Invest Like the Best with Patrick O'Shaughnessy - Martin Casado - The Past, Present, and Future of Digital Infrastructure - [Invest Like the Best, EP.280]

Starting point is 00:00:00 This episode is sponsored by Teegas. Teegas is the new digital hub for market intelligence. The Teaguez platform empowers investors and corporate development teams to invest smarter by pairing best in class technology with the highest quality user-generated content and data. Teague's content is powered by many of the world's leading institutional investors, where their expert calls are recorded, transcribed, and uploaded to the shared platform, leading to the highest quality content and datasets.

Starting point is 00:00:26 Teague has also recently acquired BAMSEC, which will allow users to see, seamlessly toggle between financial data, management commentary, and expert interviews as they get up to speed on a company. Any customer who signs up for TIGIS before May 31st will receive a free BAMSEC license as part of their subscription. Find out why a majority of top firms are using TIGIS on a daily basis. Head to TIGIS.com slash Patrick for your free trial. Today's episode is sponsored by Brex, the integrated financial platform trusted by the world's most innovative entrepreneurs and fastest growing companies. With Brex, you can move money fast for instant impact,

Starting point is 00:01:01 with high-limit corporate cards, payments, venture debt, and spend management software all in one place. Ready to accelerate your business, learn more at brex.com slash best. That's BREX.com slash best. Hello and welcome, everyone. I'm Patrick O'Shaughnessy, and this is Invest Like the Best. This show is an open-ended exploration of markets, ideas,

Starting point is 00:01:25 stories, and strategies that will help you better invest both your time and your money. Invest like the best is part of the Colossus family of podcasts, and you can access all our podcasts, including edited transcripts, show notes, and other resources to keep learning at join colossus.com. Patrick O'Shaughnessy is the CEO of O'Shaughnessy Asset Management. All opinions expressed by Patrick and podcast guests are solely their own opinions

Starting point is 00:01:52 and do not reflect the opinion of O'Shaunacy asset management. This podcast is for informational purposes only and should not be relied upon as a basis for investment decisions. Clients of Oshonacci asset management may maintain positions and the securities discussed in this podcast. My guest today is Martin Casado. Martine is a general partner in Andreessen Horowitz, where he focuses on digital infrastructure. Before joining A16Z, Martin pioneered software-defined networking and co-founded Nakira, which was bought by VMware for $1.3 billion in 2012.

Starting point is 00:02:25 health. Martin has studied, built, and invested in digital infrastructure his whole career and is the perfect person to discuss the most interesting aspects of the industry. Please enjoy this great conversation with Martine Casado. So, Martine, I think we might call this episode the past, present, and future of digital infrastructure, which is the world that lives beneath all the stuff we all spend all our days using. And probably almost no one's ever really thought about, but represents enormous markets in terms of revenue, market cap. People have heard of things like AWS, et cetera. And so I'd love to begin by you giving us a state of the union of digital infrastructure today. The major aspects of it that you think are more important. And I like starting

Starting point is 00:03:07 in the present, then maybe we'll jump to the past. But if you had to describe this to a new initiate into this part of the business world, how would you describe it and its importance? The primary narrative for the last decade has been the cloud. You have basically an old of three providers, Microsoft, Google, and AWS. And they provide all of IT, which used to be provided by central IT teams. They used to be the case that you'd have these IT teams and they'd like rack servers, they provide services. And now this stuff's just available online.

Starting point is 00:03:38 And not only is the compute, the storage, and network, which are traditional infrastructure, but higher level services on top of those. And so today, if you're going to build an app, you don't have to know a lot about infrastructure. You don't have to directsize servers. You don't have to know about how to do. to build a database. You just use those and then you focus more on the business logic. How would you put chapter headers on the stages of cloud adoption going back to, I think Azure and AWS or sort of mid-2000s, 5, 2006, thereabouts, relatively speaking, a short story.

Starting point is 00:04:09 What do you think the major eras of the cloud story have been so far? Right before the cloud, of course, everybody ran their own internal IT, right? And so they kind of write their own servers and their own wiring closets, the cloud showed up. And the early usage was what you would typically find in a technology early adopter ecosystem. It's more new projects and startups and hobbyists and the average workloads were relatively small. There was an exception to that, of course, Netflix is a very famous one, which went all in the cloud very early. But in general, that was what it was. This is like, in 2005 to 2010 time frame, it still was very experimental.

Starting point is 00:04:49 And a lot of the time, there's big discussions on whether the enterprise was, actually go into the cloud. When I ran Network and Security for VMware, which is 2012-2016 time frame, I think that was the more mainstream adoption of the cloud. So you saw large organizations, traditional enterprise, moving workloads to the cloud, very serious discussion for the Fed and the government. So it became a mainstream way of doing things. If you were a large organization, you didn't have a cloud strategy. I mean, you were even considered a laggard or a special case. So that brings us 2018, 2019. And now we're seeing a shift where the move to the cloud has implications on your finances because now, instead of you being able to buy a physical asset and internalize that,

Starting point is 00:05:34 you're basically paying a portion of your income to a third party. And so now there's a lot of discussions around how do you optimize the use of cloud. Is the right thing to go all in on cloud? Is it something that you do a portion or whatever? And so I just want to make one quick analogy. which is I always view companies going in three stages, just the product stage, the sales or growth stage, and then the operation stage. And the product stage, you're finding product market fit. The stage is you're getting to repeatable sales and growth.

Starting point is 00:06:00 You don't really worry too much about unit economics. And the operation stage is when you care about unit economics, and you go into multiple products, you do all the operation and complex things. The cloud had gone through the exact same three phases, which first was trying to find product market fit, which tended to be with the new projects, then it went to the growth phase where everybody went all in,

Starting point is 00:06:18 and it didn't worry about the implications to the economics of business. Now we're in the operations phase where we're starting to rationalize all of that. Maybe tell the story of Dropbox, which I think as an individual company is a great example of cloud isn't just some panacea. It has incredible benefits in terms of how quickly you can get going,

Starting point is 00:06:36 outsource the reliability to somebody else that's just focused on the SAWS or whatever. But from a cost standpoint, it can get really out of hand. And I think Dropbox is a good, and probably unfamiliar to most tail, of going the other direction. There's basically two trends that happen at the same time. And it's important to understand those two trends to understand what happened at Dropbox

Starting point is 00:06:54 and actually a number of other companies, too, as not just Dropbox. So the two trends are the following. The first one is Cloud, which we talked about. The second trend is SaaS. And specifically what's unique to SaaS is, is before if you were a software vendor, you have built software and you'd ship software. And somebody else would run it on their own infrastructure. So your cogs, your cost of goods as a software vendor did not include the infrastructure that it was being run at because it was being run on somebody else's infrastructure. So, for example, my startup, we built software for networking. We'd shifted other people, run it on their infrastructure. However, if your SaaS, if your product is software as a service, then part of your cost of goods is actually the infrastructure.

Starting point is 00:07:38 So if someone comes, I've got a SaaS side of someone comes and uses it, then they pay me some, and then I pay say, I do the US portion of that. That is a change of cost structure. So the are very different. So while the cloud is getting adopted, all software is going from basically on-prems SaaS. And in some cases, and there's many of these cases, it turned out that it was very tough to get software margins just because the cost of the cloud services on the back and they're so high. So in the era of shipping software, we'd all say these companies have 80% margins because you basically write the code once. And it's free to copy bits. So you just ship it to everybody else, especially in infrastructure.

Starting point is 00:08:16 There's many companies that felt like they're basically reselling a thin layer on top of AWS or one of the big clouds and then paying a large portion back to them. So, for example, I know multiple companies that are household names where they've got product lines that have zero percent margins because all of the money goes back to the cloud services hosted on. So Dropbox very famously had this situation where S3, which is the storage layer on Amazon, is not optimized for this use case of many small objects. They found that they're paying a tremendous amount. Now, they were a very large user of this specific use case. AWS was not optimized for it. They decided to build their own internal infrastructure and probably saved the company at the time

Starting point is 00:08:59 by moving off the cloud and taking it internally. Literally what they did is built the equivalent of data centers or like in AWS. They were just building some series of places that had the same infrastructure, but they controlled it and was optimized for their use case. Dropbox is an often used example. It's a very specific use case. And the things to highlight is there's a subset of the workload. Like, listen, if you're just doing a webpage, it doesn't make sense to do your own infrastructure. If you're adjusting some kind of vertical SaaS, that's not compute heavy, it doesn't. But many workloads that say if you're doing music serving, a lot of data intensive stuff, especially if you're one of the world's largest websites, then you may have

Starting point is 00:09:40 a certain workload amongst all of your workloads where optimization really matters. And this is the case of Dropbox. What they did is they noticed for the storage that Amazon, you know, this is not optimized for their use case. It was a large compute workload. And then they built their own infrastructure to support that. So the own data centers, their own servers, and then their own software on the back in order to do that.

Starting point is 00:10:03 But again, it's not like the entire company moved over there. It was the specific workloads that they moved over. There was a really interesting thing that you wrote about. the interesting concept of lost market cap of companies that were big users of the public clouds. I'd love me to walk through that concept because you mentioned maybe the save Dropbox the company. I get that that's a very special, specific case. But it sounds like there's a bigger story here of lost margin and therefore lost market cap

Starting point is 00:10:28 because of the use of public cloud. I'd love me to walk us through that. We did this analysis. It was a very simple analysis, which we said, okay, right now there's a tremendous amount of money that SaaS companies spend on cloud. And let's say if they brought the inside and they were able to drop those costs by half, which most people agree that you can drop the cost by half by bringing inside. If you could do that, what would that do to the stock price? So normally when people look at this problem, they say, well, if you bring this inside, yes, it'll save you money.

Starting point is 00:11:01 You'll say 50%. That money won't cover the team, the complexity because that's not a lot of money. But if you look at the leverage that increase in margin does in the stock price, now you can free up for a large company, potentially a lot of money, which will flow over to cash. So it could be a big win. So what we went is we looked at just public software companies. We looked at 50 of them. We looked at all of their spend, and we said, let's assume you cut that spin in half. Then we calculated their margins.

Starting point is 00:11:29 And then we said benchmarking against other public companies, if their margins were half, what would that do to the stock price? And it turned out that it would increase in the stock price by $200 billion. It's just a tremendously high number. I think we wrote $100 billion to be conservative in the actual blog post with $200 billion. So that means if you're a company that's worth $10 billion and you can reduce your cogs by a bit, you could now become $14 billion. And then you have access to that for debt and hiring or whatever else. Because these two trends happen at the same time, you have the cloud trend as well as the SaaS trend.

Starting point is 00:12:04 I don't think there had been a lot of focus on what it does to the margin structure. And so we did the first analysis says, actually it's huge and it can impact your stock price. And so I do think, especially now in this market correction, it's a good thing for companies to start looking at. I think Netflix was one of the very famous early adopters of this public cloud infrastructure. I love the evolution of product to sales to operations. It seems like right now, given what's happening in public equities at least, that every company is going to be looking for ways to save money. Do you think this becomes a huge headwin for the three public cloud providers that everyone's going to say, wait a minute, if we can do that consolidation, this is the time to do it?

Starting point is 00:12:43 The three big cloud providers are going to continue to grow and continue to preserve margins. And the reason is the following. We're still in relatively earnly endings for the cloud. And there's a lot of workflows that should be in the cloud that are not that are going to continue to move there. I also think a lot of the companies that can benefit the most from these already have very much sweetheart pricing for historical reasons. So they've got a lot of pricing power.

Starting point is 00:13:04 And so I think even if they do move certain workloads off the cloud, I don't think that the margin impact is going to be as significant just because I don't feel like a lot of the margin comes from those. By the way, there's one more trend I think that we should also talk about, which is you had the cloud and then you had SaaS, but there's also been the evolution of things like Kubernetes or technologies that enable you to run on clouds and between clouds that's also matured. So that's also matured to a point now that makes it easier for folks to run their own infrastructure and move between cloud. I do think we're at a point that companies will now have honest discussions, like should portions of our workloads be run on third-party clouds, on the Big Three, on our own infrastructure, they'll have that discussion. There is the technology for them to do it. There's enough proof points of it happening that they can have the discussion. And we'll start to see that optimization. But I don't think that the workloads that are going to be impacted, let's say, in the next five years are going to be the ones that are primary margin drivers for the Big Three. I really don't. Before we get to something like Kubernetes, a little bit more complicated of a topic, I'd love to just return to super basics around digital infrastructure in the first place. And maybe you can go all the way back to the original AWS website where I think it was storage, compute, database. You mentioned networking.

Starting point is 00:14:17 What are the base level most primitives of the digital world? What are the most important big things that actually happen? Because I'd love to understand what's changed in those areas. Like compute sounds like compute. What is changing in those three, four, five base level areas? The traditional infrastructure is compute and storage and then databases. Prior to cloud, you buy a server from whatever, Dell or IBM or HP, you'd buy a switch from Juniper or Cisco, you'd buy a storage array from whoever, EMC, and database from Oracle.

Starting point is 00:14:47 So all of those have now been basically collapsed into a software layer over basically merchant hardware in the cloud. So you can get the equivalence. Just compute by PC2. You can get very flexible networking layers where you can put security policies, and that's largely implemented in software within the cloud. And then you get these scalable services, like the database services that are scalable because they're in the cloud. And so that's the bread and butter in the cloud. For a cloud is basically you take these additional abstractions, computer and appracken storage stores that were mapped into a box, and now they're just basically software services that you can spin up and they should be able to grow up to the size of the workload.

Starting point is 00:15:25 But what has also happened in the last, say, five years is a number of services been built on top of those that are higher level abstraction. So, for example, machine learning workflows, analytic workflows, different types of databases that focus on different types of query patterns. I want to do analytics, or I'm going to do LTP, or I want to do very fast queries or time series. We have seen this renaissance of infrastructure, again, which used to be tied to a box, now being implemented in a software services in a way that's much faster than we've seen historically for the exact reason that is not confined to a box. in terms of just like what literally is happening in those three key original areas,

Starting point is 00:16:04 how much innovation has there been? And maybe it's like a Moore's Law question or something. How much has that evolved on a unit basis? And how can we expect it to continue to get better or evolve? I think we've entered somewhat of a dark ages for infrastructure innovation. And the reason I say that is just because it's become so consolidated into three companies who keep it so under wraps that the rest of the industry just does. and enjoy it or doesn't have as much access to it.

Starting point is 00:16:31 So, what's the last time you saw a big innovation in servers? And I still think there's a lot of great work to do there. Just so dominated by the big three. The same thing I would say with networking. So much of this has been consolidated by the data centers and driven by that one vision. I think this stands to reason as you have this aggregation phase where like a few companies drive probably 50% of the infrastructure market. It's just something enormous.

Starting point is 00:16:58 However, what's happening is we hit this operations phase like you and I spoke about, users of the cloud are starting to evaluate, where can I get benefits that are outside of the cloud? Are there other maybe not part of the big three companies that provide something interesting? So we're just starting to see glimmers of innovation again in traditional infrastructure stuff, like compute network and storage. And so we can talk about that. I think it's actually very interesting to talk about is now we're starting to see the innovation come back out of the cloud and be exposed to more. So we're entering again this cycle of innovation.

Starting point is 00:17:29 How will that happen? It's like up against a Death Star or something, like facing these three big companies. What do you think the best entrepreneurs will do? Pick something like crazy specific and just go after a single thread. How do you think this innovation cycle will happen? All of these companies are like very strong repeat founders. And the companies are Mighty, Fly.com, and Mosaic. So what are these companies do?

Starting point is 00:17:52 So Mighty is browser as a service. I don't know about you, but right now. Even as we speak, I probably have 30 tabs in my browser. My laptop goes slow. If you use Mighty, all of that's offloaded and you get this crazy good experience, which is great for most of us, especially as the browser does more workloads. What is Fly? Fly allows any developer to run compute workload at the CDN tier all across the world,

Starting point is 00:18:16 which is important if you care about responsiveness to the users. And what is Mosaic. Mosaic is basically machine learning as a service. So they provide the ability to consider. kind of run models very, very quickly for AI-specific workloads. So what's unique about all three of these companies is all of them are doing their own hardware. They're looking at their own servers.

Starting point is 00:18:35 They're racking and stacking. And these are very, very strong founders. All of them are repeat founders. And all of these companies have great traction. So what is happening here? I think it's exactly what we've spoken about, which is there just are across the industry certain workloads that if you look at that very specific workload, the cloud is just not optimized for them.

Starting point is 00:18:53 And that provides room for, the mighties and mosaics and flies of the world to provide something that is a very attractive proof point or performance point or whatever it is with respect to the clouds. And so I don't think the answer is we're going to see a lot of prop boxes where the end customer builds around data stand. I do think we are seeing very concrete signs of third party companies coming in and providing cloud services that are just at a much better price point or a much better performance point or much more optimized for a workload. And because the cloud is growing to size, there's enough market now for solving companies to do these. And so I think this is the very beginning, again,

Starting point is 00:19:31 of a much bigger trend. Can you say a bit about your view of what I'll call API first companies, which I think a lot of people would include in this definition of digital infrastructure. If I can hire Stripe to be my payments processor by simply inserting a API into my software that I build and care about, and then there's one of these APIs that's proliferating for kind of everything. What do you see happening here? Is that infrastructure in your mind? Or does this fit into this equation? As markets grow, the unit to which you monetize gets more granular. And my favorite example of this, and it's one that maybe a cliche is worth saying, is the car market. So way back when in 1913 Ford had a factor called the Rouge River factory. And this factory literally went in on one

Starting point is 00:20:16 side was like water, rubber, and coal, you know, and like iron ore. And what came out on the other side was cars. And the reason is there wasn't a sufficiently large market for cars to actually have suppliers. You couldn't be someone that provided wheels or whatever. And if you look at the car market now, I mean, there's companies that provide nuts and bolts and you've got multiple tiers of OEMs and integrators, et cetera, et cetera, et cetera. So it's the same thing has happened to systems historically. So in the 1970s, the same company would build literally the chip, the motherboard, the sheet metal, the operating system of all the apps. And then, of course, the OS got disaggregated from the hardware and the apps got disegregated from the OS.

Starting point is 00:20:57 So now what's happening is the application itself is being disaggregated. You take any application, you blow it up and assume the market for this application or any application is so big that independent component of applications now can become companies. So what does an application do? I mean, applications to educate users. They need access controls. They need to send emails. They need to do payments.

Starting point is 00:21:19 These are things that all applications can do. So it's almost like every helpful library in an application is now becoming a company. So much so that I remember, even five years ago, you drive up 101, the heart of Silicon Valley, in the Bay Area. And you have billboards where the entire company was an API, PubNub, SendGrid, you know, Twilio. And so this is a major movement where now you don't have to build a business app to build a company. And from an infrastructure person, this is super exciting because most of the founders of Einstein, there are technical founders that are providing technical functions that are only useful to developers. In the past, it was hard to build a business that way, but now you absolutely can.

Starting point is 00:22:01 If you're in tech at all or you're an investor or all, I definitely think you should look at an application and assume that any subcomponent does have the potential to not become a company because the market is just so large. What stage of that process do you think we are in? Fulio and Stripe, everyone knows. Turns out payments and sending messages. It's almost like the equivalent of storage and compute in application building. where do you think we are in that process?

Starting point is 00:22:24 I think we're still pretty early. I mean, on average, an application uses 17 external APIs, I think like a mobile app, something like that. But if you look at the use of libraries and open source and everything else, is still incredibly high for people having to integrate external components and management operate them. So I think that there's still a long way to go, especially as we get into kind of more complex things.

Starting point is 00:22:50 So, for example, every application often requires some sort of internal policy. You can access what? And this is a very specific computer science problem. How do you build a language or a policy language that can have accesses that allows a third party to declare a set of rules and that it gets access to those rules? This is a component in most programs that can be pulled out into a company. There's a number of companies looking at that. They're just getting started. When it comes to this developer-facing tooling, there's this open-source way of building.

Starting point is 00:23:20 and there's the more proprietary close source way of building. What have you learned about what works well in which domain? And then I'd love to also learn, like, if you're an open source company versus not, what is more or less important as you think about product and go to market and everything like that? I'm starting to be of the opinion that as we move to SaaS, and that's the primary way of consuming infrastructure, which it seems to be, that open source matters a lot less. And the reason I say that is, if I'm a developer and I'm running an application,

Starting point is 00:23:50 And I need to authenticate my users and I need to authorize their access to things. And I need to send them emails or send them SMS texts or whatever. I have two options. I could download some open source package and then operate that. Or I can just use an API that somebody else operates. The secular trend is I'm going to use the API that somebody else operates. And if I'm doing that, whether or not the code for that is open source, it doesn't matter that much to me. So let's take the case of it is open source.

Starting point is 00:24:18 So even if it is open source, there's some value there. A lot of actual code to running that service has to do the operations of the service. How do you make sure that it's high availability? How do you debug it? How do you check for perform? And that operation is going to be very specific to the actual service running. So it isn't even useful. So that would never be open source anyway.

Starting point is 00:24:40 So even if I had the source code, I couldn't really use it and operated it in the same way that somebody else could or it is running it. When it comes to dev tools, things that I am specific. specifically using in my program as I develop, like that will always be open source. That's very important. But anything that's functional and offered as a service, I think the actual value of open source decreases. And what raises importance is actually open standards, which is I still want to be able to make sure that I'm not locked in to one and I can move between them.

Starting point is 00:25:06 But that's not an open source argument. That's kind of an open standards argument. So the role of open source is absolutely shifted very, very quickly in the last 10 years, largely driven by this consumptuous SaaS. And I think that we're getting a more nuanced. view of where it's useful and where it's not. Whereas 10 years ago, there was this broad consensus that open source is great and it's going to take over the world.

Starting point is 00:25:26 That just doesn't seem to be the case in the way that we all thought. So whether or not a company is open source based or not, what have you learned about successfully building relationships between a company and a big market of developers? Like, it's a very unique market to be marketing to or selling into. There's bottom up, there's top down. There's lots of ways to do it. Just generally, what should we know about developers as a bottom? class. This is still very much an early and evolving area. There's a number of truisms that everybody

Starting point is 00:25:56 knows what we're saying. So one of the truisms is developers can have more and more budget. Another truism is that traditional sales and marketing techniques still work for developers. Another truism is developers tend to be very opinionated and have strong identities and tend to create factions that localized around certain types of technologies and so forth. The best companies in the developer space have really unlocked what it means to build communities of developers. And this requires a tremendous amount of sensitivity to their problems, their workflows, their aesthetics, things like that. And so one area I'm personally very interested in is how do you go from the art, which is developed relationships. And it's very much an art. There is no rule book for how you do

Starting point is 00:26:39 developer sales, even if there are real books that don't work. So how do you kind of systemize that and get more visibility into that. We're seeing the number of interesting companies pop up. Orbit is one that I'm involved with, which like their entire goal is to help companies build developer ecosystems and then have some visibility into those so that it can kind of move more into a traditional motion. Like I said, I haven't been involved in this now for the last while. There's a lot for the industry to learn about that. Can you say a little bit about the difference between a front end and backend developer and what you think is changing in that relationship or set of responsibilities. It's an interesting one.

Starting point is 00:27:18 We use the term developer. Like there's this generic developer out there, which is not the case. There's a number of ways you can classify some. One way I think that's particularly useful is the difference between a backhand or a frontend developer. So back-end developers tend to be like system folks like I used to be. They build big distributed systems. They build infrastructure. it's less specific business logic.

Starting point is 00:27:42 It's almost like a service operation to the front-end developers who are building like the user experience. The back-end is you're building some heavy data platform or compute platform or whatever, and front-end developers are actually developing the UI and the logic that's facing the customer. And in the past, the view was the front-end developers

Starting point is 00:28:01 was mostly about presentation and user experience and a lot of the work was back-end. What's very interesting is there's probably, I don't know, 100 front-end developers for every back-shoulding developers. So what we're seeing is front-end developers slowly consuming more and more of the stack. And this has broad implications on the type of people

Starting point is 00:28:20 that are front-end developers, has big implications of the type of infrastructure that supports them. The big trend we're seeing is front-end developers are now doing more and more of the application, which is naturally resulted in an atrophy of the back-end. But that's the distinction that, a very helpful distinction to make, and this one certainly we make on the investing side

Starting point is 00:28:36 because it does dictate a lot of like, like when technologies get adopted, where and who needs what? Going back to this notion of, so if they're the consumers of these APIs or little pieces of infrastructure, I absolutely love the Ford factory example and what happens as it matures, I think it's so clean. What do you look for as an investor when you are seeing one of these, let's say, API forward or first companies for the first time? What is your method of investigation? How are you processing a new company? To wrap us all together, we talked about a trend. So there's a lot of front-end developers.

Starting point is 00:29:06 we talked about probably 100 to everyone back end. And those front-end developers, they're building more and more of the application. So in the past, they'd have to work very tightly with the back-end, more and more. Instead of having their own back-end, they can use an API from a third-party company. Let's say they're using Tully-2-SM or whatever. The interesting thing about these API companies that offer to the front-end is that the unit of consumption really is like a function call or an API call. So they almost have these consumer-like dynamics. And so the primary evaluation criterion to answer your question and why it's so different,

Starting point is 00:29:43 like in the past, if you're going to evaluate a server company, who's the buyer, what's the go-to-market motion, you know, like what's the ACV, you talk to a bunch of the buyers, you'd see if they can build the technology, et cetera. Now with these API companies, you literally just can look at what the usage graphs are. How many users, how do they monetize them, et cetera? and it's become much more of a bottoms up or SaaS or consumer type profile. And so we stopped a lot of that approach to investing when evaluating these companies. It's much less about can they build it who's the buyer.

Starting point is 00:30:16 And it's much more about however they're using it in a practice. Then you'd be very interesting. A lot of these companies, they do. They've got these kind of beautiful growth patterns just like you're looking at the next WhatsApp. They really are almost consumer-like phenomenon. What would be the most common red flags or disqualifying options? If you're investigating one of these companies beyond lack of that nice looking usage or engagement. Well, I'll tell you what I've gotten wrong.

Starting point is 00:30:40 You know, I do come from the older era where you actually evaluate the technology. You have a thesis on go-to-market. Often we've seen these companies come in and they've got these beautiful usage graphs. They haven't monetized yet. But we're like, oh, well, I mean, who's going to pay for this? Or this is just developers. Like, whatever. And we kind of talk ourselves out of the deal because we,

Starting point is 00:31:00 we know the market better than the founder. And in almost every case, I've regretted that because the reality is, and this is an internal thesis of ours, is like the graph in almost every case, this is smarter than our theory. Like, the market actually knows what it wants. These days, if one of these API companies is doing very well and the usage is great, like I'll give an example, hugging faces is a phenomenal company. And if you look early on at the usage, this thing is a rocket ship. And you can have a bazillion theory is why you can't monetize a model and you can have a bazillion theories of why they go to market is going to work. But like the reality is the market loves it and it's a great company.

Starting point is 00:31:37 For me, it's almost like a counter thing, which is I do think that this API like makes life a lot easier. So you don't have to have a grand unified theory about how things work is you can literally just look at how this thing is being consumed because the consumption has become so bite-sized. You get a lot of early signals. I think it really boils down to- Comes down to usage. Yeah, to usage.

Starting point is 00:31:55 How should these things be priced? What have you learned about actually, building the revenue model around something that looks more usage-based. All these examples, AWS, what we started with, these API companies, they tend to be usage-based pricing. So what have you learned about that? Is that the right thing? Do you think that changes?

Starting point is 00:32:11 It feels to me that apps are per C pricing and infrastructure is usage pricing, and that's basically how it is. And if you're an info and you're not doing usage pricing, you better get there. Like, you really have to. And if you're apps, if you can get away from C pricing, that seems like that's where you'll end up. I do feel that when it comes to company building, there's a few areas where there's no simple answers. There's a lot of stuff that's systematic. Like how do you hire your sales force is pretty systematic. How do you do your org? It's systematic. But one of the things is just not systematic is pricing. Pricing is actually dictated by the market and the shape of the product. And it takes months to get it right. I'll give you three mental landmarks. And I think the rest is just actual work. And so one of the mental landmarks is pricing is often fixed by the market. And so you should look at the ecosystem and the other

Starting point is 00:32:56 types of companies and how they price, and I think you should follow that model. For example, if you're building on top of Snowflake, how Snowflake charges is going to be very similar to how the customer expects to buy. And if you're building on top of that, you're going to want to align with that. And I've been in many cases where the companies wanted to innovate on their own pricing model, but the ecosystem alignment just wasn't there, and it was just painful until they had to change. I think another mental landmark is the market will tell you the price over time, but not initially. The less that you have public or the less that you force opinion on the better it is.

Starting point is 00:33:28 I do think that a lot of early sales and discusses is just to figure out pricing. Like, that's what it is. I give a goal to reverse engineer how they think about that. The good news is because the consumption is so much higher on these, and the unit of consumption is lower, like it's per API call. There's kind of a lot of room to experiment. I'd love to move a little bit into some of the biggest reasons why all of this proliferation is happening.

Starting point is 00:33:51 And the modern data world or modern data stack, as it's commonly called in the investing circles is one really interesting place to dig in. And everyone's familiar with AI and ML and their importance in applications and the exciting stuff around Dolly and TPT3 and all this kind of stuff. I would love to focus on the infrastructure portion of the modern data stack. Maybe just give us a little bit of a lesson. What is going on here? Why is this a theme with a name? What are the important parts of it? Ten years ago, if a company comes in and they pitch their dog walking company to you or whatever, if you'd have big questions on whether they could build it, whether it would scale, can they go international? you have lots of questions about their infrastructure. You don't ask those questions anymore because

Starting point is 00:34:29 if anybody wants to build a website, even if one that's globally scale, like we've got the technology to do that. So let's say in the case of dog walking, let's say I'm an investor and 10 companies come in and pitch the dog walking company to me. In almost every case, the differentiation now between these types of companies has to do with how they handle data. How do I match dog walkers and dogs? How do I do dynamic pricing? How do I do personalization? How do I run kind of acquisition campaigns? We've moved as an industry to the businesses competing on how they use data, not how well they use software. I think it's a very, very big shift, and it's actually very, very real. Again, even if you're just like how we evaluate these companies. And so what we've seen to catch up is this dramatic improvement and maturation of tools to help with that.

Starting point is 00:35:16 Because if you look at the use of data in systems, it's very archaic relative to tools we have on software systems. So the modern data stack is just a catchphrase that folks use when it comes to like modern tooling to help build these types of systems, ones where you can provide deep insights and real-time insights into running applications. And they're normally split in one of two ways. There's the analytics side, which is generally helps with human decisions, having a lot of data and doing dashboard. And then there's the AIML side where this becomes part of the application itself and provides

Starting point is 00:35:52 real-time decisions. And so just to throw out a few names, the modern data cycle is typically typified in the kind of AIML as a world, like Databricks is a keystone company, and there's an entire ecosystem around that. And the analytics, you've got BigQuery, Redshift, and Snowflick has tended to be. And for ETL, you've got DBT and 5 trans. And these are kind of modern companies using modern approaches around this to tend to address the shift. But it's all converging now, where it used to be analytics and AINL are separate. They're all converging now.

Starting point is 00:36:20 So we're actually seeing, I think, one of the greatest battles for, technical supremacy playing off right before our eyes in the modern data stack as the tools mature. Maybe I'd like to double click on a couple examples just to bring them to life. Snowflake is one that we've had Sleman on the show. It's a very popular stock, not just company, Main Street's familiar with it because of its great business story. So what does something like Snowflake do for its users and why is this unable to explode with the sort of growth that it has?

Starting point is 00:36:49 So before you just did not have, the industry did not have a solution that was. cost-effective to throw all your data into. You had to buy like Teradata data or these big data warehouses. They're very expensive. You know, you're buying hardware. And what they did is they basically offered a cloud service, which allows you to throw all of your data. And it's cheap enough that you can throw it all there and keep it in there.

Starting point is 00:37:09 Stuff like is the independent company, like Redshift from Amazon, is larger and started this. Then the BigQuery is also like a phenomenal company. So all three of these companies provide that. You actually can treat data almost like this resource that you can continue to come back to as an organization or as an application rather than something that is outside that you've got to process once because it's so expensive and then bring in later. Are there primitives in this world, maybe it's ETL or something similar to the primitives

Starting point is 00:37:37 in the cloud world that we talked about before that you think of as the core use cases or uses or whatever you want to call them? So on the operational AIML, it's the model where you want your application to be able to make decisions and those decisions rather than being something, somebody coded. It's the product of analyzing data. It's a prediction, the model prediction. But importantly, it's based on analyzing data rather than Martin sitting down and writing a number of rules or something like that. And so that's very much primitive.

Starting point is 00:38:07 And these are used for things like fraud detection and pricing, all the things that we know of. And then the entire tool chain to create that, the entire workflow to create that is becoming the standard primitive. Also, the data pipeline, which creates these models. So like from the source of the data all the way to like the workflow to create the models is becoming a primitive. And this is kind of where ELT comes in. You've got these data sources, whether it's cars or old databases or people on websites. Those get synced into the data warehouse. They go through a set of transformation. Features are pulled out. And then you actually create models based on those. So I think that entire workflow from data sources all the way to the models

Starting point is 00:38:42 being served are the primitives that are starting to evolve. But very importantly, this is so new, we're all still kind of trying to decide which pieces, what, and what categories are going to emerge. And so So we're starting to get the glimpse of a stack for me. Part of our active diligence every day is trying to figure out how exactly that's coming together. We haven't talked at all about security as like a pervasive theme or important function across all of this. I mean, people understand the importance of security. And it seems like if anything, this is going to be something that gets more and more

Starting point is 00:39:14 important as more stuff is put into digital kind of sphere. What has been the history of securing all of this? And what is most interesting to you as an investor as you think about the future of digital security? Security is always been a very tough one for me because it's been a very fragmented market, impacts everything. You know, one thing that's for sure is that security itself is also becoming an API like we had talked about. And it's becoming something that's endemic to the billing of applications. Auth Zero is a great example of this where authentication is something. We're seeing this for policy, which I mentioned.

Starting point is 00:39:52 We think this for like data integrity and data management. And so I view the movement very similar to the infrastructure in general. This is becoming primitives and programs that are starting to get stitched in, which is significantly different than what it was in the past, which is almost this afterthought and it's a box or this application that you put in that watches everything else. I think it's becoming more intrinsic to the industry. And as you think about the ways that all of this intersects with the real world now,

Starting point is 00:40:19 which we really haven't talked about. We basically talked about digital infrastructure that leads all the way up to applications at the top end and the APIs in between and all this great stuff. But it seems also like as we mature, more of this technology will apply to the real world, too, whether that's new kinds of hardware, whether that's intersection with physical goods like cars.

Starting point is 00:40:39 How do you think about that side of things and maybe the hardware world of technology? People have a hard time grasping what, let's say, AI and ML concretely provide because it's such a diluted buzzword. So for everybody's listening, the important thing to realize is that what modern AI and ML does, which we've never really been able to do before in systems, is take unstructured data and digitize it and add it to the typical logic of the program. We've never been able to do this with vision, like objects out in the real world. We've never been able to do this with natural language in the level of actually we can.

Starting point is 00:41:19 We've never been able to do this with voice or speech. That technology just hasn't existed. And so we've never been able to big programs around them. And now we can. And it's a sufficiently different workload that two things happen. First, it pushes software into the realm of the physical work. We can now see things and interact with things. And we're talking quantum leaps of accuracy improvement.

Starting point is 00:41:44 It also drives the type of hardware and software that we build because the workload is so different, right? So we've seen tons of innovation all the way down to the ASIC level, Mosaic as a company's building data setter just focused just on this type of stuff. So I think that this really is a massive impact on infrastructure writ large, not just the infrastructure, but also what sorts of application software can go after. It's very cool to consider what all that might mean. I mean, like self-driving cars is like the obvious constant. an example of what computer vision might allow us to unlock. Obviously, Cloud had this crazy impact on the services you consume. It's unlocked innovation by reducing friction. As you think about what's going on in the digital infrastructure world, period, what are you most excited about

Starting point is 00:42:30 in terms of what it might unlock in the 2020s or over the next decade that maybe we're just starting to think about? Any problem that human beings go after that's been outside of the realm of software is currently in the realm of software. And this is, farming, agriculture, oceanography, like you name it. And so I am a tech optimist and tech maximalist. I think that part of our job is to solve problems. IT has really been limited to IT, like information. And now I go from IT to just tech. You look at any industry, any industry at all, and I think that it'll be touched by this. That's to me just tremendously exciting. What's interesting, I would just say very quickly is we're still asking the question, are these still software companies? Are there something

Starting point is 00:43:17 different? So you could say, this is just software going after agriculture, and now you still have software business. Or you can say this is still an agriculture business. Or you can say it's something totally different. That's a question I'm personally very interested in. If we think back to the people behind all these innovations, are there any things that you've learned about, I guess the first meeting, the first two meetings, the first three meetings for screening out people specifically, and their potential to deliver against building one of these transformative technologies. Hard to think about being a founder is just managing your own psychology. And a lot of that is you have to have this balance between diluting yourself enough

Starting point is 00:43:58 to like not run away screaming, but not diluting yourself so much that you'll run off a cliff. And this sounds like a very high-level answer, but having lived this so significantly in my own startup, and now living it with others. My biggest early sensitivities is, is the person tackling this optimistic enough to push you the hard times, but self-aware enough to know and realistic enough to know to make very practical decisions? I actually think that's the biggest balance for all of us. How do you test that? How do you try to get a sense for that without having worked with them yet? There's two types of evaluation that you've gone to. One of them is all of these things come out in the state of the business. So it's a later stage company, you kind of don't have to. If it's a problem

Starting point is 00:44:43 they would have materialized. And I do believe that later stage investment can largely be relatively mechanical. Like, you should do the work. Diligiless leaves you all the market works, this and that. You know, early on, I think free-form discussion is just incredibly important because anything that's rehearsed is just not an indication of reality. So I personally just like to spend a fair bit of time talking about things. And often I like to talk about things that aren't actually the company itself. So I love to talk about experience backgrounds, lessons learned. Probably my favorite line of inquiries, lessons learned, especially if we've got a shared background. Like they work in the company, I know the company that work the startup that I know

Starting point is 00:45:20 the startup. So I spend an enormous amount of time there. And mostly what I'm looking for, I don't really care about right or wrong answers. I really do care about self-awareness, you know, an appreciation of the level of sophistication of problems that are being solved. If you put your investor hat on, I guess you're purely selfish investor hat, meaning you're just trying to maximize returns. And you could somehow have a crystal ball that would reveal some information about the future, which is currently uncertain, where if you knew what the future was going to hold, be super valuable to you as an investor trying to earn a return, what would you ask of that crystal ball? What would you want to know about the future that you're not sure which way it will go? I am very curious about where crypto lands. And I think, there are three potential views, right? On one end, on the most negative and bare end,

Starting point is 00:46:09 folks are like, this is all fake. It's just a Ponzi scheme, yada, yada, yada. On the extreme other end, it's a total reformation, not just of technology and companies, but an organization. This is like everything. You don't just have routers. You've got crypto routers. You don't have just storage. You have crypto storage. You don't just have businesses. You've got Dow's. You don't just have money. You've got defy. Like, everything changes. And then there's a bit of a middle view somewhere. This is a continuum which says, you know what? There's something very innovative there on the ability to build networks.

Starting point is 00:46:40 There's a number of primitives that are very innovative on the ability to build applications. There's a number of innovations on how you offer new services to consumers where you don't know the endpoints. There's a lot of great primitives consumption and monetization layer. Just like social was primarily a consumption monetization layer. In that future, that layer is on top of a lot of systems, but you still have, traditional computer networking storage, you still have traditional clouds. You still have to know all of those things. And it's something that's added to that.

Starting point is 00:47:10 And I think the answer to that question, given the amount of money that's already involved, is more than this. And I don't think anybody knows the answer. I tend to be in the middle where I think that there's a real innovation there. I think there's real value. I think it's a real unlock for a lot of new business applications in these cases. But I think that infrastructure itself, a lot of the traditional model still applies. you still have to build databases, you still have to storage,

Starting point is 00:47:34 you still have to understand the trade-offs of asset. A lot of these things still apply. Obviously, distributed systems, some of the smartest people in the world, are working in distributed systems, not necessarily like crypto networks, but just like the ability to distribute state or update state constantly faster, smoother, or whatever.

Starting point is 00:47:53 As an infrastructure person, when you look at the current technology in crypto networks, maybe the dominant three or four, what are you watching or interested buyer, looking at. The consensus mechanisms, the scaling ability, like what are the dimensions that you as an infrastructure person are keyed in on today? The crystal origin solved a very important problem that's traditionally not been solved practically, and that is allowing basically an anonymous set of people with no prior trust relationship to have strong guarantees on something, right? Originally

Starting point is 00:48:27 it was a ledger and then it's become more to generalize the compute. That's a very, very, very real innovation thing. And that unlocks very, very interesting business use cases like we've mentioned. But distributed systems is one of those things that you just can't paper over with a thin software layer. You can't hide it under an API. You've never been able to. There's entire languages that just help programmers manage distributed systems. What's important is what developers end up using or what distributed paradigms they end up using because that will drive the capabilities of the system. So if everybody says this is a purely distributed world and everything I write must be purely distributed, that will have some implications from the type of systems that you can

Starting point is 00:49:15 build. So the thing that I'm most interested in is a old distributed systems guy is what are the nature of the applications? Is this going to land in the realm of purely distributed stuff? It's only embarrassingly parallel applications like defyce and embarrassingly parallel application. Like there's other things that are very similar parallel, or is this going to go more to the model of general compute? Is that something people are going to do, or people are going to build like the AWS and crypto? The answer to that is actually very, very significant. You could say, well, listen, traditional distributed systems are great for building AWS, and this is going to just be the consumption monetization layer, or it actually is going to cause

Starting point is 00:49:47 innovation in the way that we do distributed programming in the future. I don't think that that's clear yet where that's going to land. When I first had Chris on, maybe it was a year and a half ago now, to talk about that left end of the spectrum, it's so compelling. There's so many aspects of it that sound like it will unlock so much, and it just seems like the pace has been frustratingly slow in terms of just actual outcomes. Do you think that that's the fair characterization? There's the promise of crypto. That's what it means to the users of crypto, and then there's the mechanism you used to get there. Those are two totally different things. And I think a lot of

Starting point is 00:50:22 the promise is actually being realized. The amount of innovation, the amount of developers, the amount of platforms, I think that has been tremendously rapid and very exciting. But I think sometimes we conflate the innovation of crypto, what it means to the users and innovation with the actual mechanisms to get there. There's this grand complication. And if you can get that with traditional mechanisms in combination with these newer innovations distributed systems, I think it's still a win. And that to me is the big question like, where does that land? So I believe that you're still going to have traditional switches and routers to make the internet. And then I believe that you're going to still have traditional storage layers and things like that for some large subset of the workload. And then

Starting point is 00:51:09 there's going to be certain portions that are embarrassingly parallel. Now, Chris and I think we're on our crypto fund, they would say, you know, Martine, like you can distribute all of those things and we're going to, and it's democratized. It's going to disrupt the traditional infrastructure. And that very well may be the case. And again, sitting on the board of all these traditional infrastructure, as I'll tell you, they're selling it to all the crypto companies which suggest there's a lot of adoption of traditional infrastructure. As we wrap up, step back and think just very big picture within the world of digital infrastructure, kinds of companies, trends, you know, et cetera, where do you think you're most bullish relative to maybe your peers and most bearish relative

Starting point is 00:51:44 to your peers? Here's one of the most bullish. I think we're just getting started. I literally think that like infrastructure is basically the beginning and the end of the, technology. Honestly, I believe that the future, like vertical SaaS ends up becoming horizontal thin layers on top of core infrastructure components. And the infrastructure unlocks the capability of these apps to do new things. I just feel like we're in early innings of this. I think we've got decades of massive, massive growth. And I think a lot of the value, it's not most of the value will recruit the infrastructure. So I just think on the entire space, we hike bullish.

Starting point is 00:52:24 As far as being bearish, I tend to be the most skeptical that infrastructure businesses look like software businesses. AI businesses tend to look more like services businesses. I think infrastructure businesses tend to have lower margins. I tend to be quite skeptical. Not skeptical, but I tend to be a little bit more conservative on the actual profile of these. Martin, this has been so much fun. I've learned a lot in our conversations just about how things are being built in the modern

Starting point is 00:52:50 digital world and all the guts matter a great deal. and I think I've done a nice job exploring those today. I ask everyone that I interviewed the same traditional closing question, what is the kindest thing that anyone's ever done for you? Oh, man, my entire life has just been people taking big bets on me. You know, every major move has been someone taking a big bet on me. Every major positive inflection in my life. And this is from PhD to being invested in to being hired,

Starting point is 00:53:16 whatever is someone taking a huge, huge bet on me. Boy, I'll never forget that. I really think that the major reflection, points, but then through the kind of others, not through like any sort of skill of my own. Fantastic. So many people say something like that. A good reminder to pay it forward. Martinez has been so much fun. Thank you so much for your time.

Starting point is 00:53:33 Thanks, much. If you enjoy this episode, check out join colossus.com. There you'll find every episode of this podcast complete with transcripts, show notes, and resources to keep learning. You can also sign up for our newsletter, Colossus Weekly, where we condense episodes to the big ideas, quotations, and more, as well as share the best content we find on the internet every week.

Invest Like the Best with Patrick O'Shaughnessy - Martin Casado - The Past, Present, and Future of Digital Infrastructure - [Invest Like the Best, EP.280]

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.