Recommended: Listen to this story

The Story: Early-stage AI companies are being funded left and right, but few of them have made it to the growth stage yet. Most AI startups are in the nascent stages, and only time will tell if growth stage venture will welcome the best of them in the coming years.

The Expert: For more insight into how a growth stage VC sees the rise of AI, we sat down with Jill Chase, partner at CapitalG, Alphabet’s independent growth fund. Founded in 2013, Capital G’s focus is on larger, growth-stage technology companies, and invests with an emphasis on profit rather than only strategically for Google.

Jill, which AI startups are poised for a breakout year in 2024? Whether those are household names already or unknowns with a revolutionary product.

Jill: For sure. It's a great question. I think we're gonna see a lot become clear over the next couple of months and the next year because there have so many massive funding rounds into really exciting companies. And a lot of those companies as a result are taking the next few months to really hunker down, build their models, build their products, and then we'll sort of see them come out with big releases over the next quarter or so.

So one I'd love to talk about is a company called Magic.Dev. They are building an AI software engineer. We actually led an investment in them almost a year ago and as described they are one of those companies that has been hunkering down building their own foundation model to power an AI software pair programmer. So very excited about that. You know, I think we'll see some exciting stuff from them over the next few months. One to keep an eye on.

On the other side of that coin, which AI startups do you think will falter this year and experience setbacks?

Jill: I'm gonna give you a little bit of a non-answer because I think it's a huge question. The interesting thing about this category is it's incredibly compute-intensive to build some of these big foundation models, and they're still binary bets.

Companies like Adept AI and Character.ai and Inflection AI—we're putting hundreds of millions of dollars into these businesses as VC as a whole, and they're going away and sort building these large foundation models to power either consumer enterprise use cases, and the bet is that these models will be powerful enough to really be game changers on the true use case side of things. That's a huge bet. Building these models is not easy and I think we'll get a of information next year as to whether some these companies were able to build truly game-changing models that result in game-changing end applications.

It's hard to predict who's going to do a good job of that and who's not, but I am excited to see. I think we'll get a lot information on that over the next year.

Companies like Microsoft, Box, and Adobe have implemented unique cost structures for their generative AI features. Microsoft charges a subscription, Box offers users 20 queries per month but they pay more if they go over, and Adobe has free and paid plans. Where do you think AI companies will find the most success in turning a profit in the coming years? By viewing generative AI features as selling points for existing services, or viewing them as new products?

Jill: Yeah, it's a really good question. And I think it comes down to there's this answer that sort of underlies the business model of AI companies and the question that I think is really relevant is like what are the unit economics of leveraging some of these large language models? And then how do you parlay that into a business model for whatever you're offering.

As an example what you saw with OpenAI’s recent demo day, you saw sort of a lot of the cost of using their LLMs go down quite significantly. And what that does is it allows for companies building on top of these APIs or companies building infrastructure to actually have more compelling unit economics for their own businesses because they're able to leverage these APIs in a more cost effective way.

I had that question at the start of this year. I think it was actually the prediction I made at the beginning of 2023, but I think we've seen that the cost of training these models and using these models at the inference player has gone down quite dramatically. And so I honestly think it's going be easier for these businesses to make money than it seemed like it would have been at the beginning of this year.

Jill, you focus on growth-stage investing. AI startups, however, usually only receive funding in the early stages. What have you seen the most successful AI companies do to continue receiving funding into the growth stage?

Jill: I think that this category is so early on in its evolution and we're still a little bit in the technical error of these AI startups. And the interesting aspect that's unique about this category is again the compute intensity of the businesses if they're building their own foundation models.

And so what you're seeing from a fundraising dynamic is businesses, especially those that are building their own models, will raise capital. You know a seed or a Series A round, oftentimes with just an idea and an exceptional team. But it almost looks like a traditional sort of series B business from a valuation perspective.

Then, on the back of a demo six months later, they'll be able to raise hundreds of millions in capital again oriented towards building these large foundation models. But the bet is honestly still binary in many cases because oftentimes you're effectively funding sort of a bet on building a model that's gonna work, but you still don't know a) is that model gonna work? and b) is the use case that they're building for gonna be something that people want and something that people are willing pay for?

It's interesting how it relates to the growth investing market because growth investors are seeing Series Bs or even Series As or Cs that are at rounds that look like, you know a billion dollars plus from a valuation perspective, but the companies themselves from a maturity perspective look a lot more like Seeds or Series As.

And so I think a lot growth investors are having to make hard decisions about whether that's a profile of risk that they're just willing to underwrite or that’s something that still doesn't feel comfortable and folks are trying to wait it out and see when there's real scaling from a revenue perspective.

At CapitalG, is that a risk you are willing to take? Investing in a company with a really high valuation early on in the series A or B stage, but that shows signs of being a growth stage company due to that high valuation?

Jill: The way we've tried to approach it, and an example here is you know, we led the Series A of Magic.Dev at the beginning of the year, and I think what we see as an exciting way to get involved in the space is to just go earlier because the reality is in between some of these sort of quote “seed or Series A rounds” and then the quote “Series B or C rounds,” like not a whole lot of the business has been de-risked.

If that's going to be true and we're already making a bet on the team and the market and the excitement, then why not go a little bit earlier and partner with some of these folks from the beginning particularly in categories that we think are, you know, incredibly exciting and can be truly revolutionized by AI.

And so that's sort of why we ended up leading the Series A in Magic. It's a category that we've been following for a very long time. We believe that AI will fundamentally change the way that folks do software engineering. We met Eric and the team at Magic, and we fell in love immediately. And we were willing to lean into that risk early on knowing that it would be a bit of a longer ride than it is with some of our more sort of growth-oriented investments that we make.

What’s your most controversial take? An opinion or prediction about AI that your investing counterparts disagree with?

Jill: I talk about this a lot. I think that the research aspect of AI is like severely under-discussed. I think that investors are tempted sometimes to sort of say like, “oh yeah, leave the research to the researchers and then you know we're going to focus on business models and use cases and all that.” I think what people you know, this is my hot take, I think what people are missing is that the research aspect of this field has like a direct relationship with what companies are going to be successful…

Research questions that are being actively worked on by some of the smartest people in the field have a direct relationship to the likelihood of success to some of these companies that are building on the infrastructure layer.

Venture capital is all-in on AI. We talk about it basically every day on Venture Daily. But when an entire industry pushes its chips in for one sector, there are inherent risks. Is venture over-committed on AI? What’s your biggest fear about AI that if it were to happen, it would make the innovation significantly less valuable?

Jill: I've been reflecting on this a lot because it sort of comes down to like I think most investors are having to make a decision right now, which is, do you sort of lean into the risk and lean into some of these higher-priced rounds because you believe there's something fundamentally different about the category, or do you kind of wait for things to pan out and believe it's a bit more like the internet bubble where there were only a few winners, but there was a period of craze where not a lot actually panned out.

I think in order to really lean into some of these rounds that are quite high priced, you kinda have to believe one of two things, or both. The first is that like the market for AI use cases is just bigger than anything we've ever seen, and that this technology is gonna unlock use cases that people are willing to pay for in a way we've never seen anything like it before.

Or you believe that companies can scale from a revenue perspective a lot faster because of AI than other businesses and so growing into their valuation will not be quite as hard.

The example there obviously would be OpenAI scaling to a billion in revenue over quite a short period of time. If you believe that other businesses can do similar things, then you're willing to pay a higher valuation sooner because it's not quite as much a concern (like when will this business grow into their valuation?).

And so I think that's what most growth investors are trying to get a sharp perspective on. We're doing that by studying some of the other periods of sort of mania and excitement that have happened over the past few decades, including the internet bubble, the mobile bubble, etc., and really trying to understand how is this similar or different?

Experts predict AI will be responsible for making more than 90% of all online content by 2025. Do you expect that consumers will embrace the AI takeover, or will there be a lot of pushback from people who fear how pervasive the tech is already becoming?

Jill: I think it will really depend on the different end category. Because I think in some categories it's kind of fine for AI to do the first version of content generation, and then to have folks come in and you know do the high value add work to edit that content or to figure out to make it useful for people.

And that's because the risk associated with AI generated content in some categories is lower. I think in other categories, obviously, there are more things, like healthcare or financials, or things like that. It's a lot more concerning to have AI generating content without like significant regulation or thoughtfulness on top of it.

I think it will really depend on the category. I think for most business use cases, I agree with that statement. Probably in five years, 90% of content will be AI generated. I think for some more highly regulated Industries or categories where the risk is higher for not having high human touch, then that number will be a lot lower.

*Stay informed about the three biggest stories in venture capital and tech news every weekday morning. 2-min reads only.