TheChek News Flow

The Takeaway

The timing couldn’t have been more awkward for Thinking Machines Lab.

On Wednesday, during an all-hands meeting at the AI startup, its CEO, Mira Murati, announced that she had fired one of Thinking Machines’ co-founders, its Chief Technology Officer Barret Zoph, for poor performance and speaking with competitors, according to a person familiar with the matter. During the meeting, two more Thinking Machines researchers, Luke Metz and Sam Schoenholz, dropped bombshells when they separately posted in the company Slack that they were quitting, the person said.

Murati appeared to be taken aback after the messages from Metz and Schoenholz, according to the person, while some Thinking Machines staffers were stunned by the rapidfire disclosures of the three departures. Their unease began to grow later in the day after the OpenAI executive Fidji Simo announced that her company had hired the three researchers, all of whom had previously worked at OpenAI before leaving for Thinking Machines.

By Thursday, two more Thinking Machines staffers, researcher Lia Guy and engineer Ian O’Connell, told colleagues that they, too, were leaving the company. Of the five departures, four joined OpenAI.

Ordinarily, the departure of about 5% of a startup’s staff—Thinking Machines employed roughly 100 people prior to the defections this week—wouldn’t necessarily be a source of serious concern. But those departures included two of the company’s six co-founders and came after a third co-founder, Andrew Tulloch, left for Meta Platforms in the fall.

The events rattled investors in Thinking Machines, according to some of them. It could make it difficult for the one-year-old startup, which has released only a single product so far, to complete an ambitious funding round that aims to value the company at $50 billion—more than five times its last round.

That funding effort is likely to test whether a burst of employee turnover is sufficient to spook private investors in speculative AI startups. The company’s existing investors include Andreessen Horowitz, Accel and Nvidia, the AI chip giant. Thinking Machines could seek funding from its existing investor base or new ones (one possibility is Google, which is the startup’s primary cloud provider).

Though Thinking Machines currently has just one product—an application programming interface that allows developers to customize their models—it has bigger product ambitions. The company has discussed building a product for consumers, such as an AI-powered assistant consumers can speak to, as well as models customized to key performance indicators, or specific business metrics that companies track, typically related to revenue or profit growth. Executives have also said the company plans to start releasing its own in-house models this year.

Murati has told employees they should focus on developing technology that is substantially differentiated from products made by OpenAI and Anthropic, one of the people said.

Thinking Machines is generating some revenue, according to a person with knowledge of its financials, but it’s a long way away from larger and older labs like Anthropic and OpenAI, which are generating billions of dollars in revenue a year.

The success of the Thinking Machines funding will also say a lot about investors’ confidence in Murati, the charismatic former OpenAI Chief Technology Officer who left the company in late 2024 to co-found Thinking Machines. Murati engendered so much loyalty among OpenAI colleagues that 20 of them left OpenAI initially to join her at Thinking Machines. Investors scrambled to write checks for the startup even though Murati, during early fundraising conversations, shared only scraps of information about the company’s plans.

There was some irony in the timing of OpenAI’s hiring of the Thinking Machines staffers. Murati, too, had left OpenAI when it was in the midst of a critical fundraising process. And her hiring of former OpenAI colleagues at Thinking Machines shook staff confidence at her former employer for a time.

It’s possible Thinking Machines’ rivals could pounce on the company during its moment of vulnerability. Prior to rejoining OpenAI, Zoph had been speaking to leaders at other rival AI labs in recent weeks, including Meta Superintelligence Labs, which has aggressively recruited OpenAI researchers, according to the person with knowledge of the situation.

Zoph didn’t respond to requests for comment. Meta and Thinking Machines declined to comment. In the Thinking Machines all hands meeting, Murati also said that Zoph had lied to the company in recent months about a personal relationship with a former employee, the person familiar with the meeting said.

Simo, who is OpenAI’s applications chief, told staff at her company in a memo that she didn’t share the concerns cited in news reports that Zoph was fired for “unethical reasons.” Wired and The Wall Street Journal earlier reported some details of Zoph’s firing.

It wasn’t so long ago that Murati was able to resist the kind of raids that OpenAI performed on Thinking Machines this week.

She had recruited some of her former OpenAI colleagues with an unusual but attractive offer: Thinking Machines stock options with strike prices near zero, which they could exercise nearly immediately after joining. Such an arrangement would have been particularly attractive to the founders and early employees who joined before investors valued it at $10 billion, since their equity would have immediately been worth a lot more within weeks or months.

Zoph was one of those early employees. He had joined as Thinking Machines’ Chief Technology Officer, after a multi-year career at OpenAI, during which he ran post-training, the crucial last step of model training during which models are readied for release.

Murati had managed to stave off past raids on her staff. Over the summer, Meta had offered multiple Thinking Machines researchers large compensation packages. Those researchers declined.

Murati looked like she had enough capital after raising $2 billion less than a year after starting the company, an nearly unprecedented level for a startup that hadn’t released a product.

Researchers who flocked to the lab in large part came due to Murati, who was known at OpenAI for her ability to wrangle researchers with big personalities. Thinking Machines leaders also told them that they’d be able to decide what lines of research they pursue and would face less bureaucracy compared to other larger AI labs.

How easily Murati’s startup recovers from the losses isn’t clear. Zoph and Metz were both well-known AI researchers: Zoph for his work with post-training and Metz for being part of the original team that would develop what would later become ChatGPT.