The Swyx Mixtape

See my notes here on DX Circle!
Audio Source: https://a16z-live.simplecast.com/episodes/a16z-infra-1-2iEyBTf5

a16z on Infra

Introductions and Backgrounds [00:00:00]

Martin Casado: [00:00:00] So this is the a16z infrastructure show. This is actually the very first one.  Where are we going to be talking about infrastructure companies investing in them, building them products? It just turns out that we're three GPS at a16z, and we all have a lot of experience in info.

Like all of our companies were infrastructure companies. We do a lot of infra investing. So the way that we're going to structure this session is first, we're going to introduce, our backgrounds in context of that. Many of you know, us, many of you have worked with us. But we do want to give you a sense our relationship with infrastructure and how we went through it.

So we'll and we each go through our own kind of bios that way and I'll orchestrate that. Okay. Th then we're going to talk about why infrastructure is different. This isn't B2B, this is an enterprise, this isn't vertical SAS. It's specifically infrastructure. And it's my it's my favorite topic and my favorite area.

So we'll go through that. We've got a lot of great questions on Twitter, and so we're going to try and get through those. And then if we still have time, then we'll open up to questions. Uh, for everybody else. So that's rough, we plan to have this every two weeks and we want to cover everything as things go, we want to cover category creation and we want to cover open-source and we want to cover the shifts and go to market and the cloud and investing and everything else.

So we're going to go ahead and start with our intros and just, yeah, very quickly. So for the, if you don't know, so I'm Martine I'm a GPN Andreessen Horowitz, and. I actually want to start by introducing Ben actually having him introduce himself, but I want to let you know how I met Ben. So I was doing my PhD at Stanford in the networking space, and we spun out and we started a company called this Sera.

This is the software defined networking space. And we started it right before. Like the great recession, the nuclear winter had set in and we were struggling. And one of our investors who was also on our board was Andy Rachleff, who was actually a professor at Stanford at the time, but he's this super famous investor from benchmark.

And we were at the, we were at the bottom of the recession and we're like, we had a professor as a CEO and he needed a new CEO and we were talking with Andy Rachleff and we're like, who would be the best person on earth? It's well, there's this guy that just sold the company to HP.

And his name is Ben. And I had never heard of him at the time. And and we should talk to him. So this I, this was in 2008, so I got introduced in the band and he came in, of course, Ben's you have so insightful and he'd just done it. Like he just built a company at the same thing. And uh, so I asked Ben if he would be our CEO, actually.

Ben, do you remember what you said in response to that? I don't. I don't. I said, no, you said I'm too rich.

You're like to be CEO of a company like this. You gotta be piped in. And so instead, however, Ben and Mark invested Ben ended up joining the board. And so much of what I've learned has been with Ben. And so I thought it'd be great if Ben, most of you know him, but it'd be great if he just gave you a quick rundown of kind of his background with respect to infrastructure that we'll move on to David.

So Ben, if you wouldn't mind. 

Ben Horowitz: [00:02:36] Yeah. Sure. And that, that did bring that a, man's got to know his limitations. If you know how hard a job is if you're going to give somebody a really, really hard, nearly impossible job, it's it really helps if they're not rich, I have to say, yeah, that's a good hiring tip as well.

At some point, rich people are just like, this is too hard. I'm going to the 

Martin Casado: [00:02:54] beach. Things we learned multiple times, I think, in, in our 

Ben Horowitz: [00:02:57] careers. Yeah. Yeah. So yeah kinda my career, actually, my, probably my career in infrastructure, I started way back when I was an engineer at Silicon graphics and kind of the first.

I was working on, we were, we had to put, we had an operating system called IRX or truss Unix-based and we were the first we had built the, the original multiprocessor machines. And so there was this task of having to. Put semaphores on all the Colonel processes and so forth so that they went collide and you wouldn't have all these weird race conditions, which was I would say probably the most complicated engineering job I ever had, but.

Eventually then eventually went to a company called Netscape where I was in charge of kind of the web servers and then we needed a directory kind of project. And so we that's where we. I got the idea to  popularize LDAP and kind of make up the directory standard.

And that was like my big infrastructure thing there. And then Mark and I founded a company called LoudCloud, which was one of the first or the original kind of cloud computing company started much too early. Ironically, because there wasn't enough infrastructure like principally, there was no virtualization, for example.

And so you couldn't really do cloud computing in that way, but the tools that we had so we transformed that company into a company called Opsware and that's the one that I sell to HP, which made me to to rich, to take Martine stock. 

Martin Casado: [00:04:21] And since joining Andreesen, you've done a lot of infra investing actually so comfortable.

It'd be great to just go through that a bit. 

Ben Horowitz: [00:04:28] Yeah, sure. Made a bunch of introduced structure investments. The first  investment I've made is on a company called Okta, which was very familiar to me from my directory days. And more recently I've invested in a company called Databricks.

And then, most recently one called any scale and Databricks is infrastructure for AI and big data. And any scale is a new way to, how do you get the kind of processing power that you need, at the growth rate that you need. Now that Moore's law is definitely not going fast enough to support the hunger of AI and it can turn it can basically Make the cloud look like your laptop and make it very easy to program in parallel.

Martin Casado: [00:05:10] Awesome. All right. Cool. So I was one of Ben's investments. He joined, we were definitely for a journalist Sarah networks. I was gonna say there was another one, sorry about that. Yeah we so very honestly I learned so much of what I know by having been on my board.

We ended up selling the company to VMware where I ran that business. And when I left, there was about a $600 million business and end to end, that was about an 11 year journey. And then I joined Andreessen Horowitz where I also focus on core infrastructure. And as part of that, I'd always heard of Dave

But I met him and I don't know if you ever have these moments where you're like, I dunno, you meet someone new and it's like talking to a long lost brother, but I feel like he's lived this parallel life to me as far as the company he's built and what he's done. And so super happy he was able to join as well.

And do you, if you're cool with it, it'd be great to get your personal journey through infra just to set the stage before we actually dive in. Sure, sure. 

David Ulevitch: [00:06:03] And this is great. This is a favorite topic. So. I was one of those people probably like you Martina and who fell into infrastructure early in my computing career.

And I had an internship at a mom and pop ISP in Sydney, Diego because I had demanded to our family that we get real internet. Not like AOL or prodigy much to the dismay. I think of my sister because all her friends were, her prints were on like AOL, but when I wanted internet, so we got that. And then I was able to get an internship in eighth grade at this company and learned all about this.

This is a company that's run on Sonos and later on a Solaris, which are just a slightly marginally better than IRX that Ben was talking about. But only marginally and So I learned all about routing and networking and Unix, and started my career as a mediocre or programmer there. It was great to understand how appearing works and how the internet worked.

And I really, at that point fell in love with the internet and how it cross borders and how, you had this asymmetric opportunity to either write code or do something that could reach a whole bunch of people. And then there was sort of one technology in particular that I fell in love with, which was the DNS or the domain name system.

And I started buying domain names. I would eventually went off to to college and built a domain name management sort of service because I started to own a bunch of domain names and all these other tech people started to use my free service. And we had the founder of Rackspace using my service.

We had distinguished engineers at sun and chief scientist at Microsoft. And at that point we became big enough that we needed to build a network. And so I started to learn about things like any CAS networking and how the internet really started to come together from a commercial standpoint. And we got Rackspace that data centers, things you don't have to do any more.

Cause we have AWS and GCP and Azure, but we started to build out a global network. We got IP addresses allocated to us, and I really had an incredible experience learning how to scale up and infrastructure. And that eventually paved the way to building a company called open DNS, which was a cybersecurity company that started out.

As a consumer company to give people a faster and safer and better internet experience. Many, many millions of people wanted that none of those people wanted to pay for it. They'd rather go buy a cup of Starbucks once a month, instead of paying me three or four bucks a month for a better internet and safer internet.

But it turned out a lot of businesses were using our product and the timing, this was We started the company at the end of 2005. And so by the 2007, 2008, the iPhone was out. People were bringing all kinds of wireless devices into the office devices, you couldn't install traditional end point security onto.

And so a lot of it administrators were using open DNS to provide better protection as a network service, without having to install endpoint software, because we were delivering all of our security as a service using the DNS. And that was a novel innovation that we had. We'd come up with.

And so all these businesses were using it. None of them were paying us and we needed to make money. And so in 2009, we went through a major transition to becoming an enterprise cyber security company and starting with all those millions of free users, eventually figuring out what two ones are businesses.

And we'll talk about some of that today, but it started to go down that sort for enterprise, go to market and. That's what ended up becoming a decent sized enterprise cybersecurity company. We sold to Cisco. And like you said, we were you and I have a very mirrored career path. And in some ways I went on to run the security business at Cisco, which was a great experience working with really great people.

We had about 240,000 customers. It was really an exciting opportunity to both operate and lead at scale, we did a bunch of acquisitions and I was very happy after three and a half years in Cisco to to get a call from someone that people that know a16z well know, Jeff stump is on our team here and talk to him and Ben and you, and a whole bunch of folks.

And they've been here ever since now. It's two and a half years, and I invest in enterprise software, but a little bit of, a little bit of the infrastructure stuff. I think the one investment that we've announced in this space is is census, which sort of operated the, at the data layer. We can talk about that later.

What is Infra? [00:09:51]

Martin Casado: [00:09:51] So I think maybe for starters, especially since we've got a very varied audience, it's probably good to cover, like what infrastructure actually is. So what do we mean by infra? Why is it exciting? Why does it make sense to actually have a separate, talk around in for us?

It for traditionally has meant that the stuff that apps are dealt on. And so it, traditionally has been compute network storage. These days is everything from dev tools to security, to API APIs. But think about it is the underlying layers that we use to, to build modern compute apps.

And, it actually is quite different. Like companies and infra are, tend to be quite different than other, and that, they tend to be like quite technical. They tend to be horizontal. They they tend to have, complex insertions. They tend to be the product of. Real tactical work.

They tend to be a little disconnected from business users. And so if you look at an infrastructure company, they often look quite different than, a typical app company. But they've really had a moment in the last couple of years. If you heard names like Okta and snowflake and GitHub and Databricks and do on MuleSoft and Datadog, et cetera, et cetera, et cetera, these are all core infrastructure companies.

Will Infra Ever Be Over? [00:10:56]

And so I think, actually, it'd be interesting to hear your thoughts. Ben and David, which is, it's always been these underlying. The underlying technology, but you always have to wonder is is the day of infra ever over like steel? Are we ever done, like building like the concrete and like the rebar and whatever, and it's just about apps or is it going to continue to be a growth area?

Because this year, last year just been phenomenal for infra. And if you have any opinions on that, it would be fun to talk through it. 

Ben Horowitz: [00:11:20] I think we're very far away from Infrastructure being over and it being done. And what happens is the needs of the applications change.

And then the infrastructure has to change. The kind of, one of the first big shifts in our careers in infrastructure was the shift to cloud computing, which completely changed the requirement. As one of the brilliant things about Nycira networks is that networking got completely, just got devastated in terms of all of its architectural flaws. Once you move into the cloud environment. And, we used to talk about, okay, East West traffic versus North South traffic and. You know what Cisco could do and what kind of and so forth. And then, but that happened at every layer in the stack, storage was broken, the operating systems were broken every kind of piece of infrastructure wasn't built for that kind of an architecture.

And so it has to be completely rebuilt and, and then as, things have scaled, then things have to be upgraded, on, on the other side, as we move from just procedural programming so AI, and now we're, the data is the program. There needs to be a whole tool chain to support that.

And then, a lot of people are going, Oh, wow. Like this NFT thing is an amazing craze and it seems like it popped up overnight. But the, those of us who are infrastructure, people know that this thing has been, decentralized computing. And, trustless state, across machines is something that's been worked on since 2009.

And and the first kind of cut at NFTs for things like crypto kitties blew up the infrastructure, so the infrastructure needed to be revamped. And so anytime there's a new important application need, I think, it ends up causing an evolution of the infrastructure. And that evolution of the infrastructure is, maybe the best thing to look at.

If you're trying to understand. Stan what's going to happen in the future because, like I say, a lot of people are like shocked and amazed by NFTs and they just think it's some wackadoo thing. But, for those of us who have been like an infrastructure where Oh yeah, this is coming and we're surprised it took this long.

And now that it's here, we know exactly what it is. So it's a a great way to fortune tell what's going to happen in apps world as well. 

Martin Casado: [00:13:32] Yeah. Yeah. I was, I always find it interesting to walk through like the major trends that we track and infra, particularly with respect to this, because it just drags on so much behind it, an early, I think an early lesson that I learned building an infra company as.

It just seems like lower layers of the stack never go away, but there's always something that you build on top of them. And so it's almost like we're just gaining in complexity over time. And so I want to talk to a few of the trends that we track at a 16 Z and then, and then David, as I end up, it'd be great.


Trends - Data [00:13:59]

If you want to add any of them, but it'll just give you a sense of the disruptive power of infrastructure, how big the markets are, and how it, it opens up future potentials for say what's companies or investment. One of them, this is going to sound like a cliche.

But but it, I, there is a lot deep there and that is that as data. And what do we mean by data? Data is just so central to how we build businesses, right? And again, like this is one of these cliches, which everybody has been saying. It we've been saying it for 20 years, but like we have these massive proof points recently of, large companies, large startups, snowflake Databricks, for example that are centered around data, but then applications and.

And all of in front of is really to empower applications of using data for differentiation. In fact, actually, if you look at the application space, you're like okay, like what differs like three different SAS? Those let's look at three different, like whatever, three different SAS apps say that they're consumer apps.

And you're like, okay, to build one of those things, Yeah, you have to understand distributed systems, but we know that yeah, you have to build a big cloud service. You have to know that, like these things are hard things to do. But they're pretty well understood, but to actually do good personalization and actually to do good prediction, to actually use data in a way that's differentiated, that's almost where a lot of the battle has moved to.

And as a result, we're seeing this kind of explosion  of companies and tooling around data. And what they're trying to do is they're trying to enable those that are building SAS apps, the ability to to do, whatever on the business side they're wanting to do with kind of higher accuracy, whether this is wait time prediction, whether this is personalization, but this is pricing, whether this is fraud detection.

And so from an investor standpoint, there's probably too many companies to track and you always have seems an infrastructure. You've got these kinds of explosions and then these contractions, right? Like you see an explosion of apps. Companies, and then they consolidate, we're seeing an explosion of data companies and they're going to consolidate.

So if you look at data as as a market segment, it's growing about twice as fast, the cagar is the rest of infrastructure. So it's a major area, it's a major area of focus and investment. And again, you've got dozens of companies that underlie it. So that's one trend that we're very focused on.

Trend - SaaS [00:15:59] 

Another trend that's quite interesting is, when David Ben and I got into the game decades ago, like software companies were actually software companies. Like they shipped software that's where I cut my teeth. There's you'd write code and you'd ship it. And you built companies around that.

Like you built companies around having releases and getting them out and being able to debug remotely and supporting, complex HCLs Hardware compatibility lists. It's in different environments. And that was really what the companies looked like. These days it's very tough actually to find Harbor company or sorry, shipping software companies, almost every company is becoming basically a SAS company or an operations company.

And it's just a very. Different type of company, right? The way that you write software, the way that you support the software. It all is becoming operations and there's huge impacts to infrastructure because rather than having something that I can give to somebody else, that's easy to debug and run.

The question is how can I have high up times internally? And listen, we've seen massive increase in, or like changes in personnel. Like for example, SREs now are very much in demand. There's change in the programming style to do this change in development methodology, changing in the tool set and so forth.

So this is also another kind of cross vertical area. That's very interesting. Do you mean be, want to talk a little bit about the pro-sumer ization of infrastructure or bottom up, because that's an area that you focus on quite a bit.


Trends - Security Analytics, Billing, Metering [00:17:14]

David Ulevitch: [00:17:14] Yeah, actually I was thinking, as I'm listening to you talk about these things, I was thinking that, all of these trends are talking about have the.

These massive derivative sort of downstream effects, right? So when people talk about the massive amounts of data, that creates opportunities, not just for the storage companies and the database companies, but then it creates new opportunities for the analytics companies. It creates a new surface area for security companies to focus on data privacy and data governance and all this regulatory controls around data.

And that all of those standalone, all of those derivatives, sorry. Sort of our categories of their own. When you talk about every company is becoming a SAS company and we don't ship perpetual license software anymore. One of the things I think you and I have heard over and over again, especially when we talk to infrastructure companies and I'm sure Ben sees this as well, is that all these companies are now dealing with major issues around billing and metering of their customers.

And, AWS may have a major metering solution for being able to do billing and metering. But all of the SAS companies building on top of these platforms have almost nothing, right? There's, they're older tools around subscription billing, but not a lot around usage based billing. I'm sure companies like Twilio have had to build their own.

Then, you mentioned the SREs and all this operational sophistication that companies need that creates whole new opportunities around management tools. Around, across cloud and private cloud sort of management tools that people need. And so all these major trends end up coming up with all these derivatives categories that ended up getting created that are brand new surface areas for new companies to be made.

And many of them are just very large, even on their own. That's what I was thinking about as you were. 

Martin Casado: [00:18:46] I do think it's worth it. And just a moment, we'll get more in kind of the weeds and questions that are, I know, on, on many folks minds, things like, open source and so forth.




Trends - Prosumerization of Tech [00:18:55]

But I do think it's worth talking through what makes it infrastructure companies. Different like how we look at it on 

David Ulevitch: [00:19:02] the, on the bottom, on the bottoms up point, you were asking about, developers have way more power and control than they've ever had before in terms of the buying decisions of technology, especially for technology companies.

And that has changed. I think the entire go to market for infrastructure companies. You're no longer a sales person reaching out to a CIO to arrange a golf game, to talk about. While your database is better or why your data center is better. Like all those days are gone. Now you're basically doing content marketing to get somebody excited about the technology that you've created.

The developer is just going to download it. Maybe it's open source at the heart of it. They're going to download it and play with it. Or they're going to set up an account and start using it right away. They have to have time to value that's measured in minutes or hours. And then eventually you're going to start tracking all the analytics and data of how people use it.

And that then eventually feeds over to a sales person who tries to go create a much larger opportunity, but that transformation has been, I think a, not a major shift in how every infrastructure company has ever gone to market before compared to how they're going to market today.

And it's been a major re-imagining of even the stack and how sales teams and marketing teams come together inside of infrastructure companies. 

Martin Casado: [00:20:12] One thing that's super, I think maybe unique to infrastructure, which is you almost have this weird barbell construction with the team, which is most infrastructure companies require a kind of a deep understanding of the market or technology, right?

Like you had to understand DNS, Ben had to understand the cloud. I had to understand the networking. And so the people that do it tend to be the, tend to be, technical founders, very focused on that. But at the same time at the same time, there's like actually tends to be a much more difficult go to market as well.

Just because, often you're selling something that's low level, that's relatively technical. Normally you're actually somewhat removed from the business, value. You're like, yes, I get the case of software defined networking, yeah. makes things operationally more flexible or like whatever it is.

It's very difficult to actually tie that to our revenue. And so I think one of the things that's tough about being a founder for an infrastructure and even an investor is you do need someone that, both has to be very good at go to market and has to be very good at tech and product.

And and we see this in a lot of the founders and then everybody up here, and I don't know, Ben, I don't want to put you too much on the spot, but you saw, like my evidence. So I just, I just look how naive I was early on. I was just so tech focused and later I just had to learn, go to market and it took about 10 years and this and that.

And I'm just wondering do you have any thoughts about how you bridge that gap either as a founder or as a board, or, how do you think through that? Because I think it is probably the broadest gap between technology and go to market in the industry. 

Ben Horowitz: [00:21:37] Yeah, that's a great lead up. I would say, one, it varies a little, so it depends where in the infrastructure stack you are. So in a Sarah was, I would say the worst case scenario in terms of go to market challenge, because it was infrastructure that.

Was basically at the bottom of the stack. So if something went wrong with Nicera, then you basically get fired as the customer. And that's always a tough place to start with a new company. You really have to be. Comprehensively good at product marketing sales, lead generation story.

Every everything's gotta be exactly perfect to get that off. These days, just no matter how compelling your value proposition or how great your technology is that ends up being a thing. But it is always. I would say challenging because you are going, you're the underpinning and if your stuff doesn't work and if your company doesn't last and all the things that people have to trust about, you have to be in place and, Oh, by the way, if you ever, God forbid have some kind of breach, you'd just die right there.

A lot of the things that I look at are, most of what I do with the infrastructure companies that I'm on the board on are focused on and, helping the founders, CEO get the right kind of partners, I would say on the kind of sales and marketing side To do that product.

And I always assume that the founder doesn't quite know what they're doing. And like really one of my favorite stories on this was, Okta has got off to a very rough start because of this go to market problem. And what happened was, they had come from salesforce.com and salesforce.com had a very kind of specific go-to market idea, which worked for an app, which was, Hey, we have.

One 10th, the features of CBO, the on-premise product. So we're not going to go sell to those guys. We'll sell the guys. They can't get to because it's too heavyweight. And so they start at the bottom of the market and they work their way up. So Octa tried that, but it turns out. Little companies don't care that much about their security infrastructure.

So it was a terrible place to start. And so just, me, a lot of the things we worked on is, okay, not only are we selling to the wrong customer, but because we're selling to the wrong customer, we have the wrong channel. We have the wrong pricing. We have the wrong marketing, like everything is wrong and we have to make that shift.

And of course, once they made the shift, they went from being a company that looked like it. Wasn't going to make it to a very successful company, just highlight how important, go to market ends up being usually what makes an infrastructure company succeed or fail. Yeah. It's 

Martin Casado: [00:24:08] it's just so you would hope that this is the challenge for those that have spent, whatever their schooling to learn business.

So they understand all the challenges, but in France almost, primarily product and tech founders, right? Like David  like DNS, and it's interesting when you see I remember when I first met David I'm like, listen, this guy is like a head of sales, right? All he thinks about as good a marketer you've talked with Ali or any of.

Any of these founders for these companies is just, such a big focus. And even like D David, th just to pull you in really quickly, I remember you're like the last three years I've been like a student of go to market. Like, how did I. Did you know, and now as you take board seats, for example and you work with, infrastructure founders do you feel tempted to drag them into, go to market or focus on product or, 

David Ulevitch: [00:24:51] yeah, I think I look, I think there's different phases of the company journey.

And I think, that when I think about my own experience and then. The companies that we work with, and then we spend time with people. Basically you have the technical founder hat, which is you need to have a technology that works. That's really good. It's differentiated at solving a real problem.

And it really, 10 X way to what the world looks like in me, to Ben's point about Salesforce. I may not do everything that the incumbent solution does, but it does something else just in a dramatically better fashion that, so you wear that technical hat and product hat and you build that, but then eventually.

Every founder sort realized. I was like, just because you build it does not mean that the customers will come. And so I found that out in multiple painful ways, because every time you solve that problem, then you want to Uplevel your company. Then you have to resolve that problem. And so figuring out and spending time with customers and really putting on that sort of sales CEO, hat, I think is one of the best ways to actually become a great CEO.

Because it means that you're listening to your customers. It means you're understanding exactly what the market is telling you. People always ask questions, like how should I price my product? What, how should I package it up? A lot of those questions actually get answered just by being in the market.

And you, as a CEO of, I think of an infrastructure company, you can't outsource that. Like you can get help from your product team, from your product marketing team, but you actually can't outsource it. You have to be out there with your customers, especially if you're aspirational and who you want your customers to be.

You can spend time with your current customers. That's great. But if you want bigger customers or you want a different segment of the customer base or a different vertical to go after, you have to get out of there. So I was spending, this is like the period of my life, where, and I think, Ben and Martina and lots of people that are listening can relate to this.

You ended up getting the, United global services and you realize that's actually a status you'd never want to get, because it means just spending too much time 

Martin Casado: [00:26:34] on the road, seven years, seven years of golden services, man. 

David Ulevitch: [00:26:37] It feels really good when you first get it for cancer.

Yeah. 

Ben Horowitz: [00:26:41] Yeah. That's a. It's very bad for your health. It turns out. 

David Ulevitch: [00:26:44] Yeah, that's right. Yeah. Ultimate Pyrrhic victory and, and but, so I really enjoyed the experience of getting out in front of customers and recognizing that the software and the services we had built in the infrastructure we had built to deliver that service.

Was powering some of the biggest companies out there are some of the companies I respected the most. And so that, and then we're actually one of the most interesting things about that journey is you actually find out your whole company really loves that experience too. Like engineers that are working on really hard engineering problems, love finding out that, you went from having your biggest deal, be like a hundred thousand dollar.

Annual deal to closing a customer, that's paying you now a million dollars a year. And like the satisfaction of knowing that, the code that you write and the services you've stood up are now I'm getting paid for in that way, because people are finding that value. Like I found that really satisfying and exhilarating, but the only way to go through that journey is I think to be as close to the customers as possible then brought up like, 500 really important threads.

You talking, he mentioned channels. He mentioned pricing. Like all of these things are things that you just have to spend a lot of time on. And the worst part about all of them is that once you actually get one working and get it right, it's only going to work for a while and you create more Slack somewhere else in the system where you dedicate your attention, but then once you then go, tighten up the Slack somewhere else, you have to come back to pricing.

You have to come back to packaging. We have to come back to your channels and distribution strategy. And that, that for the, the right kind of founders, I think is really an exciting part of the company building process.



Deplatforming and Decentralization [00:28:13]

Martin Casado: [00:28:13] Awesome. Okay. Let's I'm going to, so I think we should get a little bit into the nuts and bolts now.

There's one topic I want to go too into David. I actually, so this, we've had an amazing few years in infrastructure. Like many of the top enterprise companies are like core and for companies, it's data it's, whatever, but Going forward. There's also a lot of opportunity, both in frontier tech.

Like whether it's satellites, but we're also just seeing like the traditional players being blown up. And I know one that you've been focused on is listen. There's a lot that's going on with kind of regulation and compliance. And when you see companies taking political stands and you and I hit this and I helped run a CDN, many years ago.

And. You ran a DNS server and this has always been an issue. But often when you try and deep platform, infrastructure is what reshapes to provide hosting. And so love your thoughts on the current environment and what that might mean to infrastructure.

And those in this audience that are thinking about building infrastructure companies. Oh, yeah. We'll call 

David Ulevitch: [00:29:10] this a leading question, but I know, my views on this, so I'll share them widely, which is that, I think companies have a right to operate. They want the way they want to operate.

But when I looked at what was happening, around the election, when AWS kicked parlor off their services, like I, and I say this about actually any opinion about parlor or, and, I guess I've only seen AWS do it. They want what it really is to me signaled, it was like, this is a great.

Infrastructure opportunity because for a long time, I think people felt like the lower layers of the stack were fully baked. There's no room for innovation, Azure GCP, and AWS have cornered the market on standing up hosting and infrastructure and services at those low layers. But, if you're running a service like parlor and you can just get thrown off by one of those services that, you just get thrown off by the other two as well.

Then all of a sudden. There's going to be people that are now like fully invigorated and have been now catalyzed in action to go re-imagine. How do you build truly like Bulletproof globally, distributed, decentralized hosting and infrastructure solutions. And we know that in the crypto world we're seeing this across a number of of the crypto initiatives to create file hosting, file distribution content distribution, and all of these things now are going to, I think, start to get a ton of R and D effort because one of the great things about the internet has been that it's been decentralized and, as these Cod platforms of urgent and offered a lot of flexibility, they've also now centralized a lot of that control that the internet used to de-centralized.

And so I, I get excited about the idea that, maybe parlor, which, good writtens are not good writtens, but it may end up being the greatest catalyst to re-imagining infrastructure that we thought was done fully baked in that no room left for innovation. And I'm sure someone's going to start to tie together new pieces of new ways.

And I'm excited for that. 

Ben Horowitz: [00:30:56] Yeah, and really interesting on that, parlor, there's a lot of things that would be very hard to run on a decentralized service right now. Like Netflix kind of comes to mind as something you would rather not try, but parlor could definitely work. So that is a, like a really interesting idea.

Like parlor is something that you could probably decentralize right now 

Martin Casado: [00:31:16] for sure. I just love the info when this whole thing went down or just funny to see the different personas and yours is definitely like the, I think the infrastructure founder view, which is, this goes down and some people like, this is good, parlor's a bad thing.

Others are like, Oh, this is council culture. This is terrible. This shouldn't happen at central ship. But my reaction was yours, which is this is totally a failure and infrastructure. That's right. 

David Ulevitch: [00:31:38] Yeah. And we know it can be done the pirate failures than online the parent days, like two decades now.

And this is a site that has been legally shut down so many times. It's it's still on there. Totally. And we know what can be done, but it can be done better in a more scalable way. And in a more general purpose 

Martin Casado: [00:31:52] way. So my sensibilities were totally offended just because I'm like, how is it possible that you can actually shut something like this down?

Whether it's politically the right or the wrong thing to do, I was less concerned with, it's just, when we ran a set CDN back in the early two thousands, there were these sites that like, Interpol would show up, every couple of months asking to take it down and whatever, and they would just pop up somewhere else going for, and so in this day and age, especially with crypto, the fact that you actually could cancel something means that like it's time to rethink.

The stack not by the way, because of any political agenda, positive or negative. It's just like infrastructure is meant to be ubiquitous in everywhere. And this is just, I think either a failure of imagination on the case of partner, or maybe we've gotten to a place where infrastructure, isn't what it used to be.

And that's 

David Ulevitch: [00:32:35] exactly right. That's exactly right. There's still room for innovation




Open Source is becoming Irrelevant [00:32:38]

Martin Casado: [00:32:38] Let's talk opensource. I actually, so I have a bit of a, I think I'm a contrarian position on opensource. Ben, it'd be great. If you could backstop me on this one or contradicting or contradictory. Sorry.

Here's the thing. For those that are listening, what is open source? So open source has been around for quite a while. It's the idea that, if you're writing code, the code should be, visible for. For everybody so they can look whether it's, to modify it, to build a community around it, to make sure that, it's correct to to address bugs, et cetera.

 This is a massive movement and, there's just been a lot of, focus on open source as the way to build these businesses. Yeah I can't help, but think that actually opensource. Is actually becoming a little bit irrelevant and not for the reasons, people say meaning it's fine to build an open-source company and it's fine to have open source, but in the era of SAS, Whether it's infrastructure satisfied in the era of you're actually running a SAS service, it's not super clear the role that open-source plays.

Certainly not as clear as it did say 10 years ago. So if I'm shipping you software, if that's my business model, I'm shipping you software open source is something that people can get and like download and run. And it's a great kind of go-to-market channel. You get, the actual users of the open source become your customers.

But if I'm building a SAS service, even if I open source the actual functionality of it, there's so much of the operations, which is actually really hard about this stuff, that is an open source, right? If you think about AWS as an open source, there's nothing open source about it. Get hub, isn't open source snowflake, isn't open source, but these are these kind of developer focused companies that are all very huge, that, and we had this question on Twitter and I think it's a great question. Like where does the source fit? It yeah. Going forward. I actually think, listen it's still like what it used to be. However I do think it's it's diminishing and relevance because the consumption model is becoming a service and listen, David, I know you've got an opinion on this.

Would love to hear that. And Ben would love to hear yours as well. 

David Ulevitch: [00:34:28] I'll just say mine really quickly. Cause I, my view that people are conflating multiple things here, which is, I always view SaaS as basically a way of outsourcing IT. And so when you sign up for a SaaS service, you're effectively saying, I don't want to pay somebody inside my IT organization to run something, on-prem manage it, install patches and updates.

And so it's a little bit orthogonal and I agree with you that. People are paying for SaaS, whether it's open-source underneath the hood or not, they're not paying for open-source, they're paying for it basically to not have to pay to manage something themselves. And in fact, they often pay a premium for that.

I think the open source angle is often I'd go to marketing. What's a marketing way to get to build awareness to give people competence in your product. But I don't think that open source. Component isn't important at all. Like I'm agreeing with you, but it's really because of the framing to me is about outsourcing it.

Not about whether they're paying for software or not paying 

Martin Casado: [00:35:22] for something. It's just, I've just felt like there's so much industry focused on open source. And this, I have a long personal history with open source, I've funded it, I've worked on it. I think it's fantastic. It's transformed the industry.

It will continue, but if you look at a new company in the infrastructure space, It's most likely they're going to monetize with a SAS service. Not all the case and if that's the case, it's just not super clear how relevant open-source it's like you can have it and it's great. And yeah, it's great from maybe a marketing community standpoint, but it's not clear to me that going forward, like this is the thing to rotate on.

Ben, did you have a comment?  

Ben Horowitz: [00:35:51] it's funny when you think about the history of it, right? Because the thing that made open source destroy. Closed source were these big internet services because if you were running Facebook or Google, there's no way you can take a closed source piece of infrastructure and build Facebook on it.

Because if there is a bug it's way, way, not okay to be dependent on some software company to patch it and ship it to you. That's just too long, a low. And so that moved everything to the thing that made everything open source was running these big services. But when the open source became services, Themselves then the kind of value of the open source went away.

And I agree with David, it became a marketing tool. It actually is. Weirdly analogy is like an industry where, but when it went to streaming, then, the song is almost more of a marketing thing for building the brand of the artist than a way to make money. And. In a way open source is more, just a branding thing for the company and not how they make money.

They make money on the service. 

Martin Casado: [00:36:54] Yeah. I mean it just, for everybody listening, we love open source. We're huge fans of it. I'm on the board of maybe seven open-source companies and they're all great. It's just, we used to hand ring as founders and investors so much about, Oh, can you do open source and still monetize it?

And we just don't worry about that anymore. If you have an open-source company, great. Don't worry about monetizing it because we know you're going to most likely monetize as a, a SAS offering. And that just doesn't count isn't cannibalized because nobody wants to operate their own infrastructure.



Infra Company Margins [00:37:22]

So I actually want to take this moment to dovetail a little bit into another question on Twitter. Which is another thing that we actually obsess about a lot, which is the margin question. So yeah, I feel like so many of the, so many of the the way to think about software companies, the unit economics came from the days of like shippable software, right?

You're like, Oh, 80% margins. Cause you ship it and people run it. But more and more in particularly for infrastructure, like we've mentioned. Like the companies are monetized as services. I think AWS, or whatever platform, any scale Databricks. And the thing about, a service, especially in infra is, the Mo the margin structure actually can start to matter.

We've been doing this internal study, hopefully it'll turn into a blog post soon, but it's interesting if you take a an average. SAS company that's built on top of the cloud. The amount of contribution that cloud has to margin to the actual cogs is significant, right? It can be like double points of margin.

And as long as the company is growing, that doesn't really matter. Especially in this macro economic time or debts, basically free, like these companies that get these great valuations cause they're growing very well. But while they're doing that, whether they're taking on VC funding or. Or whatever, they're writing all of this unoptimized code, that's running on these clouds.

And then let's assume that at some point in time, the growth slows down then margin really starts to matter. And if you've got multiple points of margin, Like the impact to the market cap of the company is significant. So let's say you've got a $20 billion company that's doing this.

The difference in 10 points of margin could be, billions of dollars in market cap. And when, and then, so you hit these very interesting situations where it totally made sense to use the public cloud early on hundred percent because you're growing very quickly. Then let's say, you saturate your market and you go public or whatever, then you slow down.

Then all of a sudden, like it's hard to pay that margin tax. And then when you have actually pretty good arguments for repatriating like Dropbox, did you basically say, listen, it will never make sense from a cashflow perspective to build a data center, but this is probably 10 billion of dollars of market cap that we can save.

Oh, sorry. I'm just, David wants to do a question. Oh yeah. So maybe, yeah. Yeah. So let me explain a little better. What I mean by this, which is if you're offering a SAS service. And let's say you're running it on AWS or you're running on the cloud. And let's assume that you're doing some level of compute while you're doing it.

Like you're paying AWS. So you actually have a co like a cogs cost, like a cost of goods. So to bring on a new customer, it's expensive. On your AWS bill, for example, that hits your balance sheet as margins. And when you're a growth company, when you're invested like, by David or Ben or myself and your invest, when you're being investigated by a VC, you don't really worry about that as much because you're growing very fast and you're mostly valued on growth, but as soon as growth slows.

It really matters and there's probably, tens of billions of dollars right now, trapped in kind of cloud margins. It's something that we're very focused on. And so I don't think that we fully rationalized how much margins are going to play a role going forward, but you have to believe with all these SAS companies, IPO, knowing that there's going to be a reckoning and Ben, I know, you've been watching the data space as well.

There's been a lot of discussions around margin. I'm wondering if you think is there a big reckoning coming where this has to be reconciled? Can we just grow forever? Is this like a non-issue. 

Ben Horowitz: [00:40:34] I think that, it's funny, the data space is a little bit special with regards to your ability to get out of the monster clouds because of the data gravity problem in that let's say you're an analytics company, you have to be proximate to the data because the thing that's more expensive than the regular, storage and compute is bandwidth.

And specifically been with. Out of the cloud. So if you are an analytics tool and you have your own data center and you're talking to say Databricks or snowflake or Redshift and another cloud that's going to be a very expensive proposition for your customers to move that data across. Whereas moving data within a cloud of course is nothing.

So I think that's a special case, but I do think that. On the other hand for things that don't necessarily interact with a lot of other kind of components in the cloud yeah, like it's real, it is a gigantic amount of money at scale. So I do think that's going to. FIA thing for certain companies.

Here's 

Martin Casado: [00:41:30] this thing I'm just kidding, just from an industry standpoint. And here's what I'm curious on. We're in this very special macro economic time where debt is very cheap, right? And the public markets are valuing. Companies based on growth. And then of course, on the private markets, we value companies a lot on growth because it's so hard to even calculate margin.

It's just so early, and they're just mature. And so let's assume that there's this whole cohort of companies in the last five years that are funded by VCs. They get to scale, they go public, they all have great growth and then they start to slow down and then you find that.

There's you know, 10 points of margin plus captured, by the cloud, this could be hundreds of billions of dollars of market cap, I just wonder if there's going to be a reckoning here. And does that mean that we're going to see a layer that optimizes that? Are we going to see repatriation or, it's a very open question in my mind.

Ben Horowitz: [00:42:21] And so what is the thing that causes debt not to be cheapest? Also, that's a macroeconomic question. These countries are printing money, like crazy good. It would basically destroy the country to raise interest rates now. It's a very. Dangerous given how much money we owe, but 

Martin Casado: [00:42:39] I feel so much, it's an odd thing to say, but I feel so much of the current state of infrastructure is being dictated by exactly this, which is, it's a weird thing.

It's basically yeah. We're printing money. And a lot of it's going to AWS because we don't care about margins. That's basically the collection of it because we're just focused on growth. Uh, but if the growth slows and or infrastructure interest rates go up, I do feel we're going to have Like a significant portion of the economy is going to be trapped in this layer.

And I think it's very interesting to see whether, we can dig our way out of it. Are we going to see optimization companies? Are we going to see repatriation? This is going to blow up infrastructure yet again. I don't really know the answer. 

David Ulevitch: [00:43:14] I don't know the answer either, Martin. One thing I think people underestimate is that there is a level of competition in the infrastructure space and people forget that pretty, but it's easy. It's easy to actually remember it when you need to. So I think most of us who are on boards have gotten an email at one point or another from a company that we've invested in.

It says, Hey, great news. We found out about these things called reserved instances. And all of a sudden we've lowered our Amazon bill by 25%. We're saving millions of dollars a month. And you're like, great. You figured this out now. It's fine. It's good. But the same thing happens when they say, wait a minute, we just got an incentive to move to Azure and we can save 30% on our bill.

'cause like to use the Bezos quote, " your margin is my opportunity" that will eventually get used against all the major cloud players. And even the old school providers, like the Equinixes of the world are now starting to offer more and more sort of dynamic like services elastic like services.

And I think that the margin issues and tends to solve as there's contraction in the market. And that people find ways, whether it's an overlay tools, the pride optimization, and as the virtualization to us continue to get better and better. You're just going to see more and more people be able to move more nimbly and that'll push pricing down.

Martin Casado: [00:44:26] I think. I'm going to take the, I'm going to take the counterpoint on that, which I think is totally reasonable. A reasonable thing to say is it's not an issue right now. And once it becomes an issue, like we can move workloads around or optimize here's okay. Listen, you're all you all are hearing it now.

Here's my prediction, which is there's so much money at stake. I think fractions of, a trillion dollars, right? So much money. If the macro economic environment changes that we're going to see a thermonuclear detonation on the cloud, it's just going to be like, Holy shit. We've got just, so much money in market cap trapped on this layer.

And yes, we could do these 20% optimizations, but Oh, by the way, if I have a. Focused. If I built my own data center, for example, I can drop it by a factor of 10 and we're going to see just like the cloud kind of decimated PCs. I think there's a chance that we see it moving back. Who knows, but I do think we're talking of numbers of that size, right?

David Ulevitch: [00:45:15] Yeah, but I'm okay with people building their own data centers. We, when you and I were having that discussion about parlor and infrastructure, I think that the more structure we have out there and the more choices, the better 

Martin Casado: [00:45:25] that was good, every section stays relevant.

David Ulevitch: [00:45:27] There's a related point, Martine that in the discussion of sort of margin, which is that. Usually the cost of the company is incurring to run. Their service is not at all related to the way they price their offering. In fact, usually in the in fact, the best companies don't connect those things at all, because usually the customer has some axes of pricing that might be users or storage, but the actual cost of the service might be based on compute cycles or processes or something else.

And th that's actually, I think where companies need to be a little bit more in tune with what their actual costs are and what the incremental cost of a customer is, and that they can certainly do independent of the macro environment. 

Martin Casado: [00:46:05] I dunno. I just think everybody talks about technical debt and we're like, Oh, it's bad  but like it's all fun and games until it's a $200 billion problem. And somebody looks like, literally somebody looks at the, like the U S economy. And it was like, Oh my God. Again, like some percentage of market cap is being owned by AWS or whatever.

And then I think that people get very motivated to make big dramatic changes. And we'll just see.