AI neolabs are raising huge sums. Most have little to show for it
The “neolab” era began in June 2024 with Ilya Sutskever’s Safe Superintelligence (SSI). Radical Ventures counts 29 neolabs founded in two years, which PitchBook says have raised a combined $23bn. By The Economist’s count fewer than 35% have released a commercial offering, and only slightly over half have released any code, often research demos. SSI, last valued at $32bn, says it won’t ship a product before “superintelligence”; Nvidia added $5bn after SSI said in July it had research worth scaling, taking its total raised to $8bn. Mira Murati’s Thinking Machines Lab raised $2bn at a $12bn valuation, launched Tinker and its Inkling model, and reportedly has annualised revenue above $200m while targeting about a $40bn valuation. Reflection AI (open-weight Beam model, $800m from Nvidia) is reportedly in talks on another big investment or a sale. Typesafe’s router model Jev is the only neolab model in OpenRouter’s top 20; it raised $40m before launch and $870m this month, though copycats have appeared. Nvidia has made 32 investments in startup model-makers, more than anyone, and AMD paid $8bn for Fei-Fei Li’s World Labs in September. The counterweight: OpenAI and Anthropic also took years to commercialise, so patience may pay; the article’s concern is that, unlike biotech, neolabs have no milestones and need to show results more often. Revenue and valuation-target figures are reported or from people briefed, not disclosed accounts.
Why it matters
This is Dealroom’s core territory: a defined cohort (29 labs, $23bn) of AI startups whose valuations far outrun products, with Nvidia as the dominant strategic investor. It frames the question LPs and VCs are asking about AI rounds right now, and the commercial-release ratio is exactly the kind of signal Dealroom can track across the cohort.
Executive takeaways
- Cohort scale: 29 neolabs in two years, $23bn raised (PitchBook/Radical Ventures); fewer than 35% have a commercial product.
- Valuation without revenue: SSI at $32bn with no product planned; Thinking Machines reportedly eyeing ~$40bn on a reported $200m+ run rate.
- Nvidia is the kingmaker: 32 startup model-maker investments, plus $5bn into SSI and $800m into Reflection—vendor financing risk worth watching.
- Exits are starting: AMD’s $8bn World Labs deal and Reflection’s reported acquisition talks show strategic buyers are a real outlet.
- Product traction can be fast but fragile: Typesafe’s Jev went from launch to an $870m round, yet copycats followed within weeks.
- The Economist’s verdict is “patience, but prove it”—it does not predict a bust, and notes OpenAI and Anthropic also commercialised slowly.
What The Economist may be missing
Likely thin: a full list of the 29 labs with round dates, lead investors and valuations; any European neolabs; down-round, secondary or structured-term dynamics; and how much of the $23bn is compute credits or vendor money circling back to Nvidia. It also doesn’t separate primary cash from headline valuation targets.
