Startup funding may have plunged so far in 2023, but one sector seems to have missed the memo: generative A.I. So far in 2023, valuations for buzzy artificial intelligence startups continue to soar. And despite the prices, venture capital investors are not only willing, but excited, to keep fueling the fire.
A host of A.I. startups have raised cash recently, including OpenAI rival Anthropic; A.I.-powered presentation startup Tome; and Typeface, which uses A.I. to generate marketing content for companies. The sector is capturing the interest of venture capital investors, even in an otherwise chilly private market. That’s perhaps one reason why the rise of A.I. valuations looks so startling: Generative A.I. “is where the outliers are happening right now—It’s even more pronounced because that’s juxtaposed not with a hot market, but a cooler market,” David Beisel, cofounder and partner at NextView Ventures, which has several A.I. portfolio companies, told Fortune.
Beisel likens the current moment for generative A.I., a transitional shift, to “a browser moment” or “a social networking moment.” And given one look at the data, he’s clearly not alone.
According to PitchBook data provided to Fortune, the median pre-money valuation for generative A.I. firms soared to $90 million so far this year, based on nine deals PitchBook has tracked through March 29, up from $42.5 million for 2022. Those big numbers are boosted by big deals for the likes of Anthropic and Tome, and early-stage A.I. startups addressing use cases like software, customer experience, and media generation are seeing high valuations. While it’s still early days in 2023, that data “suggests that investor momentum will be on early-stage startups in this new wave of A.I., rather than supporting late-stage vendors” and those using older, more classic models, Brendan Burke, a senior emerging technology analyst at PitchBook, who covers A.I., told Fortune. In fact, in a recent report on generative A.I., PitchBook analysts predicted that at a 32% compound annual growth rate (CAGR), the market could reach $98.1 billion by 2026.
Some of the bigger valuations are for startups in the so-called “foundation model" segment, like Anthropic, which reportedly recently raised $300 million at a roughly $4 billion pre-investment valuation. As venture investors note, those types of companies are cash-intensive because the models are pricey to build and train. Meanwhile, that category will likely have only a few winners compared to other parts of the A.I. space that might be less concentrated.
When faced with those hefty price tags, some VCs can see two sides of the same coin: “I would say yes to both questions: Is it justified, and are they too high?” Beisel quips.
His reasoning is that if investors believe we are at the beginning of this new era, they are thinking not just about how they can make a profit on any one investment, but "how do I get into the early big companies that will then set me up to...develop a reputation within this domain?" He likens it to investors in the early social platforms like Friendster and later iterations like Facebook: "If you as a VC had a reputation for investing in one of the earlier ones...the fact that you were in it helps you with getting into the next one with those entrepreneurs." In other words, being known as a top VC with all the big A.I. deals can create a pipeline for future A.I. startups.
Others like Gaurav Gupta, a partner focused on A.I. at Lightspeed Venture Partners, point out that for the large language model providers like Anthropic or OpenAI (and even Stability AI as an open source platform), the frothy valuations could look reasonable if you think about these companies as dominating a category in the same way that Google, Amazon, and Microsoft dominate the cloud. “They became the computing engine for everything out there and they generated these massive, massive businesses,” he told Fortune.
Still, it’s quite the gamble for VCs: “There's gonna be a lot of money lost,” Gupta says, “but a few players will win.”
VCs are also looking to invest outside of those top dogs, putting cash into smaller startups addressing consumer applications that use these models. From Gupta’s perspective on the ground, the “most active area” right now is in those application startups—types of companies that, for example, could make it easier to code or conduct legal research (Gupta points out Lightspeed portfolio companies Typeface and Tome). He says there’s lots of excitement around those types of companies, and rounds are getting done at “twice the valuation you normally would see” for the seed and Series A stages.
To Erin Price-Wright, a partner focused on early-stage A.I. investments at Index Ventures, some of the rounds getting done are “kind of crazy,” she told Fortune, but she believes that overall, the hype is warranted. She argues that in 10 years time, all software will have A.I. at its core.
Of course, there’s a fear of missing out, or FOMO, element fueling the prices. Startups that are early to the market and perform well are often the ones that succeed, and there’s a “race for people to identify those companies early on and put money in them and to be part of that wave,” Lightspeed’s Gupta says. “If you jump in too late, and you wait for too much validation, the prices could be even higher.” He notes that VCs may also be looking at what happened with the valuations and success of large language model companies like OpenAI as a justification: “It's very expensive to get in now, so VCs are…using that as maybe some level of proxy,” he suggests.
Nevertheless, Gupta makes his case for why the prices are worth it: With the recent release of GPT-4, OpenAI’s latest system, and other innovations in the space, he argues, it's “becoming more and more clear that all of these existing software categories can be disrupted by the tech.”
Not all A.I. startups are created equal
It can be easy to paint “A.I. startups” with a broad brush, but the data shows there’s a divide between which companies are getting those lofty valuations and which are, like other areas in tech, seeing a decline.
Per a recent PitchBook report on A.I. and machine learning broadly, “2022 challenged the highly funded AI vendors that are unlikely to drive the field’s future,” as “VC funding for the vertical declined more than it did for IT overall, falling 34.9% to $78.0 billion in VC deal value.” So far this year, global A.I. and M.L. median post-valuations for startups have been a mixed bag, depending on the stage: Venture growth valuations plunged so far in 2023 (based on only a handful of deals), while there have been incremental increases for other stages like seed and early VC (see chart), per PitchBook data.
“On the whole, A.I. valuations are still coming down for the bulk of legacy vendors that haven't adapted to the latest techniques,” adds Burke, like those who “commercialized the more classical machine learning techniques and those that use off-the-shelf models instead of developing their own.” He explains that some companies focusing on classic techniques have "come under pressure on revenue and valuation growth even as A.I. adoption has grown quickly" as they can "prove slower to adapt to new trends" compared to cloud providers like Google or Microsoft. He highlights enterprise A.I. company Iguazio’s acquisition for $50 million after previously being valued at over $167 million, and A.I. platform Dataiku’s down round, dropping from $4.7 billion Series E to a $3.7 billion Series F valuation, per PitchBook data, as examples.
It's creating a “big bifurcation in the market between innovators and legacy startups,” he notes.
“Not all A.I. companies can achieve high valuation premiums,” PitchBook’s Burke said, “only those with highly qualified founding teams and original foundation model innovations, which is a very narrow subset of the A.I. industry. Other companies are being held to the same standards as software startups.”
‘I don’t know how much crazier it could get’
Many VCs say that in this tougher environment, deal prices and terms tend to swing in their favor—meaning VCs have more leverage to get better deal terms and a lower valuation than startup founders. But for “anything that's A.I. or generative-A.I.-related, VCs are price takers,” says Beisel.
While it may be too late for VCs to get in on the ground floor, at least at a reasonable price, Beisel believes it's still early enough in the shift for there to be ample opportunities moving forward.
For Index Ventures’ Price-Wright, 2023 is a “rubber meets the road” time for A.I. startups, with more “scrutiny around, are you building a real business? Do you have a business model that makes sense [at] scale?... These questions are going to start to get asked, I think, more and more.”
The problem for some VCs, as is always the case in venture, is what happens if their big bets don’t turn into the next big thing. Gupta notes that many of these A.I. companies are raising outsize seed rounds—not just something like $3 million, but “some are raising $10 million or $20 million out of the gate,” even without products or customers, he says. The danger there, of course, is that “if you invest at these high valuations and you're writing larger checks, then you're gonna regret if you hadn't picked the winner” because “suddenly you have a hole to fill in your investment strategy.”
But will generative A.I. valuations continue to get frothier? “I don't know how much crazier it could get,” Beisel says, and other VCs like Gupta and Price-Wright see valuations likely moderating in the next year. But the shift, Beisel notes, is being propelled by large platform companies, as in Microsoft’s partnership with OpenAI and Google’s Bard project (as well as Meta). That will spur “more and more headlines,” notes Beisel, and persistent headlines “[spin] up VCs.”
And right now, he adds, the “majority” of deals that come across his desk have the word “A.I.” in them.