Tim Higgins (2:51)

And I'm Tim Higgins. This is Bold Names where you'll hear from the leaders of the name companies featured in the Wall Street Journal today. We ask, will the big bet on artificial intelligence pay off?

Martin Casado (3:11)

Martin?

Tim Higgins (3:11)

Yeah, thank you so much for joining us. Hundreds of billions of dollars are pouring into artificial intelligence right now. Help us understand where that's going. What's where exactly? Is that those dollars flowing? Is it into programs, is it product development, is it the picks and shovels? Where are you seeing that money go?

Martin Casado (3:31)

Well, well certainly there's a mix. The vast majority is going into actual data center capacity. So this is GPUs, this is real estate, this is power, this is H vac systems to cool them. Then of course there is the standard software costs of the teams themselves and everything else. But it's really dominated by the infrastructure.

Christopher Mims (3:53)

And you, your focus is infrastructure. So at Andreessen Horowitz, which obviously is a tech investment firm, since we go to a broad audience, want to make.

Martin Casado (4:03)

Sure we specify that I run the early stage infrastructure fund.

Christopher Mims (4:06)

Yeah, but you got more than a billion dollars to invest. And when you say infrastructure, what do you mean?

Martin Casado (4:12)

Yeah, so it's specifically computer science infrastructure and primarily in the software ecosystem. So we broadly define it as we invest in the stuff used to build apps. You know, we invest in stuff used to build the stuff. And so this is the traditional verticals within infrastructure are compute, network, storage, databases. But now you have a number of others like dev tools, security, you know, AI models, et cetera. So this is computer science infrastructure. Everything from let's say the chips all the way up to the actual apps that a, that a technical person would use. Yeah.

Christopher Mims (4:47)

And for those who are not familiar with, I feel like even people who think they know what venture capital is are probably not familiar with the modern structure of venture capital. What does it mean to be a Series A? Like what series are we talking about?

Martin Casado (4:58)

So, so, so we call ourselves venture investors because these things you read are, are actually pretty fluid. So we invest from first money in which would be called a seed, all the way to, you know, pretty late Series Bs. Normally we invest often, sometimes even pre idea if it's a second time founder that we know, often pre product, sometimes pre market production, sometimes early market traction, and then all the way up to like say, you know, a product is in market and they're doing pretty well, but they have not seen, you know, repeatable growth over six quarters. Like at some point in time it moves to growth investing, which is a separate fund in a separate team. So we tend to evaluate technologies, teams, team, market fit, market trends, things like that.

Tim Higgins (5:45)

I kind of think of it as you're one of these guys who getting excited about the people involved or the wild idea that they might have to change the world and seeing the potential for the future of that.

Martin Casado (5:55)

Right. Without actually having the monetary data to prove it.

Tim Higgins (5:59)

Right.

Martin Casado (6:00)

So like late stated investors, they abstract the entire world as a financial spreadsheet and that tells all the truth about the company. Where, you know, we get very excited about founders, we get excited about certain markets before they show up.

Tim Higgins (6:14)

I want to talk about some of that excitement. I noticed on X earlier this year you posted something that I thought was really insightful. I live in San Francisco, so I'm seeing some of this too. You're talking about what it's like to be in the Bay Area right now. And you, you're kind of talking about how it feels like the late 1990s during the dot com boom and a quote, just Silicon Valley in peak disruptive glory. What do you mean by that? Peak disruptive glory?

Martin Casado (6:39)

So these, these movements tend to be much more than just technical movements, right? So they often are culture movements. And it was the same thing with the early web. You had kind of a lot of kind of these wild websites that weren't really practical but they were super cool and people loved them. Like, do you, I don't know, do you remember Hamster Dance? Oh yeah, yeah, absolutely. A website with hamsters dancing. And like, you know, often companies kind of get reconstructed because you have a whole new set of like the rules are being written because it's a new technology, right? So like when the PC came, it was an entirely new technology wave. So you didn't have like mainframe people from IBM starting these companies. You had, you know, normally a younger cohort or a younger generation. They didn't really know how to build businesses. The buyers were entirely new. And so like, you know, there wasn't a playbook for them. So you kind of remake companies too. You know, a lot of stuff that was almost anathema, you know, five years ago is being recreated again both from the, the company side and the consumer side. And then of course, you know, you've got, you know, capital flooding in and all of the externalities of that you've got entirely new user behaviors that people don't know what to deal with. Right. I mean, like, I mean, what is AI good at? It's good at like creating new things. Like computers haven't been creative, it's great at creating emotional connections with humans. Like computers haven't been good at that. And so it's just kind of this melo of tech and culture and company building and insight. And again, yet again, Silicon Valley is at the epicenter of it. Just like it was for the Internet.

Tim Higgins (8:13)

Absolutely. And I think when you use the dot com boom, I think a lot of people think of the dot com bubble. And so I think you're seeing the good parts of that, the good comparisons of that era. And it sounds like what you're just telling me, all the, the potential, all of that kind of wave of new ideas and how it can be implemented. You know, are you have any confidence that history isn't repeating itself with the bubble part of that boom?

Martin Casado (8:41)

So all markets go in waves, right? I mean, we just saw this with COVID Right. I mean there was a, you know, there was a, I would call it a speculative wave where equity values went super high and then, you know, it dropped dramatically after that. And so that will always happen. And I, I predict this will happen again at some point in time. That said, I don't think that the fundamentals are anything like dot com. There's a number of things that were very different. Right. So the first one was most of the infrastructure was being provided by WorldCom, right? Yeah. A lot of which, a lot of debt.

Tim Higgins (9:22)

Right.

Martin Casado (9:22)

Which had $40 billion in debt.

Tim Higgins (9:24)

Yeah.

Martin Casado (9:25)

We also had 9 11, which happened in 2001. And so, and, and back then there's a lot of users on the Internet. There weren't a lot of people that were paying for it. We hadn't even really figured out a business model. So if you compare that today, kind of every one of those is different.

Tim Higgins (9:44)

Right.

Martin Casado (9:44)

So the companies that are investing in these data centers have hundreds of billions of dollars on the balance sheet. Like, I, I don't know the answer to what I'm about to say. But what do you think? Do you think Meta is spending more money on VR or AI?

Tim Higgins (9:58)

Probably has been VR, but now with the AI and the checks they've been writing, I mean, Zucks has talked about spending maybe $600 billion by 2028.

Martin Casado (10:07)

Yeah, yeah, yeah, it is. It's a ton of money. Right. But like these companies spend lots of money on infrastructure historically. And so like maybe they're inflating it, but these companies have, you know, great balance sheets, great cash flow. And so, like, the fundamentals of, like, who's funding this is, is quite different.

Tim Higgins (10:24)

I want to get in the fundamentals in a second. You know the thing, though, put your investor hat on, because I think there's some confusing signals out there, right? Because on one hand, Mark Zuckerberg, CEO of Meta OpenAI CEO Sam Altman, they have talked about the excitement, but they've also had kind of their own version of suggesting some of the excitement might be premature. Here's what Mark Zuckerberg had to say on the Access podcast. I do think that there's definitely a.

Martin Casado (10:52)

Possibility, at least empirically, based on past large infrastructure buildouts and how they led to bubbles, that something like that would happen here.

Tim Higgins (11:02)

And so on one hand you had some warnings. On the other side, you have a lot of money going out there. What should people make of these conflicting signals and noise sorts of situations?

Martin Casado (11:15)

Know, listen, I, I think it's very important to be sober about, you know, how, how long it takes for technology to be adopted. And I think everybody, especially leaders of companies, try and temper it because otherwise, you know, expectations will outpace actual adoption. But I would say that articulating expectations for humans is very different than actual operational planning, where you've got to be, you know, three to five years ahead because it takes so long to build these out. And so I, I wouldn't conflate a CEO trying to set expectations for a market versus an operational plan, which is what they've been in the business for, for the last two decades.

Christopher Mims (12:07)

After the break, why Martin is skeptical that a bursting AI bubble would have disastrous consequences for the economy.

Martin Casado (12:15)

It's very hard for me to see how just because you could have a speculative bubble, absolutely. This somehow denotes that we're going to have a systemic issue. Remember, we overvalued things in mobile. We overvalued things during the early cloud boom. All of these things we overvalued that did not result in systemic collapse.

Christopher Mims (12:37)

Stay with us.

Tim Higgins (12:47)

There are two major concerns driving the conversation around a possible bubble, right? One, that AI is not transformative as we had hoped. The other is that all this money is pouring into its future, these data centers and energy and chips, and it can't possibly lead to enough productive gains to justify that level of spending. I want to set aside the question of whether the technology is developing at the progress, at the levels that we've seen and just go straight to that money point, because I think for A few years.

Martin Casado (13:21)

Now.

Tim Higgins (13:21)

One of the proof points behind why the AI spending spree wasn't the dot com bubble was because companies weren't relying on huge amounts of debt to build out that infrastructure. But as of late, we've seen that debt, right? We've seen some of that debt starting to show some flashing red lights. Right. And I think of some of the numbers that my colleagues at the Journal have talked about where they see OpenAI talking about a trillion dollars for data centers. Consultants at Bain have estimated that the current wave of AI infrastructure spending would require 2 trillion in annual AI revenue by 2030 to justify that. So that seems like a lot of red flags there or flashing red lights. Why are we not in a bubble?

Martin Casado (14:08)

Well, I just think we need to define bubble here. So like what you're saying is there may be a systemic collapse, which is a very different statement than saying we're in a speculative valuation bubble. So like often when we talk about bubbles, it's a speculative statement like, oh, we're overvaluing a set of stocks for whatever reason. Right. And we see these all the time. So I would say, listen, if you look at valuations of companies, as I mentioned before, they wax, they wane, they may be overvalued over a period of time. Right now, it tends to be in the long terms, these things get pretty well justified. So for example, if you look at the valuations for dot com, even though they were totally crazy, if you look at, you know, it was like the primary growth driver of the economy for the next 20 years, they're actually pretty well justified.

Tim Higgins (15:01)

Yeah, I mean, it seems like.

Martin Casado (15:05)

You.

Tim Higgins (15:05)

Know, you talk about how AI is the biggest innovation since the Internet, but it took a little bit of time for that to correct the ROI to be there. Right? Yeah, clearly Amazon.

Martin Casado (15:15)

Yeah, but I want to actually talk to this notion of systemic collapse.

Tim Higgins (15:18)

Yeah.

Martin Casado (15:18)

So the dot com was, the dot com collapse was really kind of a fiber glut and a fiber bust. As we mentioned before, you have one company that managed demand. It was that one company was massively levered and it was cooking the books. And you had 9, 11. So you had all of these things happen. And yet even in that when fiber went down, it created a systemic issue in the financial system. Even then, the glut only lasted about four years. Listen, the question is not are we over investing relative to near term demand? Probably the question is, are we over investing relative to long term demand? And if so, do we have the economic reserves to stop some sort of systemic unraveling? The fundamentals are so different and so much better this time than they were during the late 90s. It's very hard for me to see how just. Because you could have a speculative bubble. Absolutely. This somehow denotes that we're going to have a systemic issue. Remember we overvalued things in mobile, we overvalued things during the early cloud boom, we overvalued things during the early SaaS boom. All of these things we overvalued that did not result in systemic collapse. And so I think we need to really tease these two things apart. It's like, yeah, yes, like valuations often get the timing wrong. They wax and they went, you know. But you're asking specifically, does this mean there's going to be systemic collapse? I see absolutely zero indication of that and I think I can back back into that conclusion via, you know, both historical precedents on the use of bandwidth and data plus the, the, the fundamentals of the industry that's funding these things.

Christopher Mims (17:01)

Let me just interject on the systemic collapse bit. So you know, one of my favorite columnists happens to be at Bloomberg. Shout out to our colleague slash competitors there, Connor Sen. He's says, you know, every, every time is different. People like to say every time is different, but that's true. Every time that we have a, a bubble or a systemic collapse, it is different. So here's some things that I think are relevant that are different now. Right?

Martin Casado (17:29)

Yeah.

Christopher Mims (17:29)

We've never had the level of concentration in the stock market in terms of the proportion of the value of the stock market represented by the top five or so companies that we have now, all of which are tech companies. So you know, the stakes are higher.

Martin Casado (17:43)

Right.

Christopher Mims (17:43)

Like there's lots of cash on the side, but people are also pouring cash into the market as a whole. The current level of investment implies just with a basic back of the envelope calculation, that the total amount of revenue generated by, let's say AI, in all of its manifestations at all these companies needs to increase something like 40x in the next four years to justify the current level of investment. Isn't it possible that that gap is so large that we could have a significant dip in markets and, and, and in investment in this infrastructure in a way that people look back on and say, oops, that really was a mistake.

Martin Casado (18:27)

It's totally possible. There's another, there's an interesting thing in the way that you phrase it that I thought was absolutely correct and I think it's worth calling out. So the companies that are implementing a lot of this AI spend have existing businesses. Right. And the AI portion has to grow 40x, which is a lot, but it is actually not a lot relative to their businesses. And if you look at their existing businesses, you'll actually see this is the largest shift in budget we have ever seen. Right. It's going from basically one side to the other side. So I would, I would, I would agree that you need the AI portion to grow, but what I would disagree is you need those companies to grow. And this is so often confused when people talk about these numbers. I mean, you have companies that their entire job, like meta, is like, okay, we launch new technologies and new things and then we shift spend, budget, user behavior onto those things. And so again, let's not confuse something like the Internet, which was kind of like a net new behavior, net new spend, net new companies drove basically all of it with existing companies that are shifting spend from one column to another column, which is actually what we're seeing coming up.

Tim Higgins (19:37)

It's a heady time for AI investors. So where does Martin see the opportunities?

Martin Casado (19:43)

We're very excited about the fact they're going to have new companies. But I think even when you ask questions, you're like, in your head you're thinking, open AI, that's your model. But I will tell you, that's not what the landscape looks like. That's one company. And the state of the art models is a very small subset of the long tail of, of AI companies.

Tim Higgins (20:00)

That's next.

Christopher Mims (20:13)

So you're, you're an optimist, right? You, I've heard you say that, you know, especially now is not a time for, for zero sum thinking, you know, and that makes sense.

Martin Casado (20:24)

Clearly growing. Yeah, yeah, yeah, yeah. The market's clearly growing. I mean, listen, it's not optimist. I mean, the realistic view of tech in the last 30 years is that it keeps growing.

Christopher Mims (20:31)

True, true, true. Although not every idea pans out, right? Not every sector. This one is clearly already generating a lot of revenue, as you said.

Martin Casado (20:41)

Yeah.

Christopher Mims (20:43)

That said, on your investments, when do you expect to see real ROI on the things that you've been investing in over the past couple of years?

Martin Casado (20:51)

You know, so this is such an interesting question, probably for a different reason than you're intending, which is a lot of the best companies don't go public. Ah, right.

Christopher Mims (20:59)

Which is kind of a new thing, by the way. Right?

Martin Casado (21:01)

It is a super, It's a definitely a new thing. I mean. Yeah, and, and so I understand.

Tim Higgins (21:06)

And that's because, and I just slowed down there. That's because there's so much capital out there that they can use that money to grow without going to the public markets.

Martin Casado (21:13)

Exactly. It's a very. Like, the, like many of the best companies don't go public. And so I actually think there's a very interesting discussion to have on exactly this question, but I think it's orthogonal to AI, meaning. I think that's kind of the, like, the, the dominant trend there. I mean, these AI companies, many of them have like, you know, they're profitable, they're at scale, and so that they could. But there's a bigger question of, well, if there's enough money in the private markets, why would we, you know, it's just a lot of overhead that I don't need. And I can actually be more aggressive as a company without doing that. And so I, I think, you know, from my vantage point, that kind of shift is something we're also trying to understand. Like LPs are trying to understand that. We're all trying to understand what that.

Tim Higgins (22:00)

That means, because how do you make your money? Because either traditionally it's been through going public or getting acquired by a bigger company, that, that has been usually the path.

Martin Casado (22:10)

Yeah, that's right. Well, but, but it's also interesting because actually the answer is very simple, which is, well, we can just sell it to somebody else later stage and we can cash out. But if it's the best companies that are doing that, why would you ever sell? Because they're growing very nicely. And so it just. All I will say is like, like, it doesn't pose a problem with liquidity. It just changes how you think about this in a way that I don't think we've institutionally had to think about it. Like, you know, you end up with these kind of captive markets, you know, that you have exposure to that are growing really well. And so, like, should you sell those? Should you not sell those? What are the expectations of our investors? And I don't have good answers for you now, but I will say this is. I would say every VC is having exactly this discussion to try and understand, like, what is the right posture?

Christopher Mims (22:59)

Where do you see the biggest opportunities for investors like yourself in AI? I mean, without giving it away to your frenemies at Sequoia or whatever.

Martin Casado (23:06)

Oh, no.

Tim Higgins (23:07)

Who are big listeners of this podcast? I have no doubt Sekai is great.

Martin Casado (23:14)

Every time you have a new capability and a new behavior, you get new companies, and that's what's exciting. Like, a lot of times it's like you've got this tech but like, the technology isn't disruptive enough or it isn't a new behavior. And so it's very, very hard for to get like new iconic companies. I mean, that's actually been the story of AI previously, like before this wave. AI has been a thing since the 60s, right? I mean, I took my first AI course in 1999. So then where are all the AI companies? We've had 30 years of this, like, where are they? Why hasn't there been any AI companies? And the answer is because the economics have been terrible. It's like they give you a 20% gain and the product kind of sucks and it's been okay. And so it's been a technology that like large companies can use to get 20% better. I can, you know, I know your preference 20% better. I can detect fraud 20% better. Like that's been the story of AI until this generative wave. The generative wave is like, this is a totally new behavior and it's a thousand times better than the traditional way. And when you have those disruptions, then you end up in, you know, in these super cycles where you have new generational companies. So I would say generally this is very exciting because you're going to definitely have new companies that rise up, like the OpenAI's or the Anthropics or the Cursors or whatever it is that are clearly going to be new. And then, and then when you like as like particular markets, we kind of break the world into two pieces. There's like, there's the state of the art large language models like the OpenAI's and you know, in our view those require tremendous amount of capital. But then there's all of these other companies that people don't talk as much about. The ones that do, like image diffusion or video diffusion or speech or music or whatever, these are all generative AI. The companies are great. And so we invest heavily in those too. And so we invest broadly, like you do in any cycle. We're very excited about the fact they're going to have new companies. But we do kind of segment the market based on, I think a lot of the time, I think even when you ask questions, you're like, in your head you're thinking, OpenAI, that's your model. But I will tell you that's not what the landscape looks like. That's one company. And the state of the art models is a very small subset of the long tail of AI companies. So we're very interested in the long tail. In addition to, of course the OpenAI's, etc.

Tim Higgins (25:36)

You feel like we're at a tipping point for AI, that the economics are, you can see the economics are coming, that this is real businesses here on the horizon.

Martin Casado (25:43)

I mean, yeah, for sure, 100%. I mean, but we've seen that this isn't even like for some uses cases of AI. I just don't think this is a controversial take. Like the economics are great for a number of use cases. Now listen, I think we just tend to think pretty muddy about a lot of these things. So for example, the statement that there's no long term defensibility is very different than can you build a profitable company? And the answer is we can definitely build profitable companies. We have those that have great growth, like those exist today. And there's a number of examples of those. Then people say, well are they long term defensible? Maybe. But I will say like, you know, traditional defensibility does pertain to, to these companies. So if you can build a two sided marketplace or long term integration or whatever it is, you can get defensibility. And so I just feel like a lot of the times we kind of move the goalposts on simple questions, but we can be very, very clear that yes, you can build a company today that is AI based, that is profitable, that grows at levels that we view is incredibly healthy.

Tim Higgins (26:52)

So what story do you think we'll be telling about today's AI boom 20 years from now?

Martin Casado (26:58)

Do you remember what the first video for the web was like?

Christopher Mims (27:03)

The subject of the first video for the web.

Martin Casado (27:05)

The first, the first, the first video on the web.

Christopher Mims (27:09)

You got me.

Martin Casado (27:10)

The first video was a coffee pot.

Christopher Mims (27:12)

And oh, the live webcam of the coffee pot. I do remember this.

Archive Audio / Narrator (27:16)

Cambridge University coffee machine gained international stardom with more than150,000 people around the world avidly watching it.

Martin Casado (27:24)

And the reason, you know, someone put it up as a researcher I think in Cambridge, and he's like, I want to see if the coffee pot has coffee in it. Before I like go down to like waste my time to go get some coffee.

Archive Audio / Narrator (27:35)

All I have to do is to click on a little button that says coffee machine like this. And eventually I get a picture on.

Martin Casado (27:43)

My workstation and it was like this total sensation. And so if you, if you look back, you know, in the mid-90s you had these things that just looked like toys and they were so silly and you made fun of them. But the reality is, is that video of a coffee pot in no small way became Netflix and people could see that, right? People could see it. And so there was, you know, like, like we kind of like would make fun of all of the excitement about what seemed like trivial things. But it all turned out to be true in the long run. And I think that that will be the story of this realm, which is we see a lot of like anime and a lot of silly use cases and then we tend to poo poo this as like, oh, this isn't serious stuff and where are the enterprise use cases and blah blah blah. But like, this is what the future always looks like.

Tim Higgins (28:33)

Well, Martin, thank you for coming on.

Martin Casado (28:36)

It was such a pleasure. I appreciate it.

Tim Higgins (28:40)

A quick note. The coffee pot video that Martin referenced was the first live webcam video on the Internet, not the first video on the web.

Podcast Host (28:50)

Thanks for listening to the A16Z podcast. If you enjoyed the episode, let us know by leaving a review@ratethispodcast.com a16z we've got more great conversations coming your way. See you next time. As a reminder, the content here is for informational purposes only, should not be taken as legal, business, tax or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16Z fund. Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see a16z.com disclosures.