“Can investors make 10-figure AI valuations pay off in markets no one is sure exist?”
That was the title of a panel featuring prominent AI investors at last week’s TechCrunch Disrupt conference.
The discussion—which featured Sequoia‘s Sonya Huang, Greylock Partners’ Saam Motamedi and Benchmark‘s Miles Grimshaw—largely skirted the provocative question. Still, the panel’s name underscored an open secret among investors: Most of the AI startups grabbing valuations in the hundreds of millions have hardly any revenues to speak of.
Ask any VC, and they are likely to say that most Series A and beyond AI rounds are now valued on future promise, rather than actual sales.
Tomasz Tunguz, founder of Theory Ventures, estimates that there are fewer than 25 pure AI startups generating more than $10 million in revenue. Furthermore, he suggests that about 95% of generative AI companies have less than $5 million in annual recurring revenue.
Many investors say it is too early to expect meaningful revenues from generative AI startups. The technology went mainstream in November when OpenAI made ChatGPT available to the public.
“There’s not been a lot of time to develop a product and sell it,” said Tunguz.
Minimal revenues have not stopped investors from backing 74 generative AI companies at a valuation of $100 million or higher since 2022, according to PitchBook data.
Investor appetite for bestowing high valuations on nascent and unproven AI businesses comes as the rest of the VC ecosystem is going through a sharp slump.
The price of ‘user traction’
In many cases, VCs have justified lofty AI valuations by pointing to high usage rates of the products. Their hope is that free users will convert to paying customers.
In April, LangChain, an open-source library that helps software developers connect to large language models, raised a $10 million seed round led by Benchmark’s Grimshaw at a valuation of $45 million, according to PitchBook data. A few weeks later, Sequoia’s Huang reportedly came in with a preemptive Series A offer, valuing the pre-revenue company at around $200 million.
On the sidelines of the conference Sept. 20, PitchBook asked Grimshaw whether the valuation made sense, given that LangChain had no paying customers at the time.
“That’s OK. It’s all about user traction,” he said.
But banking on free users becoming paying customers can be risky, according to Sandhya Hegde, a partner with Unusual Ventures.
“Many high-growth, low-revenue startups have raised growth rounds based on the signup rates,” she said, adding that the problem is how many of these companies have extremely low retention of users.
Costly computing
VCs have also been writing large checks because AI companies can require lots of capital for pricey computing power.
"$One hundred million post-[money valuation] for a company that’s not ‘live’ yet—that’s crazy. But then you realize that the minimum they can raise is $30 million because it’s expensive to build what they are building,” said NEA general partner Vanessa Larco.
In general, these larger deals go hand-in-hand with high valuations. But Larco noted that some capital-intensive, pre-revenue AI startups are willing to accept higher dilution, which allows them to raise more money without inflating the valuation even further.
While many generative AI companies require a lot of computing power, not every AI company is expensive to build. For instance, infrastructure startups like LangChain aren’t training models and, therefore, have smaller needs.
Investors realize that backing AI companies at the prices they are currently is not without risks.
As Greylock’s Motamedi put it during the TechCrunch panel: “It’s a very exuberant market, and I think a lot of money is going to get lost.”
