Thinking Machines' $2 billion seed prices talent, not product

Mira Murati's lab raised $2 billion at a $12 billion valuation five months after launch and with no product. Early AI capital now buys frontier teams.
Thinking Machines Lab has no product and no revenue, and on 15 July it closed a $2 billion seed round at a $12 billion valuation. Mira Murati, formerly OpenAI's chief technology officer, launched it in February with about 30 researchers and engineers, most of them from OpenAI, and with OpenAI co-founder John Schulman as chief scientist.
At the seed stage, AI capital now prices people rather than traction. A seed round used to buy time to find product-market fit. This one buys a frontier research group and the compute to keep it busy, at a valuation that used to require years of revenue. The cap table explains part of the price: Nvidia, AMD, Cisco and ServiceNow are among the investors, and chip makers invest so that the next lab trains on their hardware.
Early money is concentrating, as it did in model builders in 2022, in a handful of teams from three or four labs, and the cost of senior researchers rises with it. Returns on rounds like this depend on the next round or a sale. The same week showed what a sale can now look like. Google paid about $2.4 billion to license Windsurf's technology and hire its chief executive, and Cognition bought the rest of the company, an exit built as a license and a hire rather than an acquisition.
Two markets are forming for founders. Those coming out of frontier labs raise at prices once reserved for public companies and inherit expectations to match, while everyone else competes for the same researchers against packages they cannot pay. The better ground is applied work, where a small team with deep domain data beats a larger research group, and the tools the new labs need from day one: evaluation, data pipelines, training infrastructure and the open-source ecosystem Thinking Machines says it will feed. Windsurf carries a warning as well. When a buyer takes the leaders and a license, the remaining team and shareholders live on what is left, so change-of-control terms matter more than a year ago.
Talent rounds stand or fall on what the talent ships. Murati said the first product arrives in the coming months with a significant open-source component, and delivery on that timeline, with customers paying for it, is the first proof. The next seed rounds for teams leaving OpenAI, Google DeepMind, Anthropic or Meta are the second. If prices above $1 billion before any product become routine, lab founders are a new asset class, and further license-and-hire deals would confirm that large platforms prefer to buy teams rather than companies.
Early-stage AI is splitting into research teams priced like companies and companies priced on revenue. Fanaura sides with the second group, where patient capital is paid back by customers rather than by the next round.
Building an AI-native company? Apply in five minutes. →
More from Fanaura
Building evaluation or training tools that new AI labs need from day one? Send us the project.
Fanaura invests its own capital and the capital of its limited partners (LPs), usually through a separate SPV for each deal. If we see a fit, we get in touch.
Get the news by email
One short note each weekday. No marketing, and you can unsubscribe at any time.