A new class of AI company is raising some of the largest checks in venture history on little more than a thesis and a famous team. Twenty-nine "neolabs" — startups founded by researchers who left established AI labs — have raised a combined $23 billion, according to a count by venture firm Radical Ventures cited by PitchBook. Yet fewer than 35% of them have released a commercial product, The Economist's tally shows.

The numbers describe an extraordinary bet on reputation over revenue. Just over half of the 29 labs have released any code at all, often only research demos, and the biggest names in the category have been explicit that products can wait.

The Sutskever Standard

The neolab era began in June 2024 with Ilya Sutskever's Safe Superintelligence (SSI). Within four months SSI had raised $1 billion, and less than a year later it was valued at $32 billion — all with no public product roadmap. The company has said it will not ship anything until it reaches "superintelligence."

Investors are unfazed. Nvidia added $5 billion to SSI in July after the lab said it had research worth scaling, taking its total raised to $8 billion. Mira Murati's Thinking Machines Lab raised $2 billion at a $12 billion valuation, has launched its Tinker toolkit and Inkling model, and is reportedly doing more than $200 million in annualised revenue while targeting a valuation near $40 billion. Reflection AI, whose open-weight Beam model was backed by $800 million from Nvidia, is reportedly in talks over another major investment or a sale.

One Working Model Among the 29

So far, only one neolab model has broken through to genuine market traction. Typesafe's router model Jev is the sole neolab model in OpenRouter's top 20 rankings — and the company raised $40 million before launch and another $870 million this month, even as copycats emerge.

Nvidia has become the era's kingmaker, making 32 investments in startup model-makers — more than anyone else. AMD, not to be left out, paid $8 billion for Fei-Fei Li's World Labs in September. The pattern echoes classic venture behaviour: enormous capital chasing unproven research bets staked on the reputations of star researchers.

The Case for Patience — and the Skepticism

The neolabs have a historical defense. OpenAI and Anthropic also took years to commercialise, and the counterargument is that patience may pay off. But the concern, as Dealroom's analysis puts it, is that unlike biotech — where milestones mark progress — neolabs have no waypoints at all, and they need to show results more often.

Whether frontier-AI reputations translate into revenue remains the open question. Until it is answered, $23 billion sits on research promise rather than product.

By Hamza Faisal