Ben Horowitz 谈 AI 投资：AI 泡沫、经济影响与 VC 加速

You know, if you want to change the world, you have to believe you can change the world. What you’re really trying to find is whether they’re literally the best in the world at a thing, and that’s always the thing that’s worth investing in, as opposed to, “They’re pretty good at a lot of things, and I can’t figure out what they’re not good at.” We just have a higher concentration of talent here than is probably possible in a company, in terms of sheer IQ.

There are a lot of VCs, but very few who can actually help you succeed as a company. Being one of those, I think, is still quite a special position.

So I’m going to start with more about how you manage the firm, and particularly the lessons that you’ve learned over the years, obviously extrapolating lessons as a founder as well, and then how we think about running the firm on a day-to-day basis.

The first question I’ll start off with is from your book, but it’s relevant to this conversation, in part because when you wrote that, in technology businesses, you rarely know everything up front. We’re sitting in this massive AI wave right now, and it’s still incredibly early.

Yeah, but the difference, of course, between a mediocre company and one that’s magical is often the difference between letting people take creative risks and then holding them too tightly accountable.

There are many ways we could take that question, but maybe first start with how you manage a group of GPs. What’s different about managing GPs versus a company, and what’s the same?

It’s pretty different from a company. With a company, there are functions and very specific outputs that you’re driving toward. The people in the company are different, too. We just have a higher concentration of talent here than is probably possible in a company, in terms of sheer IQ.

If you look at Chris Dixon, Martin Casados, Alex Rampel, and so forth, these guys have all run companies. It would just be very hard to have that many people with that high an IQ on an executive staff. If you have somebody like Martin, who is probably the best architect in networking software in the last 20 years, plus a really talented investor and so forth, I’m not really telling him or giving him that much direction.

I’m more helping him understand the process—how the process of the conversation affects the process of investing, how you work your way to the right answer, and how to take the right amount of risk. The biggest mistake we make is getting too wrapped around the axle about some weakness that a company has, as opposed to focusing on what they’re great at and how great they are.

It’s not as though everybody is great at something. You could talk yourself into being great at something, but what you’re really trying to find is whether they’re literally the best in the world at a thing. That’s always the thing that’s worth investing in, as opposed to, “They’re pretty good at a lot of things, and I can’t figure out what they’re not good at.” That’s generally a much worse investment.

It’s about orienting around that, helping think through the platform and the personnel, figuring out how to deal with conflicts, how to close deals, and that kind of thing. So it’s very different. I would say it’s also about understanding when people run out of gas. To be good at investing in technology, you really have to be deep in the tech, and I think it’s very possible that as people get older, they get less into it sometimes. At that point, we’ve got to make a change.

But sticking with the topic of GPs, how do you also think about accountability? One question that has come up is when to promote the right people and when to manage out the right people. Ultimately, the vertical leaders are making decisions, but you’re also making decisions—you and Marc are making decisions at the firm level as well. What’s your thinking and framework, and how has that evolved over the last 16 years?

I think it’s evolved a little, but the main things are kind of the same. I think it’s dangerous in VC to wait for the outputs because they’re so far out—to wait and see whether somebody has a great portfolio after 10 or 15 years before deciding what to do with them.

That’s such a long time, and you can make a lot of bad investments in that time frame. You could also miss out on a lot of good investments if you don’t put that person in a position to do more. So I really try to look at, at the point of attack, how they’re showing up: how good they are at finding opportunities, how good they are at winning those opportunities, and what we think the general quality is at the time of investment.

Some work out and some don’t, but it’s not all magic. You kind of know how great an entrepreneur Meera is or how great an entrepreneur Ilya is. Those are pretty special people, so if you can win that deal, that means something, whether or not those companies work out.

Maybe just to focus in on verticalization: I think verticalization was probably one of the seminal points in the history of the firm and changed its structure. You and Mark have talked about this, and now, in retrospect, it’s clear that it was the right decision at that point in time.

How do you avoid some of the pitfalls associated with verticalization—strengthening communication across verticals and making sure you still have that connectivity as the firm continues to size and scale?

The most important observation—and this was a conversation Mark and I had with Dave Swanson, the late great Dave Swinsson, back in 2009—was what Dave said at the time. He said an investing team shouldn’t be too much bigger than a basketball team. A basketball team has 5 starting players, and the reason for that is that the conversation around the investments really needs to be a conversation.

I always had in mind that any investing team really shouldn’t be much bigger than that size. How do we maintain that? The only way to do it is to verticalize. The other thing that was happening simultaneously in the industry was that software was eating the world, so we had to get bigger in order to address the market. But I didn’t want teams to be bigger than a basketball team, and that led to the vertical structure.

I think, in effect, the most important thing is that those teams should be good. We handle communication across the teams in different ways. If the teams are very close—AI infrastructure and AI applications, for example—we have people from each team going to the other team’s meetings, so there’s good, hardcore connectivity that way.

In addition, we’ve got a kind of management meeting for that group that we do, and then the big thing is the GP off-site. We take everybody away for 2 or 3 days twice a year, with not much of an agenda.

I do think David Haber has this thesis that opportunity lies at the intersection. Everyone not only knows culturally but is also economically incentivized to see everyone win. There isn’t that level of politicking that can be prevalent at other organizations, where it’s very zero-sum and there’s protecting-your-fiefdom-type behavior as well.

Yeah. That’s kind of a cultural idea that we have at the firm. The feedback we get from people who come from other firms is that we have less politics than firms with 10 or 11 people.

It’s just a cultural thing. Either politicking gets rewarded, and then you have everything from coups to infighting all the time, and people don’t like each other, or politicking gets disincentivized. That’s what we have here.

Another point around culture: I’m always in awe of the fact that you always hear the gossip around the firm, even details about minor things, where I’m like, “How do you know that Ben and Marc commented?” You often say that whenever you run something, it’s got to be in the details; that’s the only way to do it.

Maybe I’ll ask the question slightly differently. How do you stay on top of the details? How do you hear all this? How do you find the subtle balance without also micromanaging? I think there’s also the creative process of letting things run their course. How do you balance those 2 things?

If you think about my job, a lot of it is setting the direction and then making decisions when things get into conflict or we’re not sure what to do. If you think about what decision-making is, what makes you good at it, it’s a combination of intelligence and judgment—or judgment, which is a combination of intelligence and knowledge.

So what do you know? And then how smart are you at turning that into the correct judgment? The knowledge in an organization tends to live with the people doing the work, not the managers, I would say. What are the deal partners doing? What are the individual people on the IT team doing? What are the accountants doing? What are the people doing when we go visit LPs?

That’s where the knowledge lives: in talking to people who are at the point of attack, so to speak, or talking to entrepreneurs. So I try to spend enough time in the team meetings, and I just end up knowing a lot about a lot of things. Plus, I’m a founder, so if something gets f’ed up in the firm, somebody calls me. That happens a lot. They’re like, “Okay, I’m going to tell Ben this. He’s not going to like this.”

This is a key thing for leaders: You never want people to think, “Oh, we shouldn’t bother them with that,” because it took me 14 seconds to resolve it. Generally, people aren’t looking for you; they’re just looking for clarity. A lot of what an organization needs is clarity, not correctness. If you have clarity, you can move.

Yep. Last question on verticals. We have 7 verticals today. One topic that has come up is: How do we know these are the right verticals? And maybe give some examples of vertical ideas that you’ve resisted starting—ideas that sound good on this premise but either don’t have the right technological legs or even the entrepreneurial capability around them, and that we’ve decided not to pursue.

They’re really designed around the market and where the entrepreneurs are. We try to match up to that: If there’s a big cluster of important entrepreneurs who are going to create multibillion-dollar companies, do we have a team that’s going to win those deals? Different categories end up having fairly different needs. The needs of a crypto entrepreneur, a bio entrepreneur, and an American Dynamism entrepreneur are very, very different. You have to have a product that matches that market.

In picking markets, you want to not be too early and not be too late, right? So it’s a little bit of an art. I think what we’ve seen is that I’m very confident those are the right markets because there’s lots of very interesting activity in all of them. Now we’ve got to perform in each market. It’s not a given that just because we show up and we're Andreas and Horowits, we’re going to win that market. We have to evolve the team, evolve our thinking, and make sure that we win. But I think the markets we’re in are pretty clearly very good markets.

We’ve had a few markets pop up that people have proposed and that we haven’t pursued. I don’t think we got that serious with them, but the other one was ESG kinds of things—cleantech, green energy, and this and that. We thought the right lens on that was much more American Dynamism, because it wasn’t weirdly constraining and was much more oriented around the economic outcome, as opposed to “do good by doing well,” or whatever the phrase is. Those things can lead you into very weird decision-making.

Investing is hard enough without introducing criteria other than whether this thing is going to be a giant company and make a lot of money. You want to have a focus on that. I think the beauty of American Dynamism would be that it’s maybe good for America or whatever, but there are so many opportunities in that space. The U.S. really does have to modernize the way it does defense. We really do have to get much better intelligence in public safety to keep everybody safer. We’ve got to solve the energy problem, and we’ve got to solve the rare-earth-mineral mining problem.

There are very good problems to go dig into. If you look at something like, “Okay, we’re going to come up with an alternative energy source or an alternative fuel,” will any of those work? Maybe. So we never did ESG.

Yeah, I remember, actually, as you were going through that. Very early on in our discussion around American Dynamism, I remember you pushing the team and asking them, “Hey, is this a marketing message, or is this real technological, transformative change?” They went and did the work around it, and it was very clear, now in retrospect, that there was real tech change happening, especially on the supply-chain side and the defense side, and in how people actually engaged with the government.

Oftentimes, when people ask us this question, it’s both a combination of whether there’s a real technology change, because that’s when you make and generate venture returns, and whether the entrepreneurial talent is actually there to build it. Ben Horowitz

Yeah, ADA is a good marketing idea. When they presented it internally, they presented the marketing idea, and I was like, “Well, I want to know what the fund idea is. How do I make money? We have investors. We’ve got to make money.”

It’s a great marketing story, but we’re not doing all that. The fund is going to be less than the marketing in terms of its focus. It’s going to have a tighter focus.

Then we ultimately zoned in on 3 core vertical areas where there was actually a tech change happening. Maybe switching gears: Mark and I believe the best thing society can do for a person is give them a shot—a shot at life, a chance to contribute, a chance to do something larger than themselves, and a chance to make the world a better place. That's the best they can do.

Can you elaborate on this and how it’s driving how we are evolving as a firm, particularly looking ahead as people think about a set of funds that are deployed over the next 2 to 3 years but ultimately have an impact over the next 10 to 15? How do you think about that as you think about leading the firm?

I think it’s important to put the work that people do into context, and we’re in a super-special position. What I wanted to get at was, if you take a step all the way back and say, “What’s been good for humanity?” historically, what’s been good for humanity is when people have a chance to do something larger than themselves, contribute and look.

There are many systems and ideas like, “What if we could make utopia? What if everybody could be equal?” This and that, and you know, that’s ended up doing the opposite. If you look at the history of communism or what have you, it’s kind of everybody having an equal chance of getting no shot. That’s much more what occurs.

You really want to enable contribution. The rise of America coincides with the rise of a free-market, capitalistic, rule-of-law system. If you look at the history of the country and the history of humanity, the rise in wealth, lifespan, and the population size of the Earth all grew spectacularly in the last 250 years. America has been very important in that.

America today is still, I think, very clearly the country and the system where people are most likely to have a shot—a real shot—at life. We’ve done some things to screw that up and so forth, but that’s certainly still the case. For America to maintain its importance in the world, it has to win economically. It has to win, which means it has to win technologically. It has to win militarily, which means it has to win technologically. Our job is to help the country win technologically.

It’s not only important for us; it’s important for the country, and it’s important for humanity. For our people, it really helps them go, “Okay, these things matter. Creating these opportunities matters.” To give you an idea of some of the things it leads to, Jen and I were just in Mexico. A lot of that was catalyzed by a junior person on the team saying, “What we’re doing is so important, and we need to help with this alliance. We need to help secure the border. We need to help with our own defense manufacturing. We’ve got to help with energy. I’m going to get this meeting.”

Then we got the meeting. If you want to change the world, you have to believe you can change the world. That’s a lot of what it was about.

It seems like Little Tech M&A is opening back up. What’s your view on whether that is here to stay and whether that might actually expand to larger opportunities as well?

AI is such a disruptive phenomenon that every company, every incumbent, is under threat from AI in general. A lot of the ways that you deal with the threat is that you just acquire the DNA of the future.

And so I think there's going to be a lot of M&A because I think that people need to reconstruct how they work if they're going to survive. So if you go back, say, 3 or 4 years, I think people believed that the big foundation models would be these giant brains that could do anything better than anybody. It has not played out quite like that.

The way it's played out is that the big models do provide a very important infrastructure that all of our companies end up building on to some extent. But often, for any particular use case, the long tail of not only scenarios, but the long tail—or the fat tail, I should say—of human behavior ends up itself being something that you have to model and understand very, very well.

If you look at Cursor, Cursor consists, as I think of it, of 13 different AI models, all of which model different aspects of how you program, how you speak to a programmer, et cetera. Those models end up being very important—so important that they in fact released their own foundation model specifically for programming and coding. So they have a coding model that you can swap in place of Anthropic or OpenAI if you want, or you can use the OpenAI or Anthropic models with their other set. That has gotten great adoption.

It's kind of going, well, maybe the application behavior is actually, in some ways, more important than having the biggest model trained with the most GPUs. It's not clear exactly how that plays out, but currently I would say that the complexity of the application itself is very high and is not subsumed in the foundation model. I think these things are not as straightforward as they appear, and the benchmarks can be misleading.

I think this is also showing up in every aspect of AI. We've seen a great post that Justine Moore from our team did on how there's no god vid level video model, which I would encourage you to read. It gets into how different use cases end up needing different models, which is, again, not what we thought 4 years ago.

For sure. Yeah, that actually goes back to what I started this webinar with: your quote from The Hard Thing About Hard Things, “In technology businesses, you rarely know everything up front.” So as folks are figuring out everything that's happening in AI, the calibration of benchmarking is changing, and also the expectations around the utility of those things are changing. But also, the founders that are building them are changing.

It is a very fun time, to say the least, but also one in which we're learning in real time. That can sometimes be deeply unsettling, I think, for folks who may be anxious about where valuations are, and the market and the environment as a part of that.

Okay, I am going to switch gears to the future of VC investing, which is very related. Some believe leaner, more efficient businesses will allow founders to retain more of the cap table. Do you worry about that at all when it comes to ownership and what you've been working with GPs on in terms of expectations of ownership in this new environment?

I think that what we've seen is we're getting pretty good ownership. If you look at a lot of the recent investments, we're at 20% or better. There are ones where we don't get to that level of ownership, but those companies get so valuable so fast that it's been fine. There have always been special companies with very, very special founders at a moment in time where, okay, that is what it is.

But for us, for a lot of the core infrastructure things, the core applications, and so forth, the ownership has been pretty reasonable.

A question around just the VC landscape today: when you and Mark started the firm 16 years ago, there weren't nearly as many VC firms. There are now 3,000-plus VC firms running around. There is still this rate limit of great entrepreneurial talent, but increasingly there are more sources of capital flooding in, whether that be on the retail side or elsewhere.

How do you see the power dynamics between LPs, GPs, and founders evolving as you think about the future of the firm? Is there anything you're doing at the firm to prepare for that?

Well, I think it's still very hard to build a company. If you're just an engineer, an AI researcher, and you've invented something and you're jumping into the world, it's a very competitive world. Having a financial partner that can help you build the company—does the initial valuation matter more than the partner?

I think most entrepreneurs who are smart realize it's a partner. There aren't that many good ones. There are a lot of VCs, but very few who can actually help you succeed as a company. Being one of those, I think, is still quite a special position.

The one area where we're upping our emphasis is this kind of entrepreneur who's just starting something and doesn't quite yet qualify for VC money, with our Speedrun accelerator. One of the reasons why we're emphasizing that so much now is that, with the new tools, it is possible to convert an idea into a product much more easily than it has been in the past.

We just want to make sure that we keep a very, very close eye on that. The brand is translating into that accelerator. We're just getting a lot of talent in there, and so that phenomenon is something we want to make sure we're on top of.

Building on the earlier question about being multiple winners in AI, why did prior cycles of technology not play out the same way? Why were there only Google, Amazon, and a small number of concentrated winners in those product cycles, compared to what you think about in what's happening in AI?

We don't know quite what's going to happen yet. But I would say AI is a new computing platform, so you have to look at how many winners there were who built applications on computers. That's the order of the size of what this is.

In the internet era, if you say, “What are all the businesses that got built on the internet?” it was actually a reasonably large number of businesses—from Meta to Netflix to Amazon to Google and so forth. Those were very, very spectacularly huge winners.

I think in AI, the products are having an even bigger economic impact. I think there are going to be more companies that are worth over a billion dollars and over $10 billion than in the last era, from what we've seen so far. It's a very big design space. It's an enormous design space, like one we've never seen before in technology.

Nobody's asked me this yet, but I've gotten a lot of questions about the AI bubble. I think one of the reasons why people are so worried about it being a bubble is that the valuations have gone up so fast. But if you look at what's going on underneath in terms of customer adoption, revenue growth rates, and so forth, we've never seen demand like this.

We've never seen valuations rise like this, but we've never seen demand rise like this either. We are in a bit of a brave new world, at least from anything that I've seen in my career. We'll see how it plays out.

Even the NVIDIA multiples aren't outrageous, particularly when you look at the growth rate and the size of the earnings and so forth. They're not historically nuts, to the point where people would be claiming a bubble like that. People think, “Is that growth fake? Is it this?” From what we're seeing, no—the demand is very intense.

We'll see how it plays out, but I think this is a bigger technology market than I've ever seen. We'll see how many companies actually win it.

I'm going to ask you a few fire-round, lightning-round questions, Ben, and then we'll close out here. What will be your most-played song on Spotify rap this year?

I think it's that Young Thug song, Do you know how it feel to see your face on the news? called you a boohoo. Yeah, it's that one really touches me, like particularly crunch.

I love it. I will say I think “Follow the Leader” is going to be one of my most played.

Yeah, that's a great song.

Wrote the piece and also saw Rock in person early this year at your event, and then also—

George Clinton. It's been fun to dig back into the database of songs that I've never really listened to.

Yeah, no, that's an all-time great song. Indeed. What's one AI tool that you use every day?

For sure, Grock and Chat GPT I use every day. I've also been playing with Vio and Nana Banana a lot on a daily basis. Those are the big guys.

Okay. And because we asked Marc, we have to ask you: do you plan to be cryogenically frozen?

No.

Same answer as Mark. Do you plan to go to Mars? Same answer as Marc as well, despite your background. All right.

Look, I'm trying to stay healthy so that we don't have any generational transfer, but I don't necessarily believe in living forever. I don't. It's my destiny.