Generalist venture capitalists didn’t use to pay close attention to the niche area of artificial intelligence known as generative AI.
In recent months, however, the growing buzz surrounding the technology has reached a fervor the likes of which the VC market hasn’t seen in years.
The release of tools like DALL-E 2, Stable Diffusion and ChatGPT has captured a wide array of investors by demonstrating that generative AI is now ready to be applied to various commercial applications from advertising and law to writing software code.
Generative AI has been around for years. But in the last 18 months, the language models that power most tools in the field have become stunningly good at creating art, writing and making other content with a human-like touch.
Access to these innovations had been limited until late summer, when OpenAI and Stability AI made their so-called foundation models available to developers at an affordable price or in an open-source environment.
Since then, the number of existing and new companies building products on top of these technological capabilities has exploded, and VCs are taking notice.
“My peer group of AI-focused investors have never been as busy,” said Sandhya Hegde, a partner at early-stage firm Unusual Ventures. “There are many talented founders trying to build companies using this technology.”
Seed deals explode
Hegde said that Unusual Ventures had backed a couple of such startups recently, and she estimated that over 50 other generative AI companies raised seed funding this past fall.
In 2022 alone, investors bet at least $1.37 billion on this crop of VC-backed companies across 78 deals—almost as much as was invested in all of the previous five years combined, according to PitchBook data. That’s a rare bright spot in a year that saw deal values slump in most categories.
Seed investors are eager to back generative AI companies in large part because they see what large valuations these companies could get once they graduate to their Series A stage and beyond, said Jeremy Kaufmann, a partner at Scale Venture Partners. Scale led a $10 million Series A for Regie.ai, a startup that automates blog posts and sales emails.
In December, Runway, a developer of image-to-text and image-to-image generation tools, raised a $50 million Series C led by Felicis at a $500 million valuation. The company has annualized revenues of about $1 million, according to a Forbes report, indicating that investors are willing to assign generative AI companies frothy valuations that were commonplace in 2021 but are now considered exorbitant for most VC-backed companies.
Other recent deals include a Series A for copywriting company Jasper, which garnered a $1.5 billion valuation in October. That same month, Stability AI closed a round led by Coatue, Lightspeed and O’Shaughnessy Ventures at a reported $1 billion valuation. In November, Descript, an audio and editing company, was valued at $575 million in a deal led by OpenAI Startup Fund.
Most of the seed-stage startups being funded have just a semblance of a product and only a few users. But investors are willing to pay high prices for Series A and later stages because many believe those companies will attract customers quickly as soon as the product is ready to go to market, Hegde said.
Despite her excitement about the sector, Hegde acknowledged it’s too early to tell which new companies will catch on with customers or whether the initial users will stick with existing products.
Ease of build worries some investors
Much of the hype among VCs has focused on the rapid adoption of OpenAI’s ChatGPT bot, which gained 1 million users in the first five days after its release in November.
“The only [deals] that people were doing quickly in the fall were generative AI startups, and nothing else was getting done,” said Vanessa Larco, a partner at NEA. “The question is, does enthusiasm translate into Series As, Bs, Cs and Ds?”
In the meantime, she said she expects a lot of “noise” in this sector because generative AI tools are relatively easy to build with the help of one of the foundation models.
That’s a cause for concern for James Currier, co-founder of the seed-focused venture firm NFX.
Currier said that he had already invested in four generative AI startups before the market heated up, but he has since backed off of additional investments.
“If you look at other major technological shifts like the internet, mobile or crypto, there [was] a lot of skepticism,” he said. “True believers had an opportunity to create a startup in those areas when others didn’t believe.”
But with generative AI’s usefulness becoming apparent to everyone, Currier said, he’s concerned that hundreds of similar companies that are hard to distinguish from one another could be built.
Not all investors worry that it will be hard to create a differentiated, defensible product in generative AI.
“I think just about every firm already has one investment in this space,” said Oana Olteanu, a partner with SignalFire. Last year, her firm backed Tome, which uses natural language processing for legal services. Olteanu said that she also led SignalFire’s seed investment in an undisclosed pre-product generative AI startup.
Learning from past exuberance
Olteanu said that she believes that foundation models could eventually become a commodity, and most of the value will be realized by the companies that customize the models for specific use cases.
It may be years before it becomes clear what type of companies—incumbents such as Google, startups, or foundation model developers—capture the majority of generative tech’s upside. Meanwhile, some investors say the widespread belief that the VC market was overheated in 2021 is helping them keep a cool head when investing in generative AI.
“I think the [current] market conditions are overriding the typical hype behavior now,” Olteanu said. “I believe many other VC firms, not only us, are very careful and diligent in analyzing the opportunities.”
Currier, who is currently spending all his time on AI opportunities, said investors aren’t paying elevated prices for generative AI seed deals. “In some cases, valuations are lower because the founders are desperate to get going,” he said. “They see how fast everything is happening.”
Correction: An earlier version of this article incorrectly described SignalFire’s portfolio company Tome, also known as Electric Tome Co., a startup that uses AI for legal services. A separate startup with a different business model also does business as Tome. (Dec. 29, 2022)
Featured image by Je-nar/Shutterstock