After pouring billions into AI startups with growing enthusiasm, venture capital has found a new favorite category. Rather than a product or a market, it bets on the research capabilities of elite scientists and small expert teams. It centers on research labs valued at a billion dollars or more that have little to sell. Such labs have earned a new name: “neolabs,” and the numbers now attached to them are getting hard to ignore.
What exactly is a neolab?
According to Menlo Ventures partner Deedy Das, a neolab is a pre-revenue startup working on long-term AI breakthroughs, usually carrying a valuation of $1 billion or more. His tally in February was 50. By May, the list had grown to 63.
The idea is simple. Instead of chasing immediate profit, these startups prioritize long-term research and building new AI models. Investors are not buying a roadmap or a customer pipeline. They are buying a team, a research thesis, and a chance at a breakthrough. Some labs raise hundreds of millions, or even more than $1 billion, before they launch a public product.
The numbers behind the hype
The scale is vast. Das’s May count put the collective valuation of those 63 labs above $300 billion, on roughly $48 billion raised. Other trackers are more aggressive. Paul Bonnet of 20VC counts 102 neolabs that have raised $70 billion over three years, with a combined valuation near $320 billion and less than $1 billion in revenue. A separate dataset puts $94.3 billion into 119 labs since 2024, and only 16 of those labs disclose any revenue.
Different methods give different totals, but the picture is the same: a very large amount of capital sits in companies that have barely begun to earn.
Some of that money has produced visible, public output. Thinking Machines Lab, which raised $2.0 billion, has released a training API and an open-weight multimodal model. Others are far quieter. Safe Superintelligence, also at $2.0 billion, has put out no public model, product, or research system. That is the range investors are underwriting, from shipping labs to labs that have shown the market nothing.
Why the checks keep coming
The question arises: if neolabs have so little to sell, why do investors keep funding them? The simplest explanation is FOMO, the fear of missing out. After the phenomenal success of OpenAI and Anthropic, the fear of missing “the next one” drove an extraordinary wave of capital formation, and nearly every major VC firm now backs several neolabs, many of them yet to earn meaningful revenue. Funds that missed the early bets on OpenAI and Anthropic see neolabs as a way to get into the next wave of frontier startups, in a market where the best deals are hard to access.
Deal structure is further fueling the boom. Tranched rounds let different investors pay different prices for the same company. It helps founders raise record sums at very high valuations before they have a product. Baseten, for example, raised $1.5 billion in two tranches, one at an $11 billion valuation and the other at $13 billion. The headline valuation can look steadier than the economics beneath it.
On top of that, compute costs explain part of the size. Building an AI lab takes enormous numbers of GPUs, so these companies must raise huge sums at the start to do serious research. A lab cannot start small and grow into its cost base the way a software startup can.
The underwriting problem
One of the trickiest questions is how to value a company with no revenue. The blunt answer is that nobody has a settled method. For the 103 labs in that dataset without disclosed revenue, the market relies on proxy signals that may or may not translate into commercial success: how quickly valuations are marked up, researcher headcount and pedigree, compute under contract, and usually private evaluations.
Proxies like these measure sentiment and credentials, not commercial traction. A star researcher can raise a fortune, and a rival can raise a fortune on the strength of a different star. When the signal is reputation, valuations can climb together without any new fact arriving about the technology.
The math that makes skeptics nervous
Bonnet’s analysis shows how steep the climb is. For investors to see a 10x return, the neolab cohort would need roughly $500 to $900 billion in implied revenue. At the low end, that is 17% more than OpenAI and Anthropic produce today. At the high end, it is about seven times more.
In other words, the cohort has to generate more revenue than OpenAI and Anthropic combined produce today, while most of its members have yet to find their first product. A few winners could carry the portfolio, as in any venture strategy. But with this many labs chasing similar breakthroughs, the odds of each one becoming a winner shrink. If enough academics cluster in one area, several neolabs could end up aiming at the same market.
History adds a note of caution. Commentators have compared the moment to the dot-com era, when many companies folded before they brought in any revenue. The comparison is imperfect, since these labs hold real research talent, but it is a reminder that a strong story is not a business.
The counter-argument
Defenders of the model have a serious case: frontier AI research is expensive, slow, and uncertain, and the field’s biggest winners began as research labs. OpenAI itself began as a research lab and later became one of the world’s most valuable startups. If some researchers believe today’s dominant models are approaching a ceiling, funding alternative approaches is a rational hedge.
There is also an alternative exit path that does not require revenue at all. Frontier labs valued at $1 trillion or more can use their balance sheets to acquire or acqui-hire neolab talent. A lab that never ships a product can still be worth a great deal to a buyer who needs its people. A recent example points that way: Safe Sign Technologies was acquired by Thomson Reuters twenty months after founding, as that company’s first pre-revenue acquisition in 174 years.
The founder behind that deal made the case that real value can exist well before anyone measures it. That may be true for some labs. The question is how many.
What this means for CXOs
For enterprise leaders, the lesson is less about the investors’ returns and more about vendor risk and signal quality.
First, funding is not traction. A lab with a billion-dollar valuation may have impressive researchers and no product you can deploy, so evaluate on what ships, not on what was raised.
Second, expect consolidation. If acqui-hires become the main exit, some labs will be absorbed into larger players, and their roadmaps will change with them. Any dependency on a young lab’s model needs a contingency plan.
Third, watch the talent market. A Stanford computer science professor said this is the most turnover he has seen in academia in his decade of teaching. Neolabs are drawing researchers out of universities and incumbents alike, which shapes who builds the next generation of tools.
The bottom line
The neolab boom is more nuanced than a simple bubble-or-breakthrough debate: it is neither a clean bubble nor a clean breakthrough. It is a concentrated bet that a handful of research teams will make the revenue numbers work years from now, financed at prices that assume many of them will. The arithmetic asks a great deal, the valuation methods are thin, and the exit routes depend on a few giant buyers.
None of this means the labs will fail. It means investors are paying today for outcomes that have not been demonstrated yet. For readers who work in technology, the practical stance is to take the science seriously and the valuations cautiously, and to keep asking the question the whole category invites: what has this lab actually shipped?
