AI热潮放缓后会发生什么

You recently wrote about 4 buckets of investing styles that might be applied to venture, private-market tech, let’s call it. Can you enumerate those and say a sentence on the ethos of each?

At this point in my career, after doing this for a decade and hundreds of deals, I sit in these discussions and realize people aren’t communicating very well. I did this exercise where I asked myself, “What are the mindsets people bring to investing?” I think there are basically 4 types of mindsets.

1. The Four Investing Mindsets

The one we were talking about before is the traditional finance mindset: the correct proxy for the health of a business is the market it’s going into and its finances. What are the SaaS metrics? What is the size of the market? That’s a proxy for the entire business. Clearly, growth investors do this, but more and more, you see early-stage Series A people do this, which to me is a little bit discordant.

Then there are what I call productive investors. You believe you’re investing in a productive asset, and the productive asset is what’s creating value. If you believe in that, as it creates value, it will create a market. And as it creates value, you’re going to have to find ways for the business to catch up to that value.

The type of diligence you do if you’re a productive investor is more like: What is the technical approach? What are the adjacent markets? How fast is this likely to evolve? How close is it to capturing value? That sort of stuff. Is this the right team to build it?

Then there’s what I call frontier investing. I think this is the romantic, idealized version of VC from the ’60s and ’70s, which is kind of YOLO, man. It’s like founders know better than VCs, and they’ve seen something that I don’t see, and they’re going to go out there and do it. Who am I to second-guess them? They almost look like talent networks.

I always say Chris Dixon—I mean, remember when he was investing in nootropics? He invested in Soylent, Oculus, Coinbase, and then crypto. It’s like every science fiction trope from the ’70s and ’80s you invest in, and you just don’t overthink any of these things. So it’s almost like the strong form of the productive investor.

The last one, and it is always so surprising how prevalent it is, is speculative investors. They don’t even care about the asset. They just want to time a market. You’ll see this in crypto and in secondary markets. They don’t know about the underlying asset. They probably don’t even know about the actual business metrics, but they’re like, “Well, this looks cheap. It’s clearly going to go up in value.”

Yeah.

“Oh, this is the top of the market. I’m going to get out.” They’re trying to time markets. They’re actually just speculating. They’re not really interested in whether the business will grow.

Again, the problem is that if you have all of these archetypes, you can justify any deal or not justify any deal. If you don’t know what your beliefs are, you’ll always justify whatever personal bias you have.

If you think about yourself as a productive investor focused on disruptive technologies, I love this idea that you should not look at some of the traditional things that so many investors talk about. Talk about what you do look at. You look at so many companies. What is your method for assessing whether someone might be creating a truly disruptive piece of technology, a valuable asset?

I only do computer science investing, and I only do infrastructure investing. I think infrastructure as an industry has a certain shape, certain buying behaviors. The technology looks a certain way, and the industry acts a certain way. What you’re trying to do is take a technical approach and map it into that techno-strategic framework.

There are a number of approaches. One of them is to ask, “What are the leading companies doing internally?” This will give you a glimpse of how platforms might evolve. Or perhaps the team doing this did it in another context, and now they’re bringing it to this context.

How much can you diligence without just using the numbers as a proxy? You can talk to customers: Will they use this? You can see if there are adjacent budgets that are growing and could arguably fragment to become this. Or maybe you have early signs of willingness to pay, but they haven’t figured out the business model.

Every tech wave—I’ve been in the industry for 30 waves—people wring their hands about margins. I remember the cloud, and I remember mobile, and I remember the internet. Clearly, now people worry about margins. Maybe there’s a willingness to pay and a lot of interest, but we haven’t figured out the COGS yet or the actual viable business model. You’re using partial indicators, not full indicators, to do this.

My team has a lot of technical people and product people. Many have PhDs. We don’t second-guess the tech. The founder will know the tech better than us, but we can ask: “How does this tech fit into the broader platforms and fit into spending? How does this tech fit with our understanding of stickiness and defensibility?”

Then we look at early market numbers and say, “Hey, listen, a financial investor would puke at this, but it actually shows that the customers are really interested in figuring something out. Will they be able to figure something out?” There’s a lot of diligence you can do. With frontier investing, I don’t think you do any of that. It’s about the people: Is this the best team to do this crazy thing? Then you just YOLO.

You were saying before we hit start that nobody understands network effects. Can you explain what you meant?

I’ve realized that even the most common terms you use have no consensus around them. Network effects is my favorite example. I asked you before—

I messed it up.

No, you didn’t mess it up. You said what about a third of people say. I asked, “So what is a network effect?” And what did you say?

I said, “A product that gets better as it adds a marginal user.”

Right. To me, this is almost a tautological definition because most products will get better as you add a user, right? They’ll find a bug. They’ll provide feedback.

Yeah.

They’ll tell other people. That doesn’t say anything about defensibility. But other people, when they use the term network effects, actually mean the original meaning of the term from the ’70s: a telephone network where, when users join, they interact with each other. That grows superlinearly, and that’s where you get natural monopolies.

Other people use the term network effect to mean an actual flywheel, a self-reinforcing flywheel. Not just that things get better, but that it self-reinforces in some way. This is where you hear the term “data network effects.”

But even in that sense, you don’t know if that self-reinforcement is superlinear or not, so you don’t know if it makes the company defensible. You have this very common word that’s on 70% of pitch decks, where a founder is sitting in front of you giving a pitch and nobody has consensus on what it means.

So how would you define it?

I think the root of the confusion is not a definitional problem. I think the root of the confusion is that conceptually, we conflate the mechanism with the shape of the moat. There are many things that could create a moat, like a brand, scale, a two-sided marketplace, or a network effect. But reasoning about that moat is only really useful if it has some properties that make the company defensible against somebody else.

If that moat grows superlinearly with scale, like a classic Metcalfe’s Law network effect would, that’s really defensible. If it grows sublinearly with scale, it may not be useful at all. For you and me to have a conversation about defensibility, I think the first thing we need to do is not conflate the shape of the moat with the mechanism that caused it. First, we should say whether the moat we have is superlinear, linear, or sublinear with scale.

A lot of things actually contribute to a moat. For example, as we mentioned, network effects or scale effects. What’s a scale effect? To me, a scale effect is when, as the company scales, it can improve unit economics, normally because it has pricing power, for example, or maybe CAC goes down because of brand. What’s a two-sided marketplace? That’s very different to me from a network effect. That’s where you’ve got self-reinforcing supply and demand.

There are many mechanisms, and then there’s the shape. To have a discussion, you want to make sure the shape actually provides long-term defensibility. The last thing I’ll say is that it’s almost impossible to have these conversations in isolation, because we’re actually talking about competitive equilibria. If you have a network effect, maybe the competitor has a network effect too. If they have more capital than you and grow faster, that probably impacts the shape of your moat. You may not grow as fast marginally because they’re stealing your customers or something like that.

I just think we need very basic concepts. We need to not conflate the actual shape with the actual mechanism, and then we should enumerate the mechanisms, because a definition like the one you just gave doesn’t really help anybody.

Can you give examples of superlinear, linear, and sublinear, just to bring it to life?

Yeah, for sure. A common view of where network effects came from was from the ’70s. I forgot the name of the economist who came up with it, but he worked at Bell Labs, and he noticed that for train networks and communications networks, the interactions among users grow with the square of the number of people who join. As a user joins, they can communicate with many other people, so the actual number of connections grows more than the number of users. Therefore, the value grows more than linearly as users are added.

It’s the same thing with destinations on a highway or destinations on a train. If you have 2 competing companies, one takes $5 million of funding and the other takes $5 million of funding, but one is marginally better, marginally larger—this one is a little bit larger—for the same amount of investment, it will grow faster—

Mm-hmm.

—than a challenger. The growth actually increases over time, and that’s why you get a monopoly. Whereas sublinear means that if I’m a leader and I invest the same amount as a challenger, I grow marginally slower than the challenger. That would be sublinear.

This is clearly very important when it comes to moats, because if you have 2 competing companies and there’s a sublinear moat, the challenger will always catch up to the leader, because the leader has to spend ever more money to grow, all things being equal.

Favorite recent company examples of these different types?

How many actual network effects in the Metcalfe’s Law sense of superlinear do you think there have actually been?

Very few? I don’t know.

I feel like this almost never happens. I think we can all agree that telcos—that’s why we had to break them up. We could probably all agree on LinkedIn: they haven’t had to change very much, and they seem to have a permanent status. Social networking, too. All of these examples are about the interconnections of people.

I would say GitHub was another one that had a—

Which was effectively interconnections.

It’s interconnections of users, right? I think there aren’t very many examples, but when you have them, these are very durable, high-value companies. That, to me, would be a classic network effect.

Other companies probably have more scale effects. For example, people will say Uber. I’m not sure that would fit the definition of a network effect. Maybe it’s a two-sided marketplace, because you have the drivers and the passengers. Maybe you could use that, but even then, you don’t have a lock on either of them. All drivers drive for everybody, and all passengers could use anybody. That was more of a funding scale effect, because they could raise more money and therefore the unit economics went down.

2. Growing Markets Diminish Moats

My meta observation is that the network effect thing was more of a personal annoyance, and me being pedantic. But I actually think defensibility and moats aren’t really even worth talking about most of the time. Let me explain why. Here’s my experience, for what it’s worth: if a market you’re investing in is growing, companies aren’t going to want to compete. There’s a lot of new white space that gets created, so what they’ll try to do is move into other white space.

Talking about defensibility is kind of a second- or third-order thing if the market is growing. And if the market is slowing down or shrinking, normally it consolidates anyway, independent of whether there are moats. I do think we spend a lot of time as investors hand-wringing about defensibility when, in practice, it doesn’t really matter, especially in the growing markets we invest in.

We’ve seen this with AI. I remember early discussions about models. I can’t tell you how many investors were like, “I will never invest in a model. It’ll never accrue value. It’s not defensible.” That’s fine, but as long as the market is growing in a large market, it’s going to accrue a lot of value. I remember this about the neoclouds: “They’re never defensible. The quality of revenue is bad.” I think we trick ourselves as investors by thinking in terms of defensibility when, if there’s a growing market, it’s kind of not even worth considering.

Do you think that’s a unique privilege of your stage of investing? And then once you get to a later stage, public, someone buying the Anthropic S-1 probably needs to think about defensibility.

It depends on your thesis about the size of the market. If you believe we’re 0.1% penetrated, this has got a long way to go, and value will accrue. If you’re liquid, you can go in and out of the market if value accrues.

I think the nature of venture has really changed in general. If you read Arthur Rock, it was really frontier-like investing, and people didn’t think about quality of revenue. They didn’t think about defensibility so much. It was people, technology, and new markets.

Over the last 30 years, the tech market has grown so large. The amount of money available to investors has grown so much, and the size of private companies has grown. A private company today rivals public companies from just 10 years ago. So, on some level, VCs have actually become public investors. A lot of the people who have entered VC have come from more business school and traditional finance backgrounds. I think a lot of the mindset comes from that world, but it gets applied too early.

The financial mindset gets applied too early.

A traditional investing, financial mindset gets applied too early. Very often, for example, I talk to a junior investor, and they’re like, “Oh, the quality of revenue’s bad.” I’m like, “Homie, we didn’t even know how to monetize the internet.” Business models always lag disruptive technologies.

If you truly believe you’re investing in a disruptive technology, if that’s what you truly believe in, A, you should believe it creates TAM, so why would you worry about market size? It’s creating market size. And B, it’s disruptive, so we don’t even know what the pricing model is. Remember how long it took for us to go from perpetual to recurring? Years of hand-wringing, et cetera.

My caution is that most investors should know what kind of investor they are. Maybe you’re a financial investor. That’s how you view the world. But then you’re not actually evaluating tech. You’re not evaluating the disruptive power of tech. You’re literally looking at the quality of revenue, which, by the way, is going to naturally move you toward non-disruptive technology, because if it’s disruptive, you probably don’t know what the business model is. You probably don’t know what the TAM is.

On the other hand, if you believe, “I invest in tech, and tech creates value,” and that’s the type of investor you are, and you do that early, you should not consider these other things.

Hmm.

I think it’s less about whether there’s a right way to invest. No, these are all legitimate, coherent, right ways to invest. It’s more about not being confused in your basic approach, because otherwise you become incoherent. You’re like, “Oh, I’m a disruptive tech investor, but I’m looking for a good NRR.” It doesn’t work that way.

How would you describe the state of the union today in your world of infrastructure? Obviously, it’s changing really, really fast.

Yeah.

There are new kinds of companies, but there do seem to be some layers or categories.

Yeah.

How would you describe what it looks like today?

For me, the definition of infrastructure is that the differentiation is actually technical. I don’t view SaaS or apps that way. If you invest in a SaaS or an apps company, it’s a technical investment. It’s a business model innovation.

For us, as technical investors, it has to sit in the traditional technical stack. The traditional technical stack for infrastructure is compute, network, storage, and databases. Each one of those has a subcomponent, and they all have buyers. We’re seeing that being reformed a bit with AI, and all of them are being disrupted.

Now you also have models, which I would consider infrastructure. They look more like ASICs than traditional software, given the capital needs and competitive pressures. I think they’re disrupting the entire application stack above them. We don’t even know if they will layer. Maybe you could argue that with Jeff[?] you can put them in software, but traditionally, they’re their own separate entity that you can’t marry to software very well.

I’d say we’re at a point of maximum disruption. But all of the infrastructure-supporting software, whether it’s vibe-coded or not, or whether it’s used to build models or not, doesn’t change the major components. We do have a bit of a traditional mindset of compute, network, storage, and databases. But we also throw up our hands and say, “This is all changing so fast.”

For example, when you worried about dev tooling in the past, you worried about ergonomics for a human. Now you don’t really worry about that. A lot of the design decisions we made as technologists were based on the idea that a human was going to be using this, so it couldn’t be too complicated. You’d trade off things like safety or reliability. We don’t really have to do that anymore because the models are doing it. I think this is the biggest refresh of computer science infrastructure we’ve seen, maybe since the internet, maybe ever.

Would the shorthand be that each of those five portions of infrastructure is being refactored for agents? Is that oversimplifying?

I split it into 2 pieces. People are vibe-coding a lot of stuff. There’s more software being written than ever before, and that stuff has to run somewhere. That’s not agents; it’s just normal software. Maybe an agent created it, maybe a human being created it, but it needs databases and compute. So that’s one thing: it’s the product of agents.

The other thing is that we’re building these new systems that have AI at their heart, and then the question is how that changes things. Those are the 2 pieces. Just assume way more software is being built and has to be run and deployed somewhere, and then consider that we also have systems that actually create agents.

It seems like each new wave of technology requires a new kind of database. Are you seeing that across all the areas you care about? Because of AI, are there whole new entrants?

Yeah, I think so. A lot of infrastructure is getting a bit sleepy, and now, because agents are very sophisticated, that’s no longer the case. For example, take OLTP. What is OLTP? OLTP is a transaction database, so think of the Oracles of the world. That’s a pretty mature industry. You didn’t see a lot of movement, maybe a move to the cloud or something like that.

But now we’re seeing really interesting companies that are saying, “You know what? The hard thing about using a database from a programmer’s standpoint is that databases always have their own languages. They’ve had their own languages for technical reasons, and it’s really, really hard to have them work with normal programming languages.” One reason is that if you change that language, anybody who knows the databases wouldn’t know how to use it, and you’d have to train a whole new generation of programmers.

Now we’re seeing databases come out that are an entirely new way of thinking about databases. They have way better properties. You don’t really have to train anybody, because the agents already know how to do it. I think these things are path-dependent. In some sectors, a new entrant is going to come and change things. Databases, OLTP, may look entirely different in 5 years. In others, the incumbent may figure it out and say, “Hey, listen, we’re just going to add a few things.”

Networking, of course—I did my PhD in networking, so it’s an area of special interest. It’s been so sleepy and boring for the last 15 years. Now, because the clusters we’re building are so large and have such rigid performance constraints, we’re seeing net-new networking companies come out for the first time since Arista. You’re seeing changes in the materials used to build them. We’re seeing optical come up, so there’s a lot of interesting stuff in networking.

Compute: the whole notion of what abstractions you run an agent on—is it a container, a VM, a durable workflow, something like Temporal?—is changing. I do think we’re at a point of maximum disruption where many parts of the infrastructure stack can be rebuilt.

Patrick O’Shaughnessy

Can we spend a few minutes on models? Anthropic’s S-1 came out this week.

Yeah.

Patrick O’Shaughnessy

Obviously, it gives us a beautiful view into the business itself.

Yeah.

Patrick O’Shaughnessy

But I’m actually more interested, given the infrastructure stuff, in what you think about the state of these things today. I mean, the debates around closed versus open, the public frontier versus the private frontier that the companies keep for themselves, movement of talent, opportunities, things that don’t exist that should—whatever’s on your mind about these things and how you see them evolving. I’m fascinated to hear.

I think the biggest question is what happens if and when the market slows down. We’ve got 3 years of data about what happens in the existing, very big, fast-growing market. What happens there is that being on the frontier means you’ve got tremendous pricing power. Even if you’re just an epsilon better than the next guy, you’ve got tremendous pricing power. There’s no demand problem. You’ve got basically infinite demand.

These frontier models that are the best tend to get 80%+ of the market, dollar-weighted, and that’s happened for 3 years. But you could argue that’s happening because we haven’t saturated the market at all, and people are always going to want the smartest thing. Under that view, there’s nothing structural that’s going to change as long as the market’s growing. You could argue this is going to go on for quite a while because the penetration point is 0.1% or something like that. They say 3% of the enterprise, but actually, budget-wise, it’s like 1% or less. There’s just a lot of room to grow, given how useful the technology is.

The more interesting question to me is what happens if the market starts to saturate in pockets. If that happens, things shift to open source. Will you lose pricing power on the frontier? One thing we don’t do a good job of accounting for when we model out the numbers for these companies is what happens if you invest in a model and then you don’t get the next model out really quickly. Let’s assume Anthropic stumbles and they don’t release a model for 6 months. What do you think the lifespan of the previous model is?

Patrick O’Shaughnessy

A very short period of time.

It’s very, very short, which is entirely different from what we’ve ever had in software. In software, we’ve always assumed some long tail. But what happens if you put out the SOTA model and everybody loves it, then something internal happens? I don’t know if it’s regulatory, technical, a breach, whatever, and then you slip 3 months. It could be the case that you lose the entire market.

In my experience, and looking at a lot of these model graphs from apps companies and pitches, I’d say the lifespan of a model is 3 to 6 months. That means you have to release the next model. For this conversation, I think it’s a very interesting question. We should tease it apart: what happens if the market continues to grow and doesn’t saturate? I think we already know, because we have 3 years of data. It just means the frontier wins, and they do a great job making great, great models.

Then what happens if either the use cases saturate and you don’t need more intelligence, or the market starts slowing down for another reason? Nobody knows the answer to that. I think there’s a very real world where the value comes crashing back down into open source, and we’ll see a shift that we’ve never seen because software’s always had a long tail of retention.

Patrick O’Shaughnessy

On this trip, I’ve been asking a lot of people what percentage of software coding tasks are already saturated, and the typical answer is a very, very high percentage. That’s been the biggest market. And yet it still seems like people are spending like crazy on frontier tokens to do coding, which seems a little at odds with the idea that I could do it with a six-month-old model or something vastly cheaper.

Well, these are 2 different statements. This is a great thing to dig into, but we need to be very careful. Statement number 1 is, “I’m not using more tokens. I have enough tokens. I’ve automated 95% already, so I can’t automate any more.” I think we’re getting pretty close to that. Comment number 2, “Therefore, I can use a cheaper model,” doesn’t follow from comment number 1.

Patrick O’Shaughnessy

Hmm.

So if you and I are competing startups—Martin’s startup and Patrick’s startup—we’ve got competing startups. And because we’re in competitive equilibrium, are you going to choose a dumber model to compete with me? In competitive equilibrium, you wouldn’t. You would always pay 10 times more for the smartest model.

Patrick O’Shaughnessy

You would, yeah.

Yeah, always, always, always. There’s this very open question of whether you’ll saturate tasks so that you don’t need a smarter model. I think it’s very hard to do that in a Darwinistic economic equilibrium when you’re actually competing. Some of them may, but at the end of the day, it’s about whether you can get software out faster, make it better, and compete in a marketplace. That will necessarily push you to the more intelligent thing.

That said, clearly things will slow down because you’re not spending more tokens. So either you have to increase prices or you’re going to have to go broad. But in my experience, the pricing is maintained, for sure, because of the competitive equilibrium. You’ll always want the smartest thing, and then we’re just seeing a broadening use of these outside of code, in general knowledge work. I’m a huge Grok bot user. It’s my favorite thing ever. I think we’re starting to see AI take over more and more of these areas.

Patrick O’Shaughnessy

Say more about your personal experience with that: what you’re doing with it, what you’re learning from that experience.

My new Grok is just so funny. I don’t know—

Patrick O’Shaughnessy

Maybe you’ve got a home automation thing going on.

Oh, I know. It’s so good. I just feel like I’ve waited my entire life for—

Patrick O’Shaughnessy

For the power.

…this power, this technology to show up, right? I want Star Trek. I want to go home, and I want my house to greet me. I want to talk to it, have the face show up on the monitor, and have the whole thing. We’ve never been able to do it, and now it’s kind of trivial. I use Grok bot for a lot of stuff. I use it for personal finances. My big hobby is modular synthesizers. I like making music.

Patrick O’Shaughnessy

Oh, cool.

But with the 1970s-style patching and tweaking.

Patrick O’Shaughnessy

Yeah.

It helps me with a lot of that stuff, so I use it for a lot of standard things. My new passion project is that I want Grok bot to embody itself. I say, “I want you to be the living AI in my house.” That’s what I want, and I want it to help me make this happen, just like Star Trek. And it’s literally—

Patrick O’Shaughnessy

Buy stuff.

It’s literally buying stuff. I was like, “Oh, let’s try this.” Yesterday, I bought a few hundred dollars’ worth of gear that’s coming to the house, and I’m going to have it embody itself so that—

Patrick O’Shaughnessy

What gear is it buying?

I don’t even know. It’s these weird open-source projects, really obscure stuff. I know some home automation stuff, like Zigbee lights, but this is some really obscure stuff it’s decided to buy. I just have this vision in my head of that vintage view of what the future looked like, the ’70s view: you walk into the house—

Patrick O’Shaughnessy

The Jetsons kind of thing.

Yeah. I just want that. I’ve basically been feeding it pictures and descriptions, and I want it to create that by itself.

Patrick O’Shaughnessy

Hmm.

I’ll let you know how that goes.

What do this and your other experiences using this technology tell you about what’s coming, since you’re probably in the 0.0001% of creativity in using it?

Yeah, this is it. People keep thinking of these things as software, and it’s not. We keep mapping it to things like search, and it’s not. We’ve created general, flexible technology that can bring us many of these future things we’ve always dreamed about. We just didn’t have the technology to implement them. I feel like we’re so constrained by our presuppositions about what software can do that people aren’t doing as many of these things.

I’ll give you another example. I worked with Fei-Fei at World Labs, a great company, and one thing I realized from what they did is that they created a model that would generate 3D scenes. They released a model called Atlas a few months ago. It would generate basically any scene, but it also supported dynamics, so it would move. I realized we’re this close to being able to step into a movie. Imagine—Miyazaki probably wouldn’t be happy with it—but imagine having Spirited Away running, putting on your VR glasses, and walking in. You’re there.

Did you ever see Sleep No More, the immersive—

Of course, yeah. I went to it.

Why isn’t there a Sleep No More in VR that you can walk around in, where the whole thing is going on? Just for those listening, Sleep No More is an immersive play that’s 3 hours long. It was in this three-story hotel, or a building like that, in—

Yeah, in New York.

New York, and you’re running around trying to figure things out, while Hamlet is being played out or whatever. We’re starting to realize how powerful this stuff is. We can actually make a lot of these tropes and visions of the future from the ’50s and ’60s. I just think the biggest constraint is no longer the tech.

Mm-hmm.

It’s literally our creativity, and that’s because we’ve been beaten down by 30 years of software being shit. Let’s be honest: software has sucked, technology’s been hard to use, old people hate it, and it’s most often broken. Now we’re actually at a point where this stuff could start working.

Mm.

For me, I’m so excited about what this brings.

It’s tempting to say you could just freeze the models at their current capability and still get all of what you described, so there’s a diffusion question I’d like to explore. Before we do that, what, if anything, do you think is missing from the models? How do you hope, wish, or dream that the models themselves can be improved?

I think, again, if you step back and look at what we know as an industry, anything in the current tech stack that’s digital can be solved with enough capital. Anything you can simulate on a computer, you can create RL environments for, whether that’s 3D using Blender, general application use of a computer, or code.

You and I have talked about this before. When it’s all said and done, the most interesting thing about AI—the meta thing—is that you can take a model, point it at a problem, and it will solve that problem. Maybe it’s expensive, but at least you can solve it. In the past, you’d be blocked in engineering and you’d die. Now you can direct money at problems and solve them.

I feel very confident saying that anything in the digital world can be solved because you can create an environment, use RL, and train on it. I feel very comfortable that anything you do with a computer, AI will be able to do. Maybe it doesn’t do it now, but you can see that path happening.

What I don’t know is whether the same thing applies to the physical universe. Sure, you could set up a lab to make speakers or whatever, to do that—

Yeah.

—but are the scaling curves the same? Is the distribution the same? Anything to do with robotics, the physical world, science, and medicine, I have no idea—

Mm.

—if it follows the same laws, so I think that’s still an unlock. But I feel very comfortable that the current digital side is done.

How have you processed all this interesting security stuff? Breaches. I’ve heard it called “breachpocalypse.” It’s obvious that we’re entering a very interesting era in cyber. What do you make of all this?

I lived through the internet on the ground, with formal studies in security, and it was way worse than this. By this time with the internet, we had worms that had caused billions of dollars in economic damage. We’d taken out critical infrastructure. Remember Windows 95? You’re probably too young.

No, I remember, yeah.

Okay. Remember Windows 95?

Just barely, but…

It used to be that if you installed Windows 95 and connected it to the internet, it would be infected before the installer was done. It wouldn’t take it over. I’m not joking.

Yeah.

You didn’t even have time to patch it. It was so chaotic, and there was so much mayhem. This is what happens every time you get a new technology that changes how we run computers and changes the security posture. We’re definitely dealing with another one of those moments.

We can look to what happened with the internet. There were a few years of mayhem. You had breaches that caused real issues, and then the industry evolved. Normally, the way you think about security is: will the good guys be able to spend more money than the bad guys? If the answer is yes, which is typically the case, you’ll come up with a technical solution. I think that will be the case here, and you’re already starting to see it.

These vulnerabilities showing up are good. That means somebody used the model to make things safer by pointing out the vulnerabilities. I think that’s going exactly as you’d expect.

Mm.

But I think the net impact is that software will be much more secure—

Mm.

—because AI is also securing it.

Can you talk about capital? There are very interesting things happening in capital markets. There’s clearly a lot of demand destruction happening. We’re not funding things we would have funded in a different environment because the cost of capital to fund them is too high.

That’s it.

And the cost is rising because the returns are so high on anything AI. I was looking at this company this morning. If you and I had seen the deck 5 years ago, we’d have said, “This is an amazing software business, and we should fund it.”

Yeah.

And now it seems to pale in comparison to other things we could put our capital toward.

Yeah.

Capital itself feels like this interesting variable in the whole equation, as a constraint. I’m curious how that feels to you.

I sit on the boards of a lot of great companies that have the problem you’re describing: profitable, growing 40%, 50%, 60%, at scale, with $300 million, $400 million, $500 million in ARR. What do those companies do? You can’t really IPO because they’re subscale, but they’re great companies.

I think about my own personal portfolio. I’m like, whatever, I’m in muni bonds, man. I’d rather be in these companies, right?

Mm.

Everybody would be happy with an asset that grows 40%. You’d be happy having your personal money in that if you could guarantee it. I still think these are very good companies.

I feel like it’s more a question—again, this is my naivety, not understanding this—of whether we’re going to see a shift in the market so people can get exposure to this asset class. I think the risk profile is very, very different from AI. Nobody knows what’s going to happen to AI when the market slows down. Nobody knows. It could be easy come, easy go. It could keep going forever. Nobody really knows.

The only free lunch I’ve ever seen in investing is diversification, and I’d love to personally have—

A basket.

—a basket that includes these slower-growing assets that don’t disappear.

Are most of the companies you have in mind, though, especially exposed to this disruption risk? There are lots of companies with that rough profile that seem terrifying in the face of how easy it’s going to be to create software.

I’m an infrastructure guy, so for me, more software means you need more infrastructure. We don’t vibe-code databases yet. You do if you’re a data lake company.

Would Snowflake be a well-known example of what you’re talking about?

Yeah, it’s a great example. It’s not just building it. It’s really hard to have the long tail of features and the long tail of performance characteristics. Operating these things is really, really tough.

Let’s say Snowflake is growing 30% to 40% and the economics look good. I don’t think they’re at any risk of being disrupted by AI. Maybe by Databricks, but that’s a pre-AI type of thing.

Again, I’m very naive when it comes to these things. I’m a very simplistic thinker. I don’t think AI is going to disrupt these. I think they’re good businesses. More and more, I think the industry is going to want to wrap these up and get exposure, and I don’t know what that looks like.

It’s kind of interesting—we’re even seeing SPACs come back. I get a lot of calls now about companies I sit on the board of, which are good companies, where these SPACs are like, “Hey, listen, we can make this stuff go public.”

I think the financial markets have to figure that out. A lot of companies will be disrupted by AI, but a whole bunch won’t. I just think the market doesn’t know what to do with them right now.

Most meetings I have around here, at least in the last 3 months, spend half the time on compute. Across categories, the product just needs more compute than is available to accomplish its goals.

This is another way of asking about the capital thing. You guys manage a ton of money. Everyone’s deploying their funds very quickly. Each company that’s working requires lots of capital. I’m trying to get a sense of whether capital itself is the constraint. If we just airdropped another trillion dollars into all the funds, would that be a good thing?

I said this before, but I’m going to reiterate it because it’s so important. I really think the big disruption of AI is the fact that you can throw capital at problems. In the history of tech, you haven’t been able to do that because it was always an engineering problem, which just doesn’t scale. The Mythical Man-Month is the Mythical Man-Month for a reason.

Ten years ago, if I had 5 founders show up and they asked for a billion dollars, and I gave it to them, what would happen?

It sits there.

Or they would detonate, right? They’d hire a bunch of engineers, and then you’re really screwed. You can’t throw capital at it.

Can you explain The Mythical Man-Month? It’s a very interesting idea.

The Mythical Man-Month basically says that you can’t increase the speed of engineering. It’s like a natural law of physics.

It’s sublinear.

It’s very sublinear, and it has to do with organizational communication. That’s one of the conclusions. The takeaway for those listening is that this comes from, I think, Fred Brooks in the ’70s. You had these software projects and wanted to get them out faster, so you’d throw more people at them, and then they’d go slower.

Engineering has always been this bottleneck. In my experience—and I had my own startup for 10 years, and it ended up being a very large business—this is totally true. You can’t scale engineering.

To me, the biggest shift—an Industrial Revolution-level shift with AI—is that I can take 20 people, give them a billion dollars, and they can use it. Will they make $500 million or $2 billion from that billion? I have no idea. But the fact that you can productively use that money is so new and so interesting.

At some level, the capital markets need to figure this out because these companies are taking this money and using it. If you look at Anthropic and OpenAI, they’ve probably raised more money in aggregate than the entire downstream ecosystem combined, by a factor of whatever. They’re mechanisms for turning money into growth.

That’s one side. But there’s another side, the debt side, which is quite a bit different from the equity side. I do think that currently—I’m not a growth investor, a public markets guy, or a debt guy—but here’s what I’m seeing on the ground: we’re running out of available debt.

If you’re the Norway pension fund, are you going to lend to a random neocloud or buy T-bills? I think we’re at a point where you don’t have enough money to go around, so it’s going to favor investment-grade companies.

Mm-hmm.

Those that aren’t investment grade will have a harder time.

So scale wins again.

Scale will win, and it’ll be harder to raise debt. If your business model requires you to raise debt, there just isn’t enough debt to go around.

I know you were talking about capital in the sense of general capital. You’re going to see a ton of liquidity. That feels fine to me. I don’t know these debt markets, but I definitely know it’s harder to get debt now.

I think this will mean that if we want to have smaller players that aren’t investment grade and can get debt, they’ll need to prepay, which means their customers will need to prepay. Over time, this all flows back to the availability of cash and capital—

Mm-hmm.

—because you do have limited amounts of debt.

This reminds me to ask about Neo labs, Neo clouds, inference providers, and companies. I don’t know if you call these infrastructure, or what part of the stack you—

Yeah, yeah. I view those all as infrastructure.

If you take those categories in turn, lots of them have been funded. Very few of them are at scale. I’m curious how you think this plays out: what roles they should and will play, and what the market feels like. These things are growing unbelievably fast in most cases.

Yeah.

I’m very interested in these companies, and I’m curious what you think of these categories.

Well, let’s put Neo labs aside, because I think that’s the least well-defined of all of them.

Sure. I included them, to be clear, because so many of them seem to be building their own—

They’re becoming Neo, they’re becoming Neo clouds.

Well, every pitch seems to turn into some version of, “We’re building a token factory with some degree of vertical integration.” Some Neo labs are going all the way down to buying land and power, and all the way up through the token. You know what I mean? So there’s this weird—

Yeah. In my world, that wouldn’t be considered a Neo lab. That would be—All neoclouds want to be token factories, so if your output is a token of an existing model architecture—

Compute or something, yeah.

Yeah, you’re a neocloud. They all want to do that, and they’ll all verticalize. The actual limitation right now is hardware supply. That’s it. It’s not capital; it’s not debt. That’s the real limitation—and power.

To me, a neocloud is one that provides the physical infrastructure that ends at tokens. They could provide GPUs, fine-tuning services, or tokens. I think of it as a data center that actually has the GPUs in it. That’s what a neocloud is to me. A Neo lab, to me, is: “Give me a billion dollars. We’re going to come up with a new model architecture.”

Train a model.

A lot of these business plans are, “If we solve ASI, that’s our business model.” Literally. To be fair, that was OpenAI’s original thesis, and to some extent Anthropic’s: if you solve the intelligence problem, the business model part is easy. So to me, an AI lab is about new architecture and model training. Sometimes they’ll buy their own infrastructure and their own GPUs, but that’s not what they’re building toward or selling.

The real constraints right now are power, chips, and materials. With that sort of limitation, 80% of these things are going to go to the scale providers. The other 20% are going to be hard-to-reach, niche-type places. Maybe I have a relationship with a random government that has the power.

It’s kind of funny: I met this company that’s a GPU marketplace, but it’s the most low-level version of one. Many of their GPUs are in converted chicken sheds—chicken coops where people on farms are finding ways to provide them. The market is so large that there will always be a significant portion that it just doesn’t make sense for the SpaceXs to serve. SpaceX isn’t going to create a marketplace for chicken-coop GPUs. I think that’s what will end up happening: the 20% is large enough to support a very wide, fragmented array of folks, while 80% of the market will be these homogeneous things like SpaceXs.

I heard this story recently that someone asked Larry Page how much of compute would be orbital—

Yeah.

—and he said “five nines.” The reason was that it’s just physics. Eventually, it’s all going to be in space.

The space thing actually pencils out, but the reason it makes the most sense is regulatory.

Just because it skirts regulation?

Yeah. There are 500 proposed pieces of legislation to stop data center building in the United States alone, including in Texas. If you can’t build a data center in Texas, I don’t know what we’re doing anymore. If that regulatory pressure continues and you actually can’t break ground—and I’ve heard from many of the top data center people that it’s very tough to break ground right now—then either you go international or you go into space.

Your comment on these software companies growing 60% reminds me to ask about prevailing valuations in the market. As a productive investor focused on disruptive assets being created, even if the disruptive asset is truly incredibly valuable and great, every one of those has a price that will ruin the return on investing in it if it’s too high.

So this is another one of my bugaboos. I don’t know how you can be a VC and not believe tech creates value, which is the first conversation we had. Now I’m going to have a second conversation: I don’t know how you can be a VC and not believe markets are efficient. What are we doing? We’re capitalists; you have to believe markets are efficient.

I think private markets are very efficient. Private markets figure out the price, so I just don’t try to bet against the market. A lot of people think private markets aren’t efficient, and that you can find the deal and find the non-consensus deal. All of those deals were incredibly competitive. I didn’t win by looking for a deal or whatever. I pay the market price, and maybe I’ll get a little discount for negotiating, but I never really worry about that.

As a productive investor, I just trust that the markets are pretty smart, and I trust that they’ll value things on a risk-adjusted basis. I believe my diligence reduces that risk for me. I don’t do it based on market value. I’m not trying to find a better price; I’m just trying to find a less risky asset.

Doesn’t it follow, though, that the approach would be to try to index the private markets? You’d just try to own the market weights of—

Oh, I would love to, but you can’t. I would love to, of course. This is where this all leads, and this is kind of how we think about investing. Of course, if I could have an index of every competitive deal, my dream would be to get 5%—

Is it, though, in Cognition when you buy Cursor?

I’ve gotten 5% of every competitive deal. I really believe private markets are super efficient, and I think the consensus is actually smart, because I’m a capitalist and I believe in markets. If I could have 5% of every competitive deal where you’ve got legitimate investors in it, I’d be happy.

But you simply can’t do that. You can’t do that for competitive reasons or capacity reasons. As an investor, you have to figure out, how do I not run to the competitive bus stop? Even more importantly, how do I not run out of our ability to keep track of and help these companies? I do take the board seats; I get very involved. That’s just impractical.

The way we think about it—which you’ve heard before, but I’ll restate it—is that for every legitimate sector, which we define as one with good founders, we don’t second-guess the founders. If they’re risking their futures, the opportunity cost, and their family’s wealth, then it’s a good sector.

We do the work to decide which is the top company in every good sector. That means looking at the team, the tech, the monetization potential, the market growth potential, and all of that stuff. Then we choose the best one, and we assume the market is efficient. That’s the approach. It’s kind of an index, but we’re segmenting it by market. Sometimes we get it right and sometimes we get it wrong, but that way you’re indexing the interesting markets.

Yeah, there’s this thing in indexing called stratified sampling. It’s from my old quantitative days. You can buy a small subset of the index and have very low tracking error to the index, because what you’re describing is effectively stratified sampling.

Okay.

You get exposure to the best, biggest, or whatever per category, and that will perform similarly to the others.

Yeah, yeah, okay. I think venture capital early on—and I wasn’t there; I just read the books—believed that tech created value. Often there was no business to evaluate. I just don’t see so much hand-wringing about bubbles or this or that. The market price is the market price.

To me, there’s always this huge dissonance when I’m talking with other investors. They’re always decrying the bubbles, which nobody can ever predict. By definition, if everybody’s calling it a bubble, it’s probably not a bubble. It’s almost a bubble when everybody’s investing because they think there’s still upside. And then they don’t somehow think private markets are efficient.

I remember I got VC-canceled because I said, “Non-consensus investing is dangerous.” I once wrote on Twitter, “Hey, listen, I think it’s dangerous to invest against consensus because basically there’s follow-on capital at risk.” After I said that, many prominent VCs came out and said, “Oh, all of these deals that we’ve done were non-consensus.” I think there’s something very tied to the ego and identity of investors: they have to believe they can truffle-hunt and find the non-consensus thing, that they’re smarter than everybody else.

But I think the reality is that, if you look at our portfolio, the ones that returned money were actually very, very competitive in general. I just think these markets are quite efficient.

I love the idea of market efficiency and private markets being one area where you disagree with lots of well-known private investors.

Yeah, apparently a whole lot.

Apparently all of them. Are there any other areas like that where you feel—

Yes.

—especially alone?

Yes. There's another one that drives me nuts. This is my last one, I promise.

Okay.

Another thing that drives me nuts is this belief that if you put money in private markets, it doesn't somehow grow the TAM of private markets. I love Roelof. I think he's fantastic. He's an incredibly intelligent guy. He's a great board member. But he gave this talk at TechCrunch Disrupt. Do you remember? Many investors believe this, so this isn't to call out Roelof, because I'm actually a huge fan of his. But he was like, “Venture capital is not an asset class. There are only 20 startups that matter, and that's it. If there's too much money going in, you've overfunded private markets.”

To me, this is ludicrous. Capital creates opportunities and value in markets. Even though there is so much more money that has gone into the private markets in the last 20 years, the value in the private markets has kept pace, right? You've gotten massive private companies, tons of opportunities, and very large exits.

This zero-sum thinking about private markets is so bizarre. We're venture investors. We create value, like tech creates value. We put more money in it, it'll grow. You might be surprised—or maybe you won't be—but I've been very surprised at how common this mindset is. Very often, I'll talk to a pretty senior GP who's been investing for 20 years, and they'll be like, “Oh, there's too much money.” I'm like, “What are you talking about? There should be more money in there. We need more ecosystem, more value, more investment.”

So I do believe money in private markets creates TAM. It creates the investable TAM. I think it creates value, and there's more opportunity to deploy money now in private markets than there ever has been before. A very simple example is companies not going public. If there's more private money available, why would you even go public? Again, we need to all step away from zero-sum thinking. I think we all need to believe that technology creates value. Now, especially with AI, we need to believe that capital enables technology to create value. If that's the case, we should deploy more money in private markets and create value.

You've been on fire. In NBA Jam terms, you're the player who's on fire and can jump really high. Between Cursor, Fei-Fei Li's company, and OpenRouter—all of these in a very short period of time—well done. But I'm more curious about the lessons you've learned about M&A and what's going on in this market, having been a part of maybe the 3 biggest major-scale M&As in the last couple of months.

3. Strategic M&A Runs On Trust

So, actually, I have a history with strategic M&A going back to my company. When I sold my company to VMware, it was a $1.26 billion exit, which at the time was the high— I think I beat David Sacks' Yammer.

Who's counting? Yeah, right.

It was kind of funny, because Catalyst showed me this chart. It was just a chart of revenue multiples at the time of exit, and there were 2 outliers: Yammer—

You and Yammer.

—and your company. They were really close to each other. I was marginally better. Just a small bit. At the time, I think that was the highest enterprise software acquisition by revenue multiple. Then I was part of a deal that beat that, which was the Tabular acquisition by Databricks. And then, of course, Cursor—I think it was the largest private acquisition ever. Fei-Fei's acquisition was also a very large acquisition.

I've just been a part of a lot of these strategic M&As, and I think my comments only apply to strategic M&A. I've noticed a few trends. One of them is that there's almost always deep personal alignment, because it's a real risk for both sides. Think Lisa Su and Fei-Fei Li, or Michael Truell and Elon Musk, or Alex and Patrick. You almost always have people who have the same visions of the future and think about it the same way, whether in terms of markets, infrastructure, or data. That's almost always present. There's some existing alignment.

The second thing is, you have to look at it from the buyer's perspective. What the buyer is doing is basically buying an option on a large market. In the case of VMware buying Nicira, they're buying a $1 billion option on a $40 billion market. For that to make sense, you kind of have to solve the inductive proof for them. You have to show them the base case, which is, “We have this business that sells into networking, and we can get that budget.” Then you have to do the inductive step: “You guys have virtual machines, so if we attach this to a virtual machine, we can sell more. And therefore, if we're part of you, we can sell a whole bunch of these things.” Because they're buying a future option on a market—

The company's job is to show the inductive step.

The company being acquired, or both of them? The discussion—

The acquirer has the—

Yeah, the discussion for the M&A has to be that you have the base case, you have the inductive step, and therefore you can get this into a large market. In my experience, that conversation is one that only founders can have, or maybe investors who really understand what's going on strategically.

Those conversations are almost like really high-stakes enterprise sales. I've been a part of a lot more auction-y type M&A. You never have those conversations. You're looking at the financial metrics and the spreadsheets, and they're buying an asset, et cetera. They have a very different flavor.

But I think that, especially with the timelines going down in tech, we're going to see more and more strategic exits. So I think it's probably important for people to get pretty smart on how these things work, and they're very different from what people assume. Most people assume you should hire a banker, have an auction, and get the highest price. Hiring a banker is fine. You can do that if you want. But the idea that you're going to go to the apex predator, someone like Mark Zuckerberg, and play poker—you're kidding yourself, right? These things are built on trust and telling a story, not playing games or playing poker.

Mm.

I just think there's a mindset you need to have if you're going to engage in these things that's quite different from the intuition.

So can you tell, from your perspective, the story of the Cursor acquisition, since that's the biggest one and you're such an involved investor? Give us the story.

We were just about ready to get a raise done. We'd raise a few billion dollars. That was getting done. But this was about the time that Mythos came out, and if you worked backwards, you realize it probably cost—I don't know how much it cost—but it probably cost, say, $3 billion to make Mythos. So we're like, “If Cursor is going to build models, we need to raise more money and get access to that compute.”

Then, of course, that kicks off strategic discussions. There aren't a lot of people who have access to that much compute, so then you have those conversations. And then, of course, there's quite a bit of interest in the company as a result, just because the company had by far the most users and by far the most data.

Again, this is one of these great examples where, in the end, SpaceX had the compute, Cursor did have the data, and it had a team that had very successfully built models and built this practice. Code was very important, so there was just a lot of alignment. It always helps to have a raise, because that actually helps set the value and the price of these things. So that was actually a fairly straightforward deal.

I'm curious if there's a similar lesson from OpenRouter or from Fei-Fei's company, World Labs. Is there anything else shareable from those 2 experiences that might offer lessons about this kind of strategic M&A?

So I'm going to go back to what I said: for these strategic M&As, the most important thing is the intuition, especially for a lot of investors. There's a deal going down.

We need to play games. We need to leak it to the press. It drives me fucking nuts. We need to get other companies in. You don't realize you're literally dealing with the most sophisticated people on the planet. I just don't think you walk up to Zuckerberg and say, “Oh, I’m gonna beat you in poker, dude.” That guy will fucking take you for a sucker.

Every one of these deals—and these are world-record deals—the companies, in very good faith, engage with the buyers. They didn’t play any games, and they knew they were worth a certain amount of money because they had done raises or there were market comps, and there was a lot of trust. Having now done so many of these deals, I’ve found that approach for strategic M&A is the right one. I’ve been part of deals that size that have fallen apart because somehow a board member got into the founder’s head: “Oh, you need to get competitors. You gotta leak it to the press.” God, the OpenRouter deal drove me nuts. It got leaked. Did you notice, by the way, that World Labs did not get leaked?

Yeah, interesting.

It didn’t get leaked because we didn’t tell anybody. It was me and a few other people. OpenRouter was leaking constantly, right? I think a number of VCs think, “Oh, this is how you drive up the price,” or, “This is how you get it done.” You see these things leak all the time. That destabilizes deals. Again, I think the most important thing is a structural one: your intuition as an investor, in particular, that this is some commodity asset and it’s about playing poker. It’s just wrong.

If it’s a strategic deal for a lot of money, I think you need to view it very professionally. I think you need to develop a trust relationship. I think you need to do it in good faith. You should not leak these deals, and you should realize that the people who are counterparties to this are the most sophisticated people in the industry. The founders in every one of these cases treated it that way. Michael Truell treated it that way, and so did the team. Alex and Chris treated it that way for OpenRouter. And for sure, Fei-Fei treated it that way.

How are you seeing the role of investors evolve, both literally across the many boards you’re on and in terms of whether investors generally of your type are doing a good job or actually helping companies? The stakes are so high, growth is so big, and the capital amounts are so big that all these investors are overseeing—it’s kind of crazy. Some commentary on the role of the investor.

One thing I will say is that venture is not what it used to be. The startup ecosystem is not what it used to be. If you look at the largest private companies right now, they would rival any public companies from 20 years ago. Stripe, Databricks, SpaceX before it IPO’d, Anduril—these companies are massive. One of the problems is that what it means to be a startup has shifted so much that now you have this undefined quagmire of who should probably be public investors investing in early-stage startups. I just don’t think those intuitions are right, and I think that’s the root of the confusion.

Maybe it makes sense to start delineating a bit more on the private side. Or, if you have later-stage people investing in earlier-stage companies, maybe they need a little more self-awareness. For example, you wouldn’t put me on a mining company board. Remember 2021? We’d have these PE investors investing in these Series B companies. I’d sit on the board with these people, and it was literally like putting me on the board of a mining company.

Yeah.

They have no idea. They’re trying to reason about the company through burn.

Mm.

Dude, you’re gonna be miserable here, because all we do is burn cash for a long time. It’s more about the blending of the lines, because private companies are a real asset class for late-stage investors. A lot of this is just growing pains. How much larger do you think the market is now than it was 30 years ago?

A thousand times.

So of course the industry’s gonna have to mature around that, and of course you’re gonna have to view it like a later-stage investor. But I think there’s this funny identity crisis with a lot of later-stage investors who view the world that way just because they’re in private markets, and their VCs think that—for example, I would never invest in a late-stage company.

Mm.

I don’t think it goes the other way. What would I do?

Yeah.

I would go there and tell you how to run a finance operation. It’s just not me.

Mm.

So I think that’s it. It’s basically the encroachment of traditional banking and traditional finance into venture, which absolutely belongs there. I just think sometimes it reaches a little bit too low, and then you have these issues. You have these issues in M&A, where they’re like, “Whatever, let’s run an auction, and let’s do all this other stuff.” You have it in the operations of a company, where you’re 20 people and they’re ready to burn, or they’re over-rotating on metrics for disruptive technology. I just feel like we need to bring back traditional, productive VC for some subset of this, and there’s enough market for everyone.

Has anything changed about how you meet somebody for the first time? If I came and watched 10 meetings with you in a week 10 years ago, when you started, and then I came today and watched 10 meetings with you, have you changed how you conduct your early interactions with founders?

Yeah, a lot. I think most investors go through this, especially when you come from the operating world. You have to learn how to listen as an investor, and you also have to learn how to ask the questions you really care about so you can answer the questions for yourself. It’s a very delicate balance. I do feel like a lot of meetings with founders tend to be pretty performative on both sides. You don’t know what to ask, so you’ll be like, “Who’s your competition?” A lot of it is just asking questions for the sake of asking questions, but it doesn’t actually inform your decision.

Especially if you’re an operator, you’ll hear something you disagree with and want to fix that problem, but that’s not your job. I’ve learned, and I’m not perfect, but the primary job is just to listen. A founder will tell you what they believe. If you disagree, you disagree, but that’s what they believe. So often they’ll say something like, “Well, I would do it better, and therefore it’s a good investment.” It’s truly an art. You have to make a decision based on that meeting.

There are a lot of silly questions you could ask that aren’t really going to inform that decision. “What if Amazon does that?” “What does OpenAI…” Whatever, you can answer that yourself. Maybe you can figure out how they think, but there are key questions you really want to know. Knowing which ones those are during the meeting and getting them in is hard to do, and I think I’ve really refined my ability to do that.

Maybe a fun place to close is just to ask what has you most excited. We’ve talked about lots of the magic of this new era of technology and things that weren’t possible before, bottlenecks that are now gone. I know all of that excites you, but if you had to step all the way back and describe why this job is so incredibly fun right now and what you’re most excited about, what would you say?

I’m more and more of the belief that this is an Industrial Revolution-level shift. For me, the Industrial Revolution took us from basically a zero-sum world to a non-zero-sum world, where value is being created—net new value is being created. I think this is a very similar shift. Before, to solve problems, you had to solve very complex engineering problems, which basically didn’t scale and were very tough, and then you’d have to maintain them. That’s why all software has always sucked.

Now we’re basically turning this into a capital problem. I don’t know about the real world yet. I think we’ll get there, but I think we’ve turned this into a capital-resourcing problem, and we’re capital allocators. I just feel like we’re in this very, very strategic position to help drive the future in a much more direct way than we ever have before, because I think this is a scale-up capital game that causes that to happen. I don’t believe private markets have limited TAM. I think they have unlimited TAM, and the more money that goes into them, the more value will be created. I just feel like we’re in this very privileged position, very key to the innovation chain, and we’re both very lucky to be here.

I love the every-two-year schedule that we’re on. I hope we get to do it again in another two years. I can only imagine what the world’s gonna look like two years from now.

We’ll have our Grok bots talk.

Thank you so much for your time.

Thank you.

Patrick O'Shaughnessy is the CEO of Positive Sum. All opinions expressed by Patrick and podcast guests are solely their own opinions and do not reflect the opinion of Positive Sum. This podcast is for informational purposes only and should not be relied upon as a basis for investment decisions. Clients of Positive Sum may maintain positions in the securities discussed in this podcast.