Thinking Machines Lab

In July 2025, a startup that had not yet shipped a single product raised $2 billion at a $12 billion valuation. What investors were really buying was a team, led by a founder who had spent six years inside OpenAI.

Mira Murati was born in Vlorë, Albania, in 1988. At 16 she won a scholarship to Pearson College in Canada, then studied mathematics at Colby and engineering at Dartmouth. Product roles followed at Tesla, where she worked on the Model X, and at Leap Motion. She joined OpenAI in 2018, became its chief technology officer in 2022, and ran the company for three days as interim CEO in November 2023. She left in September 2024 and launched Thinking Machines Lab in February 2025.

“A mission buys a young company time. What it builds in that time decides whether the next offer finds a team or a list of names.”

The test came within weeks. WIRED reported that Meta had approached more than a dozen people at the lab, with packages on a scale rarely seen in hiring: most between $200 million and $500 million over four years, some with $50 million to $100 million guaranteed in the first year, and one above $1 billion over several years. These were reported multi-year figures that vest over time, not cash handed over on day one, and Meta disputed the details, saying it had made offers to only a handful of people. Even so, the report's central fact stood out. At that point, nobody had said yes.

The loyalty did not hold. In October 2025, co-founder Andrew Tulloch left for Meta. The Wall Street Journal reported that Mark Zuckerberg had first tried to buy the company and then turned to recruiting Tulloch, with a package that could have reached $1.5 billion over six years if every bonus and stock award paid out in full, a description Meta called inaccurate and ridiculous. In January 2026, co-founders Barret Zoph and Luke Metz went back to OpenAI, and Murati named Soumith Chintala the lab's CTO. In July 2026, co-founder Lilian Weng announced her exit too. Four co-founders gone in under ten months. And Semafor reported in September 2026 that Tulloch was leaving Meta, less than a year after he joined.

And yet the lab kept building. Its first product, Tinker, a tool that lets researchers fine-tune AI models, arrived on 1 October 2025. In July 2026 came Inkling, its first open-weight model family. In September, The Information reported that the company was in talks to raise more than $1 billion at a valuation of around $40 billion, below the roughly $50 billion that earlier reports said it had sought. Talks are not a closed round, and a valuation is not a business: whether products and revenue can carry numbers like these is still untested.

The tidy version of this story, that her people all turned down Meta's money, was accurate only until October 2025. The fuller version is more useful. A mission can keep a young team together long enough to ship, but it cannot keep everyone forever, and it does not remove the risk that a company is really a bet on a handful of names. The lesson is not that vision beats money. It is that a mission buys time, and what you build in that time, depth in the team and a product that survives any single resignation, decides whether the next offer finds a team or a list of individuals.

Sources: Fortune, TechCrunch, eWeek, Computerworld, OfficeChai, The Next Web, Let's Data Science and RuntimeWire, which in turn cite WIRED, the Wall Street Journal, Semafor and The Information. All compensation figures are as reported.