Sequoia合伙人David Cahn谈AI、国防领域的赢家，以及全新的0至1亿美元打法

I do think we're in an AI bubble. You can see the fragility. Everybody can see the fragility. The thing that I think is more interesting is who's going to survive the bubble.

Consumers of compute benefit from a bubble because if we overproduce compute, prices go down, your COGS goes down, and your gross margin goes up. The lesson that punches you in the stomach in venture is you can't make a company succeed.

How would you respond to the idea that Sequoia was asleep at the wheel when it came to defense, not being in Helsing and Anduril, the two clear market leaders in the category?

I would say, "Ready to go."

David, I love your writing. Our episode last year was one of the most downloaded shows. I had the CMO of Meta tell me that it is the single show that he has forwarded to more people and cited more often than any other.

Not to make you nervous or set the pressure for this episode, but thank you so much for joining me again, dude.

Thanks for having me, Harry. You're always very kind.

Now, the year of the data center sounds wonderful. We had an amazing discussion last year. What did you predict last year, David, that happened and that we're seeing in action now?

I think there's really—so, we talked about this concept of steel, servers, and power last year. If you remember, rewind to the summer of 2024: the big conversation at that time was compute, models, and data. That's what everybody was talking about.

I had this view that everyone was underestimating the physicality of these data centers. I'm on the front lines; I'm talking to people every day. You talk to people, and they're flying electricians to Texas, trying to buy out generator capacity. Generators are sold out until 2030. How do you get in line, and how do you do that?

I had this sense that people were thinking very abstractly—in a bits perspective—about AI, but they should be thinking in an atoms perspective about AI. I think that prediction came true in 2 ways.

The first way is that the best trade of 2025 was the AI power trade. A lot of Wall Street people made a lot of money betting on the fact that power was going to be the constraint. You hear Sam Altman now talking about gigawatts every day. He's not talking about dollars anymore, right? We're moving away from dollars and toward gigawatts. I think that transition has fully happened in the last year.

The second way I think it was right is that, funny enough, now, a year and a half later, you see this on the cover of The Economist, on the cover of The Wall Street Journal, and on the cover of The Atlantic. The mainstream media has now really picked up on this narrative that the physicality of AI is what translates to GDP.

GDP is an imperfect metric, and it generally captures physical things more than virtual things. GDP is now picking up all of this construction boom that's happening, all this steel that's getting created, and all of the physical stuff that's happening in AI data centers.

You're seeing these stories, which I think are true: AI is now one of the biggest contributors to GDP growth in the United States. I think that's the second way in which that prediction has played out.

Does its contribution to GDP growth run counter to your $600 billion question in terms of where the revenue will come from? You know me—I just go rogue and off-grid, but it's much more fun.

Well, the $600 billion question—and maybe just to remind folks what that is—is basically a very simple equation. If we invest—and this was in 2024, when I wrote this—$150 billion in Nvidia chips, that's about $300 billion of data center investments. To pay that back, the person using the compute needs to earn a 50% gross margin. So there's about $600 billion of revenue that needs to get generated.

If you redo that analysis in the summer of 2025, it's about $840 billion. So it's grown, but it hasn't grown dramatically. The question behind the question was: Is the customer's customer healthy?

We know that the customer is healthy. We know that people are buying all these data centers. We know that people are building these data centers. We know that those stocks have all gone up. We can see that. But is the customer's customer healthy? Is there actually an end user for this compute? I don't think that's been answered.

The question last year, which was the valid question, was: If everyone's spending all this money, why hasn't it shown up yet? People haven't put the shovel in the ground yet. I literally wrote a piece last summer called "AI is shovel ready." The shovel is going to start hitting the ground.

Now the shovel is hitting the ground. We're mid-construction on a lot of these projects. One of the predictions I made last year, in addition to saying it was going to be the year of the data center in 2025, was that we were going to have construction delays and issues building out these data centers.

The Information has done a very good job of reporting on this, but I think we're at the beginning now of seeing some of that play out as well.

Are we going to see a mass proliferation of delays in data center construction, do you think?

I think we're going to see variability. One thing I'm always interested in as an investor is that there are winners and losers, and there's variability. I'm very skeptical whenever anyone tells me that everybody is going to win, everybody is going to lose, or everyone is going to do anything. There's always variability.

Imagine a race. You have a track race. There's somebody in the front and somebody behind, and someone's faster than the other person. I think with data center construction, one of my core perspectives that I've been developing over the last 18 months of writing about this is that construction itself is going to be a moat.

The ability to build things is hard, and I think we underestimate that. I think we continue to underestimate that because we say, "Oh, well, it's fine. Everyone's going to do it. The timeline is 2 years." But there's a lot of complexity that goes into that.

By the way, the complexity compounds when everybody is doing the exact same thing at the exact same time and everyone is trying to buy from the same vendors. I've written a lot about the AI supply chain for that reason, because you really need to care about not only the fact that Meta and Google are both building a data center, but also: Who's the person they're calling, and who's the person that he's calling?

You've got to follow it all the way down the supply chain to get to the core of what's really going on.

There are so many things I want to unpack within those. I do want to go to what you did not predict or foresee that did play out and surprised you.

I think there were 2 big misses last year. The first big miss was these big talent acquisitions. If you had asked me the probability a year ago that, if you're a 25-year-old recent grad from an elite university who is perceived to be an AI expert, you could get a $50 million pay package right now—and if you're a brand name that everyone recognizes, you could get a $1 billion pay package right now for a single individual—I totally did not see that coming.

If you had asked me a year ago to predict that, I would have said you were crazy. Sometimes I do think the beauty of AI is that reality is stranger than fiction, and a lot of crazy things happen.

Before we move to the second, do you think those pay packages are justified?

I think they're symbolic of this desperation in the ecosystem. We need to eke out progress. We need to prove that all these investments are worth it.

There's this logic that gets really abused in the venture world and in the tech world: "If I increase the probability of making $1 trillion by 1%, that's worth a ton of money. That's worth $10 billion." Sure, that's true, but it's very easy to overestimate what the 1% is. Is it 1%? Is it a hundredth of 1%? Is it a thousandth of 1%?

Our brains are very bad at reasoning at that scale of number. To the extent that you believe hiring this very impressive researcher increases the probability that you win by 1%, I can totally see why you would justify a $1 billion pay package for an individual.

That said, I think we are psychologically biased to overestimate what that percentage contribution is. It may be the case that there are broader macro variables, which I'm sure we'll talk about later in this discussion, that are actually driving progress in AI and that no single individual can change.

I'm very upset looking at these pay packages that my mother didn't push me toward a more engineering-heavy design. Does everyone feel that way? I think that's probably the universal reaction to seeing these packages, man.

I'm like, "Mom, you should have done better." Bad parenting. You encouraged me to do English. Really? War and Peace doesn't quite make it, does it, when you're getting paid three and a half billion by Zuck?

What was the second?

I think the second one is that, you know, one thing we talked about on the podcast last year: I predicted that Meta was going to do really well, and I think that prediction was clearly false in a 12-month time horizon.

I thought that the vertical integration Meta had was going to be an advantage. I think that Meta's $100 million packages are coming in large part because they haven't performed as well as they thought they were going to.

The reason I thought Meta would do well is that it was vertically integrated and founder-run. I continue to believe that, in the fullness of time, it is possible—and I think the dramatic actions that Zuck is taking represent this—that I will be proven right over a longer time horizon, which is to say that Zuck is going to fix the problem.

It’s amazing what founders can do. He’s so focused on this. He’s spending all of his time on it. But I think if you look back a year ago at the prediction that Meta would do well, I think you would say, “Wrong.”

Have you changed from a buy to a sell on Meta?

I think the dramatic action that Zuck’s taking represents just how deeply invested in this he is. I think it also shows us what founder CEOs can do and why founder CEOs are different from non-founder CEOs. There are all these studies that say if you just invest in the basket of founder CEOs, you will outperform the basket of non-founder CEOs, and I think what Zuck is doing represents that. So I remain optimistic about Meta long term.

You said that vertical integration was part of your thesis. I totally agree with you, and I was probably shaped by hearing you, to be quite honest, David. You said to me, “Data center and model teams need to be coupled,” kind of going to the vertical integration element. Do you stand by that? How do you think about that when hearing that today? And does the fact that OpenAI and Anthropic don’t have that vertical integration challenge that thesis?

Well, I think the simple version would be that OpenAI and Anthropic are now steel, server, and power companies. That’s a big change that’s happened in the last 12 months. So I actually think that, in many ways, OpenAI and Anthropic are becoming more and more vertically integrated every day.

You’re seeing a lot of announcements around them developing their own chips. Every day you hear Sam Altman talking about gigawatts of power and procuring his own power. So I think you will continue to see the big labs moving vertically down the supply chain. That’s been one of the biggest trends of the last 12 months.

Do you think we will continue to see that? We saw Poolside recently announce a 2-gigawatt data center that they’re building out in conjunction with CoreWeave. Do we think all model providers will need to be vertically integrated in this way?

I think competitive pressures will push all of the model providers to spend more time on this and to have teams focused on this. So I think the answer is yes. I do think that this is a trend that is going to be durable.

When we think about where we are today, everyone says bubble. You’ve heard it. I’ve heard it. It’s on my TikTok. Do you think we’re in an AI bubble?

I do think we’re in an AI bubble. I also think, to your point, a year ago when we had our last conversation, it was a very contrarian thing to believe that we’re in an AI bubble. Today, it’s a very consensus thing to believe we’re in an AI bubble.

Sam Altman, Vinod Khosla, Jeff Bezos—some of the biggest AI bulls—have now come out and basically said, “Hey, we’re in a bubble of some sort or another.” Each has their own perspective on exactly how that’s going to manifest. So I think right now the bubble conversation has reached full consensus.

The thing that I think is more interesting is who’s going to survive the bubble? What’s going to come next? I think there are 2 components to that. Number 1: Who are the winners and who are the losers?

If you remember from the dot-com bubble, a lot of companies from the ’90s still did well. Amazon still became an amazing company after the dot-com bubble. So I think there’s an opportunity for winners to continue to do well after the bubble.

The second thing that’s really interesting is just timelines. I’ve always said my core belief is that in 50 years, when you and I are 80 years old, AI is going to have completely changed the world. It’s going to dramatically reshape everything about society.

If you take that time horizon and you say, “Okay, AI is this tremendous technological innovation. It’s probably the most important thing that’s going to happen in our lifetimes. It’s going to be among the most important things that’s ever happened in human history and in the history of this planet.” Right? So it is this amazing thing, and yet the market is implying some probability that all of this is going to happen in such a short time horizon, with a very specific chipset and all of this stuff.

I think unpacking the tension between AI as a long-term winning trend and a long-term generational change, and a short-term market cycle that will incinerate capital, is the second area that’s really interesting.

How do you balance that as an investor today, David? Play the game on the field—the Bill Gurley quote—but then also have the awareness of the long-term impact that will come over multiple decades?

I think it’s tricky. The one benefit I have is that I’ve been investing in AI for about 8 years. For me, this is not a 12-month thing where you’re running and have this FOMO to get into AI.

I started investing in AI in Weights & Biases’ Series A when everyone said deep learning was going to be tiny. It was a year after the Transformer paper came out, and everyone said deep learning was a tiny market. Why would you invest in this company? Of course, they had a really nice exit to CoreWeave recently.

I invested in Runway ML when Stable Diffusion hadn’t even been born yet. Everyone was saying, “Oh, Transformers is the only way.” Of course, Stable Diffusion introduced a new model architecture.

I invested in Hugging Face, and I still remember the first meeting I ever had with Clem. He had launched this Transformers library. It’s funny now—Transformers is on the tip of everyone’s tongue—but at that time it was NLP, by the way. It wasn’t AI at that time.

He had this amazing Transformer library, and for folks who are steeped in AI, it was a successor to BERT and this old school of NLP models. I just say that to say that when you take a long enough time horizon in AI, over the last 8 years, you have more opportunity to find investment opportunities.

It’s not about finding 10 investment opportunities. At least for me, I don’t need to find 10 investment opportunities this year. I’d like to find 1 or 2 investment opportunities a year that I really love.

This year I’ve invested in Clay, which I think is an amazing application-layer company we can talk about. I invested in Juicebox, which is building an AI recruiter that has tremendous love. I think you can find exceptional AI companies that I believe will do really well over the long time horizon and will continue to succeed for decades and decades to come.

One thing I ask myself before I make every investment is, “Is this company going to succeed in spite of market volatility?” If the only way your company is going to succeed is by raising infinite capital in a cheap capital market, that’s very difficult. If you have real customer love and you’ve built something that people absolutely need, you’re going to be able to navigate through any market environment.

By the way, we’ve seen that now with all of these 2021 companies navigating that environment. Some of them came out really strong on the other side. Look at Databricks: $60 billion, now a $100 billion valuation. So you can come out the other side of market cycles if you have compelling product-market fit, a great team, and a great founder.

So, David, when we play out your question there of the winners and the losers, just so I understand that, who do you think the winners and the losers will be when we look back on this last 12 to 18 months?

I’ve had a very simple framework for this. It’s actually, I think, probably the first thing I ever published in AI: “AI’s $200 million question,” way back in 2023. The framework is this: Consumers of compute benefit from a bubble because if we overproduce compute, prices go down, your COGS goes down, and your gross margin goes up.

So I’ve had the view that you want to invest in consumers of compute. Producers of compute—imagine you’re producing any commodity asset. If other people produce a lot of that commodity asset, it doesn’t matter. It has nothing to do with you.

You might be running the best operation possible. You might be an amazing businessperson. But if everybody else starts producing the same commodity asset, prices go down. It’s very hard to control your destiny in commodity businesses.

By the way, this is why commodity businesses tend to trade cyclically and tend to trade at lower multiples than non-commodity businesses. So I think if you’re a producer of compute, you’re fundamentally in a commodity business, just like an oil company is in a commodity business. That’s going to trade a different way, and it’s going to have more cyclicality than if you’re in a non-commodity business consuming the commodity, consuming the energy, and producing intelligence on top of that.

I think if you’re consuming this raw resource, which is power, and you’re producing intelligence and doing something that people love with that intelligence, those are the businesses that are going to do well on the other side of this market cycle.

Are 3 of the best businesses not commodity businesses in the form of Google Cloud, AWS, and Azure?

I love this question. So let’s talk about it. I think it’s really interesting. One thing I’ve written a lot about, and you and I have talked about this, is game theory and these big companies. One of my core beliefs—or one of the things that I think is underestimated in the market—is that we’re living in an anomalous monopoly era.

And it's funny because there are so many comparisons to the Industrial Revolution, and in some ways we're living in this new Gilded Age. We have these 7 companies, and they represent 40% of the S&P 500, which is just mind-blowing. They have these amazing monopolistic businesses, and these businesses are cash cows. I think people extrapolate from that and say, “Oh, all businesses are monopolistic.” I think people have a mental model that implies too much monopoly and not enough commodity.

What I think people underestimate about the big tech companies is that when the big tech companies were founded—when Google was founded—nobody thought it was going to be a monopoly. Think about YouTube selling for $1 billion. That would be crazy if you had known how big all of this was going to be. Nobody knew that Google was going to be monopolistic. You can build monopolies when they're hiding in plain sight. Nobody can see them.

And so you build this monopoly and you don't have that much competition. AWS is the same. You mentioned AWS. Nobody knew that the cloud was going to be this tremendous opportunity when AWS started doing this. To their credit, that's why they have the biggest market share in the cloud business, and that's been very durable for them.

I think when nobody sees the monopoly, you can build a monopoly and then extract margins on the other side. But AI is so different. Everybody knows that AI is going to be big. This is the irony of AI: everybody knows AI is going to be massive. But if everybody knows something's going to be massive, then everybody builds companies. And if everyone builds companies, there's tremendous competition.

I think the difference between the AI era and the big tech era makes sense. Everyone is over-indexing or over-training on the big tech era because that's the era we live in. But the difference is that these monopolies are not hiding in plain sight. We all now know that if you build an amazing tech company, it can be worth $1 trillion.

In 2000, if you told people that they could have a $1 trillion tech company, they would have laughed you out of the room. I think the market environment in which these companies are getting built is dramatically different, and monopoly profits are unlikely to exist.

By the way, a final point on this: that's good for us. That's good for everybody. We shouldn't want monopolies to exist. Monopolies are bad for the consumer. The consumer wants to get things for free, and the consumer wants to get things for the cost of capital. To the extent that there are not monopolies in AI, that's much better for how AI is going to evolve in a healthy way than if it evolved in a monopolistic direction.

You said that consumers of compute will win. I like that. But, respectfully, it feels relatively accepted in venture ecosystems, for sure, in a way that your bets before weren't. Weights & Biases wasn't. Runway wasn't. Hugging Face was kind of a weird community play at one point. What do you think is obvious to you that is not obvious to the rest of the community today?

When I first started saying this 18 months ago, it was definitely not consensus. One thing that's tricky in the business of ideas is that as soon as an idea becomes accepted, it was always obvious. But in the moment where you propose a contrarian idea, everyone criticized it.

I do think it's been interesting to see the change and, by the way, to see that people who had the wrong opinion very quickly changed their opinion such that they weren't actually wrong. The idea game is a tricky one.

The second thing I would say to that is, while people say they believe this—and you and I talked about this on the podcast last year; you probably remember this—everyone says they believe this, and then you look at these PitchBook charts where it's like, “Where are the dollars going?” I think probably 80%+ of the dollars in AI are still going to producers of compute, not consumers of compute.

I do think you're right that it's an accepted narrative, but the producers of compute consume so much more capital than consumers of compute. If you are in a capital deployment strategy and you're trying to deploy as much capital as possible, you have to invest in the producers of compute.

I think that's one of the dangerous things in investing: there's almost an incentive to invest in people who consume more capital because they're calling you every day. The people who don't consume capital don't want to raise capital. Some of the best investments are those companies that don't want to raise capital.

When Sequoia invested in Zoom, they didn't want to raise capital, right? They were profitable. They were doing really well. Those are the businesses that I think, as an investor, you really have to focus your time on.

I spoke to Sonia on your team beforehand, and she gave me a fantastic question. She said, “If this is a game-theoretic bubble, is there a coordinating mechanism for the spending to stop and the bubble to pop?”

I love game theory. My basic framework on AI—and this is actually kind of how I write all these pieces—is that there are 10 players around this big chessboard, and they're extremely powerful. Each of their moves affects the other people's moves. It's recursive, and you sort of have to think first-order, second-order, third-order. How does my move affect other people's moves? These are very sophisticated players doing this.

The simple answer to your question is that it's not coordinated. That's the beauty of the invisible hand. That's the beauty of people's incentives. These are big companies that are acting out these incentives, and so I think until the incentives change, the behavior is not going to change. There is no coordinating mechanism.

I do think that's one of the surprises. It's always the surprising fact of capitalism: everyone wants to believe that everything is coordinated. It's easier for our brains to grok everything being coordinated, but I actually think it's pretty uncoordinated and incentive-driven.

If we think about what you said earlier, it is definitely a bubble, and we're seeing this consensus across the different visionaries in our ecosystem. If it's a bubble, does it pop or does it deflate? How do you expect that to play out?

I'm a student of Nassim Taleb, and I will lean on Nassim Taleb's perspective. He's a hedge fund investor and philosopher, and he's written Fooled by Randomness, Antifragile, and The Black Swan. I think these are books that a lot of folks will be familiar with and really influential books in the investing world.

His philosophy—and he says this in Antifragile—is that it's really hard to know if a building is going to fall down, but you can see when it's wobbly. You can't really predict when the wobbly building falls, but you can notice the fragility. My perspective on AI right now is that you can see the fragility. Everybody can see the fragility.

Can I ask you what specifically makes you say you can see the fragility?

The circular-deals dynamic is probably the main thing. When I think about why this AI bubble narrative went from contrarian a year ago to consensus today, I think the main thing driving the consensus is these circular deals and the big tech company dynamics. Let me unpack that.

A year ago, hyperscalers were holding up the AI ecosystem, and everybody felt very comfortable with that because everyone knew that these were very robust businesses. Microsoft and Amazon specifically were driving the vast majority of the AI capex growth, and they were explicitly saying, “Hey, we're going to buy out your generation capacity for 5 years. We're going to sign a 20-year lease on this data center, and we'll back it up with our credit.”

They were basically putting themselves in front of all the risk. The way I thought about it a year ago, and wrote about it a year ago, is that they're almost grabbing the hot-demand hot potato and saying, “It's ours. Don't worry about it. We've got this covered.”

A year later, Microsoft and Amazon have really stepped back. This started—and The Information has done a really nice job reporting on this—at the beginning of the year. There was this big public announcement, or leak, or whatever you want to call it, where Microsoft walked away from 2 data centers. It sent a message to the market: “Hey, we're not stepping up. We're not going to take all the risk on everybody else's behalf. We're not going to be this risk absorber in the ecosystem anymore.”

Then what happened later this year is that Oracle obviously stepped up and took on a huge amount of the compute demand, and CoreWeave has really stepped up and taken on a huge amount of the compute demand. You have this shift from Microsoft and Amazon to Oracle and CoreWeave.

The second-order effect of that is that Oracle and CoreWeave are a lot smaller than Microsoft and Amazon. They simply can't absorb as much risk as Microsoft and Amazon could. The chip companies are now stepping up and saying, “Okay, we'll absorb some of the risk. We'll put in the capital to finance this build-out, where the demand on the other side is not so clear,” because, of course, the chip companies also get to book this as revenue.

Their cost of capital is very low. One might even say their cost of capital is negative in some of these deals. It's the cheapest capital available. You're moving from expensive capital from these big tech companies to cheaper capital from the chip companies themselves, who get to benefit from circularity.

I think that's probably been the biggest change in AI in the last 12 months. I think that's something a lot of people have observed. It's fairly obvious, and I think that has changed a lot of people's minds.

Do you think these deals are priming the pump, so to speak?

I think all of these deals are now priming the pump. You basically announce the deal, it's 10% or 20% funded, and then you have to go raise capital to fund the rest of it. Everyone announces these deals in gigawatts, not dollars, anymore, and I think most people don't know how many dollars a gigawatt is.

The rough math is that a gigawatt is $40 billion to build out. Jensen says it's $50 billion or $60 billion if you use the next-generation Vera Rubin chip. Let's say it's somewhere between $40 billion and $60 billion. So, 100 gigawatts of power buildout, which is what people are talking about now, would be AI's $8 trillion question. Then 250 gigawatts of power is AI's $20 trillion question.

We've totally upped the ante, and the magnitude is just much, much bigger. But of course, that's not funded, and so I think the funding for these deals is going to be an important thing that has to play out.

How do you read them? When I hear you speak now, I feel very concerned. Is there even enough capital supply in the world for these? We've heard about Sam Altman and the $1 trillion that he needs, requiring the same energy as Japan. You're looking at that and going, “Well, not even the sovereigns have enough money for that, actually.”

We're living through this amazing moment, and I do think it's precarious. The entire capital market is just AI, right? Forty percent of the S&P 500 is these big tech companies, and they're all basically trading on AI. Private capital is all targeted at AI. I do think the world's capital machine is directed in a single direction.

I think the risk is that it's all focused on a very constrained period of time. I actually think that, in the fullness of time, it's not that risky. These things are going to play out. We're going to get these amazing AIs. AI is going to be amazing. We're going to get these huge technological breakthroughs. Tremendous revenue is going to get created, and it's going to be a big driver of the economy.

The problem, and the simple way to think about it, is that it's all B100s and H100s. What if it actually takes 3 years and it's the Rubin chips that get us there, or it's the Feynman chips that get us there, which is the 2028 chip, right?

I think, again, it comes back to where we started, which is the physicality of AI. You can't just say, “Oh, I'm going to upgrade my chip. Great.” It's not my fingers; I've upgraded my chip. No, you have a giant warehouse sitting with these chips, and they might be legacy chips. Maybe it's going to take us 10 years to get there instead of 2 years to get there. I think that is the risk that the financial ecosystem is taking on, whereas, as an AI investor and an AI believer, I'm saying we actually just need to spread that risk over a longer period of time and a greater number of bets.

Oracle is one of the biggest players that we've seen enter the market, as you mentioned. When you look at their debt-to-equity ratio, traditionally considered very, very high, do you not think they're out over their skis?

One narrative that I have been thinking about a lot is this idea that debt is going to unwind the AI bubble, which is to say that a lot of these AI investments are debt-funded. The problem with credit is that credit unwinds, and when you have a credit unwind, a lot of bad things happen. Actually, that's not the way it's going to play out, which may be surprising.

I think the reason people are so anchored to this debt narrative is that 2008 was a debt and credit unwind, and people understand how messy credit unwinds are. What's interesting about this AI buildout is that, for the most part—and let's put Oracle aside, which maybe has some debt—for the most part, the AI buildout today has been equity-funded and cash-funded.

Every bubble looks different, and every unwind looks different. I think we always over-index on the lessons of the past. What I think is going to be interesting is that, if the bubble unwinds at some point, it's going to be an equity unwind.

What that looks like is that 40% of the S&P 500 is basically a bet on AI. To the extent that the bet unwinds, stock prices go down. What's different this time, again, versus 2008 is that more Americans—a greater percentage of Americans' net worth—is in equities than I think ever before in history. People are going to feel this in the form of their equity portfolio going down, more likely than some credit unwind where the banks get affected and all of that stuff.

Are you as concerned as I am by the concentration of value in the Magnificent 7? Again, if I'm pushing you on company specifics, dude, I mean, I really—I’m not a journalist in any way. I have zero desire to get a clickbait answer. But I look at the concentration of value in the Magnificent 7 as a class or cohort, and I am worried.

I was sitting down yesterday with Alasdair Nairn, who's the author of the book The Engines That Move Markets. It's one of the all-time great tech-investing books. We were talking about AI and markets, and he made this comparison to Japan in the '90s, where, basically, if you didn't invest—if your portfolio was not leveraged to Japan in the '90s—then you were the best-performing fund in the '90s.

He said—and this really surprised me—that Japan was basically 43% of the equity market and the U.S. was 41%. It was a really huge percentage of the market, and that really unwound.

I think you have a similar dynamic here, where the Magnificent 7 are just a humongous portion of the market. These companies are great. They have cash machines, and they're going to do fine. But I do think we should be concerned that these companies represent such a huge fraction of the market and that any change in the AI narrative really affects them.

I want to discuss something you mentioned earlier in the conversation. We mentioned the concentration of value in the Magnificent 7, and a lot of that is predicated on the belief that AI will impact GDP meaningfully. We touched on it earlier. Masa said that he thinks we'll see a 5% GDP impact. How do you think about and respond to the magnitude to which AI will impact GDP and productivity levels?

I think Masa makes an interesting point here, and I actually agree with him fundamentally that AI is going to affect 5% of GDP. Probably where I disagree with Masa is that I think he used the n trillion dollars. I think that's the number he used. It's going to disrupt n trillion of GDP. Then he says his next assumption is that there's going to be a 50% profit margin, and then it's going to be $4 trillion of economic profit.

I agree with him that it's going to affect 5% of GDP, maybe more, in the fullness of time. But I think this comes back to the point we were discussing earlier, where people overestimate the monopolistic nature of businesses. We're living in this sort of unique Gilded Age, monopolistic era, and that is not the steady state of business.

I found this McKinsey report recently that said, if you look at total global GDP, 1% of global GDP is economic profit above the cost of capital. I think that's surprising, and I think it again confirms this intuition that I think some people have, which is important: for the most part, GDP accrues to regular people—working people who get wages and salaries.

It is very hard to sustain an economic profit above your cost of capital. Again, to moralize for a second, that's a good thing. I do think that's really good, and I hope that the economic benefits of AI accrue to everybody and not just a few companies.

In terms of overestimations, I was just chatting with Rory O'Driscoll from Scale and Jason Lemkin, who we have on our weekly show. They said that the biggest problem today is that we're seeing this overestimation of demand. They were specifically talking about legal, where every law firm is looking for an AI provider today because they've been told, “Look for an AI provider.” That will not be the case next year and the year after.

It's an atypical market cycle, where 100% of the market is looking for a new provider or a provider, whereas normally it would only have been 5%. Do you think that's a fair description?

I think there are a number of things that are being overestimated. The most important one is the timeline. You've probably seen a lot of commentary in the last few days about the AGI timeline getting pushed out. This is something I've been talking about for the last 4 months, because a lot of the leading indicators were there in June and July, but this did change over the summer.

It makes sense why everyone's talking about this right now. In June or July, Andrej Karpathy at Y Combinator said, “Hey, we're in for the decade of agents,” as opposed to AGI in 2027. A few weeks ago, Richard Sutton was on the Dwarkesh podcast and basically explained why. Dwarkesh, I think, has been doing a good job of fleshing out why the current technology paradigm is not enough, potentially, to get us to AGI.

Sam Altman came out, I think also in June or July, and said, “Hey, it's going to be a more gentle singularity.”

I've actually been surprised by how gradual the change has been, as opposed to being this crazy change.

For me, there's this contrast between what I think of as the lunchroom conversation at these big labs. You have these 25-year-olds sitting around at lunch saying, “AGI is 100 days away.” “No, it's 200 days away.” “No, it's 300 days away.” The highest-status person is the person who says it's 100 days away because they're the most aggressive.

You contrast that against the true thought leaders and godfathers of AI—the people who really invented this category. People like Richard Sutton, Andrej Karpathy, and Ilya Sutskever, who said in December that pretraining is dead. Those people think, “Hey, the timeline's actually 20 years, 30 years,” and so on. I think that contrast is probably the biggest thing that's being underestimated.

The irony is that it's actually the forward-thinking leaders who led us down this path. The path we're on was invented by these people who are raising the most concern or saying the timeline is longest. It's the people who've been in AI the shortest who are saying, “Hey, it's going to come tomorrow.” I think there's an experience curve: these things are just hard, and they take time.

By the way, I want to say this because it's so important: if this happens, it's a cataclysmic event in the history of our species. It doesn't really matter if it happens in 200 days or 50 years. What matters is that it does happen.

I almost feel apologetic because you're so smart and intellectual, and then I'm like, “Yeah, well, venture, baby.”

Kingmaking is a real thing. Making one person the anointed winner with a large amount of capital, distribution, and brand, à la Harvey, is a very real dynamic that we're seeing play out. How do you balance the importance of kingmaking today with the long cycles—the decade-plus timelines that we're talking about there?

I don't believe in kingmaking, and that's maybe a controversial thing to say. You'd think, “Oh, Sequoia should be able to kingmake companies,” and that would be, by the way, really economically valuable for our LPs if it were true. I don't think we believe that's the case.

If anything, some of the hardest-learned lessons in this business are that you think your capital is going to change the business. It doesn't. Fundamentally, the founder has to be amazing, the idea has to be amazing, and product-market fit has to be amazing. Maybe we can help them navigate a few difficult decisions along the way, and we like to think of ourselves as company builders. But the lesson that punches you in the stomach in venture is that you can't make a company succeed. The company has to already be successful.

The second-order effect of that is that you should be humble, because the company succeeded not because of you. The company succeeded because of the founder, and maybe you helped a little bit. You can't make companies succeed as a venture capitalist. Ego gets in the way when people think they can, and I just don't think they can.

So you don't think, in a market like Profound, that Sequoia and the subsequent quick round have helped them significantly get great talent, great customers, and subsequent funding, which then widened the moat between them and the plethora of other people? I'm sorry—I love you, but I respectfully disagree.

I think there are flywheel dynamics in venture, for sure. I'm not saying that having us on your cap table makes your company more successful. I'm not saying that having a great, brand-name VC who's going to work really hard on your cap table doesn't change the probabilities. I just think it changes the probabilities less meaningfully than people think, on average.

You use Profound as an example because I was in the pitch when they came to the investment committee. The business was ripping. It was an amazing business. They had tons of customers lining up at their door to buy the product. We're lucky to be in business with them, and we're grateful to be in business with them. I hope that we can shape their journey in some way.

If there are 5 engineers who join and Sequoia can help them join, phenomenal. I think the number one way that companies benefit from having us on the cap table is talent and recruiting. We can talk more about that; I'm fascinated by recruiting and recruiting dynamics. I do think Sequoia helps with that, especially for folks who are more mimetic, where the brand name really helps.

That said, I resist the idea that, “Oh, I'm going to put $20 million into this business, and now it's the Sequoia company in this space, and suddenly it's going to succeed.” No, it doesn't work that way. We've learned that the hard way, and in our investment committee conversations, we really resist that because I think that's how you make mistakes in venture.

It's so funny. I remember when I interviewed Doug, and he was saying, “People think that because we're Sequoia, everyone just comes and says, ‘Here you are. Here's my deal. You must have it. Take it.’” He's like, “I wish. I would love that. It's not how it works. I have to fight and fight and fight.”

I'm like, “Yeah, your biceps are bulging, Doug. I totally believe that you have to fight for every deal. It's all good.”

You mentioned a couple of companies that you work with. The common critique posed to consumers of compute is margins, margin structure, and unhealthy margins. Do margins matter today at this entry point of AI or not?

I think they matter, and the companies I've invested in typically have reasonably high margins. They're a directional indicator of how much product you've built on top of the foundation models. They are not absolutely important.

I remember investing in a company many years ago that had a 30% gross margin, and now it has a 70% gross margin. Gross margins go up over time. One thing that you viscerally experience as an investor is that plenty of companies that get critiqued for having low gross margins end up being super-healthy businesses in the long run.

Snowflake was one of the big indictments against Snowflake in the early days: that it had a low gross margin. Obviously, it's a very good business. If you have a real product that delivers a lot of value, and there are reasons why, as you get bigger, the cost is going to go down, then you can build a healthy business. In AI, there's such an obvious reason: the cost of compute just keeps coming down every year. The trend line is very clear.

I would even go to the extreme and say that, although I haven't invested in any of these companies, I can imagine how some of these companies with 0% gross margins are going to work. The companies I've invested in typically have higher gross margins than that, and I think that's indicative of the amount of product that they've built.

At the end of the day, our job is to invest in companies that become really successful, not to be super-smart about analyzing them. Sometimes the instinct to criticize a gross margin can get in the way of money-making.

The thing I've learned from Doug, or the thing I admire most about Doug, is that the job is to make money at the end of the day—for LPs, for founders, for everybody. That's the business that we're in, and I try to keep that as the goal at the end of the day.

I have something called WWDD, which is “What Would Doug Do?” In a tough situation, I'm like, “Hmm, WWDD.”

Framework.

Margins is one. Growth rates is another. The companies are just growing so much faster than we've ever seen before. I had him on the show from GC, and he said, “Triple, triple, double, double.” I say, “Go—come back when you've got something better.”

Bryan Kim said recently it causes a lot of FOMO. If it's $2 million in ARR in 10 days, come on. How do you feel about this growth-rate-on-steroids requirement from VCs, and how do you feel—is triple, triple, double, double dead?

I think of it as the $0-to-$100 million club. It's a variation on this: the best AI companies right now are going from $0 to $100 million of revenue very quickly. I don't think you have to be at $100 million in revenue, to be clear, but as an investor, you want to believe the company is going to be one of those companies.

Companies that are on or have crossed that trajectory include Harvey, OpenEvidence, Clay, and Juicebox. These companies are growing really, really fast.

The reason why it's important is because, to your point, there's so much demand for AI right now. For the best companies, it's the best indicator we have that they've built something really useful. We've talked about this a number of times in our partner meetings at Sequoia.

You look back at the internet: there weren't that many people on the internet, so these companies could only grow so fast. Right now, everybody's on the internet, and everybody wants to buy AI. If you have something really good, it's going to get adopted really fast.

And so I do think, to the point of playing the game on the ground and adapting to what you see in the market, the biggest thing that we've seen in the market is that the companies growing from $0 to $100 million are the companies that have smashing product-market fit. I'm happy to invest in a company with $2 million that is smashing product-market fit. But I would tell you that the companies with smashing product-market fit are growing faster right now.

And, by the way, they don't always have to grow faster. The goal is to invest in something that, in 20 years, is this amazing public company with billions of dollars of revenue, and that is still the first-order thing. But I think you don't fight the tape. You can't ignore the traction on the ground.

I always say I don't care how long you take to get to $1 million in revenue, but I care desperately about how long it takes for you to go from $1 million to $50 million.

Yeah. Yeah.

There's a lot of data that indicates that that is a very good leading indicator, for what it's worth. The data I've looked at suggests that that is a historically good algorithm.

One of yours is UiPath, and he's a dear friend of mine, Daniel. It took 9 years to get to $550,000 of ARR.

I wish I'd invested in him in the first few years. I got to work on the investment when it was later stage, but obviously it's an amazing story, and I think it's one that should inspire people.

One thing I try to talk about with founders is that I want to inspire founders to understand that it can take a long time, because Silicon Valley sometimes has such a short-term time horizon. I look at Juicebox. This company started 3 years ago. The CEO was 22—sorry, he's now 25. He had finished Harvard in 3 years. The CTO dropped out of Dartmouth; he was 19.

They took 3 years. They were always focused on recruiting. They had an initial music app in college, and they evolved that into the recruiting market. They spent 3 years figuring out what the product should be. Now, of course, it's growing really fast, and they're really good founders.

One thing I've learned—and this incentivizes me to invest in companies like this—is that people like David and Aamod, the Juicebox founders, who've been through the founder journey and the pain, understand how hard product-market fit is. I think in the fullness of time, they are better founders for it.

That scar tissue, even though it's really painful, does pay dividends long term. And I think for founders who are listening, who are in year 1 and things are hard, that's painful, and there's nothing that I can say that's going to make that less painful. But we would love to invest in you as you figure it out, and we're super patient.

There's this false narrative, I think, that all the good companies raise the seed, then raise the A, then raise the B, and it's all in 12 months. That's not really how most of these companies work.

Clay spent many years in the wilderness figuring out what their product was going to be. It's funny we've been talking about Juicebox and Clay. Sequoia invested at the Series A, I think, in 2019. The company spent 3 or 4 years in the wilderness really figuring it out.

I look at Kareem, and I think the man is enlightened from this experience. It's a super painful experience. Varun ended up joining as a later co-founder. It's an amazing combination.

The company completely changed from the Series A, and then I led Sequoia's investment at a little north of $1 billion, which we are doubling down on in the growth stage of the company. Obviously, the company has continued to rip and has done really well.

So I think the default narrative of, “Oh, I'm going to start the company and then 12 months later I'm going to be successful”—at least in the case of 2 of the investments that I'm most excited about—that was definitely not what happened.

The reason you come on the show is because I stalked the shit out of you. I spoke to Varun from Clay and David from Juicebox before. I told you, I didn't have 1 person not respond to my calls or messages about you, which is very, very rare, dude. That's testament to you.

You mentioned there that, for founders who are in a hard period, we don't want to present this false picture of it being easy. Completely true. But we are seeing these very quick, successive rounds. If we look at, say, Rillet or Profound, do you worry about them?

I remember Pat Grady once saying to me that his biggest challenge is that when he does a deal, everyone else wants to put in money at double or triple the price, and that really stuck with me. Do you worry about these very quick, successive rounds?

I think we try to find the right balance, and, to be honest, this is a conversation I have with a lot of founders. This is a very active conversation. We're all having these conversations all the time.

We're obviously in a market where capital is very abundant and very available, so I see the argument for why people want to take the capital. I think one lesson we've learned is that more capital does not make a company more successful. Capital is fuel, but capital does not create the engine.

I think this is a tension. I think this will always be a tension, and I think this is definitely a tension for companies right now. We learned this the hard way in 2021: getting overcapitalized has downsides.

The biggest downside, in my opinion, is that it leads to this sort of internal perception of, “We're winners. We're so successful. We're so great.” The only thing that makes you a winner is having tremendous product-market fit and having customers who love you. So I think that's a tension.

Some founders—and I've seen some founders do a great job of this that I've worked with—really act like the money is not in the bank account, and they behave diligently. The team size doesn't grow too fast and all of this stuff. But I think that is the exception, not the rule.

I think it's not the founders that you should be most worried about; it's the engineer who joins the company the day after the billion-dollar fundraise with very little revenue. That dynamic is tricky.

I admire the founders navigating it. I don't think there's an easy answer. I wish there was. I don't think there's an easy yes-or-no answer, but I think it's a tension we should be talking about. As company builders, it's something that we really need to think about.

Speaking of Pat quite a lot, poor guy—it's like an advert for Pat. He taught me something that was really interesting, which was 2 questions that are a framework for amazing insight from founders. He said, number 1 is: What does everyone think they know that they actually get wrong?

If we apply that to AI and what we see today, what does everyone think they know that they're actually getting wrong? I guess I would say—and this is a really hard lesson, and it's something I've learned from a lot of my mentors in this industry, because one of the things I really try to do is learn from people who've been doing this for longer, who are smarter, who are more thoughtful—that one lesson I've learned in this business is that anything multiplied by 0 is 0.

I think one of the tricky things in investing is just to say that market volatility doesn't matter in the long run if you have a great business. But if you overextend yourself and then some crash happens and you go bankrupt, you're bankrupt. There's no way out of that.

I heard this phrase recently: “Momentum has its own reality.” I think there's this sense of everyone living in this reality-distortion field of momentum. I think of it almost like a slingshot. You pull the slingshot back, and then you release it, and it sort of has its own momentum after that.

That's sort of a fundamental law of physics: things in motion stay in motion, and things at rest stay at rest. So I think the thing that people are so confident in is this reality-distortion field that comes from momentum. When that reality-distortion field goes away, you need to survive that.

One thing that I hope I can be to my founders is a partner in making sure that we survive those moments, navigate those moments well, and position ourselves well against that. I think the most prudent, or the most sober, investors can actually be really helpful.

Your job as a founder is to be maximally aggressive, and you should do that. The investor should hopefully be giving some advice, helping think these things through, giving some perspective, understanding a broader time-horizon perspective and a broader data set of companies. Then you sort of navigate to the right end destination.

I don't think people are thinking about this concept of anything multiplied by 0 is 0, because the time horizon is so compressed into this shorter period of time. That's just something that I think a lot about.

The final one that Pat taught me, and then we'll move to talent, which I do want to touch on for a quick fire.

The other one, yeah.

A very, very, very good guy. What is no one thinking about that everyone should be thinking about? For me, one thing I think is astonishing is that no one is thinking that if you overfund your engineers with the capital that you're stuffing into these multibillion-dollar companies, you may not get an equivalent level of productivity as when they didn't have multiple billions of dollars. Give a nerd billions of dollars: a nerd buys 5 cars and a boat. A nerd's not so productive. Sorry to be so blunt and direct, but it's the same with companies.

I think companies underestimate 23- and 24-year-olds. I think this is something that people really, really underestimate, and I think this is more true than ever right now in AI. I meet probably 200 or 300 young recent college graduates every year, and the reason I meet them is that I want to recruit them into my companies. A lot of them are founders. This is the population that I learn the most from because I know that my blind spot is going to be that somebody started using ChatGPT when they were 18 and I didn't. They're going to have a different perspective, and that's the perspective I need most in my life.

In any case, I introduce some of these people to companies, and the companies are like, “What's their skill set? Why should I hire them?” I think this is something that people are not thinking enough about in AI right now: ChatGPT has been around for 5 years. Nobody has more than 5 years of experience in AI. The playing field is super level. In a changing and dynamic market environment, dynamism, slope, and the ability to learn are more valuable than ever.

The thing that inspires me and the thing I spend a lot of time thinking about is, in a company like Juicebox, how can we get the very best 23-year-olds in the world working at this company? That's a big part of my job, and I spend a huge amount of time on that. I'm there 1 day a week right now at Juicebox just working on this. How do we get the best people in the world inside of these companies?

Maybe 10 years ago, in the era of software, a senior software engineer or a staff software engineer had more experience than an L3. Architecture is hard, writing code is hard, and they were much better. Maybe it made sense that there was this old playbook for startups: “You hire the staff software engineer who kind of knows what they're doing, and you don't have to train people.”

I think the new playbook for these AI startups is actually going to be much more about hiring the AI generalist: this 23-, 24-, or 25-year-old who's really native in AI and really passionate about it. I think those are the front lines that are going to make great companies.

Totally agree, and I understand that. Do you worry about emotional maturity a little bit? I don't mean that patronizingly, but Jesus, I'm 29 now, and when I was 22 or 23, I did some things that I would not do now.

I think hiring always has trade-offs. One thing I believe more generally speaking, because it's worth saying, is that I really believe in trade-offs. Everybody wants the free-lunch thing. When you don't know the trade you're making, then the negative is hiding from you. There's no such thing as a trade without negatives. There's no such thing as a decision where it's all positive and there are no negatives.

I talk about this a lot at Sequoia. It's hidden risk versus visible risk. When you hire a 23-year-old, there's a very visible risk: they're emotionally immature, they don't have any work experience, and it's very obvious what negatives you're taking. When you hire someone who's more experienced, it's less obvious what risk you're taking. It seems to be lower risk.

Every decision is a risk, right? Maybe the risk you're taking is that they're not going to work as hard. Maybe the risk you're taking is that they're less AI-native. There's always a risk. I think people have this tendency to favor hidden risk. By the way, price is a hidden risk. You don't perceive it as a risk, but it is a risk.

People prefer hidden risk over visible risk, and I prefer visible risk. I want to know exactly what risk I'm taking. By the way, I'm a huge risk-taker. I started investing 8 years ago. I love risk. I think it's important to calibrate that: I love risk-taking, but I want to take visible risks where I know the risk I'm taking.

I think herd behavior and consensus mentality are about hidden risks. The risk is just beneath the surface, and you're not paying attention to it, whereas I want to take risks that I can see. In the hiring dynamic, when you hire a 23-year-old, it's super obvious why you shouldn't hire them. Yet sometimes that's okay because the reason you should hire them makes up for that and more.

Completely agree from the employer side. On the flip side, when you think about advice to them, if you were advising your younger sibling on choosing their first job, I saw on LinkedIn that you said, “Follow the smartest people a year ahead of you.” That moniker of advice may not be relevant anymore. What advice would you give to them?

This is the biggest learning because I've met with 200 or 300 young people a year. I have a very big data set, and I think I've probably spent more time than anybody at Sequoia on this specific thing. My biggest lesson is that the way young people choose their careers is through what I call the mimetic algorithm.

The mimetic algorithm is: what did the people 1 year ahead of me in school who I thought were the best go do? It's a recursive algorithm, right? What did the people a year ahead of me do? Those people chose based on what the people a year ahead of them did, and those people chose based on what the year ahead of them did.

One reaction to that would be negative: “That's so mimetic. They should think for themselves.” I actually don't have that perspective. I'm fine with it. I think it's a reasonably good algorithm.

When I graduated from college, Palantir was the hottest company to go work for. All the really smart people went to work for Palantir. Going to work for Palantir would have been a great life decision at that stage. Before that, in the early 2010s, Google and the big tech companies were the hot places to go work. Those companies were all 10x over the 2010s. Some of them, I think, even 25x. It was a good decision to go work at Google in 2010.

I don't think the mimetic algorithm is inherently broken. I respect it, and I think that, to your point about maturity, people are going to go through a maturity curve. They're not going to use this algorithm when they're in their 30s. They're going to evolve and change, so I have this respect for it.

That said, I do think that recursive algorithms break down in the face of dramatic new data. The dramatic new data is the AI cataclysm. AI has totally changed how the world is going to work, and it should change your forecasts on the future. The recursive algorithm of “What did the guy a year above me and the person above him do?” is actually breaking because those people didn't have this information. They didn't know that AI was going to change the world. They didn't understand GenAI.

The advice I try to give young people is just to factor that into your algorithm. Do you. Join the company that you want to join. Go to the place that's going to make you the happiest, but factor that in.

It's worth at least giving a shout-out to this group of people that I call builders in a Substack post that I did. Builders are people—most people, 90% or more—the question they're asking when they're choosing a job is, “What can I get from this job? What is it going to enable me to do? Who am I going to surround myself with? How am I going to become better?” It's a very “What do I get out of it?” perspective.

I think there's a 10% group of people—maybe it's 5%, maybe it's 1%. I don't know exactly what the percentage is—but there's this group of people asking the question, “What can I contribute?” By the way, if you contribute a lot, you generally get to extract a lot. I think contribution is a beautiful thing about capitalism: when you contribute a lot, you do get rewarded for that.

Those are the people driving Silicon Valley. When you go into a company and you're like, “Why is this company succeeding?” it's those types of people. They're the type of people who go from one great startup to another great startup to another great startup.

That distinction between these 2 groups of people is valid. There's no problem with either of them. You have to respect that a career is a very personal decision. Depending on what you're trying to solve for, you have 2 options: “What's going to grow my career?” Or, “Where can I contribute the most and therefore extract the most?” I think there's a bunch of great opportunities ahead of you. Just factor in the AI variable.

I think one thing that just frustrates me on this topic is the mimeticism that continues despite market changes in the UK. What do I mean by that? Goldman Sachs, investment banking, and consulting are still what people tell you to pursue if you go and speak at universities, which I do once or twice a week now.

Wow.

Everyone still wants to be an investment banker. When you were talking, I was thinking: What does it take to break the mimetic chain? Maybe it's AI and the proliferation of AI in popular culture and the media, but I think it's changing.

I agree with you that it's changing too slowly, and that's why I'm having these conversations. I'm trying to help, and I'm sure you are as well in these talks that you're doing. One positive thing I would say is that I've seen a material change in the last 12 months, which is interesting because I'm not saying the last 24 months or the last 36 months. It took 2 years after ChatGPT for this to really start flowing through.

A lot of the people I'm talking to are currently investment bankers who want to get into AI companies, so it is funny that way. More and more of these high-performing people want to be inside AI companies. That's why I think it's a 2-way match: these companies need these people more than ever, but I think these young people can benefit more than ever from being at an AI company.

To make the value proposition clear for the young person, maybe 10 years ago, if you joined a startup—and people didn't join startups that often 10 years ago—there was this whole experience curve. You're the junior engineer, there are a lot of people who are smarter than you, and you're going to have to learn. It's going to take 5 or 10 years to become a really meaningful contributor.

That's not really true anymore, right? You're entering at much more parity with everybody else. I think there's good reason why people are making this change.

Dude, I'm throwing in a curveball here, but I was told that you're the man who does defense at Sequoia. I say this with love, but I'm going hardball on this one: How would you respond to the claim that Sequoia was asleep at the wheel when it came to defense, not being in Helsing and Anduril, the 2 clear market leaders in the category?

I would say—and I think this ties into our conversation so far—that defense is the next AI. That's how I started getting involved in AI. I think that if the Transformer moment was the starting gun in AI, the ChatGPT moment hasn't happened yet.

There's no way around it: Sequoia was late to defense. But I think Sequoia is working really hard to catch up, and that's part of business. You don't always get things right, but you keep trying. I think we have that ethos and that humility.

Why do you think defense is the next AI? Sorry.

I think that, by the way, I'll share a little bit of how I got interested in defense. You and I know each other now, so before I got into AI, I was reading all this stuff and trying to learn from people. I think my investing style is that you spend 2 years learning about the thing, and then you start investing in the thing. I take my time to build a foundation.

My foundation in defense was reading Napoleon and Churchill, all of the history of war, the history of wars, the history of defense, and geopolitics—really getting educated. I probably spent 2 years educating myself and meeting founders. You learn a lot from founders in this space before you get involved.

The thing that I learned, and I think the thing that a lot of people who are deeper in the space than I am already understand, is that deterrence is the first thing. You only go to war because you have to. The whole point of defense is to prevent wars. Geopolitics is a real thing, and there's real competition between nation-states. That's always been the case, and that will continue.

As the world gets reshaped—and we are living through a reshaping of the world order—I think that's something that a lot of people have seen and written about. There are a lot of variables that we can unpack. I think Ray Dalio's Principles for Dealing with the Changing World Order is a really good book on this topic.

The world order is fundamentally changing, and that leads to this interesting opportunity where we have to catch up. There's 50 years of catch-up that has to happen. That's how I see the current defense moment.

This is why I say we're 2 years after the Transformer paper and we're not even at the ChatGPT moment yet. We're about 1% of the way there in catching up. We're actually so early in this defense cycle because now we have a few dozen companies, maybe 100 companies, that have new innovations. They're not integrated into the force structure meaningfully yet. There's so much more that has to happen.

I think we have our clear market leader now in the United States with Anduril, and I think there are more companies internationally that are going to do really well as well. We've crossed the chasm of this being a thing that matters. We've crossed the chasm of the government knowing this matters. You talk to people in Washington, D.C., and they now understand Palantir and Anduril. They know those businesses.

In terms of the force structure changing, the way that we actually protect ourselves changing, and U.S. deterrence changing, I don't think it has changed that meaningfully. After the ChatGPT moment, what's going to happen is that, pre-ChatGPT, if you were paying attention, you noticed it. After ChatGPT, everyone knew this was important—every American, every single person.

I do think we're going to get to a place in defense where everybody knows that this is really, really important and that we need these companies to succeed.

When you look forward to the world of AI, you've assumed that everyone will be improved with AI, will use it hundreds of times a day, and that it will be a part of everything we do, think, and say in many respects. Taking that view on defense assumes that this continuing conflict increases, not even decreases, from where we are today. That would go against human cycles. There are periods of intense conflict and periods of not.

But suggesting that defense is the next AI, I would suggest that that is the case. How do you feel about that?

I think, by the way, I'll share a little bit of how I got interested in defense. You and I know each other now, so before I got into AI, I was reading all this stuff and trying to learn from people. I think my investing style is that you spend 2 years learning about the thing, and then you start investing in the thing. I take my time to build a foundation.

My foundation in defense was reading Napoleon and Churchill, all of the history of war, the history of wars, the history of defense, and geopolitics—really getting educated. I probably spent 2 years educating myself and meeting founders. You learn a lot from founders in this space before you get involved.

The thing that I learned, and I think the thing that a lot of people who are deeper in the space than I am already understand, is that deterrence is the first thing. You only go to war because you have to. The whole point of defense is to prevent wars. Geopolitics is a real thing, and there's real competition between nation-states. That's always been the case, and that will continue.

As the world gets reshaped—and we are living through a reshaping of the world order—I think that's something that a lot of people have seen and written about. There are a lot of variables that we can unpack. I think Ray Dalio's Principles for Dealing with the Changing World Order is a really good book on this topic.

The world order is fundamentally changing, and that leads to this interesting opportunity where we have to catch up. There's 50 years of catch-up that has to happen. That's how I see the current defense moment.

This is why I say we're 2 years after the Transformer paper and we're not even at the ChatGPT moment yet. We're about 1% of the way there in catching up. We're actually so early in this defense cycle because now we have a few dozen companies, maybe 100 companies, that have new innovations. They're not integrated into the force structure meaningfully yet. There's so much more that has to happen.

I think we have our clear market leader now in the United States with Anduril, and I think there are more companies internationally that are going to do really well as well. We've crossed the chasm of this being a thing that matters. We've crossed the chasm of the government knowing this matters. You talk to people in Washington, D.C., and they now understand Palantir and Anduril. They know those businesses.

In terms of the force structure changing, the way that we actually protect ourselves changing, and U.S. deterrence changing, I don't think it has changed that meaningfully. After the ChatGPT moment, what's going to happen is that, pre-ChatGPT, if you were paying attention, you noticed it. After ChatGPT, everyone knew this was important—every American, every single person.

I do think we're going to get to a place in defense where everybody knows that this is really, really important and that we need these companies to succeed.

Do you not worry about the concentration of buyers in that world? When you compare it to defense, you have every business in the world or every consumer in the world.

What I really don't like about defense is actually what Brian Singerman told me about what makes Anduril so special, which is the complementary skill set of the founding team: whether it be GTM into defense and government, product, or intense operations with their CEO, Brian Schimpf. I just don't like the concentration of buyers, the selling to governments, and the lack of incentive for them. Do you not worry about that?

I definitely think about that. I guess my framework—and this is the thesis that I've been investing behind for the last couple of years—is that there are going to be fewer companies that succeed in defense for this reason. Defense is consolidated for good reason. There's a single customer, and so you need to serve that customer really well.

I think what makes a great defense company is being a national champion. Fundamentally, what makes a great defense company is understanding the customer, being able to serve the customer, and being able to drive what is fundamentally a nationwide transformation that needs to happen.

People talk about digital transformation. This is a digital transformation for the defense space. That's what it is. It's funny that it's a very old phrase, right? But defense actually hasn't gone through it yet.

It reminds me of Wiz, where Wiz really benefited from the rise of cloud.

And you would have said, “What do you mean? Cloud was already a thing in 2017.” But, of course, these things take time. So I think we're finally going through the digital transformation for defense, and I think there's going to be a few concentrated winners in each country.

We'll have venture-funded, equity-funded R&D companies that come out, and they'll get consolidated into the national champions. In my view, Anduril is clearly the national champion in the US, and credit to that team—just a really phenomenal company and phenomenal visionaries.

The other 2 national champions that I've invested in: one is a company called Kela, which we think is going to be a national champion. It's based in Israel. The thesis is that Israel has the best people in the world for this, and they can help defend the United States and Europe. The second company in Europe is a company called Stark, which Sequoia has now invested in over 2 rounds, and we believe can be the European national champion. Both companies have done really well, but they're earlier.

I spoke to Alon at Kela, I think. No, Alon and Hamutal—phenomenal people. Hamutal is the GM for Palantir Israel. Tying into our conversation on talent, they've become a massive talent consolidator in Israel. I think the 2 big talent consolidators right now in Israel are Kela and Decart.

I get in trouble for this, but I don't think defense is a category. A category is something big enough to support an ecosystem with its breadth and depth, and I don't think defense is. I think there's Anduril and maybe 2 to 3 more in the US. And I think there's Kela, Helsing, and Stark, but I don't think it's like SaaS, where there are 30 to 50, or fintech, where there are 20 to 30. Do you agree with me?

I do agree with you. My objective—I've probably invested in a dozen AI companies in my career, and I hope to invest in 20 more—is not to invest in 20 more defense companies throughout my career. I think it's going to be a very small handful of companies. Maybe we'll do 1 every couple of years, but you have to go after the right opportunities. You have to build in the right way, and the winners are going to keep scaling.

I think so many of the dollars going into it today will be lost. I see so many McKinsey consultants who are now VCs being like, “Oh, my cost per kill,” and I'm like, “You have no freaking idea what you're talking about.”

Yeah, we don't think that way. I think we think in terms of defending the country, in terms of having people feel safe, and in terms of deterrence. So I agree with you. I don't love that type of language.

Dude, I want to do a quick-fire round. I say a short statement, and you give me your immediate thoughts. Does that sound okay?

Perfect.

What have you changed your mind on in the last 12 months?

We talked about this a bit last time, but I can close the loop for people. I finally decided to learn how to drive, and I got my driver's license in January, which I think is funny because it's kind of like capitulating right before the trade is in the money. I was waiting for self-driving cars for all these years, and then I finally got a license. Now, of course, self-driving cars are on the streets every day.

Come on. We were aligned on one thing: why did we do that?

I encourage you, Harry, to go out and learn. It was a good experience. Roelof told me that I had to because I was having a baby, and I think that was pretty reasonable—to help my wife and drive around the baby.

Tell me, how has being a father changed you?

You know, a lot of people say this, and it's true: it just focuses your priorities. It's so important. I think it makes you less abstract. You can think about things in the abstract; your child is not abstract. Your child has needs, and they need them right now. So I think there's something, in terms of just bringing you into the present, that's really valuable about that.

What would be your biggest advice to me on partner selection?

So many people told me you had a great, wonderful marriage and that they wished to emulate it, and I was like, “Wow, fuck. Okay. Huh. Good.”

Very, very kind. I mean, I would say, I guess, pick right. My wife is smarter than me and better than me in every way.

If your wife is smarter than you, David, I'm worried about the conversations you have at dinner.

I'm excited for you to meet her. No, I think that, look, one thing that has really shaped me over the last few years, especially after getting married and having a kid, is that people talk about the importance of shared values. Every year, it becomes more clear to me that that is true. I met my wife very young. I didn't understand that fully when we first met, and I'm very grateful for that every day.

Tell me, dude, what's your biggest miss, and what did you not see that you should have seen with the benefit of hindsight?

One big financial miss is Datadog. I worked on this before joining Sequoia, but I remember Datadog was this amazing company. The numbers were incredible; it was profitable. It was one of these businesses where your mouth waters looking at the financials of that business. And I remember we lost, and we lost to Dragoneer.

I never confirmed this with Dragoneer, but the story behind it really stuck with me: Dragoneer had this list of 20 companies, and they only worked on those 20 companies. They had been spending years and years and years pursuing Datadog. It was their number-one priority, and they knew it was their number-one priority for a few years.

This was probably 6 years ago now, but it's a principle that has really shaped how I pursue new investments. If it's one of the top 5 opportunities, that's where I really want to be spending 80% of my time. Then I want to spend the next 20% of my time on the next 15. After that, I'm just really trying to focus my time. That actually shaped who I became as an investor, and I learned a lot from that.

Penultimate one, dude: what 1 technology do you think is wildly undervalued, and why?

I think people are underestimating voice as an interface for AI. We just announced our investment in a company called Sesame, which is an AI voice company—an AI conversation company. I got to work on this with Roelof. The founder is Brendan Iribe, former CEO of Oculus, and Roelof, Marc Andreessen, and Santo Politi, who's the founder of Spark, are on the board. So it's a really good company.

You might have seen their launch a few months ago. They launched this AI voice product that you can talk to. It got 1 million users in a few weeks and 5 million minutes—just tremendous product-market fit. I always had this view that we're not always going to be staring at our phones, and that's not the terminal interface to technology and to AI.

I always had this view, but we'd tried all the AI voice products and they all kind of sucked. They were not good. They were boring to talk to, they didn't remember anything about you, you couldn't interrupt them, and you couldn't really have a dynamic conversation. Your brain just said, “This is a robot.”

When Sesame came out, it was just a radically better experience. Within 10 minutes of seeing this technology, I knew that we were going to invest, and we ultimately did. I think the idea that we're going to be sitting here in 10 years talking to our AI, having a relationship with our AI, is very likely. It's a little bit sci-fi right now, and I think it's going to get less so in the coming years.

Well, listen, Sam Altman has opened the door to erotica, so you never know what's coming. We're not going to end on that, because that would be a weird ending to end on. But the thing I want to end on—I like positivity. I fucking hate the doomsday scenarios. What are you most excited for when you look forward 10 years? What is the thing where you think, “This is what gets me out of bed”?

I think this is a good place to end the conversation, because my answer is AI. It's sort of funny because we've been talking this whole time about the ups and downs of AI, the risks and the challenges, and all the complexity. But at the end of the day, AI is the most important story of our lifetime.

It's going to completely transform the world. It's going to be this event that's sort of a once-in-human-history kind of event. I think it's going to be a really, really epic ride to be on, and I'm excited to be on it with you and with everybody else, because I think it's going to change our lives a lot.

Do you know what's going to happen, David? I'm going to come to the Valley, and if it's okay with you, you're going to take me on a drive.

Okay, great.

We're going to get a photo for Ro of both of us in a car driving. I like it. Beautiful.

Dude, you're a star. Thank you so much for joining me, man.

Thanks for having me, Harry. This was very fun.