The valuation quadrupled, but the founding team is quietly walking out the back door.
The Summary
- Thinking Machines Lab is closing a $5 billion round at a $40 billion valuation, led by Accel — more than quadruple its $10 billion valuation from last year's $2 billion seed round
- The startup, founded by former OpenAI CTO Mira Murati, has been hemorrhaging founding team members to Meta and OpenAI even as it ships new versions of its fine-tuning tool Tinker
- Investors are still writing massive checks to AI labs with OpenAI pedigree, but the talent war is making it harder to hold onto the people who actually build the models
The Signal
Thinking Machines Lab raised $2 billion before shipping a product. Now it's raising $5 billion more at a $40 billion valuation after shipping one tool that helps developers fine-tune language models. The math is simple: investors are betting on the team, not the product. Except the team keeps leaving.
Barret Zoph, Luke Metz, and Andrew Tulloch — all founding members — have departed for rival labs. That's not attrition. That's a vote of no confidence from the people who were supposed to be building the future. Meanwhile, Accel is leading a round that values the company higher than most Fortune 500 firms.
"The valuation quadrupled while the founding team quietly scattered to competitors."
This is the paradox of the current AI funding cycle. Capital is infinite, talent is scarce, and the two are moving in opposite directions. Thinking Machines raised its seed round on the strength of Murati's OpenAI credentials and the founding team's work training early ChatGPT models. But credentials don't build products. Teams do. And when your founding engineers leave for Meta and OpenAI, you're left with a brand and a balance sheet.
What Thinking Machines actually shipped:
- Tinker: a fine-tuning system for large language models
- Tinker v2 (released Thursday): improved AI coding performance
- That's it. Two years, $2 billion in capital, one product with one update.
The company tried to raise at a $50 billion valuation last year and talks fell apart. Now they're getting $40 billion at a time when their technical leadership is walking. Either Accel knows something the departed founders don't, or this is a bet that Murati can rebuild faster than the competition can scale.
The tool itself, Tinker, addresses a real problem: fine-tuning foundation models is still too hard for most developers. Making it easier is valuable. But so is every other infrastructure play in AI right now, and most of them aren't valued at $40 billion with a skeleton crew.
The Implication
If you're watching the agent economy take shape, pay attention to who's building, not who's raising. Thinking Machines is a case study in how far pedigree and timing can carry you, but also how quickly that advantage evaporates when the people who know how to ship leave for competitors. The talent war isn't just about salaries. It's about belief. Engineers go where they think the most interesting work is happening.
For founders in this space, the lesson is brutal: your first $2 billion buys you a chance to prove you can build something that matters. Your second $5 billion buys you time to rebuild after your team realizes you can't. If Thinking Machines ships something genuinely differentiated in the next 12 months, this valuation will look prescient. If not, it's just another data point in the history of AI hype cycles.