Fifty-Eight Minutes

It took just fifty-eight minutes for OpenAI to rehire Barret Zoph, shattering the twelve-billion-dollar myth of the independent AI startup.

[Speaker 1]: Usually, when a corporate partnership falls apart, we measure the timeline in months. Maybe weeks. There are lawyers, there are press releases, there is a "garden leave." [Speaker 2]: It’s a slow bleed. [Speaker 1]: Right. But this week, we saw something different. We saw the collapse of one of the most valuable startups in the world play out in real-time. On Wednesday morning, Mira Murati, the CEO of Thinking Machines Lab, fired her co-founder, Barret Zoph. [Speaker 2]: She accused him of "unethical conduct." She basically said he was leaking information to competitors. That is a career-ending accusation in Silicon Valley. [Speaker 1]: Usually. But exactly fifty-eight minutes later, OpenAI announced they had hired him. [Speaker 2]: Fifty-eight minutes. [Speaker 1]: And that timeline tells us everything. This isn’t just corporate drama. This is a signal that the entire "OpenAI Mafia" model-the idea that you can leave the mothership and build your own empire-is collapsing. [Speaker 2]: Because we have to talk about the number that’s been hanging over this whole story. Twelve billion dollars. [Speaker 1]: That was the valuation. [Speaker 2]: That was the valuation of Thinking Machines just six months ago. Twelve billion dollars. And today we have to ask: what is that number actually worth when the people it was priced on just walked out the door? [Speaker 1]: It’s Friday, January 16, 2026, and you’re listening to The Angle. [Speaker 2]: To understand why this fifty-eight-minute window matters so much, we have to rewind to the dream. Go back to late 2024. [Speaker 1]: Mira Murati leaves OpenAI. She was the CTO, she was the face of ChatGPT. When she left, the narrative was very clear: The magic wasn't the company. The magic was the talent. [Speaker 2]: Right. And if you have the talent, you can get the money. Murati teams up with Barret Zoph-who is arguably one of the best model architects in the world-and they launch Thinking Machines in February 2025. [Speaker 1]: And the money followed. [Speaker 2]: Floodgates opened. They raised two billion dollars in cash. A16z, Nvidia-everyone wanted in. The pitch was that they were going to build "Scientific AI." [Speaker 1]: What does that actually mean? "Scientific" as opposed to what? [Speaker 2]: As opposed to the "ship-it" culture at OpenAI. Murati’s whole thesis was that AI development was too messy. It was too much trial and error. She wanted to solve something called "nondeterminism." [Speaker 1]: Okay, break that down. [Speaker 2]: So, when you train these massive models on thousands of GPUs, the results can be slightly random. If you run the same training run twice, you might get slightly different results because of how the chips communicate. It’s noisy. [Speaker 1]: And for a scientist, noise is bad. [Speaker 2]: Exactly. It makes it hard to debug. So Thinking Machines spent the last year trying to solve "batch invariance." They were trying to make the training process perfectly reproducible. It’s a noble engineering goal. [Speaker 1]: But while they were doing that, while they were trying to perfect the science, the rest of the world didn't stop. [Speaker 2]: No. The world got exponentially more expensive. And this is the trap. In 2024, raising two billion dollars felt like a war chest. It felt like enough money to build anything. [Speaker 1]: But then Amazon launched Project Rainier. [Speaker 2]: Right. Late last year, Amazon turns on a cluster with 1.2 million chips. We are seeing training runs for frontier models that cost over a billion dollars just for *one run*. [Speaker…