Elad Gil on AI market crystallization: 'We know who the finalists are' in foundation models and coding

Yes. Oh yeah, that's right. They're a customer. Uh well, we have a lot Gil coming into the TBN Ultradome from the reream waiting room. Let's bring him in now. Thank you so much for joining us a lot. How are you doing? Finally. Thanks for having me. Good to see you all.

We have wanted to do this interview since probably the very first week that we could guest. Took us a while, but you have your own show. But uh but thank you so much for taking the time. Um what uh what what what this week has stuck out to you.

I'd love to just get a state of the union on how you're thinking about the market broadly and then we can zoom in on uh on individual startups and trends and subcategories that you've been focused on. But have you had a reaction to this like bubble talk that's going on? Have you been thinking about this?

Uh what's been how have you been processing the information? How do you even research whether or not when something is going viral like that? Yeah, I mean I've been looking at this stuff for a while. Um because if you look at the '9s as a sort of precedent or anticedent the '90s internet bubble. Yeah.

I think there was something like 450 companies that went public in um 199 1999. There's another 450 that went public in the first couple months of 2000. And so about 2,000 companies went public. And then you ask how many of those are still alive? Like how many survived?

And there's probably a dozen, two dozen that that are still up and running. There's probably two or three of those that are really important to Amazon as an example, etc. And then 1,980 out of 2000 probably died, right? Went to zero. And they were being priced on eyeballs.

They're being priced on eyeballs, not not even account creation. It was it was such a different time like IPOing was like doing your series B. I remember there was a guy uh Bill Gross at Idea Lab in Pasadena, my hometown and I believe he took a hundred companies public and he ran an incubator.

It was like the Y Combinator of the day. One company I believe was Adwords that went to Google and became the backbone there and he had a bunch of great outcomes but it was like a bit of a machine. Uh if the IPO is no longer the metric that you we should be watching is it is it these revenue ramps? Is it churn?

Like how can we dig into understanding like where true value durable value moes are acrewing versus uh froth? We we often call it like the barnacle economy. Like if you're if you're uh you know a toilet cleaning startup and but you have anthropic as a customer and they 10x their office footprint, you 10xed revenue.

That's not exactly the type of business we want to see long term. Yeah, I think durability is a great question and it's a really hard one for this era. Um for two reasons.

One is things are changing so rapidly from a model and underlying capability perspective that if you look at every prior technology wave you look at for example Microsoft OS right they forward integrated into uh the office suite off of using Windows OS or Google forward integrated into vertical searches so they killed a bunch of companies or really hurt a bunch of companies are providing those services so we should see the same thing with the foundation model companies right it's most likely if they're going to afford integrate they're already doing it in code maybe they end up doing it in customer support in sales sort of all the big categories there'll probably be some effort at some point.

Now, they may or may not succeed with that, which is a different thing, but there's durability in the face of competition from the big folks. There's durability in terms of like, well, people keep using your product or we get subsumed by a startup.

And then there's just things that are just running up that clearly are never going to really work. And um the real question is, you know, how do you identify each one of those classes of companies and how do you think about them as either a founder or an investor?

Right now, we've been we've been noticing there's sort of like three types of three buckets of companies that are trying to like, you know, craft an AI narrative around whatever market they're in.

If we're talking vertical markets, smaller markets, not the foundation model layer, you have the legacy Fortune 500 company, uh, you know, career CEO in the seat who's maybe paying a consulting firm for some AI transformation plan and, uh, maybe the stock's not doing so well.

Then you have the startups that are, you know, we we we go to YC demo day. We talked to five companies that are building in the same space and it's they're complete green field project. Amazing place to be. Must be super fun to just have be puppeteering 25 cloud code instances and codeex agents to build your thing.

But then we talked to a lot of founders that have they started their company 5 years ago, 10 years ago. They have a serious business, but they're still in founder mode.

And we always find it a little bit hard to bet against those guys who are like coming back in re-energized, they have the balance sheet, they have the customers, and they can kind of take a second crack at it. Um, do you like what nuance would you add to that framework? Do you think it's on a per industry basis?

Is it all about the founder? How do you think about that? Yeah, I think there's basically three viewpoints on that. Um, I think there's a very small number of singular founders who just make amazing things happen, and that's Elon Musk, right? Like who else would go to space and build cars and do all these things?

Honestly, I think Arvin to perplexity is one of those where in anybody else's hands I think perplexity would be a dramatically smaller business and I think most other companies in his hands would do better. He's very good, right? But that's very rare.

So I kind of put you know amazing founder aside because even great founders if they're a terrible market tend to get crushed.

Um and so I think there's a second piece of it which is there are a bunch of these AI markets that have recently really crystallized where you know I used to say a year and a half or two years ago that the um more I learn about AI the less I know and it was the only market that I ever felt that way because you know you learn new stuff and you know more right you do better you can predict stuff um and I think that changed over the last six to nine months where suddenly at least for certain areas it's really clear who the finalists are.

We may not know the winners but we know who the the final contenders are. We know that for the foundation model market um we know it's anthropic, open AAI, Google perhaps XAI, Meta, a few others, MRO what but you know it's a small list.

We know for coding it's cognition cursor and then the foundation model companies and then Microsoft. Um and you can go through sort of vertical by vertical. There's a bunch of verticals now we know a bridge for healthcare and maybe you know there's there's a handful for each thing.

Um but then there's a bunch of markets where it's clear the market's going to be important and there's tons of players but we don't know who the winners are. that's financial tooling, maybe that's sales enablement, maybe that's accounting, you know, you can come up with a list.

Um, legal legal feels like legal feels like it's already solidified except a handful of these more vertical specific, you know, uh, specifically we we we were joking uh there's people doing injury, you know, personal injury law agents, you know, ambulance chaser agents, but uh it feels like the categories are uh solidifying.

I guess the question I have is you've backed W winners and basically all these categories. Where do you feel underexposed from an investment standpoint? What do you think you didn't quite anticipate? Is it is it like energy possibly? I'm sure you have bets there, but where do you feel underexposed?

That's a really good question. And I feel like the two big trends of this era so far have basically and by era I mean like two years you know it's not it's been your errors you know of recent last two weeks you know modern millennial last two weeks. Yeah exactly the last two weeks of this era. Yeah exactly.

Um I think that there's uh the two big things are basically defense and you know I was very early involved with Anderell and Leather D and have participated basically in every round of that company and then I'm an investor in um Seronic and Helsing and then Kayla in Israel.

Um, but very very little defense actually over like a 10-year span of investing, right? I did Ander only for like seven years because it was like the only company that I thought was just going to keep going forever and I think it's the the you know a generational winner there. Um, but uh uh and then there's AI. Mhm.

And AI means everything now. It means consumer, it means rollups, it means electrical s it means Yeah. natural gas turbines are AI index, right? Yeah. Exactly. Yeah. Trains have to move GPUs across the country. Yep. FedEx throw that in there. Yeah, I love it. Yeah, FedEx is my favorite AI company.

So, I think um you know, those are the the sort of two obvious trends now. Six or seven or eight years ago, they weren't obvious. Now they are. Um and you know, honestly, obvious trends often go longer than you think.

I remember with the social networks, there was um a bunch that didn't work out and then MySpace and Friendster and eventually had Facebook and LinkedIn and Twitter and then at that point everybody said it's over. Yeah. Everything in social saturated.

But then we had WhatsApp and we had Instagram and we had Tik Tok and you know just it just kept going.

Um AI is in a much earlier version of that right now where um I think we're at the very very early days of this massive wave and so to some extent I'm underexposed to AI in general because I think it's the biggest thing that's happened in you know 20 years or longer. Lad Gil says he's underexposed to AI.

You're very very humble. The chat says everything important, a lot is in. Yeah. Uh what is what is uh what's not AI? When do you get a pitch where they use a the word the letters AI on their website a lot, but you're just like, "Hey guys, this is this is just this is just regular enterprise stuff.

" Well, I actually bring this back to like what's durable in the face of AI. And a good example of that is Ripling, right? Ripling is a amazing company. um they uh cross- sell a dozen different HR products and AI can make some stuff better, but nobody's going to do like the AI first rippling y and suddenly win.

Right now, the the main threat maybe to a rippling or deal or sort of related companies is if company headcount actually goes down because of AI, then they have fewer seats they sell, right? And so that's maybe how there could be a headwind from AI for these companies.

But but maybe that means AI maybe that means we just get a lot more companies, smaller teams, right? Yeah, it's quite possible. Yeah. So, I I just think like that's the when you see that you're like, okay, this is very durable in the face of AI and that's great, right? And so, a lack of AI means AI can't displace it.

And so, as long as it's working, it's actually more interesting in some ways. Yeah.

question I sort of ask myself is uh you know using a company like Ripling for for example uh even if companies start needing less people there's still they can they I could see them transitioning to kind of valuebased pricing around what is it what's the value of like running your HR department right is it 3% of revenue is it 5% of revenue is it 1% of revenue like either way they're going to make money if they're providing like infrastructure that function.

Yeah, it's a great insight because I think one there there's two or three things that are underappreciated about this AI wave.

Um I think that the first thing is that um the the capability set has shifted dramatically, not just in terms of what these models can do, but the fact that you can just ping them with an API and something that's accessible to everybody.

And I think that's actually very underdised, right, relative to the the prior world. I think a second thing is that the markets are oddly open. Like legal never bought any software.

it was really hard to sell into legal but because of AI suddenly Harvey can exist right um and then uh the third thing is that uh a lot of this is about what you're saying which is the tams of markets are shifting from seatbased pricing or seat-based value to labor you're replacing human labor and so you're looking at for example customer support it's not zenzas how many seats can you sell to customer support reps it's how much can you augment and do work for customer support reps.

It's the labor market versus the software market. And I think that's very underappreciated when you think about market size for some of these things. You're really going to miss the size of these markets and how big they are.

You know, the services economy that we looked at on my team in terms of like where AI could intervene is about $5 trillion, right? So, it's a lot of GDP is accessible to this. And so then you ask, okay, is it is it a workflow that's specialized to customer support like whatever?

Is it a roll up where you're buying assets and changing them? Is it a different approach? Like how do you sort of span all the change that's coming because of this?

Uh Ken Griffin gave a talk earlier this week and he was saying that in 1999 and 2000 it was very obvious that the internet was going to change the world, change the way that our economies run, yet it still took 15 years for it to actually have an impact. Uh and he was comparing that to today.

The difference of today is that we have the internet so we can deliver these products instantaneously to the entire world.

Do you think that uh do you think that this time can be different and we can uh as an industry unlock the value of this technology on a shorter timeline than than the internet took because we just didn't have you know the internet is the greatest distribution engine in history. Mhm.

Yeah, it's an excellent question and I think um both things can be true simultaneously which is we're seeing real revenue for these companies, right? Cursor is rumored to be in the high hundreds of millions of revenue.

Um you know Azure added something like two or three billion of AI revenue per quarter from sort of a cold start two or three years ago, right? That's amazing. That's like $10 billion run rate plus just off of AI revenue, right? So it is working. It is being adopted.

But the flip side of it is it'll probably take a decade, right? And so I think both of those things are true. And I think the biggest impediment to adoption isn't the technology. We could do so much stuff with the technology right now. It's organizational process. It's workflow management.

It's all the stuff that happens when a big enterprise uses anything. And they're like, you want me to change my tooling? You want me to change my people? You want me to, you know, my processes? And that's what's going to slow it down. And that's slow down every technology way.

But to your point, we have massive distribution. It's already everywhere in some sense, right? I don't know that you guys probably know the number. You just talked to Sam Alman, right? What's the number of people using ChattPT per month, but that's a huge impact already. Yeah, it's interesting. They haven't said that.

It has to be north of a billion because they're already they're reporting 800 weekly active view. If you have 800 weekly, you have to have over a billion. And that feels like such a new cycle, but maybe he's just keeping it in his back pocket for when needs a good bit. Bit of a wild card question.

And I didn't plan this, so I didn't mention it beforehand, but no, it's not it's not bad, but I just think it's interesting. If you couldn't be an entrepreneur and you couldn't be an investor, what hyperscaler would you want to work at where, you know, be an executive at? That's interesting. Oh, that's so interesting.

I don't know. I could make arguments for two or three of them. Um because I think there's such different problems to be had and different assets or, you know, Google, for example, just has such amazing assets relative to this era, right? They have the most data. They have the most compute.

They have they have amazing cash flow. I mean, they're just like input. Yeah. Amazing stock. Um, so there's amazing things they could do. Um, obviously there's uh crazy stuff Microsoft can do on the business side, plus with GitHub and uh, Copilot and everything, you know.

So, they should really be driving a lot of the coding future in my opinion if if they um make the right moves over time.

um you know and so you can kind of go through one by one and say there's really interesting things meta in terms of social I think there's opportunities everywhere um what give us a give us an update on AI rollups you have some investments here from my understanding but how how do you see the category evolving we see new com you know new teams coming together to attack various uh various markets with this strategy almost daily now But what's your view?

Yeah, I think um so it's back to if you look at services in the US, it's 3 and a half to 5 trillion of it uh is sort of labor that to some extent could be augmented or displaced by AI. And so the idea is um can you uh there's certain types of companies that are going to be very slow to adopt software or AI.

And so there's two things you can do with that. can wait and build a software company that will take a really long time. You can actually buy those companies, implement the AI changes and dramatically change their margin structure. And that doesn't mean letting people go.

It could just mean you make people five times more productive for certain types of roles. So you can look at different businesses where you know 80% of the cost of that business is repetitive white collar labor. And so you can help augment or automate stuff for people.

Um and so I've looked at a few dozen of uh teams or companies doing this. I ended up backing two of them. And um really you need three things to make this sort of strategy work which is truly transforming a business with AI and then scaling it up.

Uh the first is you need a great um AI person obviously right you need to be able to implement the technology. Second you need a great PE person.

You need to buy assets properly understand your envelope just like a SAS company has its ICP or like customer profile that goes after you almost have like your M&A profile like what fits in my pocket of stuff that I want to go after.

And then lastly, you need somebody who's operationally great, who can rework the organization against the AI because that's often the hard part, right? You actually have to get people to use this stuff in order for it to be implemented. And so very, very few of these teams have all three of those things.

And many of these teams are basically doing traditional PE rollups. They're not really using AI, but they're raising at AI prices and then they're buying at PE prices. And so this arbing and so I've tended to avoid arbitrage. Yeah, it seems like a great deal with the founder.

In another world, they'd be doing a private equity fund with two and 20 and then here they can go out and raise and dilute 20, you know, raise 20 on a 100 and then they own 80% of the of the business. Yeah. Remarkable.

Yeah, it's a it's a very smart thing to do and if I was a PE person, I totally could do that and start doing AI, but most of these things aren't doing AI, right? And so, but a handful of them that are are going to be massive, right? Imagine an AI dener, right? It's just it can really be transformative to big sectors.

Yeah. And it changes something from a services margin to a software margin. Business with software leverage. Yeah. So it changes the characteristics of the of the business. Sort of sort of flashing back in your career.

One of my questions I've always had on my mind is uh uh that you kind of like created the solo capitalist idea. People have kind of, you know, put that label with you. Was that just a happy accident? Did you were you deliberate that you didn't wrap what you were doing in a firm?

There's obviously people that start with similar scales but wrap it in a firm with a brand and uh like how thoughtful was that? What were the considerations? Do do you like what how that play? Yeah, I think um you know for a while it really was just me. So it wasn't some strategic move to you know do something.

It was just I was on my own doing stuff and I you know um and then eventually I brought on people for back office and finance because like I think that's really important right? You want compliance. You want things to be proper and all that. Um uh and then I got this moniker and I never asked for it, right?

Like I actually am happy to be called whatever as long as I get to be involved with the most interesting technology and technologist in the world. You know, they could call me I don't know what a carpenter. I don't care what they could call you big VC, big venture capital. That's the one thing I don't want to be called.

Other than that, I think um what we do is different from like traditional VC too, you know, like I don't I I uh we do traditional investing and we do traditional venture, but I actually think we do a bunch of other stuff and we do interesting projects around um you know, things like uh one person on my team who's working as an investor as a technical background is actually driving uh uh AIdriven translation of the world's thousand most important books that are off of copyright and we're working with um a few really big foundation mobs on that.

So we do stuff like that too just because it's interesting. So I hope it's never just a venture fund. But do you think when do you expect uh AI to transform venture capital? I think it's notable that the firm today and the activities of the firm look quite similar to prehat GPT.

Maybe you can write a investment memo faster. Maybe you can uh seem like you prepped for a board meeting better. you know, if if you just drop the deck in and and ask for a summary, but uh it doesn't feel like it has changed the profession at all yet.

It's still uh finding and winning allocation and and um you know, but yeah, I think it can help with some aspects of diligence to your point like it can pull competitors and things like that. Um to some extent it depends on the market, you know, there may be weird uses that nobody's done yet.

So, an example would be, have you ever done the prompts where you ask the um AI to like uh cold face read somebody and tell you about their personality? No. You you upload for this? Yeah. I mean, just just analyzing somebody's personality with a picture, even just any picture of their face. Yeah.

It's it's pretty I mean, this is like the the skits is called physomy. Sure. Sure. Yeah. Yeah. Yeah. But there's stuff like that where I've I've done that just for fun with my friends, right? I'm like, "Hey, what what is this person like? " and my friend's sitting there with me, right?

I'm not secretly trying to psychoanalyze them or something. And then I'll ask it to give me a detailed breakdown of those characteristics and why.

And it'll say, "Oh, this person looks like they have a genuine sense of humor and they're warm because of the way that the crow's eyes around their eyes exists in this way versus somebody who fake smiles because it doesn't get to the eyes, so there's no wrinkling.

" And you're like, "Wow, that's actually like super interesting, right? " And so it'll break down the sense of humor. It'll break down uh how likely that person is to be loud or quiet. the like the aggressiveness like all this stuff.

Yeah, you can imagine there's there's a somebody could create an EQ co-pilot for the new like meta display glasses.

You can just walk around and I and if I'm maybe I'm really high IQ, but I don't read people that well, I could look at John and say, uh, which is probably probably inverted, but I could look at John and be like, "Oh, John is John's very interested in the conversation, and he he clearly wants to be friends.

" So, I think there's more more to more to build there. uh give us the update on on uh on your your company with Jared Kushner. I know he's been very busy. Uh but uh but I'm excited to to hear the latest. Oh, sure. Yeah.

So, um we recently launched a company um that's called Brainco, which is focused on using AI uh as basically a platform to help transform the world's largest institutions. And so, we've been working with a number of large enterprises um some private equity um and other firms around this.

And so, um, it's just been a fun project to do, uh, with him and with Eric Woo and Lewis Vidigray and a few other people. Eric was the former CEO of Open Door. And then Lewis, uh, is, uh, the former finance minister and foreign minister of Mexico.

So, it's kind of this interesting group of people coming together to try and solve really big AI problems. So, that's been really fun.

So is this realizing how much uh the Accentur and the McKenzies of the world were were getting paid to make pitch decks on basically here's here's what you should know about AI for your business and realizing hey we could probably do that a lot better and then ultimately build software within these organizations like what is the actual it's much more focused on the yeah it's more focused on the software side of it so basically there's a common platform that's involved in terms of dealing with different forms of data dealing with eval dealing with a lot of the things that every enterprise needs to build in order to um really adopt AI and then we build vertical specific applications on top of that or in some cases horizontal applications that can be reused over and over by um similar companies in the same vertical.

So you could imagine for example for a financial industry company there's like a dozen things that every single one of them needs to build and there's some customization around it.

It's kind of funny because if you look at very large deals, right, if you're Dell or your VMware or your Oracle and you do like a tens of millions of dollar deal with with a customer, you're going to have customization against that customer. You can afford to do it, but also it's important enough to do that for them.

And so it's similar in that regards where we have common infrastructure, common platform, the same vertical applications, but then there's going to be some customization per yeah um per client just like any other giant, you know, enterprise company.

Have you thought about slicing the target customer either across vertical like we're not doing healthcare because of HIPPA just yet or we're not doing defense because of Fed ramp just yet or we we think we have a lot in industrials that we can go after or or are you more focused on uh slicing by you know market cap or size of business like yeah we're not going to work with mid-market companies.

How do you think about like where the wheelhouse customer will be?

Yeah, it's very much um the the goal is to work uh almost solely and there's going to be some counter examples of this uh with companies that are truly the the world's biggest institutions in terms of revenue, market cap, and then potentially impact, which means sometimes you work with somebody a little bit smaller, but the goal is to ask how can you use AI to really get leverage on things that are important at at sort of a massive scale.

Yeah, there was a report that JP Morgan is spending $2 billion a year investing in AI to effectively save $2 billion a year from automating. It's like when there's that much. Yeah. Yeah. They're breaking even basically.

I mean, it presumably like some of those savings are durable, but um but certainly there's a lot of money. Um, what do you how how how good do you think you've gotten at clocking uh AI pilot revenue as nondurable?

Because there you I'm sure you saw that that headline an MIT study came out that said basically 95% of pilots are are not actually delivering in value. This this felt like the year the year of the pilot. Next year is maybe the year of reality. Uh but I don't know how you see it.

Yeah, I just view that as a standard technology cycle. I thought that was a very overstated article. So, um, you know, this happened with mobile. You do the kind of crappy mobile app. I don't know if you guys remember the first BFA mobile app. It was basically like a Yeah, it was just like a web page. Yeah.

Well, the the current BFA app is still still pretty rough. So, I actually think it's pretty good. Like, you can go to an ATM, you can take out cash. I think I'm the only person who still does that. um you can take pictures of checks, you can pay with zel, you can do all these things, right, that you couldn't do before.

Um but you started off and you tried to log in and the thing would crash. So I just feel like we're kind of in that era of AI implementation. People will get it.

It'll take a decade to fully sort of propagate to your earlier point, but it really is the big wave that we're living through right now and I think it's truly transformative.

Would you ever uh would you ever uh have you ever considered an LBO of any of these companies that are in the the SAS apocalypse in the public markets that clearly have some you know you know super meaningful customer relationships and a ton of potential important place in their market but just aren't evolving their business model quickly enough.

Yeah, we've looked at that actually.

I think there's a few really interesting things to be done at scale and to your point I think part of it is just driven by can you actually implement AI and I think the if you look at the private equity industry in general they've talked a lot about um tech transformation that was a prior wave right like 10 years ago and you had all these tech based rollups like Compass was supposedly a technology company and it's a great company but it's not really a tech company right and so I think um we've we've kind of lived through the cycle before where people do almost like tech fake tech tech implementation where They claim they're doing it, they get a higher valuation and it doesn't quite happen.

They still load up the company with debt. They still run it a certain way. They, you know, the the drivers are different. The CEO is the wrong person, etc. I think there's a good version of that to be done.

I don't think it's easy, but if you do it, I think you can unlock an enormous amount of potential in companies that won't make it otherwise or won't go there. And so, yeah, I think that's a super interesting area and we we've looked at a few things over time.

Um, and part of the decision sometimes is like do you want to try and do that or you just fund a startup that you think will get there instead and uh it's a little bit of that it's kind of easier to just fund a startup because buying a company and transforming it is quite hard.

Yeah, it has to be a company that is I mean is is uh at a sufficient scale. Obviously, we just saw that the EA LBO. You don't need to go that big, but finding something that's like really a whale. Can't hurt to go that big, though. Yeah, I would like we would like to see you go that big.

I would love to see I feel like that's the next that's the next I feel like you've done it all at the early stage. You've done it all in growth. I feel like just get into the How about an Activision spin out? Activision spin out. Take Microsoft Game Pass, Xbox lagging. Come on. I know. I'm moving too slow.

You're like my mother. You're putting all this pressure on me. I know. That's that's our small ask is set the record. Wait, wait. Do you have Do you have any We like to ring the gong for people around here when they show a number. Do you have any number you can share? What's your favorite number of deals?

Number aum number I don't know anything. Do you What's a number that that quantifies your your corpus of work? Number of startups founded. my corporate. Oh, I've started um well, I've started two companies directly and there's two I've incubated. So, four of us is that's a lot of business.

Relatively modest amount for the amount of EV created. So, it's good good hit rate. Well, thank you so much for stopping by. This was a lot of fun. Really glad we'd love to do it again. We'll talk to you soon. Have a good day. Really appreciate it. Take care.

Uh before our next guest joins, let me tell you about numeralhq. com. Sales tax on autopilot. spend less than 5 minutes per month on sales tax compliance. I hate to pick favorites. I hate to say it. I that was I was I was more I was more engaged for for a lot than uh Aladd was great then was great.

Sora But uh his book High Growth Handbook Yes. is fantastic. It's a series of interviews. Yeah. Is my favorite uh favorite business book. Yeah. No, it's a great high growth handbook. Scaling startups from 10 to 10,000 people. interview some