How I Invest with David Weisburd - E402: $92B Coatue: Where the Value in AI Will Accrue

Episode Date: July 13, 2026

Everyone is asking which AI company will win. Lucas Swisher thinks investors are asking the wrong question. The biggest opportunities won't necessarily come from picking a single model or application. They'll come from understanding where durable advantages are created across the AI stack. Drawing on Coatue's investments in companies like OpenAI, Anthropic, Databricks, and SpaceX, Lucas explains why talent compounds, why data infrastructure may outlast today's application boom, why companies become harder—not easier—to disrupt as they scale, and how AI is reshaping the economics of software, semiconductors, and enterprise technology.

Today's guest is Lucas Swisher, co-head of growth investing at CO2, investors in Open AI, Anthropic, Databrox, Cursor, and Harvey. Today, the biggest question in AI isn't whether the technology works. It's who captures the value. Lucas explains why the biggest opportunities in AI may not be where investors are looking. Without further ado, here's my conversation with Lucas. So if two years from now, GPT7 comes out,

and it's dramatically better than GPT today. Are application companies cooked? I think this is a great question. It's obviously something we think about a lot. I mean, we're very large investors in Anthropic, OpenAI, and also some of the application companies, right, cursor, which was just announced that it's acquired by SpaceX, thank you, as well as other companies like Harvey and Open Evidence.

And I think we think about the world that it's not black and white like this, right, that if Anthropics models or Open AIs models continue to get incredibly better and you continue to see improvements over time that all of a sudden the application layer vanishes. We're not sort of doomsdairs on the application layer. How I think about it is it's almost like a bar chart. If you're to imagine like the X axis are all the use cases, right? So you think like legal, medical, HR, finance, all the way out to like some operating

system for car washes. Obviously, the language models are going to have some advantages in some and some advantage and not really have other advantages in others. And how do we sort of think about like where that bar chart fills up? Where does Anthropic and Open AI get 100%? And where do they get 30% of a market? What's your framework on figuring out which application verticals will get disrupted versus those that won't? Two things that I typically think about here.

One is how close is the application to the model itself, right? What categories are the application actually just the output of the model? And I think on one extreme, you can think about coding as of this, right? It's talked a lot about historically Oz cursor have a right to exist. And the reality is encoding, the output of the model, the application, is really tied very closely to the model. In that part of the market, you really think maybe the LLMs themselves will take the vast majority of the market. I think you can contrast that to something else where maybe there's a lot of meat in between the application that you're trying to build and the model itself. That may be that you have to do very deep integrations.

And maybe that there's a lot of compliance risk. It may be that there's unique data that you have to access. All those different things may end up giving you some advantage to be an independent company. But I think the biggest overarching thing for all of this is, are you really in an independent situation where the application-oriented company or the application layer company is going to need to use multiple models at the end of the day. You just take AT&T as an example. They may, in legal, not really want to use just one model for a variety of reasons. Maybe Anthropic doesn't stay at the frontier or Open AI doesn't stay at the frontier. They want to be able to use multiple models.

We've heard a lot about token maxing, right, and token minning over the last few weeks. They want to be able to control cost. They want to be able to control compliance. They want to be able to control security. And at the end of the day, a legal team at AT&T may just want to have a lot more control than one LLM provider is going to allow them. So I think about it like a spectrum, coding on one end of the spectrum and maybe like carwash operating system on the other end of the spectrum, where on one end, Anthropic may not want to focus on car wash operating systems. They may focus a little bit on legal and they're going to focus a lot on coding. And they really have a right to win there. The way that I think about it is, is the data itself a network effect.

And is the model actually improving itself based on the inputs? So you talked about Harvey, we're investor in Lagora, their competitors in the legal space. As you feed in more legal information, the models get better. Why? Because legal is game theory. Once lawyers come up with some provisions and other lawyers will come up with other provisions and it's contextual to what's going on in the market versus something like coding may or may not have this perennial right answer. That's three quarters of the way.

there. I think a lot of it does have to do with the data that you're able to get from the end user. And can you build a flywheel around that in order to train the model so that you end up getting better over time? But I do think at the end of the day, a lot of it does just come down to enterprise decision making. Enterprises want to be able to use multiple things. If you use the analogy back to the cloud, for example, very few large enterprises went all in on AWS. Very few enterprises went all in on Azure. They really use all the different clouds. Some clouds spike on different things. Microsoft has fantastic distribution. They're very strong in security. They're able to scale exceptionally quickly and they do really well with large enterprises. AWS did really well

with developers and startups. And so I think you'll see a similar dynamic, both in the clouds, but then also the application layer that sits on top. And the reason there's the spikiness is because there's inherent tradeoffs into the different models. Do you expect that to continue to evolve in the LLM space or do you see this one right model emerging? I think we continue to see tradeoffs. I think one of the big misconceptions in AI are things that a lot of folks believe is that we're heading toward a world where the language models themselves are completely commoditized.

This is completely wrong. What you're seeing is if you look at just the actual demand curve and you look at where folks are actually adopting, they're incredibly hungry for the frontier. If you look at what's happened with Fable over the last week, I think this is a fantastic example, right? There's been a huge Twitter storm since Fable came offline.

Please bring back this frontier intelligence for us. This is really important. it gives evidence to the fact that people are really hungry for the frontier that keeps getting better and better. Obviously, you don't need the frontier for everything. You're going to be able to use cheaper models, smaller models for certain tasks. But currently, we see that, one, the models continue to improve at a very rapid pace. I think if you look at the last 18 months, you can see that the frontier has gained meaningfully over open source, for example, and that they continue to develop at a more rapid pace, primarily because, one, they have access to great talent,

they have access to great compute. But then also at the same time, right, they are training the incremental model on a model that we're not even using. So Anthropic has access to a model that we don't have access to to train their incremental model. So it's sort of like a sprinter that's able to run at a really fast pace,

but that pace keeps increasing while I just get to run at the same pace behind them. In certain use cases, especially in consumer, there's only so much benefit you could get from LM. There's only such a good Mexican restaurant you could get in town On the enterprise model, there seems to be this hunger for this next frontier model. You can really clearly see this example between the enterprise and the consumer over the last six months. I think the consumer models, the consumer offerings have gotten very cheap. Oftentimes they're given away for free.

And it's because the benefit to the consumer of the model, so far, at least, the benefit to the consumer hasn't really changed that much. You also have a lot of wars going on in consumer between different companies with pricing pressure. You're seeing less of that in the enterprise because the frontier matter. more for that return on invested capital. Now, delineating use cases, it really widely varies, right? It really matters. Developers, for example, seem to really care about the frontier. The ability for agents to operate at longer time periods, more long form, runoff and work

for 24 hours versus the eight hours prior, that makes a huge difference. Doing a really simple task in and around HR, like querying a database about what is my PTO policy this month, maybe you don't necessarily need Fable 5 for that. maybe you can use a haiku or a sonnet for that query. So I think it depends based on the use case, but we really do see the hunger and this insatiable demand for that frontier. I want to push back a little bit on what you said about coding and maybe the LMs will win those coding wards.

Coding, it feels like on the surface is this finite problem where you have the right answer. But the deeper down you go, the more infinite possibilities and the different ways that the product could interact with either business or consumer. Is coding truly something that's going to be to be to be? disrupted by the core elements? If you look at how it's being used today, almost certainly, I think if you were to look at last year,

and I like to think about, okay, what's happened over the last six months, nine months, 12 months? If you were to look six, nine, 12 months ago, really these models, what they were being used for is to auto-complete, right? They were auto-completing a line of code. They might help me with something like that. Then toward the end of last year, we got to a point where, okay, the agent can actually go off and write a big block of code for me and bring it back.

And now we're at the point where the innovation is such, that I can send agents and their agents off. They're spawning other agents, subagents, and they're going and writing entire programs for me. And often for 10, 12, 14, 16 hours at a time. So I think that's why the power of that application in particular really does reside within the language model itself is because those innovations keep driving you to the frontier.

The other way you can look at it is just adoption curves and revenue. If you look at the application companies, the independent application companies versus the language models themselves, and you look at Anthropics revenue, which has been widely publicized to be over 50 billion of ARR now, and you compare that with the sum of all of the independent application companies. It's like 10 times bigger, right? And so I think you can see the evidence that people really clamor for that sort of frontier technology. And then in particular, in the coding use case, it matters a lot.

One year from now, who's in the max of it? This is a really interesting question because we talk about this all the time internally at COTU, right? obviously we are half private investors and half public investors by the assets that we manage. And at the beginning of every technology wave, you sort of see whatever the Mag 7 is of today, completely shaken up. Obviously, in the last generation, we had a fang, and then we had manga. Now we have the Mag 7. And I would posit that the Mag 7 is already over.

If you look at just the Mag 7 and how it's traded over the last six months, the NASDAGs up, I'd call it 18% rough justice. No single stock in the Mag 7 is up that much. So how can that really be the Mag 7? And it may just be that the time for those stocks as a bundle is over in some ways. I think if you think about the Mag 7 of the future, what does this actually look like now? It seems to be there are a number of categories where, you know, there are new companies that may fit very well. You have the chip companies that have done exceptionally well over the last six months.

You think about chips. You think about memory, the accelerators, the CPU companies, companies like TSM or Broadcom. You have private companies that are very, very, very. very close to becoming public companies. SpaceX, obviously one of them, the two-language model companies in Anthropic and Open AI, and then companies like ByteDance, which I think people often forget sometimes, but they have net income that's getting close to $100 billion, right? That's an incredibly large scale from a profitability perspective, still growing 20, 30% year on year. And then, of course, you have a number of other companies that have done very well in the public markets as well. So

I would almost maybe posit that the Mag 7 may be over. Maybe you have like an AI MAG8 or an AI8 or something like that. And maybe it looks completely different. It's the three company or the four companies that we think might go public soon. Bite Dance, which has a fantastic AI business and AI model, Open AI, Anthropic, SpaceX, Amazon, which is benefiting greatly, Google,

Nvidia, and maybe a TSMC. And so instead of focusing on, all right, there are some great businesses from the past, right? You're honestly seeing a lot of the growth and a lot of the growth in the valuations and the value of companies coming from a completely different subset of businesses now. And I think that, whatever you want to call it, like an AI-8, those companies seem to be benefiting more than the Mag 7 of the past.

If you had to remove one company from the Mag 7, what would that be? It's so tough. All of these businesses are so exceptionally well-run. But I think the way that you would think about it is, all right, who's best positioned in an AI world and who's worst positioned in an AI world? And I think there are some, a couple of companies that have a lot of upside potential and a lot of downside potential, right? I'd say you think about Apple like this. they have an incredible platform, four and a half trillion dollar company,

incredible business, so much durability, we're using the devices all day long, incredible mode, but not really an AI strategy. There hasn't really... Completely missed AI. There just hasn't been an AI strategy yet.

And so you roll that forward six months, who knows what happens, but five to 10 years, other people are certainly going to work on other devices and ecosystems with intelligence behind them. So maybe that's one. I'd put meta in this camp as well.

Obviously, they haven't quote unquote missed the boat on AI. They're investing in incredible amount in CapEx, an incredible amount in bringing great folks like Alex weighing over. We were investors in Scaly. I know. But you see sort of the upside and the downside of what they've been able to accomplish so far. They sort of levered to this open source ecosystem. Open source hasn't been really successful from an adoption perspective here. There are rumors and articles about culture challenges within meta. And you see it in the stock price. You've got a company that's worth less than SpaceX now, right, which just went public.

And so I think there's a ton of upside there because they've been investing a lot. But there's also downside because they sort of haven't figured it out yet. There's six layers in the AI stack. Power, chips, data centers, data infrastructure, and applications. Five years from now, which ones are going to have the highest ROI? It's such a good question. Also something, as you might imagine, we think about a lot at KOTU trying to connect all of these dots across the public and private worlds.

You look at all of those different categories and to really stand out to me. The first is the data infrastructure layer, and the second is the language model layer. If you think about some of the other layers, chips being constantly disrupted, you think about the data center, there's questions on terminal value, all the other layers, application we've already spoken about a little bit, right? There's a real bull case, but there's also a real bear case on. If you just take data infrastructure and the language models, those feel like the core backbone of the foundation of the future in AI and the enterprise

and really what we're building. So take data infrastructure first. a company like Databricks, fantastic business. We've been involved with them since 2017 over many, many rounds. They've really reinvented themselves to almost be this context layer for all of the data in the enterprise that AI needs to interact with Inquiry. That's a fantastic place to sit. And what do we know about database business models?

They're fantastic and they outlast many, many waves in technology. Think about Oracle from two generations ago, not even cloud, but also the on-prem world. Those core databases are still around. So in terms of staying power five years out, that seems like a fantastic layer, especially with a great CEO that understands how to adapt. And then you think about the language models themselves. And I think about a couple different dimensions there. One is we're always trying to watch the talent flows.

All of the talent is flowing to the language models. Some to the applications as well. It becomes self-fulfilling. It becomes a little bit self-fulfilling around where the talent flows go. Right? They have talent and they have capital. Those are the two inputs around AI and intelligence. and it seems like they have this compounding durable advantage at scale now.

Not dissimilar in some ways to what you saw with the cloud businesses of the last generation, but maybe a little bit on steroids just in terms of the black hole that they have sort of amounted to in and around talent. And so we really like those two layers. We actually like the entire ecosystem. At Kutu, we invest across the entire ecosystem from power to land to chips and semis, down to the data infrastructure to the language model and the application, all of the different layers we've invested in. But we're just more careful and want to be more mindful of the changes that happen in a lot of the kind

of periphery. But those two categories, we feel very strongly that five years out, those are great bets. Most investors don't lose because they lack information. They lose because they can't separate signal from noise. Every day, thousands of earnings calls, SAC filings, expert interviews, and market updates, compete for your attention. The challenge is in finding more research. It's finding the few insights that actually change your investment decisions.

That's exactly what I unpack in my conversation with Ryan Fenderty. Instead of talking about AI in theory, we discuss how leading investors are using it to surface differentiate research, move faster than competitors, and make more informed investment decisions. If you're an investor looking to gain an edge through AI-enabled insights, I think you'll find this conversation valuable. You can access this limited release episode by going to alpha-sense.com slash how I invest. That's alpha-sense.com slash how I invest.

Why do you think everyone's focused on the foundational market? model the LLMs. The most obvious reason is the companies are growing faster than we've ever seen anything grow in the history of technology. This is sort of the most obvious reason. I think if you just look at these two businesses, right, just take Open AI and Anthropic and you kind of shove them together. The most recent rumored revenues are over 80 billion between these two companies. That's been in the last three, four years. We've never seen anything like this growth rate, adding $10 billion a month of revenue in the history of technology. And then I think the second thing is, just the broader impact that this can have on jobs, on the economy, and this sheer market size

of what these companies are addressing. One of the reasons that I think they've grown so fast is we've moved, especially over the last six months, is we've moved from a world where these are nice tools that sit on our desktop and we can ask them to go retrieve some data or answer a question for us to actually doing the work, right? And that doing the work. They're integrated in the workflows. It's early days, but they're becoming integrated into the workflows. And what happens when you can actually do that is you can address something that's bigger than software spend. It's labor spend, right? It's actually going and doing the work. That's a leg of productivity. The technology hasn't addressed since really early robotics, since the invention of a tractor versus a plow.

Like that's the level of innovation that we're seeing here. And so I think it's those two things. It's sort of the shock value of just seeing the rapid rise of these companies and how fast they've grown. And then us knowing that we're like in the point one, point two. inning of the diffusion of this technology. And so thinking about what that actually means for broader society, I think that's why we pay so much attention to these two companies in particular. If you think about the ramifications for the broader economy, and you look at GDP growth, for example, over half of GDP growth this year has been AI infrastructure build out. Why do people have the confidence to do that AI infrastructure build out? It's because of the revenues of

these two companies. The whole economy, all of the growth in the economy, is on the back of these two companies in particular. So I think when you think about just all of the chess going on in the broader AI landscape, that's why there's so much focus on these two names. One of the reasons I wanted to get you on the podcast is you're one of the only sizable investor in both Open AI and Anthropic. You have a front row seat to these LM models. There's this idea that the LMs will be able to capture a percentage of labor.

You mentioned this. How would you further refine that model? It's a really interesting question, in particular because We're at the very early innings of this. And so maybe I'll use just an example, which is the first real use case where we can feel this happening, which is with developers, right? And I think the way that we've tried to capture this and try to understand, okay, what is the actual size of this opportunity?

What is actually happening is sort of twofold. And you can kind of triangulate it in a couple of different ways. So take the developer just to start, right? We have developers at KOTU that are spending an inordinate amount of capital on tokens. And that means that we don't. have to hire the incremental developer in order to come in and do that same task. And we see a real ROI from those pieces. So we're looking at, all right, what percentage of that overall pie do we think we can replace or augment or just juice productivity from? And that's kind of how we're

thinking about the TAMP. The other interesting way to think about the opportunity size of these things is you think about the average spend or the ARPOO of a Claude Code or a Claude Co-work. And you combine that with the tokens that people are using on top of a Mac subscription. And you compare that versus a Microsoft license, for example, which is a general productivity tool. And it's five or six times today with the models being the worst that they're ever going to be these models. And people forget, you look at these revenue numbers, like $50 billion in Anthropic. We're investors as well, but that's coming from a very small portion of the population,

essentially only developers, which is probably 0.1% of the population, if even that. And they have a hard time imagining that $50 billion has a lot of room to scale. It's really interesting. I've actually thought about AI and three waves so far, and I think this kind of gets to your point. Wave 1 was obviously the consumer. It was that explosive moment around chat GPT where the consumer curve and the consumer S curve started. And maybe we're about halfway through that S curve in terms of penetration of the consumer. The second S curve started toward the end of last year, whenever we saw the explosion in developers

encoding. We have no idea where we are in that S curve because, one, we're low in terms of the actual penetration percentage of developers, But then two, even within developers, but then also that number that they're spending per month is also going up every month. And so you have this dual S curve that you're writing just in developers. And now what we're seeing is we're starting to see that bleed out into the other use cases. It's why Anthropic and Open AI have both launched services arms and integration arms. Because then we're not stopping at coding, right? Coding this S curve that maybe we're in, I don't know, the first inning or the second inning of is we're writing that right now, obviously.

but then you think about, okay, well, what happens in legal? What happens in HR? What happens in finance? What happens in pharma? What happens in all these other use cases of the economy that we have barely touched yet? You think about, again, all the other application layer companies, Harvey, for example, it's very early in its adoption curve. It's incredible, right?

They have so much room to run because we're just starting to actually be able to integrate this technology in and address the broader pool of spend that we've now demonstrated that we can with developers. Speaking of these three waves of AI adoption, there's also almost these three waves of philosophically what AI will do to the future. The first wave was AI is going to disrupt all jobs. Now in the second waves, you actually see the more developer job listings than there are developers in the United States. So at least within developers, AI has actually increased the need for more people than it has disrupted their jobs. Where do you see this playing out?

This is another thing that I think a lot of folks really believe. leave is that AI is going to lead to massive job loss and we're going to have a huge issue in the economy. I think certainly there will be displacement. At the beginning of every technology wave, there's technological based job displacement. It's an economic concept, right, that we talk about. But there are new jobs created on the other side of that in every wave of technology that's happened so far. And all the evidence in this wave so far actually just points to people are becoming more productive. I think that's actually a very exciting future that a lot of folks don't talk about, which is, actually, I can just do more with less? Now, there is a question,

can people that are highly analytical ultimately make the shift to other things? And I think some folks will be able to. Some folks won't be able to. But the reality is the job's still there. We're just all more productive. I think that's what you're seeing with coding today, right? We have more developers at Kutu than we did a year ago. But all of our developers are way, way more productive than they were a year ago. And I think that's what we're going to see across the economy. My model of this is David Deutsch's beginning of infinity. in which he argues that innovation is infinite, meaning as you make innovations and you have other innovations, they start to innovate and integrate with each other, which create new innovations.

So the theoretical maximum of the number of businesses, the number of products, number of innovation as a totality is infinite. I love this framework. I think we don't live in a zero-sum world. We live in a positive some world. I think this is the biggest misconception around technology is that everybody just moves the pie around. Technology fundamentally grows the pie.

it makes the whole world more productive. We saw this whenever you move from a plow to a tractor. At the end of the day, it just enables you to do more. And there are new jobs created manufacturing that tractor. You see it in machining. You see it in manufacturing. You see it all over the economy. And what it enables the world to do is become more wealthy.

And wealth is not zero sum. Wealth is not a dollar bill. Wealth is a unit of productivity. And I think that's what AI is starting to enable. and I think you have the early inklings of this. Certainly there will be cost reductions in areas. I think there are really obvious places where that's going to be the case like customer support

that will be replaced by technology. But you're also going to have the creation of incredible jobs around four deployed engineers to be able to get the technology to work in production. I will tell you, being on the board of some of these AI application companies and in the room and a lot of discussions with Anthropic and Open AI and others, there's insatiable demand for that type of engineer. and the salaries are much higher than they were for that last generation.

It's certainly much higher than for our customer support rep, right? And so I think that's how I think about the world. It's not zero sum. It's very positive sum. And we all have the opportunity. And we all, AI is creating the opportunity for the world to just become more wealthy. There's a contradiction to how people think about this because there's a paradox. On one hand, they want to use AI to disrupt people.

On the other hand, they wish they had more capable people to do things. even business owners themselves, whether enterprise, small businesses, they themselves are constantly limited and subconsciously don't realize how much the lack of smart and capable people in their organization is limiting their growth. Said another way, as soon as you had much more productivity and much more people available that you could hire for these positions, people's minds will become more open towards where they could take their businesses. Of course.

I mean, we see it with many businesses. folks and businesses are constrained by the amount of people that they can hire at the end of the day. We all want to do more. We all want to build more. We all want our businesses to grow, not shrink. And at the end of the day, a lot of that is constraint for talented folks that can do what we need them to do on a day-to-day basis. And so AI is giving us superpowers. And so I think that's the key.

It's not that we're sort of barreling toward a world where, of course, everyone's going to lose their jobs. That feels like a little bit of a misnomer to me. and something that's never happened with new technology waves in the history of technology. There certainly will be displacement along the way. There are always things that are shaken up in the economy at the beginning of every technology wave. But at the end of the day, we're always better off than we were before. Other VC firms are also starting to invest heavily into AI. Do you see second order effects of this being that there's going to be much more consensus bets

if everybody's looking at the same data? I'm not sure. I think we sort of come from different places. I think if you think about someone like us, our heritage is in the public markets in many ways. We think about very big ideas. We tend to be very market structure driven as a firm. Support for today's episode comes from Square. It's all in one way for business owners to take payments, book appointments, manned staff,

and keep everything running in one place. Whether you're selling lattes, cutting hair, running boutique, or managing a service business, Square helps you run your business without running yourself into the ground. It's actually thinking about this the other day when I stopped by a local cafe here. They use Square and everything just works. Check out as fast, receipts are instant, and sometimes I get loyalty rewards automatically. There's something about businesses that use Square.

They just feel more put together. The experience is smoother for them and it's smoother for me as a customer. Square makes it easy to sell wherever your customers are, in store, online, on your phone, or even at pop-ups, and everything stays synced in real time. You could track sales, manage inventory, book appointments, and see reports instantly whether you're in your shop or on the go. And when you make a sale, you don't have to wait days to get paid,

square gives you fast access to your earnings through square checking. They also have built-in tools like loyalty and marketing, so your best customers keep coming back. And right now, you can get up to $200 off Square hardware when you sign up at square.com slash go slash how I invest. That's SQ-U-A-R-E.com slash go-slash how I invest. With Square, you get all the tools to run your business with none of the contracts or complexity.

Run your business smarter with Square. Get started today. Summer is here, which for me means occasionally trying to escape New York City on the weekend. When I get time off, the last thing I want to do is worry about keeping my personal finances organized and my budget and balance. With a little advanced planning and powerful software tools, I can enjoy the summer knowing my money is taken care of. Monarch is the personal finance app that tracks everything.

Accounts, investment, saving goals, and spending. Get your first year of Monarch core for half off, just $50 with promo code invest. What I like about Monarch is that it takes the mental load of managing your finances off of your plate. It's checking multiple accounts and spreadsheets, you could see everything in one place. The AI weekly recap is especially useful because it flags spending spikes and upcoming expenses before they become surprises. I also like Monarch's AI assistant. You could ask questions like how much I spent on travel last summer or how many subscriptions am I paying for and get answers instantly.

Use Code Invest at Monarch.com to get your first year off, Monarch core, half off at just $50. That's 50% off your first year at Monarch.com with code invest. Summer's here, which for me means occasionally trying to escape New York City on the weekend. When I get time off, the last thing I want to do is worry about keeping my personal finances organized and my budget and balance. With a little advanced planning and powerful software tools, I can enjoy the summer knowing my money is taken care of. Monarch is the personal finance app that tracks everything.

Accounts, investment, saving goals, and spending. Get your first year of Monarch core for half off, just $50 with promo code invest. What I like about Monarch is that it takes the mental load of managing your finances off of your plate. Instead of checking multiple accounts and spreadsheets, you could see everything in one place. The AI weekly recap is especially useful because it flags spending spikes and upcoming expenses before they become surprises. Also like Monarch's AI assistant, you can ask questions like how much I spent on travel last summer or how many subscriptions am I paying for and get answers instantly. Use Code Invest at Monarch.com to get your first year off, Monarch core, half off at just $50.

That's 50% off your first year at Monarch.com with code invest. We think a lot about data. Data has been at our core for a very long time. just a lot of signals and try to understand and understand what's sort of happening in the world and create a mental model of the world based on that. We come from a very different place than say someone like a founder's fund, right? They're primarily focused on the early stage. They're very founder-centric. They think a lot about, I mean, famously, most of what they think about is do we like this founder or not?

We come from very different places. And so actually, it may mean that we have a little bit more dispersion. For example, one of the places that we've been spending a lot of time as a firm is at the infrastructure layer, right? with the data center layer, the chip layer, the data infrastructure layer. That's not a place that a lot of other firms have spent a lot of time over the last five years, right? VCs haven't invested in a lot of new chip companies. My partner Thomas led to series B and Cerebrus, for example,

and we made an investment in Corweave when many VC firms thought it was not a business at all. And so I think it may actually tend to let us play to our strengths in a certain way and mean that the world becomes a little bit less consensus. Speaking of data, one of your co-founders, Thomas LaFont, shared some data on how as companies get bigger, there's a higher probability that they could 10X. This is really rattling the venture ecosystem. Tell me about that. This is really interesting. And it's a data point that comes because we analyze a lot of our own data internally.

And we look back on, okay, our track record, where have we done well? Where have we not done well? And then the ecosystem, right? The track record of the ecosystem. Where have folks done well? Where are folks not done well? And this is a counterintuitive data point. We actually figured this out two years ago.

What we found in the data is you're actually, you're much more likely to get a 10x in the pool on a percentage basis. If you're above $10 billion in market cap, then below. You're even more likely. If you can get to $100 billion, you're much more likely to get a 10x on that investment than the 10 to 100 or the 1 to 10 range. And so we found this is really interesting. And I think what it really speaks to is when you can get to scale, your advantage tends to compound. It tends to be durable and it tends to compound over time. You're a better magnet for talent.

You have more access to capital. Maybe you have a network effect or some business model that enables you to sustain yourself. And I think we're seeing this as an example with OpenAI and Anthropics separating themselves, and maybe SpaceX now, separating themselves from the rest of the language model market. Their advantage at scale has grown versus the quote unquote rest of the Neo Labs in some ways. And I think you're seeing this in revenue. You're seeing this in the talent that's concentrating within those few names. And it just gives them this flywheel and this advantage to be able to continue to get bigger.

It's so interesting because there's been this paradigm investing that ideas are cheap. Execution is what matters. And with AI, some are arguing that it's actually the ideas that are the value at it and the execution that's more of a commodity. Then you have to think about where do great ideas come from. They come from talent. So really what you should be focusing on hiring the best people, that could come up with the great ideas. That is exactly right. I mean, when you think about just the inputs around what is making.

the language models that are successful today, very successful, the Open A's and the Anthropics of the world. The key input is obviously compute. You have to have the compute. But then outside of that, it really is the talent, right? If you think about Anthropic as an example, very famously, they've lost one of the 17 original employees. It's incredible.

How many companies have been able to do that at scale? If you think about Open AI, they're able to attract talent that sometimes has left and even come back. We've seen some of this over the last three, four months or so. And that's a huge weapon for these companies. Talent really matters. This idea of ideas and execution is super interesting. I'd use another example, which is Databricks.

If you think about Ali, the CEO of Databricks, he's been able to drive this company to hop multiple waves in technology. It's so rare to be able to shift from being a company that helped move data around the enterprise, to being the core system, the core data warehouse, to now being the central repository where a lot of folks interact using AI with data in the enterprise. And being able to hop those waves, that's talent, right? That's all down to the talent. Snowflake also has the same opportunity, but if you look at their growth rates, it was at least published a couple of days ago and that Databricks's growth rates over 80% now.

I think it was published last year that was kind of the mid-double digits. Snowflakes run 34%. A lot of this has to do with team and the decisions that they've made. That's where it really starts to matter. Is this talent in an AI world matters way more frankly than it did in a SaaS world where all SaaS companies tasted like chicken? They all looked very, very similar. In this world, it's less rinse and repeat. It's more reinvention and adaptation.

And I think that's making a huge difference. So another way, Ali from Databricks and other great founders have figured out a way to break innovator's dilemma. There's been this innovator's dilemma that every new technological wave disrupts the previous technological wave. But if you're able to continue to think from first principles and recruit the very top people, you're able to continue to grow your moat. That's exactly right. It's very hard. And it's actually what we look for a lot is companies that are reinventing themselves. We found that that's one of the strongest scale is great.

We found that's actually one of the strongest signals of whether a company is going to be an enduring great public company over time is, can you reinvent yourself? Can you adapt? Can you launch new products that launch into new TAMs that were bigger than your prior, that bigger than your prior wave? Why one of the things, one of our characteristics of a COTU platform company, right, is that you have multiple products that have worked. What's the best way to predict that you'll be able to create new products in the future? You've done it already. What differentiates a company that continues to innovate? and continue to reinvent itself versus one that is not able to do that.

I just use KOTU as an example of this. I think Philippe and Thomas, our co-founders have been phenomenal at doing this, executing this in our business. I think it is all driven from the top. I think it is all driven from the management team that is willing to not sit on their, not rest on their laurels, that works harder than many of their employees, and that has a mentality of adaptation and change.

Technology is all about change. Investing, frankly, is all about change. The world changes extraordinarily quickly and being able to adapt is ultra-critical. I think that is the number one skill set. It's this paranoia, this necessity to always be reaching for something that you don't have, right? That there's like willingness to adapt and change that really drives this. I previously had Eric Becker who co-founded at Crescet, which today is $250 billion family office. He was sitting right where you were sitting.

And in between, before he started Crescett and he had several successful exits and he spent, several years going around and interviewing these family businesses that were around two, three hundred years. He wrote a game. He wrote a book called The Long Game. One of the things that he figured out is that if you're truly trying to build an enterprise that survives 100, 200 years, there's really only one source of leverage and that is culture because culture is the only thing that could lead to the right talent and talent is the only thing that could navigate you through many different cycles. That makes total sense, right? Again, there are fantastic companies that have very different cultures, I think, almost famous.

at this point. There's no one good culture. There's no one good culture. And I also think it differs based on the spaces. Would you want the same culture in LVMH as KOTU? Probably not, right? Being a great fashion house is very different from being an investment firm is very different from being a language model company. I think Anthropic does a fantastic job of this, for example, right? They are a very small team for the size of the company and they filter aggressively based on culture. And what do they really care about? They care about people that care about the mission. It's a very small team. It's think folks know that if you're working in AI, you're probably going to make a lot of money, right? I think they really aggressively filter for folks who care about the research. They care

about working at the frontier. They care about the product. They care about their users. They care about the impact they're going to have in the world. A good example of the culture being broadcast at the world, I think is Dario's essays. The two and now three sort of big essays that he's written to sort of broadcast. All right, these are the types of folks that we want at Anthropic. And I think they've held really true to that. It's why their retention is so strong. I think a lot of people, including myself, get a bit confused when they see engineers choosing between OpenAI, Anthropic, working at Google, and they see these incredible pay packages. And yet these engineers are making decisions based on culture. How do you explain that?

There's more life than money. That's the biggest thing. And especially at the incremental level, I mean, yeah, you think about these pay packages. You're going to be compensated very well at any of those three labs. You're going to be very compensated very well at SpaceX. That sort of ultimately, I think, is not the trade-off that folks think about. Some people, certainly, but I think a lot of folks really do care about the mission. If you spend a lot of time with AI researchers, you spend a lot of time with AI developers,

I think they care about something that is more than the incremental $10 million pay package over three years. That's just not something that is in many of their calculuses. And I think that's one of the ways that Anthropic and Open AI and others have done a very good job of retaining talent is by giving incredible talent incredible people access and leeway to be able to work on things that they're very interested in. Let's talk about power and energy, because without energy, none of this comes into fruition. What's the main bottleneck in power today? There's this sort of famous chart that we have inside of Kutu, which looks at energy production in the U.S. and energy production in China. And I think you can see in China, it's really a linear

curve that bends upwards. In the U.S., it's kind of like this. In the U.S., it's much more stable. And maybe it rises a little bit, but it's much more stable. And so I do think over the next five years, the next 10 years, certainly, we're going to have to have real innovations here. I think you're already starting to see some really interesting novel ways that folks are dealing with this problem. If you think about metanthropic, two of the folks that are investing heavily in CAPEX, they at their data centers are building the power capacity on site. I think that's super interesting because it's sort of a net zero or they're actually building excess power capacity. so they're feeding it back into the communities around them. And I think that's frankly going to be necessary

from a public policy perspective, but also that's one of the ways that they're already solving this. But I think on a five to 10 year basis, we're going to have to use everything that we have at our disposal. It's going to be traditional energies like oil and natural gas, down to all the renewables that have been a big focus for the last decade to new technologies or technologies that may be being revived like nuclear. Right.

And I think you're seeing a lot of actually really interesting, promising innovation. You have companies that are reinventing the battery or really reinventing storage of energy like base power, for example, which has a huge vision around this. You're seeing new nuclear companies come up like General Matter with Scott Nolan, fantastic early stage business. You're seeing new fusion technologies that may be on the horizon with Commonwealth fusion. I mean, other businesses that look like that. I think that necessity breeds innovation. There is a necessity here. We will breed innovation. We will build. I think there are a lot of companies that are both in the CO2 portfolio and not, frankly,

that are really focused on this. I think the public policymakers are really focused on this. We all sort of know this is a problem, and I think we will innovate our way to making sure that we're able to have the power that we need. Pasta Energy bottleneck is chips. InVIDIA today has roughly 80% of market share. Do you see a new chip company disrupting that, or do you see the hypers scaling their own chips?

There's almost like three things happening at once. you would have asked me three, four years ago, will Invidia be disrupted, knowing they had this incredible software platform around Kuta? And they basically had a monopoly on the chip market. I would have said, no way. They just have this incredible advantage. But I think what we've seen, and this is really started by, I think, Anthropica a couple years ago is the primary driver of this is actually you can run AI workloads on heterogeneous chips. They demonstrated that with the TPU, which is highly performant. They demonstrated that with Traneum. I think that's a really interesting example. So you have, I think, the cloud providers, right, that now have their custom chips. There's a lot of

news over the last week about Amazon selling their Traneum chips externally. Obviously, it required a lot of work for Anthropic to be able to use these chips. It didn't happen overnight. But it's certainly possible now. So you have the cloud providers as one source. You have custom silicon, right, with TSM and Broadcom, companies that enable you to build your own chips. I think you will see OpenAI, Anthropics, SpaceX has already spoken about this. You'll see a lot of the companies on the demand side of AI certainly try to build their own chips. And then what's really interesting is we haven't really actually seen new innovation at the CPU or the accelerator in like a decade. Cerebris and GROC, which was acquired obviously, and Cerebris, which just went public, a great CO2 investment, were really the last innovations.

And those were founded nearly 15 years ago. And so if you think about it, like we're sort of due for an. innovation, especially given we've started a new cycle. And I think you see really interesting companies coming up like Nuvecor, which is a fantastic team that came out of Apple that built a lot of the Apple chips that's building a new CPU from the ground up. I think you're seeing really interesting companies like Maddox and Fractyl, which are building new accelerators. Those are really strong teams. The Maddox team, for example, was one of the core operators in and around the TPU. Right. And so you're starting to see new companies. And you almost wonder, it used to take a really

long time for new chips to come up. In this world, with AI, can we get to tape out a lot faster? I think the answer is almost certainly yes, right? Especially folks that have done it before. And so, Invidia, it will continue to grow. I think the question is, what percentage of the market do they grab when you have all these competing forces now? And a lot of folks in the ecosystems whose incentive is to find something other than the GPU. It's still by far the best today. I think people acknowledge that, but this is something that we think about a lot. How would you explain to a layperson why NVIDIA has such a competitive advantage? One is they've been around for a long time, right?

They've sort of become the default. They are the default install. Two is the software platform that enables you to develop on the GPU is incredibly strong. Is that where the value is on the software layer, not the hardware? I think it's the combination of both. I think this is the mistake that a lot of folks make when they're investing is investing either in the software or the hardware. You want to invest in the systems.

That's what's made NVIDIA so strong for a long period of time. It's CUDA plus the underlying chip itself. And I think that's what you're seeing now is Anthropic, whenever they went to Google and they went to Amazon to try to use these chips, they had to spend a lot of time developing that system with them. It was a lot of co-development over a couple of years that enabled it to work. And I think that's what you're going to start to see, right? Those systems start to come online for all of these other competitors. And that ultimately may be the thing that sort of cracks the monopoly to a certain extent. I wonder how much of this is memetic.

We were also investors in Cerebris, and once you see these exits like Cerebris and GROC, the talent starts to amalgamate over these different thesises. And then suddenly, magically, these new chips are created, but it's always downstream of the most talented people focusing on a problem. Yeah. I think people are spending a lot of time thinking about chips now. That's an example of a lot of investors are now woken up to chips.

This is something that we have cooked to, who has spent a lot of time thinking about for a long, long time, in particular because the public markets, you have to pay attention to what's happening at the simile. And even Maddox's case, the fact that the market is now focused on this means, what does that mean? That means they could raise more money. What does that mean they could hire more and more talent? So even the ones, even the true believers, even the early investors are still benefiting from this. Absolutely. Absolutely. Everything that we've been talking about, all these different parts of the AI stack, all assumes that the application companies can execute. What

What mistakes are top founders making today at the application layer? The primary mistake that I think folks are making the application layer is just trying to drive to revenue too fast, right? Not thinking about the fundamental innovation that a company has, the fundamental thing that they are working on and instead just trying to chase customers and chase revenue before there's like some real durable advantage. One of the ways that I think companies that have been successful at the application earlier have been so is they've really focused on this.

Take Winston and Harvey, for example. They have access to very unique data. They've made very unique data deals. They focused both on the corporates and on the law firms, and there's a little bit of a network effect between them. And they've pursued a very specific strategy that entrenches themselves. So when one of the language models decides they want to come after the space, one, they're already embedded. And they're embedded in this network, but they have really unique access to data. And then three, they've built a harness that is able to use all of these tools plus the open source.

If you think about some of the things that are happening now, it's people are just because they know you can scale really fast, we know you can scale really fast. Folks are just coming out of the gate trying to get to 100 million of AOR as fast as humanly possible. Without thinking about those pieces, you want to take the lessons from the Harvys, which have been incredibly successful at the application layer to date, which is being really intentional about the innovation that you're having and the durable advantages that you can create. How do you square that with what we previously talked about, which is you want to hit the milestones so that you could raise these large capital pools

so that you could get the talent to help build the business? It's all a byproduct of the decisions that you make, right? obviously you want to grow fast. We all want to grow fast. All of these application companies want to grow fast. But the fact that Harvey has grown so quickly is a byproduct of those hard decisions that they made very early on to architect their products in a really specific way to go make those deals that they had to with the data providers to get access to that unique data. Everything else is just downstream. Of course you want to grow.

Of course you want to grow your sales team. Of course you want to do all these things. But if you do that before you have that sort of competitive durable advantage, you end up with a product that has bad retention. or a leaky bucket, or you're one of 50 and you're not the number one in your category. And I think the one thing we know is that being number one in the category really, really matters.

The disproportionate share of the gains in technology come from being the number one. And having that firm foundation is the most critical part to that. It's almost like you want to crawl and then you want to sprint. Yes. And a lot of people are just trying to walk from day one.

Over the next two years, what's the most likely way that investors in AI lose money? I certainly think in today's market, you have echoes of 2021. It's almost hard today to lose money. Yeah, which is what people thought in 2021. I remember investing during that period. Trees were growing to the sky.

Every company was working. Everyone was able to raise money. No matter the idea, if you were growing, you could raise money. We certainly see and feel echoes of that today. I'll just give you two examples of things that we're sort of avoiding in this market. One is pre-revenue companies that come out of the gate at very high valuations. I think that's one place where people may lose a lot of money in the cycle. Some of them will certainly work.

And the tradeoff there is these companies have exceptional talent. Of course. But it's also a way and in venture, I think that's one thing. But because you are always swinging for the fences and what we do, which is growth investing, you have to be a little bit more measured. It's going to be a part of the market where people do lose a fantastic amount of money. Is it in these pre-revenue companies at big valuations, particularly you think about like these NeoLab oriented companies where two folks in a PowerPoint will raise a billion on five billion out of the gate.

That's a really challenging. place to play. And then the second challenging place, I think, is in between the layers that we've talked about, right? You've noted that you've got land, you've got power, you've got the chips, you've got the data infrastructure, you've got the language model, and you got the apps. There are a lot of companies that play at the intersection of those, that their entire business model is in between the chip and the model. That's a very, very, very scary place. It's a very fragile place. It's a very fragile place. The world changes on you on a dime. It changes on you extremely quickly. And I think that's a scary place to be that if you're in a world where in the biggest compute shortage of all

time, you're a 20% gross margin business. What happens when the tide goes out? You could become the victim of your own success. If you're really successful, one of these two layers will disrupt you. Yeah, they certainly have an incentive to. AI is evolving so quickly that instead of asking what you've changed your mind off in the last year, I almost have to ask you over the last six months, what have you change your mind now. I think the biggest unlocked for me personally has been just how big I think this can be. I think I was at the view last year. I'm a growth investor, not a not a venture investor. And so we're as a growth investors, you're somewhat born of a little bit the west coast and you want a dream, but a little bit the east coast and some reality in the public markets. And we think a lot

about that. I think last year, we really believed that this would change the world. But the question was sort of how much. And we were in this incremental land where the models were certainly getting better, but the revenue was kind of growing incrementally or linearly. And it was really used like we've spoken about as kind of a co-pilot. These tools were even named co-pilots. And now we're in an agenic world. I think what that really unlocked for me is this true belief that the models can actually do work. They can do white collar work. And so that's been the biggest unlocked for me. And I think us as a firm and as a community, frankly, is understanding just how big the opportunity set is here is so much bigger than it felt nine months ago.

And what's kind of ironic is the folks inside the labs were telling us this, right? They've been telling us this for years. And sometimes as investors, it takes us a little bit longer to see it than the folks who are actually built in the technology. And before we started recording, you said that you are tethered in the public markets versus most venture capitalists are tethered in the private markets. How does that change how you approach investing? One of the benefits of a platform like KOTU and we are,

somewhat unique in this respect is we do live in both worlds, West Coast and East Coast. We live in between the intersection of the financial markets and the Dreamers on the West Coast. And I think in an AI world, this gives us just an incredible advantage. It gives us way more of an advantage, frankly, that it did in SaaS and that it did in the mobile internet era. Because in this world, you have to understand not just software. You have to understand what's happening with memory, what's happening with the tightness in and around memory, what's happening with the hyperscalers and their development of chips, what's happening inside of TSM, what's happening with Google.

And we're able to stitch all of those dots together and understand, or we try to understand what's happening in the world. And I think that's a huge advantage at this moment in time more than it ever was before. In the SaaS world, it sort of didn't matter. All the companies looked exactly the same. In an AI world, this really matters because the world is changing so fast. So being able to adapt and have all those data points is just this incredible competitive advantage that we have as a firm.

You become one of the top growth investors in the world, you Lucas specifically. but that was not always the case. If you could go back to right before you start as a growth investor, what would be one piece of timeless advice? I wish I would have dreamed bigger from the first inning. I started my career at Insight Partners in New York. I moved to Kleiner Perkins. Dream bigger is actually the slogan or the mantra of Kleiner Perkins, ironically.

But I think it's to dream bigger about the idea size of the companies that I was chasing. I think for a portion of my career, I sort of was chasing after some, call it more mid-sized ideas that were somewhat less transformational. And so I think it really would be to don't compromise on idea size, on Tam. Dream bigger. Lucas, thanks so much for giving me a glimpse of the future. Looking forward to doing this again soon. Thanks for wrapping me.