Summary: On January 22, 2026, The New York Times laid out in detail the chaos inside Thinking Machines Lab. It began on January 12: researcher Sam Schoenholz called a meeting with Murati, and CTO Barret Zoph plus co-founder Luke Metz sat in. The company, started in February 2025 by one-time OpenAI CTO Mira Murati, had banked $2 billion within a year at a $12 billion valuation.
On January 22, 2026, The New York Times laid out in detail the chaos inside Thinking Machines Lab. The company, started in February 2025 by one-time OpenAI CTO Mira Murati, had banked $2 billion within a year at a $12 billion valuation. Now not one co-founder besides Murati remains, and the next raise — pitched at a $50 billion valuation — has gone quiet. More than Silicon Valley gossip, this is an up-close look at the AI talent war and valuation discipline in 2026.
Told in order: a control fight broke out between Murati and three of its earliest leaders, the CTO lost his job two days later, and 58 minutes after Murati’s post went up, OpenAI declared it had hired all three.
January 12: The Demand for Control
It began on January 12: researcher Sam Schoenholz called a meeting with Murati, and CTO Barret Zoph plus co-founder Luke Metz sat in. All four had spent time at OpenAI. As the NYT tells it, the three disliked where the company was heading — the product trailed rivals, and the raise resting on a $50 billion valuation was getting nowhere. They demanded Zoph be handed direct authority over technical direction, or they would walk. Murati’s answer ran one line: you already hold the CTO title, so do that job.
Two Days Later: Fired, Then Rehired Within the Hour
On January 14, Zoph was dismissed by Murati — the NYT notes his talks with Sam Altman about a return reach back to October. OpenAI did not sit on its hands. Murati marked the exit on X with the words “parted ways”; fifty-eight minutes later, Fidji Simo, OpenAI’s CEO of applications, announced Zoph, Metz, and Schoenholz were coming back. In a headcount of roughly 100, another nine or so have departed or hold OpenAI offers, while Meta has put packages worth hundreds of millions of dollars before certain employees. Murati gave the CTO seat to Soumith Chintala, one of PyTorch’s co-creators. On January 21, OpenAI named Zoph to lead the business selling AI services to enterprises.
The Product Lagged, and the $50 Billion Round Wobbled
In July 2025, Thinking Machines banked a $2 billion seed round — Andreessen Horowitz in the lead, Accel, Nvidia, AMD, and Jane Street alongside — at a $12 billion valuation, with nothing shipped yet. Its first offering came only in October: a service for developers who want to fine-tune models on particular tasks, up against comparable features from Google, Amazon, and Microsoft. The round built around a $50 billion valuation has been stuck since, and some investors privately question whether independence remains viable. The contrast with Mistral’s CEO expects revenue above €1 billion in 2026, reported that very week, is hard to miss: a valuation with no revenue behind it is a much harder sell.
Meta Came Calling First; Anthropic Drew Closer
The NYT also pieced together the earlier threads. In the summer before, Meta executives opened contact about an acquisition: Zoph favored a sale, while Murati read the approach as unserious and held to independence. Co-founder Andrew Tulloch, too, argued for selling and moved to Meta in October. On a separate track, Murati and Anthropic CEO Dario Amodei speak regularly, and the relationship has warmed — she and his wife go back a long way as friends. Closed-door meetings, a dead deal, a power struggle, a firing: the product never made its case, but the governance risk already had.
What the Talent War Teaches
Speaking to the NYT, Menlo Ventures partner Venky Ganesan reached for Anna Karenina: happy companies are all alike; every unhappy company is unhappy in its own way. Three lessons follow. One: for elite researchers, the opportunity cost is a nine-figure package at a competing lab, and no equity story holds anyone set on returning to OpenAI. Two: the “elite team plus unannounced product” formula behind ten-billion-dollar valuations is losing its grip, as investors shift to revenue and shipping speed. Three: who holds the final call on technical direction, and whether to sell, are governance matters to put in writing on day one — a showdown meeting is the worst place to learn no consensus existed.