大多数风险投资基金都在玩动量游戏｜Michael Dempsey

I don't know if it's most. I think that a lot of venture funds are playing momentum games, and I think momentum games are not durable. Simplistically, I think all risk asset classes are really bad relative to buying the Nasdaq or S&P index over the past decade, and you need to be top decile. I think momentum is a hard place to be top decile in.

1. Price, Legibility, and the Myth of the Obvious Founder

So, yeah, I think investing is really hard, and it looks easy in certain 36-month windows. Usually, you find out it's hard in 48-plus months.

Today I'm thrilled to have with me Michael Dempsey, who's the managing partner of Compound and probably the clearest example I know of research as an actual edge in venture. Michael is a pretty rare thing in venture: a truly thesis-driven, high-conviction, original thinker who's obsessed with the craft of investing and staying in his lane.

He was early into Runway and Wave, which are some of the best seed investments of the last decade. He also runs what he's called maybe the slowest-deploying seed fund in the United States. Michael sat out most of 2021 and 2022 while most people were spraying, which earned him a lot of trust with LPs.

Michael, I've been reading your stuff for years. I'm really excited to have you on and talk about some of this stuff.

2. Why Momentum Investing Fails in Venture

Thanks for having me. What an intro. I love it.

I wanted to start with the thesis of this episode, which is that you've written that most venture funds being raised today are going to end badly, and the people running them won't find out until the tide goes out. Make the case for that.

I don't know if it's most. I think that a lot of venture funds are playing momentum games, and I think momentum games are not durable. Simplistically, I think all risk asset classes are really bad relative to buying the Nasdaq or S&P index over the past decade, and you need to be top decile. I think momentum is a hard place to be top decile in.

So, yeah, I think investing is really hard, and it looks easy in certain 36-month windows. Usually, you find out it's hard in 48-plus months.

Why is momentum the wrong strategy in venture, I guess? It's kind of an age-old debate in investing generally. You could trace it back to Soros's reflexivity versus the Buffett-like value approach. Why isn't momentum a good strategy in venture? And maybe define momentum as well—how you see it.

I think maybe defining it first: momentum is doing something because you know that there are marginal buyers at incremental price increases in the near term, not because you believe that it's durable over the long term.

The investors you mentioned have largely dealt in public markets, not private markets. The complexity with momentum is that in public markets, when momentum starts to slow, there is still a bid. In most cases, that bid will come down slowly. It might come down 10% in a given day, 15%, or 20%. Obviously, there can be pure crashes, but in most cases, the bid will come down slowly.

In venture, there's no liquidity. There's no scaled liquidity, and there's constant dilution and preferred stack that emerges. When the bid starts to soften, it doesn't soften 10% or 20%; it softens 100%. I think that creates some complexity.

Again, I think momentum is different from the more in-vogue strategy in growth, which is buying the best assets and believing that they have more terminal value than the market does. We could call that the Thrive strategy. I think that's a good strategy. That's inherently an economies-of-scale or assets-of-scale strategy, along with the ability to write the big check.

That's a highly access-constrained strategy, and it might actually be one of the most durable strategies in all of venture from the past 5 to 10 years. That's the high-level view, but I'm happy to talk through it more.

I agree with that. So, the difference is that in momentum, you're betting on future flows, whereas if you're paying expensive prices for things that you think are the winners, you're paying for the winners. I think people use the term momentum to refer to both of those strategies to some extent.

There's also a strategy that's emerged earlier-stage, at the seed and pre-seed stages, which is kind of an access-and-winning strategy: being in the best names and paying something like $100 million for seed rounds, or even more expensive than that in some cases. That sometimes gets looped into momentum.

What do you think of that strategy? I think that's something you guys avoid as well.

I think it's a fallacy to believe that the highest-priced things are the best things at the early stage. I think that's just quantitatively not true. Again, the highest-priced assets, once they start to compound in a durable way—which, with AI, you can debate what is or is not durable—might actually be quite a different form of underwriting.

I don't believe that the loss ratios of high-priced seed investments are meaningfully different from the loss ratios of low-priced seed investments. If you were to look at a bunch of data around where top-decile exits come from, it actually is pretty evenly split across valuations at seed.

Basically, the data shows that everything but the bottom quartile of valuations creates the same distribution of outcomes. I do believe that you shouldn't be trying to be a value investor at seed, but I don't think that quality is correlated at all to terminal outcome within the top 50% of price ranges at seed.

Interesting. I guess the theory would be that the best founders, or the best companies in general, are priced right. It's hard to find these diamonds in the rough, or at least it's hard to make a whole fund of them.

We've seen some people be pretty successful with this polarized strategy, where you're hunting in the backwaters for founders at an $8 million post-money valuation and making a fund that's half of those, and then maybe half of the flapping airplanes, you know, seed round, which is already, I think, $100 million or something like this. These are companies of that stature, where the founders are clearly incredible.

What do you think of that?

One can debate this, and I think we'll see in the data. I just don't believe that the most legible founders are the best ones. I don't fundamentally care if you're an amazing 20-year-old person who went to Stanford. I don't believe that your ability to build a multibillion-dollar company is 10 times more likely, or even 5 times more likely, than that of another really talented founder.

3. Portfolio Construction and Shots on Goal

I think what we're doing is correlating on-paper legibility—which creates some subset of minimum price—and meeting legibility, which creates a multiple on the minimum price, to quality. If I look at our 2 largest companies today, the founders of neither would even today be considered super-legible: the Runway team or the Wave founder.

I've seen enough of this, and even in some of our more successful newer companies, I'd say the same thing. I just don't buy it. All of this is actually coming to a crescendo at a time when we're all saying that the barrier to creation is falling because of AI.

Everyone's like, “Why? I don't want to do software because the barrier to creation is falling.” If we believe that we have a fundamental shift in the ability to create companies, technology, or whatever it is, and that the minimum viable level of intelligence is going up because we have these frontier models at our disposal, it feels very strange to me that the talent will actually narrow in where it comes from.

I don't know any other paradigm in which the barrier to entry has come down and talent has not had higher dispersion.

I don't think the Stanford thing is a straw man, though. I don't think what people are saying is that the really legible founders are necessarily just people who went to Stanford. That was the thing for sure at some point, but maybe now there's a more sophisticated, Hummingbird-inspired vision of what truly extraordinary talent looks like.

Maybe it's the person who came from a difficult upbringing, was a genius who did the Math Olympiad, got a gold in the Math Olympiad, then graduated from university early, and built a company—whatever, all these accomplishments that you can name off. I think that's what legibility looks like right now, rather than the Stanford thing.

Those are the things that get really priced. Do you still think that's the case for founders with those kinds of credentials?

I don't think that they create meaningfully higher outcomes. You might have certain pools of talent—I don't think the ones you described are—but I do think there are certain pools of talent that create floors for windows of time. We saw this during the rise of mobile, where if you were a very talented mobile engineer, you had a floor in your company because Facebook and other companies were acquiring mobile engineers for $10 million per engineer.

We saw this in self-driving, where if you had a background in self-driving from 1 of 5 places, or if you participated in the DARPA Grand Challenge, you were generally taken out for between $5 million and $15 million per team member. We see this now in AI. We'll see this in Neo labs and some of these other areas. Whether or not the investors get hosed on that is still to be determined in some of these instances.

I think you can then look at a few other things. You can look at the success rate of second-time founders building venture-backed businesses versus first-time founders. The data shows pretty much no correlation. You could look at the Math Olympiad as an interesting pool of talent that is clearly super high-intellect, and maybe in certain high-intellect problems these people are able to accumulate capital. There is some self-fulfilling prophecy here, which leads to the next-order question that people probably want to understand: Do you believe that king-making is actually possible within venture capital? That is probably the question that matters most in the legibility conversation.

Yeah.

It doesn't matter if I believe this or not if being legible allows companies to have access to capital advantages. Elon is the best example of that.

Yeah. Anecdotally, it does seem like there’s some math mafia thing going on—the Olympiad, or was it the Citadel intern class? That has the Hyper Liquid founder and Alexander Wang and the Cognition founders, Scott Wu. It does seem like there’s something there, anecdotally. Or is it something you think is overhyped right now and actually won’t be a predictor?

I think there are pockets of talent that sometimes produce really interesting people, and maybe that means 3 out of 800 instead of 1 out of 800. I think orienting a strategy of building a firm around either that or trying to continually understand what that looks like at a cohort level feels very undurable to me.

I think the Hummingbird thing is a little different. They believe they have a different view on what spiky people look like, which is actually far more horizontal than something like, “I hang out with 19-year-olds who are really smart in a few pockets of places.” Networks do sometimes compound in interesting ways. There are moments in time when networks are really interesting and good.

I think you might be able to make money by orienting a strategy around a certain type of network. It's not how I will make money, and I don't believe that the legible side of it is what creates the value. It's almost a lagging indicator in some ways.

With your Hummingbird comment, I guess you said earlier that you don't buy that—that you can build a fund around it. But maybe you don't. Is it that you don't buy that for yourself, but you think it can be done? Can you identify and develop this horizontal skill of identifying talent? I guess it's not legible, though. By definition, it's legible to you but not to anyone else. But there is a way to hone that skill of finding this sort of talent.

Yeah, for sure. If you talk to most of the people who have been at Hummingbird, left Hummingbird, or are there, they all source in very unique ways. They all think about how they find talent in very unique ways. Whether the percentage of that comes from the machine versus the fact that they recruit very well into the machine is also uncertain, but there is a commonality in how they think about sourcing humans in the world.

Again, I don't think it's them saying, “This cohort of people is interesting.” It's more, “Here are the flavors of people that we think will create possible asymmetry.” Historically, those people have been illegible and have also been very low-priced. Now that is not the case, because people are starting to orient more toward them, and there might be a bunch of other reasons why.

I think that's a strategy in the same way that what Thrive does is a very differentiated strategy. They believe there are certain types of assets that are really great, and they believe the outcomes of those assets are much larger than a lot of other people believe and that they will compound for much longer.

That's the original lesson we learned with the Magnificent 7 companies over a decade-plus: They compounded at a much larger scale for a long period of time than most people ever thought. Even after Marc Andreessen wrote “Why Software Is Eating the World” and everyone thought, “Okay, we all understand this idea of scaled compounding more,” there was a whole other tier of that.

Even believing that is a very interesting way to view the world and underwrite companies and skills. Looking at each category and deciding what you think the No. 1 asset is, then wanting to own it, is a very unique skill.

4. Research as Edge and the Problem With Obviousness

Okay, so you're saying this is the way someone can make money: Just being more bullish than everyone else or having this sort of people radar that's very well honed. But that's not how you make money, and you're someone I really respect for the way you stick to your craft. You're very disciplined about it. Maybe you could tell me: How do you see Compound making money? Where does your alpha come from?

I think our view is similar to the way some people have beliefs about terminal outcomes. We have a belief that what we do is research-centric, thesis-driven investing. We believe we get very deep in understanding the closest-to-the-metal academic and/or R&D-group research. We then ask: What is possible, or what is inevitable?

Maybe the last thing we'll say is that we think having a prescriptive view on the world and where it will go, and on the first-, second-, and third-order effects of how science and technology cascade through the world, matters a lot. For 10 years, everybody in venture has told me the same thing: You cannot predict the future. You should not try to. That is not the job of an investor. You should let the founder show you the future.

Our view is that we are best suited to partner with founders and understand investments in other areas, like public markets, if we have very reasoned views. That is only going to become more important as there is much more volatility and a much higher replacement rate of companies over shorter periods of time.

How do you think about that? You wrote a post I really liked called “On Humanity and Human Beings,” where you coined this idea of world-first thinking—the idea that technology builds a beautiful model of the future and slots human beings in last. But the most volatile thing in any model is human beings. Isn't that exactly what you just described as your investment philosophy in some sense—a world-first approach?

I think we have a lot of debates around things that might make sense meritocratically or technologically but won't work from a human perspective. One of the things where we often disagree with people in Silicon Valley, or even sometimes internally as a team, is the way people think technology will just progress because that's the way it should, because that's how technology wins: by removing the human side.

It's kind of like the Tyler Cowen view that AI is not going to diffuse nearly as aggressively as all these 30%-plus-GDP people think, because humans just don't want it to happen. It takes a long time, and people are slower. Then the San Francisco people say, “It doesn't matter what humans want. The AIs will do it all.”

I'd argue that we probably have much more of a bend toward integrating both the beauty and the complexity of humans into how we think about technological change. You can think about this on the early-stage side, and then if you look at it on the public-markets side, a lot of people have a very simplistic view: “That thing's dead. Software's dead.” It's not really how these things work when humans are making decisions and deciding how we value ourselves.

Yeah, I like that. Going back to momentum a little bit, or as an offshoot of momentum, how important do you think it is to be contrarian in venture? I think you guys are pretty contrarian, almost by definition, in some of the ways you invest. I know you don't like the word “contrarian,” but you said something—I think it was on the 2024 podcast—that AI is at its maximally obvious point today.

Anytime I see that, it typically means it's probably time for Compound to observe and not deploy too much money, right? If it was maximally obvious in 2024, it's definitely something like 2× that now.

Yeah.

How do you think about why something being obvious, by definition, makes it uninvestable? Investability is a risk-reward thing, right? It’s the price you pay and how big the thing can be. You could conceivably think of something where there isn’t enough money in venture to price in how big this thing is.

I don’t know an example now, but maybe something that extends life or makes you live forever, whatever it might be. So why is the very fact that it’s obvious something that makes you not want to invest?

I’ll say a bunch of disconnected things that maybe, hopefully, will form a picture. One, venture is the lowest-conviction asset class in finance. Because of that, when things become obvious—similar to how, when capital disappears and illiquidity drops something down to zero—things don’t scale linearly on price or risk-adjusted return. They scale exponentially because everyone moves there. There’s a lot of lack of consensus.

Two, what is it that Compound believes we’re good at? We think we’re pretty good at looking at N-of-1-style things and saying, “Okay, this is possible; this could be really, really valuable.”

In times of maximum obviousness, 2 things happen. First, there are a lot of similar things built at the same time, so you have a lot of competition from companies that all look identical. Our job, in theory, could be to pick the best one, but at the seed stage, that’s not where we have edge or alpha. It’s not what we’re best at, so I’m not going to pretend it is.

Second, there is model collapse during these times in tech. As tech has become much more concentrated in a bunch of ways and much more dogmatic around what does and doesn’t matter, you see talent start to flow into a singular place.

I ask people this all the time: if you look in traditional AI, there are actually very few ideas I’ve heard in the past 3 years where I’m like, “Wow, I never could have thought of that.” I never would have imagined a company’s shape would look like this. AI for legal is the idea people have been talking about for 30 years. Every business looks quite similar. So there is high model collapse, and again, that doesn’t allow us to operate in the best way.

I think there are some people who can operate in these massively obvious times. The last thing I would say is that we operate on 10-year time horizons, right? If something is abundantly clear today and a bunch of procurement decisions are going to happen, you can do 1 of 2 things.

One, you can say, “So much money is going to flow into these companies. They’re going to go from 0 to $400 million, and my job is just to get the thing public as soon as possible so I can get out of this because dispersion will happen. Second or final movers will come afterwards in years 5 through 10,” and that’s a real problem.

Or you can say, “I think that this is going to be the enduring business for the next 15 years. First-mover advantage wins, or maybe middle-mover advantage in some of these AI things.” I don’t know how to do that. That’s, again, not what I do and not what I think our firm is best at.

I think so much of venture is just understanding what your singular advantage is and pushing on that advantage over and over again, not trying to make sure that you’re playing the game LPs want you to play when it’s slightly outside your sphere of influence, even though it’ll make LPs happy to know you’re doing that thing.

I think we’re just fine doing the things that we think we’re good at. We tell founders all the time, “Make the decisions that, if your company fails, you can sleep at night having made.” We’re the same: make the decisions so that if we lose all the money, we can say, “We made the decisions we thought we should make.”

So I guess when a sector becomes too obvious, the skills required to win—which is, I guess, picking the best team in an obvious category—aren’t the things you’re best at. That’s why you want to avoid that style of—

It becomes paying the highest price, or it becomes flag-planting the company because you need to. You have 3 companies that are all raising tons of money. Legal AI is the best example. They can all kind of layer on each other, and all of them increasingly go against what we are best at, I think.

Yeah. How do you think about the discipline of staying focused on what you’re good at versus adapting and learning a new skill? This is something I feel like we came up with at Delphi, investing in crypto. We really didn’t invest outside of crypto for the first 6 years of the firm, or 5 years.

At some point, I got nerd-sniped by AI and wanted to start investing outside of crypto. I started learning about deep tech and things like that. As I started investing, I realized, “Holy shit, I just really don’t have an edge here.” These kids are all hungrier than me. They’re willing to move to San Francisco to be around the founders, go to the house parties, be in the polycules—whatever my edge might be. I’m just not.

Part of doing the fund of funds for me was learning from really smart managers like you about how you do it, and also piggybacking on some of the picking. But it has been a lot of reinvention versus sticking to what we were good at, which was crypto, where we learned a lot of bad habits for traditional investing.

I’m curious: how do you think about that? Have you consciously added new competencies that you think are needed?

My default view is that all venture firms are constantly decaying. So you have to fight back against the decay.

If you were to think about Compound 1.0, which we’ll call 2016 to 2022, it was predicated on this idea that we could spend a bunch of time in academic research, understand it, and then figure out what these interesting teams were that were building hard technology that people didn’t really want to underwrite yet at seed, and go do that. The understanding component was a pretty big edge.

Now, any person in the world can take an academic paper, put it in ChatGPT, and say, “Explain this to me like I’m 16.” That edge is gone. It’s not that it hasn’t decayed; it’s largely gone.

There are certain things you can do around understanding how these papers compound upon each other, what the differences are, understanding some networks, and the principles of founders who build applied research organizations. There is tacit knowledge that accumulates.

A lot of the things we talked about internally were, “Okay, now that that edge is clearly deteriorating—we felt it deteriorating—and now that more capital is coming in, what are the next-order things we need to do?” It started to become much more about first-, second-, and third-order effects.

How do these things continue to move over time? How do we make sure that we don’t just form a view that this technology is possible and, if it works, it’ll be bought or it’ll be valuable, but instead ask: if it works and there’s competition and there are early customers, how does this compound? What other adjacent businesses could exist? Is this the best way to play this version of the future we believe in, or not?

I think it has evolved to much more prescriptiveness and treating research as even more of a first-class citizen in the organization than saying, “We believe that generative AI is going to be a thing because we’ve read some of the early generative adversarial network papers in 2016.” That’s a pretty high-level thesis that is detailed relative to most, but not nearly as detailed as we would be today.

Why can’t AI do that, actually—the second- and third-order thinking?

I think because we’ve done a lot to try to get AI to do this, and there are a few things. One, I’ve seen nothing that shows me that AI is able to think out of distribution. So, by definition, through an AI model, you’re getting the maximally obvious or maximally viable ideas.

We’ve even experimented with whether we can embed a bunch of things and hopefully the embedding space will flatten a bunch of ideas, and that should remove some of that maximal viability. I also think there’s something about thinking through many types of companies, types of buyers, and types of people that we just haven’t seen AI able to do.

Maybe someday it will, but we’ve definitely tried and it hasn’t quite happened yet. Again, I think there will probably be some other new thing that we’ll have to think about over the next few years, on both sides of the market.

An obvious one today is how narratives impact multiples in public companies, right? That’s something the models are horrible at understanding, and it’s something we spend a lot of time thinking about. The other thing related to that on the private side is how narratives impact later-stage flows and later-stage investors, and their level of confidence or conviction.

That’s why we spend a lot of time publishing our writing and our research.

What do you think about AI and the way it sort of commodifies certain skills? What do you think that does to founders? I had an interesting chat with one of our managers, Abhishek from Grad Capital. He invests in the smartest Indian kids from the IITs and stuff like this, and he was telling me that he thinks there’s one year left of investing in nerds. These Olympiad and hyper-technical nerd skills are just being entirely commodified by AI, and he’s trying to think about what he’s going to invest in next. I’m curious how you think about that in terms of founder archetypes.

Yeah, I mean, I think it’s never been the case that the smartest founders are the best ones. I think you’ve had to be pretty smart, but I’m sure there are people smarter than some of these best founders.

I would imagine that some of it starts to come down to some of the stuff that I’ve written about recently around how people feel about working at an organization, and thus the leader of that organization. What do they morally stand for? What personality traits do they believe in? What is the future that they believe in? How do they believe they want to impact the world? I think that could come back into vogue in a much larger way.

I do think there is a minimum viable intelligence that is being brought up because of the models, but I don’t know, man. I’m so pro-human that I think humans are so unmodelable, so I think we have a while to run before we commoditize humanity through AI models.

It’s just—I don’t know. One example is in the neocloud space. There are these Together AI teams that are super smart on kernel optimization and all this really complex stuff. Then there were teams that were more aggressive and full-stack. A bunch of super-aggressive guys moved to Texas and secured, allegedly, 5 to 10 gigawatts.

I’m not an investor, but that’s much more valuable now, right? The fact that you actually have power and land is much more valuable than your algorithms, because that stuff is very much optimizable by AI.

And so, I don’t know, I have this thought that it’s always been important for founders to be charismatic and kind of savage—in the sense of being able to really get things done, get deals done, and be hyper-intense—but maybe it overweights even more in that direction now. It seems like a smart generalist with a lot of that charisma can go a lot further than they could before in a lot more different fields.

Mhm. Yeah, I think—again, we have a belief that is somewhat in conflict with some of the best investors. We don’t believe there are just these things called great founders. We think there are great founders for certain types of businesses, and each type of business has different things that are necessitated from the founders.

So, yeah, if you’re working in a supply-constrained, long-duration, highly business-development-oriented space, you for sure need to be more of a savage. If you’re working on something that is—

Yeah.

Super heads-down, I think attention and commitment are going to be incredibly rare on a go-forward basis. Maybe just the ability not to get distracted doing random things on the side is the main thing people should be looking at for founders. That might be the way this goes.

I don’t know, but I do think that every company type has some weird nuance to what the founder should be.

I like that. And that kind of goes into your portfolio construction, right? To some extent, you’ve told me once that you think you need 30 seed bets to really catch 1 or 2 good ones, right? There’s a lot more—or at least there’s more of a meme now around concentration, doing these big bets. I think Hummingbird and Fundomo and some of these funds have done really well with this strategy, and I think there are some others, too. I’m curious how you think about that.

I told the LPs this: our portfolio should feel riskier and riskier as time goes on. If it doesn’t, it probably means we’re not doing our job well.

That’s twofold. One, it should be riskier to them because that means the world is changing quite quickly, and LPs are not exactly leading indicators for understanding where the world is going all the time. So it should feel kind of crazy. If it doesn’t, then what are we doing here?

Two, I do think that because of that, and that’s where we think asymmetry will come from, we need a certain number of shots on goal to do that. If we look at our data and the data for some firms that are like us, we would say that 1 out of every 12 times we make a pretty good decision, and 1 out of every 20 times we make an incredible decision.

I just want to make sure, as a firm, that we have that framework in place, because venture is so much about the inputs without having any understanding of the outputs for so long. If you really believe in the inputs that you’ve set up, you should feel really comfortable just continuing to do the thing.

What gets a lot of people tied up is that they’re in year 5 or year 6 of their career, and they’re wondering if they’re good or not. They didn’t have high conviction in the inputs; they were just doing the things they thought they were supposed to be doing. Maybe they changed a bunch of things about how they behaved, so there’s no consistency. They don’t quite have the output data yet, and that creates so much volatility.

As you know better than anyone, you can’t come to a high-risk game that’s played at very high stakes with tons of fear and uncertainty internally. So I think a way to remove that is, one, to reduce noise, and two, to have an input system that you deeply believe in. You’re basically saying, “This is pseudo-religion,” with some bounds of understanding. Sometimes you have sins around religion, but generally you keep it within a band.

That’s kind of where we’ve settled, and I think we’ve been a little more concentrated over the years. We’ve had fewer companies because we only get excited so often, but generally we know that we’re certainly not going to hit a multibillion-dollar company out of 10. If it’s the 11th investment and you’re dogmatic that you’re going to be hyper-concentrated and do 10, you’re going to be unemployed in 5 years instead.

How do you think about the heuristics with that? Whenever I hear you talk about investing, I’m always struck by how analytical you are, and you have a lot of data backing some of these decisions, whether it’s the type of founder, the profile of companies, or something like that. Even how many investments you want to have per fund—there are a lot of data-backed heuristics.

I’ve historically really struggled with heuristics. I’m kind of undisciplined, and I tend to—well, it’s a weird sort of lack of discipline, but I like to break the rules. I think heuristics are really useful to keep you out of trouble, but sometimes greatness comes from when something doesn’t fit into any of the heuristics.

Maybe it’s a very broad question, but heuristics shape your view, right? If you’re even in martial arts, with the thing that I do, and you approach a position knowing its name, you go in with a certain view. It’s the Wittgenstein language-games thing: you just see what the language tells you.

If you invert the language or try not to apply language, which is almost impossible, you can see different things. I don’t know if there’s anything there that you want to pick on, but I’m curious how you think about it.

I think it depends what your goal is, right? Our goal—what our stated goal is to our investors—is to be consistently great. How we define consistently great is that we want to have consistent top-decile funds and the ability to have top-5% or 1% funds. So far, I think we’re doing a good job at that.

There are some people who are like, “There are moments in time, and my moment in time is, I’m going to go for it.” You talk to some LPs, and they’re like, “Look, if you 10x Fund 1 and then 2x Fund 2 and 2x Fund 3, we’re super pumped. Over those 3 equal commitments, that’s a great return.”

And I’m kind of like, “That’s not the strategy I want to run.” I want to be able to know what I can consistently do, with the opportunity to be very, very, very great. Again, I think even that is incredibly hard.

That’s how we’ve oriented it, and that’s why we have some of these heuristics. It’s kind of like the idea of choosing when to sell, right? Do you sell when you believe—

Another one.

You're at the top, or do you sell on the way up? In public markets, you talk to a bunch of people who are in super-high-growth names, and their view is, “I sell once it falls 25% and I get invalidated.” Again, we've looked at a bunch of that stuff, and it's pretty interesting: if you look at these hyper-reflexive names, if you actually sell at the all-time high of next-12-month revenue multiples versus once it retraces 25%, how do you do? Actually, you do pretty well if you sell at the all-time high of next-12-month sales multiples, not on the 25% drawdown. But you might miss the leg 3 years later. I don't know.

We like consistently making money, and maybe that does mean that we won't have that crazy volatility as much. But I also just think venture is so random and cyclical, with so much randomness built into even these moments, that as much as you can control, I think it actually gives you a lot more ability to operate for a multi-decade time horizon. That's kind of how everything we do is also oriented around the mental state of how we invest.

That's all great. That makes a lot of sense, actually. I don't optimize enough around that. I definitely want to get to public markets because I think it's going to be a super interesting chat.

I'm also curious what you're looking at, maybe before we move on to there, because there are some areas that you think are just hot, obvious areas—AI is one of them. You've always been really good at finding these obscure corners of the internet where interesting things are happening, both in crypto and outside of it. What are the most interesting areas right now that you're looking at? I know you're doing a lot in bio, which a bunch of us smart investors are doing.

5. Bio, Materials, and the Commoditization of Intelligence

I'd say bio is where we've invested the most heavily over the past 3 years. If you look at our second fund, which deployed from 2021 through the end of this year, it's a lot of bio, a little bit of AI, and materials science. I think probably the thing I'm most excited about is that materials science side, and just being able to build more full-stack businesses built around synthesizing novel materials and shipping an end product. Orbital Materials is the case study for that for us. That's one of our biggest investments in our second fund.

It's a foundation model built around chemistry, state-of-the-art in chemical synthesis, and one of the things that they did is that they have a material, but they shipped it as a full product, which is for cooling next-generation GPUs. It's a modular data center product. We think that is a very interesting shape of business that there will be many others in. We also are looking at some of the more, I'd say, consensus areas, but trying to understand if we assume this next few years of build is very consensus in how they're done, like energy. That's another adjacent area that we've been spending more time in.

But I would bet that a lot of our investing still will sit within bio. I think there's going to be a really interesting time to invest in AI in 2028 to 2030, once we see the dust-settling dynamics of maximally viable, incredibly cheap intelligence and the commoditization of everything. But the furthest-frontier intelligence should create a lot of really interesting companies, and I would argue we probably will be well suited to invest there.

What do you think of that commoditization of AI? You said it—I actually found a tweet of yours in 2023: “We routinely underestimate the commoditization curve of AI.” I feel like I was on this train too, and it's kind of surprised me how uncommoditized it is, in the sense of just how good and how frontier intelligence has managed to stay ahead—and for how long. Do you think it will eventually commoditize?

I think that there is a subset of tasks for which we have enough intelligence in current models that will become effectively free, if not able to run locally, in the next 24 to 36 months. You might need to post-train them. There might be some innovation, or maybe operational execution has to happen to make those things truly production-ready, but I think that is undoubtedly going to be true. I think there are certain areas that will be incredibly economically valuable that will require frontier intelligence for the next, I don't know, 5 to 10 years, and there will be a small number of people that have the compute resources to do that.

The complexity I have is that in these types of market structures, you only need 1 player who is not economically motivated by the core monetization of the core thing to really drive the price down. Everyone hates Google right now, but I think that's a pretty stupid thing to do. I think Google sitting there with the fact that people value their debt as less risky than US Treasuries is one of the most interesting things in the world to me, and that creates a lot of uncertainty.

So whether it's a commodity or not, I don't know. But I think the dynamics of, call it, 90% of the intelligence that exists in the world will undoubtedly hit some sort of multi-tiered, oligopolistic commodity. Again, you might have some settling price that is low enough, but it's definitely not going to be a premium asset.

Google seems to be, to some extent, betting on commoditization, right? By leaning into cloud and—I mean, maybe giving up on the model layers is too hyperbolic, just Twitter stuff, but yeah. Maybe, actually, let's—I'll let you respond to that, but then I want to move to public markets.

Yeah, I think Google is not giving up on the model layer. I think they ship some of the best-performing, fast, and cheap models, and I think they probably are looking at this saying they've definitely messed up. They definitely had a bunch of problems on the pre-training and post-training side, and talent was an issue. I think there's a lot of structural advantages, and yeah, I'm quite bullish on Google. Not financial advice.

6. Public Markets, Software Margins, and the Short Side

All right, let's move to public markets on this one because I think it's a good segue. You have this—and I'm sorry to keep quoting you back to you, but I just love how easy it is to do this with AI as well nowadays. It makes you look really smart. You said the thing that destroys all venture funds and maybe people in venture is scope creep, right? If this is the case, why do a public markets fund, or a liquid fund, alongside your venture fund?

I think this is the fairest criticism that anyone can have of Compound, outside of the fact that we might just be a really overpaid research organization that never monetizes. So I think those 2 things—

Pretty well.

Yeah, but you never know. We have a lot of things that we see in the early stages of technology that we, for a long time, had no way to action at all. And I think if you care about building highly durable businesses over time in the private markets, you actually have to understand the entire flow and stage of all technology companies, especially because in tech the most innovative companies are the largest ones. Interestingly enough, in bio, the largest ones are not necessarily the most innovative, though Eli Lilly, I think, is starting to change that.

I think as we spent more time and as we managed an internal portfolio, we just started to see that be meaningfully true, and we started to say that the things that we need to understand are directly monetizable, and that makes us a better investor in both ways. Understanding what Tempus is doing on the diagnostic side allows us to understand what other earlier-stage companies are doing as well—what they should be doing, what they should not be doing, how they can monetize, the customers they can go after, et cetera. There are a million examples across the board of that in our portfolio.

And so we have no desire to scale to be a full-stack private fund. We think the only performance-oriented asset classes in technology investing are early-stage venture and public markets at this point. That's where we like to compete. If we believe we have a meaningful edge that we think makes us better in both ways, we were really excited to try and continue to push that hypothesis.

And how do you—what's the thesis? What's the thesis for the public markets fund, and how do you run it? Is it a long-short? And maybe how many positions?

The thesis is thesis-driven, research-centric investing. It creates an understanding of businesses that either are going to accelerate or decelerate earnings in a way that is poorly understood by the rest of the market. That allows us to be long and short. It allows us to have a lot of time preference.

The vehicle's not evergreen; it's a closed-end vehicle, so all the money is locked up for many years. We run relatively concentrated on the long side, and on the short side, far more smaller positions and a little more breadth than depth.

On the long side, I'm curious: what are you most excited about right now, if you can talk about it?

A lot of the same themes that we've talked about around bio, around maybe next-order energy things.

And then also, candidly, some simplistic views around where we think technology is overly hated because of AI.

Okay.

Yeah, it looks like a mix of things that one would believe are very obvious for us to own, some things that aren’t, and then there’s also some really random, super-small-cap stuff because our fund is able to traffic in a lot of different types of assets that are international and maybe sub-$300 market cap, even.

Are you able to buy things like the Magnificent Seven, or Google or Nvidia, or things like that?

No, yeah, we can basically buy anything we want in the fund. But our goal is to have similar IRR to our venture funds.

Okay.

However one can do that is up to them, but most of the stuff we own is not the standard Magnificent Seven right now.

Okay. Things could change. Okay, interesting. You said software is a really hated area. What do you look for to make a software business? People thought that all SaaS was dead, right? This was the narrative on Twitter at the beginning of this year, and now software has had a great year. From there, what do you look for in the software businesses that win? Is it as simple as API and agent usage, basically?

I think there’s some stuff where the question is: can they actually transition into this usage-based model that allows them to expand revenue and probably have margin compression in the short term but margin expansion in the long term, if they’re thoughtful about it? That’s interesting to us, and I actually think that creates dislocations because the market can look at margin compression in the short term and say, “I hate this,” on a quarter-by-quarter basis, while we can look at it and say, “Actually, this shows that inference is picking up within the organization.” I think you have to understand management.

Within the organization, like spending—as in their own spending on inference?

No, within the customers that pay the organization, yeah.

And that reduces margins why?

Because usually you’re passing through—if you’re running a traditional software business in some of these areas, you’re looking at a traditional business that might run at 80% to 90% gross margin. If you’re passing through AI inference that you have to pay for—you have to pay the cloud, or you have to pay an inference API—you might run at 50% to 60%. If you look at blended margins as those things start to come together, you’re going to see margin compression.

But then you look at other things, like the Pinterest CEO. I’m sure you saw this: the Pinterest CEO, during earnings last quarter, talked about how they’ve been using a bunch of first-party models, and they now have fine-tunes of open-source models. They’ve been very dogmatic about this, and people have generally hated it. Pinterest has not performed great, and he says they ran a cost comparison, and their costs versus using the frontier APIs are 8% of the costs. That’s pretty incredible, especially because now they’re starting to see some of the ROI on the AI usage.

You could imagine a world, to the prior point, where if AI continues to make its way through, continues to fall in cost, but has performance increases, you could see a reacceleration of margin expansion. I think, related to that, you have to understand management and their ability to actually understand what’s going on in the AI world. If you’re totally blind to it, or if you have some of these CEOs who are like, “This whole thing is a sham. It doesn’t actually work. It doesn’t actually matter,” we probably wouldn’t want to invest with that management team.

There are some businesses that might violate that. There are also certain things that we believe will see meaningful mid-term acceleration of revenue, and we’re basically looking for short-term dislocations to layer into those. Again, our time horizon on the public side is 5 years, so we have time.

That’s really cool. In biotech, what are the most interesting things happening in public markets? It’s been a great year for biotech, and people are starting to wake up to it. I’m curious what you like there or what you think is most interesting.

I think there’s a subset of companies sitting on data assets that are meaningfully undervalued and will be able to monetize those data assets far sooner and at a far larger scale than any analyst covering those stocks appreciates. I also think there’s a long tail of companies that nobody covers because they’re small and have been left for dead, but that could also become quite valuable.

A lot of the stuff we own looks like things playing into a reacceleration of experimentation and biomanufacturing, as well as certain platforms and certain data-asset-type businesses. We don’t do single-asset trading on clinical-trial releases. That’s not something where we have any edge.

And on the short side, you said the next decade is about market-cap destruction, which was—

[Snorts]

Ominous. What do you mean by it, and how are you playing the short side?

I think that as markets become more narrative-driven, the companies that accumulate a lot of short-term flows are those that can sell very great narratives, and those companies are often deep-tech businesses. Historically, they were deep-tech businesses that went public via SPACs, but now they’re all sorts of different types of deep-tech businesses.

Some of these companies we meet when they’re fundraising, and we’ve met them years prior. Now they’re public, and we’re like, “We had a view that they weren’t good investments then, and we have a view that they’re not good investments now.” Others are just so grossly overvalued that we believe there’s a near-term catalyst that will show these businesses are not nearly as strong or as high-growth as they’re projecting. There are others that, existentially, we don’t think will exist within a time horizon of 2 to 4 years.

For shorting, our view is that you always need to have a clear understanding of what the catalyst is. You always want to be able to express the short in multiple ways. You never want to layer on factors that are too highly correlated if you’re going to short a bunch of different things around a singular theme.

But I think there’s an infinite opportunity set of companies, especially now that flows into these types of things jump so quickly. The first sign of invalidation, again, doesn’t send the thing down 5% or 10%; it can send it down 40% to 60% in 4 to 6 weeks. Those are the things that we often traffic in more.

How do you think about conviction in public markets? In private markets, I heard you say something like, if you believe in it and it hasn’t worked yet, you should keep trying for 5 to 10 years, which I think does work in private markets, assuming you have the time and your LPs give you the time to do that. In public markets, how do you think about invalidation of the thesis, and when will you buy back or cover a short? When will you sell a long?

We have really tight risk controls on the short side. Typically, if we’re pretty wrong on a short, it just gets closed out.

Okay.

Yeah, we don’t try to be heroes shorting things over and over again. We also have a small maximum position size on any individual short.

On the long side, I do think a lot of what we do when we write memos on public long positions is understand what the quantitative and qualitative things are that would invalidate the thesis. If, for whatever reason, the price is invalidated but none of the thesis is, what would we have to see to cut, or what would we have to see to double the position? We have those kinds of bands set up going into any position.

Again, we don’t think about sizing as, “Hey, we’re going to equally size 12 positions at the same amount.” We do have varying degrees based on where we think there’s reflexivity and where there is downside risk. There are certain things that we feel pretty certain we want to own in the near term, but we think there might be a short-term reason why they could go down 10% or 20%. In that case, you could probably try to play this with options, but you might just say, “Hey, this is my fully sized position, and maybe I’ll layer into it.”

I do think that you cannot underestimate the fact that some companies just will never catch a bid in markets, even if they continue to execute. You have to understand the 4 quadrants: businesses that look cheap but actually are expensive, businesses that look expensive but actually are cheap, and then the other 2—where you think the thing is and where it actually is.

SMAC is going to publish something on this pretty soon that's quite interesting, something we've thought a lot about.

Then, maybe on the big names, I remember when I met with you, you were pretty excited about Meta, and I'm curious how you think they've done and where you see them here, because I've also been a Meta bull. I've held it for a long time. I've been surprised by how bad their execution has been on the AI side, particularly on the model side. I would have expected a lot more.

7. Meta, Google, and the Mag 7

Yeah, I'm also surprised by things like WhatsApp just not being monetized in any way. A send-money feature on WhatsApp seems like it would be so great. It feels like leaning into stablecoins now would finally be the time for them to do that. It just seems like there's a lot of surface area being underutilized, but I'm curious how you see Meta.

Not financial advice, not all the things. I think Meta is the company that is probably best positioned to monetize AI, regardless of whether they build on the frontier or not, and I think we continually see that. They have also been pretty ahead, and maybe slightly less aggressive now, on the compute side. If they want to spin up that side of the business, that's a pretty interesting possible growth trajectory for them.

I never really had strong views that they would be pushing the frontier on AI, if I'm being honest. I just think they can monetize it with so much ease across so much surface area that every time the market gets annoyed that Zuckerberg has this “I'll just burn the whole thing down” attitude, the company struggles. I personally think that people continually underrate just how deeply embedded their products are.

There's an open question about whether this is the last turn for them, but I don't know. I think even how Reels has gone over the past 18 months has been pretty good. I just don't feel the existential dread that if you're not pushing the frontier models, you're a dead-on-arrival company as a Magnificent 7 business.

And how do you feel about the Magnificent 7 generally? I'm curious—Google, it sounds like you're also pretty bullish on. Is that just GCP and TPUs, and owning the stack that has surface areas to monetize, or—

There's so much structural advantage. GCP has just been incredible. Even the Gemini Flash series, I think, is a pretty incredible set of models. I know everyone's waiting for Gemini Pro 4 and all these other things, and then they shelved 3.5 and all the talent's lost and all that stuff, but they just have a lot of structural advantage that I think will continue to compound for a really, really long time.

Admittedly, I don't have strong views on some of these other companies. Those are the 2 that I actually have strong views on. Apple, I think, is also incredibly well positioned, and people have noticed that over the past 16 months.

I think it's very possible that the same thing we saw from going long tech beta for the past decade might continue in some form. I think the replacement and destruction that I talk about is actually a tier below in scale. I like some of the more hyper-hyped things, like Nvidia or the memory stocks, but I'm just less high-conviction either way, so I stay away from them.

Same with the neoclouds?

There are some that I really like, and there are some that I hate, actually. I think the next generation of neoclouds—this set of companies—is going to show that there are actual moats of competence and execution that will really matter over the next 5 years.

I think you'll see dispersion among them, both by their ability to execute and, maybe to your point, their ability to build the right relationships with the right people who want to finance them, back them, work with them, and then maybe how much each of them opts to verticalize over time.

Yeah, that's been my biggest sector bet. I think they're still very misunderstood. I'll be curious to ask you, maybe off-air, which ones you hate and which ones you like, to compare notes.

On markets generally, Michael Burry is pretty short and bearish, as he tends to be. Ray Dalio and Jeremy Grantham are doing the same thing, and they're pretty bearish. There are signs of a bubble brewing, or whatever.

My view has been that there will be a bubble, but it'll be much higher than here, and I don't see that much cause for concern with forward earnings and things like this. I'm curious where you sit with that.

I think I'm probably in the rolling-bubbles theory, which is that there are going to be perpetual rolling bubbles, and there's just going to be a lot of volatility. I think there's no strong semblance of what 2028 earnings will actually look like for most of these businesses, or how earnings per share will look on a go-forward basis. Do margins really change? I think there's so much uncertainty that there will just be continual whiplash.

You see it with the neoclouds, right? You can say that that was started by Leopold Aschenbrenner's Situational Awareness, Value Aligned, and some of these other funds that were super-levered on the same factor over and over again. Everyone's like, “Yeah, they got blown up and liquidated,” and then everyone went long again.

But I also think it's a collapse of confidence, and the collapse of confidence is what we talked about before: When do you sell? I sell when it draws down 20%. If everyone has a view that you sell when it draws down 20%, it's not going to draw down 20%. It's going to draw down a lot more.

CoreWeave went to 79, and then it was back over 100. Palo Alto drew down 40-plus percent on a random Mythos announcement, and now it's back toward all-time highs. These things are happening so fast that I think we'll continually see rolling ups and downs. I'm not a good enough macroeconomist to know the other side.

Yeah, CoreWeave went to 60, I think, at some point. At least I saw it at 60 at some point.

8. Crypto, Robotics, and Drone Warfare

In terms of indexes, you still think—because I have this view that the cross-correlation among the constituents of the indexes is at all-time lows, right? It's a stock-pickers' market again, or long/short hedge funds will be able to do really well. How do you think about that?

If I'm being selfish, I have a liquid book that's my part-time job, and I've been stock-picking with it. I did well in the first year, and now I'm underperforming the index, as one would expect. But I'm also very hesitant to index because I have a specific view on AI. I feel like it's more consensus now, but not that consensus. I'm curious how you think about it.

My selfish belief is that active management has been hated for a decade now and should come back. But sitting and not having to think about literally anything and owning the Nasdaq is pretty nice.

In 2024, I made this basket I called the AI Obvious basket. It was basically 30 names, pretty equally weighted. It was built around being incredibly obvious about AI, and it had some ASML, some Micron, some of the hyperscalers, Alibaba, and a bunch of these things.

That thing has just murdered everything. It has so drastically outperformed, and I literally haven't changed it in 2 years. There are times when I'm like, can you just structurally understand the theme? Should you be able to build these longer baskets around a theme that you think will be high beta to the market?

It's tough. It's really tough. We look at it a lot in terms of, within a given basket, how are we outperforming with the specific names? I do think that Citrini is kind of the middle of this, right? He builds these massive indexes of themes, and maybe that's the future for a lot of investors. It's the future of active management.

It does feel like that, actually. I met a guy the other day who's building a company. I don't know if I can share his name right now, but he's a very good investor with a very good track record in liquids, and that is his view: AI collapses the cost of asset management generally.

You need way fewer analysts and way less of that scale. The sort of Fidelity model—having a bunch of portfolio managers and then a bunch of analysts under them—you don't need that anymore.

You can extend your taste pretty costlessly if you’re an experienced investor. The end game for that is just a bunch of indices that different people construct that you can invest in. That’s kind of what he’s building: an index of what he views as the 20 best companies in the world or whatever, with a very low management fee, which I thought was super interesting.

I mean, ARK has been the greatest accumulator of capital, basically, in this thesis. They’ve also lost more money than anyone in human history, but I think that’s basically what ARK was, right?

Yeah.

I don’t know, so hopefully AI is better than them. I don’t know. I’ve got nothing.

And speaking of that, what do you think of crypto? Speaking of destroying the most money in human history, how do you feel about crypto? I remember when you first talked about your liquid fund, I think it was going to be most of what you did. I imagine that’s not the case now. Maybe it is.

We still do long-short crypto in the fund. I think we are—how would I describe it? We are still bullish on a small subset of projects that will be able to ascribe fees to the tokens and that we think will compound, with similar dynamics to what we’ve seen in technology. You will have far more power-law-looking things in crypto over time.

From an application-layer perspective, or however you call it—non-money, basically—I think that, on the venture side, we still don’t have a flow of talent that’s aggressively coming in. My hope is that will change. I do think that the things that underpin crypto still matter, and the ideas of accumulating and organizing long-term capital for strange and/or non-obviously large ideas are actually quite interesting.

That’s kind of where we are. I’m still a Bitcoin bull and still think there’s a bunch of interesting stuff that we will own in the fund over time.

So you’re bullish on Hyperliquid and Lighter, and things like this that are generating fees—maybe Ether.fi?

I think there’s some stuff that generates fees and some stuff that we think will generate fees that’s much smaller. Our purview can be really venture-looking even in this; it just has to be a liquid asset. It’s largely, call it, DeFi, DePIN, some DeSci, and maybe a few other protocols that are built on top of these L1s.

It’s just so tough with generating fees back to the token, because raising a token around anything is much harder than raising an equity round these days, I feel like. It’s reversed, right? It used to be that if you slapped a token on it, you got a 3× premium. Now it’s half or less.

Launching a token is such a ball ache, with token holders yelling at you and all the shenanigans that come with that. Why do you think that category persists? There are things like MetaLeX that we incubated, and others that are trying to give tokens more equity-like characteristics, and even redeemability to equity, which I think is super interesting. What do you think makes that category persist?

I think that, on a long-term time horizon, public equities and crypto tokens look like the exact same asset.

Yeah.

In the same way that I think private companies should go public, allow their stock to be traded, report to shareholders, and have responsibility to more than a small subset of price makers and buyers, I think that’s good for the world and is actually better for the companies. It allows you to access capital in more interesting ways.

There are also a bunch of founders who have zero desire to take their company public, and I get it.

I’m a fan of HairDAO. I actually found them earlier, when I was briefly paranoid that I was losing my hair. I went down the rabbit hole and found them. They’re an insanely cracked team, despite how silly it looks from the outside.

It’s one of the things I love about crypto. No one is better than crypto people at making fun of crypto people. AI is the opposite: they take themselves more seriously than anyone else.

It’s true. I think that, again, we are bringing up the bottom, or the floor, of intelligence and sophistication. That should actually be significantly more valuable for crypto than any other area, because it allows people who maybe don’t have traditional communities, education, or whatever it is to build durable companies with frontier intelligence alongside them, versus a Discord of degens yelling at them.

I think that’s also quite interesting over the long term. Right now, we still have a lot of regulatory weirdness and just low status in the industry, and that hurts.

On robotics, you said robotics was your biggest miss. Tell me about that.

I think it’s the biggest miss we’ve invested in. I don’t know if it’s the biggest thing I was wrong about; that was definitely VR. I love VR, and it just never came through.

Same.

I still play VR golf with one friend, and I love it. It’s amazing, but it didn’t work.

I play Population: One. I’m a fan.

Yeah, that was great. On the robotics side, I think we made some investments in 2016, 2017, and 2018 expecting the AI—the deep-learning models—to be good at on-policy learning and more generalizable. That was, in my mind, an incorrect thesis.

I think we were just a little early. We were early both in terms of how customers think about integrating robots around humans, most importantly, and in terms of actual performance scaling. We made some investments that were mechanical engineering-oriented and were good. Hyphen is an example of that; they focus on the food space.

There was a company enabling agricultural robots on farms called SparkAI that John Deere acquired. That was a decent outcome. But I think we just mistimed it.

The frustrating thing about how we invest is that, at times, things go from totally dead to super hot, and that happened in robotics. A lot of the relationships we built over many years ended up being with some of the most important people in robotics, and that was awesome to see. But when they left, they started companies and raised $500 seed rounds, so we couldn’t invest. It’s a bummer.

Now I think we’ve seen enough, and we are continuing to invest in the space because we think this time is really different. Everything we’re seeing on the performance side would suggest that, both for vertically specific robots and for more generalizable models. We were probably—I mean, by definition—8 to 10 years too early.

And on the general models, you mentioned Andrew Kang of Roubini Strategy, who I know you have strong thoughts on. He said that humanoids are a 0-to-$10 trillion opportunity, which I think I’ve heard you call insane. Why is he wrong?

Yeah. I look, I don’t know if, on a 50-year time horizon, humanoids will make sense. Sure, there will be a humanoid company that makes a lot of money. I don’t know.

I think humanoids make sense in areas where the thing you are replacing and that is being paid for is not the cost of labor; it’s the cost of human life. That would suggest that the places where you would deploy humanoids should be either military or industrial.

In the military, you’re paying for the cost of a life lost. There’s just a lot of money that you could put on that price. Industrially, it’s a life destroyed in some form due to physical labor. The problem with industrial is that it is the most heavily penetrated area for robotics in the world.

I think there will be a very large robotics company that gets built. We have an investment in a robotics frontier lab that we’re super excited about. It has a very weird approach, and we think they’re going to be great.

I think the way people talk about these things should be slightly more measured. In particular, I think he’s running a playbook that we’ve seen many times, time and time again, especially if your incentive is to draw people to buy your basket of companies.

What areas are you most bullish on in robotics? You mentioned specialized robots and some general platforms. What are the general platforms that you’re excited about?

We backed an out-of-stealth lab that is building a new learning approach for training generalized models.

They’re fully vertically integrated with hardware as well. Our main thesis there was: Can you scale your internal data faster than anyone, and can you train models super-sample-efficiently in a very low-cost way? They believe they can, and we believe they can. I think there’s a ton of really interesting vertical-specific robot companies to be built.

We’re investors in a company called Alquist that focuses on a few different forward-deployed enterprise use cases. One is retail, but there are also use cases in the data center and semiconductor spaces. They’ve basically built a platform that has many different skill sets—not fully generalizable, not fully able to do anything—but the main difference is that the thing can operate around humans in unstructured environments it has never seen before and do a bunch of different tasks.

Built around that, there’s also a bunch of software that both helps the company in the immediate term and, in the longer term, provides analysis for certain things. I think there are a bunch of areas like that. I think hospitals will have a bunch of vertical-specific robots. I don’t think that it’s going to be humans.

I think the cost curves of these things are coming down meaningfully, which means you can build awesome robots that can move, manipulate, and solve a small subset of tasks, and you can probably build them for under $5,000. That cost will only go down as compute goes down. They will enable humans to do different jobs, and they will give leverage to humans. That’s the biggest thing in the short term.

So, again, I think that we will probably make multiple robotics companies through our investments over the next 5 years. I think we’ve invested in 1 military-focused industrial humanoid business. Other than that, we just haven’t made as many yet.

Dovetailing on the military side, what are your thoughts on drones and like the drone defense space generally?

Uh, I think there’s some actually really amazing public companies that could be really, really large that are not today. And I think that my simplistic understanding of the space is that there’s a lot of budget to be allocated there. We looked at drone defense a lot.

Yeah, we looked at drone defense a lot. We don’t—I think it’s probably helpful if you can detect and catch and disengage them—but we don’t have a strong view there. We don’t have any investments there.

Yeah, drones seem like one of the most obvious things. I don’t think people are talking enough about how crazy what’s happening in Ukraine is, for instance. The ways they’re able to strike 3,000 km deep into Russia and hit refineries, causing tens of billions of dollars of damage at a cost a few orders of magnitude lower, are remarkable.

It does seem like the world hasn’t really grokked what this means for warfare. On the front lines in Ukraine, there are barely any humans there, right? It’s just UGVs, drones, and a bunch of pilots and engineers. It just seems like it’s not fully priced in. You have $3 trillion or whatever in annual military budgets, and by my estimates, 1% is drones right now. It seems like that should rise to a lot more than that, and that’s a tailwind for the entire industry.

I actually think this is a super stupid thing to say, but I think it’s because there’s not a lot of prior art around futures with military drones. I think there’s tons of prior art around humanoid robots fighting on battlefields, and there’s a bunch of drone-delivery art. That is something that has existed in the lexicon of society for a long time.

Sci-fi people don’t write about one-way drone strikes very often. I really do think so much of these things are art imitating life, which imitates art. I actually think that’s just not in the lexicon.

That’s a really interesting take. I’ve been thinking that’s sort of the area I’m probably most excited about in terms of a sector that I’ve found to have a supercycle-type dynamic. It could be a rising tide that’s lifting all boats.

It’s very hard to figure out what the winning strategy is, because it feels like these companies have to solve 3 problems at once: autonomy, the fleet-software stuff, and mass manufacturing. There are people focusing on each one of them, and people trying to go for all 3 of them. There are a bunch of risks of local maxima and stuff like that, which is hard to predict.

There’s only so many things people can be excited about at once.

It’s true. There are a lot of interesting companies out of Ukraine, too. I don’t know if you’ve spoken to any of the companies like the Iron Cluster.

I’ve heard of this company. I’ve not spoken to them.

Okay.

Yeah, I haven’t. There’s this guy who wrote this 300-page essay on basically the front lines of using drones in Ukraine for the past whatever number of years. He mentions a bunch of different companies, about half of which I’ve heard of. I’ll send it to you. It’s pretty interesting.

9. Therapy, Psychedelics, and Writing

Awesome. Last few questions, on a more personal note. We have to touch on introspection, right? It was all the rage in tech a while ago.

Yeah.

You strike me as someone who’s done a good amount of work on yourself. You’ve talked openly about therapy and psychedelic journeys, and in general, a certain sort of seeking, which I definitely identify with. I’ve done the therapy and the psychedelic journeys, too, over the last few years. I’m curious: How has it made you a better, happier person, first of all?

I think there are a lot of haters who say, “Oh, you’re just going to be neurotic and obsessed with yourself, and it doesn’t lead anywhere.” Has it made you a better investor?

It’s definitely made me a better person. I think I’m a person who definitely feels the feelings. Sometimes you need to figure out how to explain the feelings to those around you, which is equally as important, and understand which feelings to continue to feel and which not to.

I think so much of these things is about how you continue to understand the edges of your life experiences and touch the things that might push up against those edges. Sometimes it’s psychedelics, right? You do a mushroom trip or something, and you touch the edge of that experience. It’s a very interesting moment.

Other times, it’s just seeing how people move through life. I was close to someone who was incredibly environmentally aware, to a point that I had never seen before. Just seeing someone move through the world that way is very interesting. You might not fully change all of your behaviors, but it helps you understand that there’s a plane of existence that people operate on that is different from yours, and just knowing it exists is very helpful.

I think it keeps you as a well-rounded person. To the point about investing, I think it allows you to understand the humanity of how humans will actually progress and what complexities of humans can manifest themselves, versus the maximally autist, highly structured, obviously intellectual way in which humans will move through the world.

I think it’s made me a better investor. I also think it’s made me a better firm builder. When I started helping build Compound, I was 25 years old, and I think I was like many 25-year-old guys: I was so intellectually oriented and not emotionally oriented in the same way.

Building a team and building a firm where hopefully everyone else on the team continues to own the firm and helps build the firm with you requires you to be incredibly emotionally aware. It makes you a better board member, makes you a better partner to founders, and I think it’s important. I think it’s a lifelong craft. I don’t think it’s a snap-of-the-fingers, do-the-thing-and-you’re-done process.

Yeah.

And it hasn’t taken your hunger in some way? Has it taken your hunger? I think this is what Marc Andreessen was hinting at, right? That there’s a risk that this thing one-shots you and takes your hunger, and you go live in a yurt or whatever.

Yeah.

That’s what he was hinting at, right?

No, certainly not. I also think, if anything, you need to operate from a place of calmness—but calm obsessiveness, or calmness with a long-term obsessiveness.

And I think otherwise, you burn yourself out, and that's where really bad investing mistakes happen.

And I'm curious: you keep circling this moral thread in some of your writings, that we should be more morally motivated than we are right now. You ask your friends, "Why aren't we doing more?" You had this recent post as well on how tech underinvests in legitimacy. And I found the differences between the philanthropy of the East Coast and the West Coast pretty interesting.

You also did your Revery grants, which I guess is maybe related to this. What do you think we should be doing more of? And how do you think about that in your life?

I think that we and a bunch of friends and other people are in immense places of privilege, where we get to spend a lot of time thinking about the world. We get to spend a lot of time talking to people who have immense influence on the world, and we have a front-row seat to benefit from some of that change.

I think that over the past few years, many people in my life have felt similar things: "Man, this is so broken and wrong." But then it stops. And I'm like, "Well, okay, but maybe we can try. We're all relatively smart people with some bandwidth and resources and whatever. If anyone can do it, someone like us should."

I think everyone has their own battles they want to fight, and I actually think that, mostly, we shouldn't lose sight of the fact that none of this matters if we're not able to live lives that are meaningful for both our loved ones and the extended people we're around. I think it's very easy to convince ourselves that we will do that thing later or after.

At some point over the past few years, I just started to feel like I don't want myself, my friends, or my loved ones to have this view of, "Well, when we're 50 and 60 is when I'll think about taking care of the world, and right now we're executing." I just don't think those things should be in direct conflict.

In the same way that, when you're younger, you think about—at least I thought about—my career as, "Well, I just need to get to this point, and then I can relax." I think that, actually, it's much healthier to just view life as an infinite game and try to figure out what are the things that you are most morally attuned to and passionate about.

I've felt that technology as an industry has become more insular and more of an "everybody else but us" mentality. That's what's most important, because they just don't get it, or they hate us for some reason, or they're overly negative. That's not where change happens.

Yeah, I think it's been an interesting moment for at least some people in tech, where you've seen these tech heroes just pandering in an insane way that you would sort of not have expected. And I also think, at least I'm starting to see strands of this in people's thought, that a lot of us were very capitalist, growing up in venture, with the great-man theory of venture, the Elons and Ayn Rands and all that stuff.

I think the sort of acceleration and seeing some of the stuff Trump's done is, in some sense, challenging that for some of us, I would say, and making you think, you know, maybe economic efficiency isn't the only objective, and it isn't a justification in itself.

Yeah, it's a strange time, and it's a very emotionally conflicting thing to feel. I don't think there's a right answer, but I think everyone should just give it more brain space.

Last one on writing. You write a lot about a lot of stuff, not just tech, and you had this line I really liked about precision. I really identified with that—the pursuit of precision in writing, and really trying to describe things clearly, pull these thoughts out of your head in a very structured way, and put them out into the world.

You said that you can spend hours finding the exact right word and never once risk being seen: "A perfect sentence that, despite having all the right words, actually reveals nothing."

Maybe for me, I really struggle with writing about stuff that isn't in that precision mode of trying to map something out, understand something, and make myself understood. Whenever I try to write in another way, I sort of feel pretentious and get icky and have to stop. Do you have any advice for how to write in a more natural way? I'd love to have this relationship with writing that you seem to have. That's super cool.

You should default to feeling like everything you read from your past self is embarrassing in some way. If it's not, then it probably means you're not progressing enough. And if you have that first-order view of your writing, it really relaxes a lot of the constraints around what everyone will think and what you will think of yourself.

At a much simpler level, the people around all of us just want to know what's going on in our heads more. All they want is a little more resolution. I'm sure in your friendships, your relationships, your whatever, people are just looking to get a little more resolution always, because the translation from here to here is so low-resolution.

You can do therapy to make it better, you can read a lot, and you can talk a lot, and there's all sorts of things you can try. But sometimes the things that bounce around in your head can only come out on a page because there's no one there to stare at you and react to it.

I also think that is something that's a lifelong experience for me, which can also be a crutch, to be clear. There are some people who can write incredibly well and then speak with no emotion, and I think that's a different problem.

For sure.

Yeah, I don't know. That's the best advice I've got.

And so I want to end it here, because it's been awesome and I've already gone a long time. Last one: has it ever been a problem for you—the over-verbalizing? Because sometimes you can overly try to verbalize, or you're living something and you're immediately thinking about how to write about it. You're sort of narrating in your own head in some sense.

Yeah, I think that can be a problem. I think that historically in my life, though, as a younger person, I did not do as good a job putting things out so that the people I cared most about had as much resolution as I thought they did.

I do think that there is some balance of understanding when to be in the moment and let the moment sit, versus thinking that every single thought that spills through your head should be written and verbalized. Again, lifelong pursuits.

Awesome, man. Thank you so much for the time. This was a great chat. Anything you want people to know before we end it?

Twitter, MHDempsey. A lot of chaos on there. Come enjoy it. Yeah, thank you.

Thank you, man.