In the VC world, there are the rare founders who carry a certain mystique about them.
An extraordinary technical talent, or an especially hungry business savant, who appears at the perfect time in a boom cycle is able to subvert the normal mathematical rules of valuation.
Take the founder of Ceramic, a stealth startup I reported on this week. From her research with John McCarthy, one of the pioneers of AI, to her role as VP of engineering for AI at Google, CEO Anna Patterson’s résumé reads like the platonic ideal of a technical founder.
So it was no surprise that NEA, Ceramic’s lead investor, priced the startup at a $60 million post-money valuation in its first funding round, according to a person familiar with the deal. Though Patterson’s experience might be unique, the lofty valuation for a company that’s still largely in the R&D stage is not.
World Labs, a startup founded by “godmother of AI” Fei-Fei Li, reportedly rose to unicorn status in less than four months backed by Andreessen Horowitz, Radical Ventures and, again, NEA.
You don’t need a crystal ball to guess Safe Superintelligence, the company founded by former OpenAI founder Ilya Sutskever alongside co-founders Daniel Gross and Daniel Levy, will command similar interest.
Dogfighting among VCs vying to access premier founders and the hottest AI rocketships is common. A combination of AI exuberance, fear of missing out and shiny new mega-funds have created an environment where pre-revenue AI companies can almost set the price.
“There are many more preemptive rounds happening before the milestones are hit. You’re effectively paying ahead of where the company is,” said Rak Garg, a partner at Bain Capital Ventures who invests in infrastructure and AI.
For small or midsized early-stage VCs, that means often having to walk away from the bidding war altogether.
All or nothing
When seed VCs start looking seriously at a leading a deal, they will typically have a check size in mind as well as an ownership target ranging from 20% to 30% as the first institutional money in.
But problems are arising more and more where the founding team is so desirable that seed VCs have only a few choices:
- Devote a huge pile of cash to a single investment.
- Concede ground on target ownership, sacrificing potential upside on their fund returns if the startup succeeds.
- Walk away.
The third option was one that several investors I spoke to had to go with in the scorching hot seed round for large language model safety startup Haize Labs. For Haize, founded less than seven months ago by three AI researchers fresh out of Harvard Yard, the term sheets kept creeping higher and higher on valuation.
The founders eventually went with General Catalyst, with significant checks from a few luminary angels, at a $100 million post-money valuation—an extraordinary price for the first institutional check into a company.
“A lot of people are willing to let themselves go higher on valuation because of the opportunity cost of not doing it,” said Vignesh Ravikumar, a partner at Sierra Ventures focused on enterprise SaaS.
Like in any boom cycle, some of these bets will pay off— but many more will not. Founders and VCs are taking massive gambles on startups that either have little-to-no revenue or sales coming from businesses experimenting with AI that may not stick with the investment.
Rocky revenue
Many VCs are investigating the origins of AI software companies’ revenue sources more closely in due diligence because it’s proving unreliable and hard to retain. Aditi Maliwal, partner at Upfront Ventures, said she’s asking vertical AI companies who the buyers are of their software: “is it the CTO, or is it the AI innovation team? And what’s the churn?”
That combination of experimental or month-to-month revenue contracts and extremely high cash burn makes many VCs extremely wary.
Ethan Batraski, a partner at Venrock investing in developer and data infrastructure and open-source software, is one of them.
He described the cash burn profile of a typical early-stage vertical AI software play: “It’s often 20 to 25 heads, so assume they’re $25,000 to $30,000 per head per month, fully loaded.” Add to that a couple more millions per year on hosting Meta’s Llama, calling OpenAI or Anthropic, or using Amazon Bedrock, he continued.
“They’re burning on both ends of the candle, and they don’t have products that are generating high margins, and so the VC dollars are scoring right through,” said Batraski.
Featured image by Julia Midkiff/PitchBook News
Learn about our editorial standards.