.

LinkedIn profiles and internal hiring traces show Thinking Machines has hired more researchers from Meta than from any other single employer.

Diverse hires beyond Meta

Thinking Machines has also recruited from a range of other AI and research organisations, strengthening its expertise across model pre‑training, multimodal perception and systems engineering. Recent additions include:

  • Neal Wu — a three‑time gold medallist at the International Olympiad in Informatics and a founding member of the coding startup Cognition.
  • Jeffrey Tao — arrived after roles at Waymo, Windsurf and OpenAI.
  • Muhammad Maaz — a former research fellow at Anthropic.
  • Erik Wijmans — joined from Apple.
  • Liliang Ren — spent two and a half years on Microsoft’s AI Superintelligence team working on code‑focused pre‑training for OpenAI models before joining in March.

From founders to staff: the churn with Meta

There has been notable back‑and‑forth of talent between the two organisations: Meta has taken some of Thinking Machines’ founding members, and Thinking Machines has in turn hired several researchers who previously worked at Meta.

Related Articles

The agreement gives Thinking Machines access to Nvidia’s GB300 chips via Google Cloud, providing the predictable compute capacity the startup needs as it scales model training and inference workloads.

This article was created with AI assistance.