When Talent Walks: The Fragility of “Unicorn-at-Birth” AI Startups

The departure of multiple co-founders from Thinking Machines Lab is raising uncomfortable questions about a growing class of AI startups: companies valued at billions before a product ever ships.

Founder and CEO Mira Murati, formerly CTO of OpenAI, announced this week that co-founder and CTO Barret Zoph had left the company. Within an hour, OpenAI confirmed Zoph’s return, alongside fellow Thinking Machines co-founders Luke Metz and Sam Schoenholz. Based on reporting by Techcrunch.

The exits follow another high-profile departure. Co-founder Andrew Tulloch left Thinking Machines in October to join Meta. Together, these moves represent the loss of several founding technical leaders less than a year after the company’s formation.

Thinking Machines raised a $2 billion seed round last summer at a reported $12 billion valuation, with backing from Andreessen Horowitz, Accel, Nvidia, AMD, and Jane Street. At the time, the company was widely described as a “talent-first” bet: a concentration of elite researchers from OpenAI, Meta, and Mistral AI assembled before any public product roadmap. See here the founding team of Thinking Machines Labs.

That framing now cuts both ways.

In traditional startups, early departures are often survivable because value accrues to product, customers, or revenue. In unicorn-at-birth AI startups, valuation is frequently anchored to people. When those people leave — especially co-founders — the asset itself is diminished.

Talent mobility is not new in Silicon Valley, and OpenAI itself has seen senior figures depart and regroup elsewhere, including John Schulman, who left OpenAI for Anthropic before joining Thinking Machines as chief scientist. What’s notable here is timing. Thinking Machines has yet to ship a product, establish platform lock-in, or demonstrate defensible differentiation beyond its founding team.

Murati has appointed Soumith Chintala as the company’s new CTO, a respected figure in the AI community. But the episode underscores a structural risk facing early-stage AI ventures that scale valuation faster than execution.

TheMarketAI Take

Some AI startups are being valued less like companies and more like temporary concentrations of human capital. When that capital disperses before code hardens into product, the valuation thesis becomes fragile.

Until a startup converts talent into shipped systems, customers, or proprietary infrastructure, it remains exposed to the gravitational pull of incumbents with deeper platforms and fewer execution risks.

The question for investors isn’t whether elite researchers can build something remarkable. It’s whether a company can outgrow the résumé stack it was priced on before that stack starts to move again.