In September 2026, The Information reported that Mira Murati’s AI startup, Thinking Machines Lab, was in talks to raise more than $1 billion at a valuation of roughly $40 billion — a figure TechCrunch noted was still below the $50 billion the company had reportedly sought a year earlier. By then, three of its named founding researchers had already left for OpenAI, and the co-founder Meta had once tried hardest to keep had just quit Meta too. The company had shipped exactly one product. It had never shipped a second.

Six years, then one sentence

Murati spent more than six years at OpenAI, rising from vice president of applied AI in 2018 to chief technology officer in 2022, and briefly serving as the company’s interim CEO for three days during Sam Altman’s November 2023 ouster and reinstatement. She oversaw the technical work behind ChatGPT, DALL-E and Sora. On September 25, 2024, she announced she was leaving, in a written statement reported identically across outlets including Fortune and ABC News: “After much reflection, I have made the difficult decision to leave OpenAI… I’m stepping away because I want to create the time and space to do my own exploration.” Two other senior OpenAI leaders announced their own departures the same day.

Two billion dollars before a single product shipped

Murati founded Thinking Machines Lab in February 2025, recruiting a founding team that included OpenAI co-founder John Schulman as chief scientist, along with Barret Zoph, Lilian Weng, Luke Metz and Andrew Tulloch — all, like Murati, OpenAI alumni, according to Wikipedia’s company entry and contemporaneous reporting. In July 2025, the company closed a $2 billion seed round led by Andreessen Horowitz, with Nvidia, AMD, Cisco, Accel, ServiceNow and Jane Street among the other investors, at a $12 billion valuation — a figure confirmed consistently across multiple outlets covering the round. The company had no public product at the time. Albania’s government disclosed a $10 million investment in the round, large enough that it required a parliamentary budget amendment, Wikipedia’s entry notes.

Zuckerberg makes his move

Meta moved on the company almost immediately. The Wall Street Journal reported that Mark Zuckerberg personally sought to acquire Thinking Machines Lab outright for roughly $1 billion; Murati turned the offer down, according to reporting picked up by Business Standard, Gulf News and WION. Meta then began recruiting the startup’s researchers individually. Andrew Tulloch, a Thinking Machines co-founder, was offered a compensation package reportedly worth up to $1.5 billion over six years — contingent on steep performance and stock targets — per the Journal’s reporting. Tulloch initially turned that down too, before joining Meta Superintelligence Labs in October 2025.

OpenAI pulls three of them back

The attrition didn’t run in only one direction. On January 14, 2026, Barret Zoph — by then Thinking Machines’ CTO — announced he was returning to OpenAI. Within the hour, OpenAI confirmed that Luke Metz and Sam Schoenholz were rejoining alongside him, TechCrunch reported directly from the companies’ own announcements. OpenAI’s statement offered no detail beyond calling it something that had “been in the works for several weeks.” TechCrunch noted that Wired had described the split between Zoph and Thinking Machines as “not amicable,” without elaborating further, and Murati’s own public comments on the departures were similarly brief.

The one product that shipped

Eight months after founding, on October 1, 2025, Thinking Machines Lab released its first product: Tinker, an API for fine-tuning open-weight language models like Meta’s Llama and Alibaba’s Qwen without requiring customers to manage their own training infrastructure. “We believe it will help empower researchers and developers to experiment with models and will make frontier capabilities much more accessible to all people,” Murati said at launch, quoted by Fortune. Tinker reached general availability in December 2025 and was adopted by researchers at Princeton and Stanford, among others. By the time of the September 2026 funding talks, the company’s annualized revenue run rate had passed $100 million, according to TechCrunch’s reporting on the round.

The valuation that didn’t happen, and the one that might

Bloomberg reported in November 2025 that Thinking Machines was in early talks for a new funding round that could value the company at $50 billion to $60 billion — a more than fourfold jump from its July mark, achieved on the strength of a single product that had been on the market for weeks. That round never closed at that level. By September 2026, when The Information reported the company was instead in talks with Accel to lead a $1 billion-plus round at roughly $40 billion — with Nvidia discussing a stake that could account for roughly half of a $5 billion to $6 billion raise — TechCrunch and PYMNTS both framed the new number as a step down from the earlier ask, even as it still marked a fourfold increase over the company’s $10 billion pre-money mark from fourteen months before.

The $1.5 billion hire didn’t stay either

The clearest measure of how unsettled the talent market around Thinking Machines remained: Andrew Tulloch, the co-founder Meta had paid to recruit away with a reported nine-figure-a-year package, left Meta for Anthropic in September 2026 — less than a year after joining, according to Forbes Australia’s reporting on his move.

What founders should actually take from it

None of this has stopped money from flowing toward Thinking Machines Lab — the $40 billion talks are themselves proof of that. But the year between the $12 billion seed and the $40 billion follow-on is a specific, documented argument against the idea that enough capital buys enough loyalty to hold a frontier AI team together: Meta’s reported $1 billion bid for the whole company failed, its reported $1.5 billion bid for one employee succeeded only long enough for Anthropic to make its own offer, and OpenAI recovered three researchers from the company its own former CTO had just founded to compete with it. Murati’s team shipped one real product in nineteen months and is being underwritten, at a $40 billion mark, almost entirely on the promise of the next one. Whether that promise is worth the number attached to it is a question the market hasn’t actually tested yet — Thinking Machines has shipped nothing since, and the $40 billion figure prices in a second act nobody outside the company has seen.

This is an editorial profile assembled from public company statements and reporting. Sources: Fortune — Mira Murati’s career at OpenAI and Thinking Machines Lab, Wikipedia — Thinking Machines Lab, Wikipedia — Mira Murati, TechCrunch — Accel reportedly in talks to lead $1B round for Thinking Machines at $40B valuation, TechCrunch — Mira Murati’s startup Thinking Machines Lab is losing two of its co-founders to OpenAI, PYMNTS — Thinking Machines Lab Seeks $1 Billion at $40 Billion Valuation, PYMNTS — AI Startup Thinking Machines Aims to Quadruple Valuation to $60 Billion, Business Standard — After Mira Murati’s rejection, her cofounder turns down Zuckerberg’s $1.5bn offer, TipRanks/The Fly — Meta’s Zuckerberg looked to acquire Thinking Machines Lab, WSJ reports, Gulf News — Who is Andrew Tulloch? Meet the man who rejected Mark Zuckerberg’s $1.5b job offer, Forbes Australia — Meta offered this Australian $1.5 billion. Now he’s joining Anthropic, The Next Web — Meta hires five Thinking Machines Lab founders including a reported $1.5 billion engineer, Wikimedia Commons — File: Guests at the 2026 Met Gala 274 (Mira Murati).jpg, photo by SWinxy, CC BY 4.0.