From Abundance Summit: How AI is shaking up the start-up ecosystem

Note: Transcripts are automatically generated from episode audio, and are not fully corrected for spelling, grammar, and formatting.

RANA EL KALIOUBY: Hi there. It’s Rana el Kaliouby, host of Pioneers of AI. And we have something a little different for you today.

If you’re a weekly listener of the pod, you’ve probably heard me talk about investing in AI. When I’m not guiding you through this AI frontier, I spend a lot of my time as an investor.

I am the Managing Partner at an early stage venture fund called Blue Tulip Ventures, focused on investing in human-centric AI startups.

I am super passionate about identifying incredible founders of AI startups and supporting them on their entrepreneurial journey, but it’s not something we talk a ton about on the podcast. We’re changing that today.

Last month, I got to sit down with two leading investors at Abundance Summit – an annual event hosted by Peter Diamandis, where entrepreneurs, investors, and CEOs come together to learn about technologies like AI and create positive change. It was my third time and I LOVE it! One of my highlights was championing the teen program!

On stage, I was joined by two venture capital investors: Anj Midha who is General Partner at Andreessen Horowitz and Dave Blundin, Founder of Exponential Ventures.

We talked about the crazy mazy valuation of AI companies, how AI startups are scaling fast with less capital, what that means for venture funding, and where investors should place their bets in AI.

I am so excited to share this with you, so let’s get to it!

Copy LinkWhere the next trillion dollar AI company could emerge

EL KALIOUBY: We have a lot to talk about, I’m gonna dive right in. It’s really clear to me that the next trillion dollar company has to be AI. It has to be AI driven. But what does that look like? Dave, I’ll start with you.

DAVE BLUNDIN: Me first? Okay, sure.

BLUNDIN: The obvious ones are customer service, the AI voices, the avatars that you’re seeing today. Anj will tell you all about this, but they’re about to get incredibly engaging and incredibly good. There’s $10 trillion of customer service phone calls that are gonna move to AI imminently, so that alone would create a trillion dollar value company. To get to a trillion dollar value, you need about 300 billion of revenue, maybe a hundred billion on the bottom line, like Apple or Google.

So I think out of a $200 trillion add to the economy, there are lots of opportunities for that.

EL KALIOUBY: Anj, what do you think?

ANJ MIDHA: I don’t think I have a more quantitative answer than you, but my mind always goes to the stack, which is you’ve got sort of core infrastructure, then you’ve got what we’re seeing as the model layer, and then we’ve got the applications.

And if you think about the last few trillion dollar companies we’ve seen, Nvidia is obviously a great example of starting at the bottom right there. And so I think you’re just gonna see another trillion dollar company emerge at the next layer of the stack at the model level, and then at the application level.

The one I’m really excited about right now, which has come sooner than I expected, which we’re talking about backstage, is the computing interface — what we call some version of a model that can take actions on your behalf, right? Agents and so on. The reality is that we’re still kind of mediating all of them through smartphones or regular MacBooks and so on, like computers.

EL KALIOUBY: I think that’s actually very interesting because the smartphone is not an AI native interface. So there’s gonna be an opportunity for somebody to build this AI native interface, and I think you have a point of view on that.

MIDHA: Well, I think if you look at the history of computing, it’s basically there are these two lineages, right? You have the history of intelligence and the history of interfaces. And so often what happens is you have a breakthrough in the intelligence — something going from a mainframe to a PC.

And then what happens is you have something like a terminal replaced by the GUI. And so where we are now is basically we’ve had this incredible breakthrough in intelligence, which is generative models, right? But we haven’t had an interface catch up yet. What I’m watching for is an interface that captures daily context about your life.

Because a smartphone is basically living in your pocket most of the day.

It understands just a tiny sliver — it’s like drinking through a straw. The amount of context it has about your life is tiny. So some version of a wearable that sees what you’re seeing, hears what you’re hearing, knows where you’re focusing, has context about your life — that’s where I think computing is going. And I think it’s hard to not see a trillion dollar opportunity there.

Copy LinkWhy AI native interfaces may replace the smartphone

EL KALIOUBY: Yeah. Conversational, perceptual. And I would argue empathetic, but we’ll come to that. So CB Insights published a report a few months ago showing how AI unicorns are scaling with half the team size and in a fraction of the time compared to traditional SaaS startups.

I’m already seeing this in my portfolio companies. Just yesterday we had Brandon Foodie on stage. His company’s a perfect example of that. So as investors, how do you think that kind of changes the calculus around scaling risk, capital efficiency? Also, if companies need less capital upfront, how does that shift the venture model?

BLUNDIN: Yeah, well first of all, this is by far the best era for seed stage investing I’ve ever seen in my life. Been doing it for 20 years. I was very fortunate to found my first company right as the internet was taking off. And that kind of changed the course of my life. But since then, since basically 2002, nothing like this has happened until now.

And you’re exactly right.

I think the window to invest in this started maybe a year and a half, two years ago — the real sweet spot — and you have maybe a couple more years. And so if you don’t get on those cap tables in that window of time, the next trillion dollar companies are being formed right now, or were formed maybe six months to a year ago.

And if you’re not involved in them, you’re missing the window. This is it.

Copy LinkHow AI is changing startup capital needs and venture math

EL KALIOUBY: Anj, what’s your point of view?

MIDHA: I don’t know if I have a steady state conviction yet on whether — I have no idea what venture will look like two years from now. I think it changes every six months.

But I’ll tell you, two or four years ago when the Anthropic guys gave me a call and said they wanted to leave OpenAI and were thinking about starting a new lab, I said okay, how much do you need to get started? And they said they thought they could get by with five. I said okay, 5 million bucks, I think we can round it up, I’ll wire it next week. And Dario said, no, Anj, we need 500 million. I said, okay, that’s a little bit different. And so we then went and raised a seed round of a hundred million, which is still an extraordinary amount at the time. But what that taught me is, at least for the era of businesses that you would call at the bottom of the stack — sort of foundation model infrastructure businesses — they actually look more like biotech companies than they do traditional software.

Right. Because if you think about the shape of development there, it’s a group of scientists usually de-risking core research within an industrial or academic lab. There’s only a handful of talent globally who can contribute to that. It’s very capital intensive, but it’s winner take most. So the minute you have the state of the art, and you put it out, you put an API behind it, suddenly the path to a billion in ARR is just overnight.

It’s very similar to drug development in that sense. Product companies, on the other hand, have a completely different physics. We were talking backstage about a company I helped get going two years ago called Sesame that just came out of stealth. And Sesame is — quick show of hands, how many people have heard of the Sesame Voice model?

Okay, so about 10%. Great. How many have seen the movie Her? Okay, because that’s kind of basically what you guys are building, right?

It’s a team where the CEO is former co-founder and CEO of Oculus. So he has built hardware before. The CTO Kit was my co-founder and led the AI platform at Discord.

He understands voice really well, and what they showed off last week was a research preview of this model that they think kind of starts to cross the uncanny valley in voice. And what’s different about them is they put out the voice model as a research preview and it’s gotten millions of people using it in four days. But the physics of funding that look nothing like the funding of a foundation model company, because it’s much more of a product company where they actually leaned on open source language models. The one they used is called Gemma from Google.

And that model didn’t exist three, four years ago. So if you wanted to do anything interesting in product four years ago, it was extraordinarily expensive. Today you can do that largely by using open source models. And the capital required to get to your first prototype is extraordinarily less. Midjourney, for example, is a company that we’ve talked about before — they’re a hundred percent bootstrapped, and they were able to piggyback on Stable Diffusion, which is an open source image model. And I think that’s why it’s so exciting right now to be a product founder — the cost and physics of getting to your first product have collapsed overnight, so they don’t need venture. So frankly, when I see a pitch where they’re building a product vision but they don’t have a prototype in market, my first question is why not? You don’t need anybody’s permission anymore — just go build it, ship it, and start getting product market fit. Whereas three, four years ago, that was much harder.

EL KALIOUBY: Yeah, I have to ask a quick question on Sesame. Most people here know that I’ve spent the last 20 years building artificial emotional intelligence, and I really believe that that’s gonna be a fundamental piece of a ubiquitous AI interface. How do you think about that at Sesame?

MIDHA: Oof.

Okay. How long do we have?

EL KALIOUBY: Quickly.

MIDHA: The short version is they think voice is the best interface to a wearable. Their vision is a pair of everyday lightweight glasses that you just wear all day. If you think about how many people wear that interface, I think they’ve realized after years and years of attempting augmented reality, as we traditionally call it — which is a display-based AR — is a distraction. If you have a good enough, smart enough multimodal model that can sense the world for you but can talk back to you via voice and you can interact with it via voice, you actually don’t need a display for most of your life. Every now and then you’ll pull out a phone as a companion device when you really do need it. But the idea is that super lightweight glasses that look and feel no different than the ones we are already used to wearing every day with no display is kind of the hardware that gets unlocked once we get really good voice multimodal companions that actually, like you said, have emotional understanding and context about your life.

It’s like, if you could actually unlock the intelligence of somebody sitting on your shoulder, then you don’t need the phone that much.

More of my on-stage conversation at Abundance Summit in a minute. Stay with us.

[AD BREAK]

Copy LinkHow to think clearly about AI startup valuations

EL KALIOUBY: Let’s talk about the elephant in the room.

We’ve kind of already talked about it a little bit. We’re seeing AI startups getting crazy valuations, right? Like $50 million plus seed rounds. With Anthropic, it was even larger. Companies raising at a billion dollar valuation and more with barely any revenue. With my investor hat on, it’s really hard not to have FOMO.

But also maintain valuation discipline. So how do you both think about these valuations and do you think they’re justified?

BLUNDIN: Well, they’re definitely justified by the upside, but I’m not seeing $50 million seed rounds very often. Those are outliers. What’s happening there is news likes to make news, and the way you make news is by talking about the really rare outliers.

You are seeing some crazy cases like Yulia Schuberg getting a crazy founding valuation, but the much more common case — I teach at MIT, Anj teaches at Stanford —

80% of the students now at MIT want to either found a company before they graduate or on graduation day. This is a huge entrepreneurial community that wants to start companies. By and large, if it’s a great team, they’re super bonded, they’ve done a prototype, they’ll get maybe a 10 to $15 million valuation, not 50. And they’re not price sensitive. Generally, they’re looking for the perfect investor to get them credibility, distribution, or the next valuation. Then you get this rapid step up.

After that, and I think those rapid step ups are justified easily. If you look at the time to revenue, if you look at the TAMs, the total addressable markets for these companies, they’re so much bigger than anything we’ve seen before, so they make a lot of sense.

Copy LinkWhy younger founders and dense ecosystems have an edge

EL KALIOUBY: Let’s talk about the age of these founders.

BLUNDIN: Actually, if you look on screen here — when I started investing full-time 20 years ago, it was kind of a bell curve and the 30 to 34-year-old was kind of the middle. Now, if you look at billion dollar value companies in the rearview mirror and say, well, how old were the founders when they started it?

That first bar — this is out of date actually, that first bar is even bigger now — but every year that goes by the 20 to 24 gets bigger and bigger and bigger. And if you look at the heroes of the industry, the Mark Zuckerbergs or the Bill Gateses, they’re all 20 to 24-year-old or younger founders.

So I think that trend is gonna continue.

For us, the time to scale has gotten shorter and shorter and shorter too. Actually, there’s another image of that on the next slide.

EL KALIOUBY: Well, also the barrier to starting a company, right? You can build a product with very—

BLUNDIN: Yeah. I think if you’re an AI-aware 18, 19, 20-year-old, you can more than create a phenomenal company and product, but you know very little about life.

There’s a lot to learn about running a company. So what we do in our accelerator is we try and take on payroll, accounting, food, office space. Last week we bought an apartment building with 24 apartments — I haven’t even seen it yet actually, it’s a five and a half million dollar check.

I said, write that check, let’s go buy that building. One of our portfolio companies, Meco, reached a $2 billion valuation in under two years. So its valuation went up $20 million per week. So what I told our team is, look, if we saved two weeks for that team finding an apartment at age 20.

The CEO’s 21 now. If we saved them two weeks, that’s $40 million of value that we’ve created. So let’s get the building and let’s go.

EL KALIOUBY: It’s crazy. Yeah, that’s right. I love it.

MIDHA: The only thing I would add to that is I think it can be dangerous because you asked about FOMO. I’ve been through that. I understand that almost too well, because I started my last company in the middle of the augmented reality computer vision craze of 2017, 2018. Self-driving car companies were exploding, researcher salaries were exploding.

There were all these crazy arms races for people, you know, getting 30, $40 million salary packages from Waymo and so on. And I was busy trying to build as a founder — a small and scrappy team to do computer vision work. And I had all these higher and higher valuations.

And if I look back now, basically nine out of the ten companies that had all these crazy valuations from SoftBank and so on have basically died and exploded. And I think it’s very easy to fall into the popularity contest of valuations, which really don’t mean very much in private equity land or private stock land.

But what is really different is that the time to build a real business — the time to your first 10 million in revenue and real customer value — I’ve never seen this many companies get to that so fast. We are investors in a company called Cursor, which is another MIT team.

And they’re just a year or two out of school and they didn’t raise hundreds of millions of dollars at some crazy valuation. I think their seed round was like $5 million.

They shacked up in a house in San Francisco and they just put their heads down — basically a small house in North Beach in SF, and just locked themselves in a room and focused on shipping a product people love to use and they just crossed a hundred million in revenue in eight months. No sales team, less than 20 people. The valuation is, I think, a distraction, but you should — I think it’s worth asking about the speed. Valuations are sort of lagging indicators.

What comes first is product market fit and the time to that — I think — has never been faster.

EL KALIOUBY: I’m a little biased because I started my company out of MIT where I was a postdoc. But Dave, you believe strongly that MIT is the AI startup capital.

BLUNDIN: I believe very strongly in ecosystems. There’s a great Paul Graham quote that the difference between being a startup founder and unemployed is just a question of interpretation. So if you are in a community of other people doing startups, the reinforcement effect is critical, especially at a young age. But now you see the ecosystem effect regardless of schools. You see the ecosystem effect all over the place where any dense, like-minded community of people working on things share ideas at an incredible rate, but they also reinforce each other.

They give confidence to each other. And they also share leads to investors, to sales, to whatever. And that effect is unbelievable as a force multiplier. So I guarantee it has nothing to do with the talent at the other universities. It has to do with the fact that you’ve got a high concentration of people trying to build companies in a very tight area. Cambridge in particular is the densest innovation cluster in the world by far. Silicon Valley has more volume, but it’s more spread out. But Kendall Square in Cambridge, you’ve got MIT, Harvard, and then there are 200 other schools and most of those you could walk to. And so it’s just an incredible density of talent in a very small area.

So that’s what’s causing those numbers to go up.

EL KALIOUBY: We’re going to take a short break. More in a minute.

[AD BREAK]

Copy LinkWhere AI is unlocking value across regulated industries

So the theme of this year is convergence, and again kind of referencing Kathy’s presentation, how AI is unlocking value in all these other disruptive industries as well. And I’ll share an example. One of the companies I’m investing in is called Labyrinth.

They are anchored by Invisible. Francis is here, Stuart Lacey, I don’t know where you are, but hi. And they’re officially launching at Abundance and they’re on a mission to accelerate companies through the regulatory process. And that is just gonna create tons of value in industries like drug discovery, medical devices, autonomous vehicles, drones, nuclear energy.

And they’re doing this with AI and human expertise. So as a result, I actually now feel comfortable looking at industries that are traditionally very heavily regulated because I think if they can get to market faster, that de-risks it. And in fact, there’s a lot of significant upside.

So as investors, where do you think the value is in this AI landscape? What is AI unlocking? What industries are you most excited about?

BLUNDIN: What industries? God, there are so many that are touched by this. I started a FinTech company many years ago called Westmark.

We manage a little under $2 trillion of assets, and it’s regulated. It’s an RIA, and they’re loving this AI revolution because the little startups have a real hard time dealing with a regulated area, and they’re trying to deal with the SEC, right? So they have to come through Westmark. So as long as Westmark is nimble enough to keep up with it, that’s the channel by which AI gets out to the RIA world. But that gets me thinking about regulated areas in general. The hyper regulated areas like the FDA are very tough and it slows down time to market, which is kind of tragic. The semi-regulated areas like TCPA, phone calls, being an appointed insurance agent — those are really nice barriers to entry. There’s a little bit of friction, but it’s enough to keep OpenAI from just doing it tomorrow. So I love those as investment themes.

Copy LinkWhat actually creates moats in the age of AI

EL KALIOUBY: Yeah. How about you? Where do you see the value in the AI space?

MIDHA: I don’t know that AI has changed really that much where the value accrues. There’s a constant question which everyone’s always asking, which is what’s the moat. And I think that was gonna be—

EL KALIOUBY: My next question. So where is the moat?

MIDHA: I just don’t think there is a misnomer that AI models are somehow moats. They’re not — they give you head starts. They give you advantages, but the fundamental — I’m a big believer in the seven forces of value and power and moat and so on. And I think

EL KALIOUBY: Go through them.

MIDHA: They’re the classic stuff, right? It’s network effect and it’s brand and it’s sales. And you’re talking about regulatory barrier. Those fundamentals of business I don’t think have changed. Usually once or twice a decade, a new technology comes along that gives startups a fundamental speed and headstart advantage over incumbents.

But then ultimately they’ve gotta build out real value in the traditional sense. And I think teams that are accruing the most value are the ones that realize that models are not some steady state moat — they just give you a headstart. I kind of think of models as cold fusion almost.

When you have cold fusion, you don’t need to sell it — it’s the closest thing we have to a “build it and they will come” top of funnel driver, because they’re so magical. General models, when you just put them up, especially with state of the art. So if you’re a team that confuses the value between models and solutions, you’re going to end up in trouble. But if you realize that models are not solutions — models are inputs to a solution — and that customers don’t want models, they want some product, some platform, they want you to solve their problem.

EL KALIOUBY: Problem. Right?

MIDHA: The faster you do it with a model and then realize that that’s just a headstart, the easier.

I tell founders, if they wake up every day worried that the next version of ChatGPT is gonna put their business out of business, then I don’t wanna be an investor in that company. However, if the next version of these foundation models makes your product and your solution better, then that’s defensible.

In a sense, I think OpenAI has been grappling with this, where they’ve realized that distribution is what matters. Distribution is a source of value. That’s why they’ve taken a really great and aggressive stance towards trying to get ChatGPT to be a consumer brand and product.

And today it’s like top 10 consumer destination, which was not, by the way, the founding premise of the company. The founding premise was to build AGI, put out open research, and then maybe put up an API and hope the world would figure out what products to build.

And they’ve clearly, at least for three years now, been methodically investing in building a real distribution advantage. I think that’s because they’ve realized just having the top leaderboard model doesn’t grant them a license to win either.

BLUNDIN: Well, and also if you look at the revenue from just going direct to consumer, it’s 5 billion going to whatever, 15 billion. And I think that’s kind of new in the world — saying hey, we’re a platform, but we’re also gonna go direct because we can. And it’s happening in the enterprise too. If you look at one of our portfolio companies, Blitzie intends to be this incredible technical platform that can write 3 million lines of code in a single night, right?

Huge amounts of white collar automation. And then why not deliver directly to the customers too? So they’re signing up huge banks and insurance companies. One of the mega banks has $4 billion in payroll in account reconciliation and back office work, all of which we’ve sample tested it — you can automate it with AI literally tomorrow. And so it’s gonna take a while to methodically roll that out across every single use case. But that’s $4 billion saved, right? So they’re like, okay, we’ll deliver that direct and we’ll make it available as a platform at the same time.

So same thing that is happening with OpenAI on the consumer side, companies are now doing on the B2B side. Yeah, I think it’s very interesting this idea of economies of scale but also economies of scope. Because if you’ve got the data, you can easily continue to add features and expand the business applications once you’re in. Alright. So AI startups are scaling faster than ever before. The traditional venture funds are designed for a 10 year holding period.

Copy LinkHow faster scaling is reshaping liquidity and exits

So let’s talk about liquidity, right? How are these companies gonna become liquid? Are we looking at quicker exits, secondary markets, early acquisitions, faster timeline to IPO, and how is the venture model adapting to these kind of timing dynamics?

Yeah. Well, if I compare our 2019 fund to our 2021 fund — a couple AI companies in 2019, 90% plus AI companies in the 2021 fund, maybe 20 companies in the 2019 fund, 40 companies in the 2021 fund — I think the 2021 fund will be liquid before the 2019 fund.

Actually easily. In fact, I already know that. So it’s just night and day, faster time to liquidity. There are still 10 year structure funds, but when liquidity comes in, you distribute it. So it doesn’t really matter. You don’t change the fund structure. You say, hey look, we can recycle the money much more quickly.

EL KALIOUBY: Do you have a point of view?

MIDHA: My sense is that we’re dealing with a bunch of — the market always finds solutions for the greatest, best businesses. The reason we’re in this sort of weird IPO liquidity crunch when it comes to public markets is largely because of a set of legacy rules that were enacted in 2008 in the wake of the financial crisis — they don’t really track the first principles, sort of startup reality. And so what I’m finding is the best businesses, there’s no shortage of demand and liquidity in the secondary market. There are tons of buyers. We’re investors in a company called Databricks, which is an extraordinarily valuable infrastructure company.

There’s no shortage of people who want to buy stock in the company. They haven’t gone public yet, and so they haven’t done this anytime soon. But companies like Stripe have always been able to offer employees and early investors liquidity. So I think the IPO question generally for the best companies is sort of a legacy of 10 years ago.

They find ways. I think the ones who are in trouble are folks who maybe forecasted their operating plans to be dependent on having to raise from the public markets.

MIDHA: Those companies are actually gonna struggle. I’m not sure what the answer is for them.

Copy LinkThe mindset investors need to win in AI

EL KALIOUBY: Okay. Last question. So we have a lot of angel investors and LPs in the audience. Can you share one piece of advice on how we should all be thinking about investing in AI?

Anj, I’ll go to you?

MIDHA: Believe. Four years ago when I think back to that moment when Anthropic was getting going and I made 22 introductions for them for the seed round up and down Sand Hill Road, and they got 21 nos. I got calls from former colleagues — I spent a few years in venture capital at Kleiner Perkins — and I got calls from people saying, this makes no sense, this is borderline snake oil, what are you selling? Like AI, just a general purpose model that solves all language tasks.

And I was shocked by the amount of disbelief.

EL KALIOUBY: Huh.

MIDHA: But if you had believed, you get to unlock the future just earlier than anybody else. And it can just be hard, I think, for folks to realize that all you’ve gotta do is try the models out, play with the products, and believe that as long as the speed at which innovation is happening right now continues, the conclusions are pretty obvious. I think it’s disbelief that gets people wrong.

EL KALIOUBY: Belief.

How about you?

BLUNDIN: To me it’s pretty obvious that this is it. Last time we were in a time window like this was kind of 1995, 1996, a long time ago.

But it’s the same logic today that it was then — get a PitchBook account, look at every deal, and then hold. Get into your team. Look at them. Hold your team accountable and say, look, I want to be on these cap tables, right? And just pick them and say, why am I not on that cap table? Oh, I’m not in that venture fund.

Why am I not on that cap table? Oh, we had a pro rata right and we didn’t exercise. Okay. Just methodically go through them. And spend the next year making sure you’re on as many of those early stage cap tables as you can get to. And once you’re looking at the companies, you can work backward to the conduit that would’ve gotten you onto that cap table.

EL KALIOUBY: Amazing. Thank you both so much. Thank you.

Coming out of Abundance, I feel very inspired and excited about the future of AI and humanity.

I’m at so many events throughout the year, talking about AI with experts and founders.

Honestly, it’s so cool to have this podcast as a place to share those experiences and insights. One of the goals of this pod is to make information about AI more accessible!

We’d love to hear your feedback about Pioneers of AI. Rate and write us a review wherever you listen. Your input means a lot – plus it helps other people find the show. Thank you.