Pod Insights a16z Podcast Trends in and the Future of Infrastructure About this episode "Infrastructure is dead", some say, thanks to cloud computing -- and a couple large incumbents sucking out all the profits in this space. Er... not really. We're actually entering a renaissance of sorts -- a "golden era of infrastructure" -- and it's one that is biased towards startups. They're the ones who have the unfair advantage, argues a16z general partner (and former Nicira co-founder/CTO) Martin Casado in this talk.
One of the fathers of software-defined networking (SDN), Casado shares what happens when you put a "software-defined" in front of everything: No silo is safe. And it's not just storage; it's computing, networking, security, management, databases, analytics, development, and so on.
So how are enterprise applications evolving? Is it really similar to what happened with consumer tools? And what will it take for startups to truly take on incumbents in this space? In this talk, previously delivered at other industry events, Casado shares his thoughts on the future of infrastructure as well as views on how 3 important trends coming together: hardware to software; software to services; and finally, the rise of the developer...
Transcript 0:00 Speaker 0 This is Martin Casado, my general partner here at Andrees and Horowitz, and I'm going to be talking about the future of infrastructure. So unfortunately, this has the potential to be a very short talk. I mean, what is there to talk about? IT probably hasn't grown or is flat to down even in the last few years. You know, there's a lot of talk about cloud. Nobody knows what that really means, but most people think it's kind of the death of infrastructure, right? You've got one or two large players that are sucking out all of the profits, very large incumbents, all of them seem to be having problems that are sell into this space. You know, IBM has posted a lot of quarter-on-quarter losses. You know, no matter who you talk to in Silicon Valley, this is what people think, right? Whether it's an investor, whether it's vendors, whether there's engineers, whether it's operators, everybody thinks, you know, come on, we're in the armpit of infrastructure.
0:49 Speaker 0 And my view is like this couldn't be more wrong. So I don't think that we're seeing a death. I think actually we're seeing a transformation. And so just like in, you know, France, when like one king died and the next king was being ascended and coronated, you'd say, you know, the king is dead, long live the king. I think, yes, we are seeing a change in infrastructure. But I think what's coming is going to be bigger and badder and more awesome. It's like a total renaissance in infrastructure and makes me very excited. So what I'm gonna do is I'm gonna talk about this new era, this golden era of infrastructure that we're moving into and why I think it is biased towards new startups, which I believe actually have an unfair advantage. But let me start from the beginning. IT is a very large market. It's a four trillion dollar market. And we all know it's starting to be disrupted by cloud.
1:34 Speaker 0 And many think that the disruptions already happened. But if you actually look at the relative sizes, we're just starting the cusp of this transformation. I mean, cloud, if you look at it, is something like if you take SaaS and public cloud is maybe, you know, 220 to 240 billion dollar market. So relative to the $4 trillion market, we're only about 6% of the way there. And these transformations kind of have punctuated the history of computer science. And every time these happen, one of two things seem to follow in suit. So the first one is the size of the market increases. So let's take, for example, the very beginning was mainframe. Only a few companies could afford the mainframe, right? So maybe you're saying, you know, maybe a thousand companies. Then you move to the microcomputer, which broaden it to more companies and then to the PC, which kind of put a computer on most desktops.
2:19 Speaker 0 So if you look at the relative size of the markets, when, you know, you've got a few large companies to more companies to every desktop. So you say maybe you had two billion PCs out there. But then now you're seeing the shift to cloud and to mobile and then to internet of things. And if you think about it, if you go from say every PC of two billion to mobile, you've now connected every person. We keep hearing the number, something like 50 billion phones are going to be out there at a period of time. And you go to internet of things, you're talking trillions of potential endpoints that everything gets connected. So the size of the market is going to increase. So the second thing that happens in every time that you have one of these epochs is the existing incumbents from a previous era, very rarely make it to the new one, right? If you actually look at the names were relevant in the mainframe, they do entirely different things today.
3:07 Speaker 0 And so I think that we're starting to see kind of this big disruption happen now. And both the market size is going to increase, but also the companies that are going to persevere, the companies that are going to rise to the challenge and become the new incumbents are those that are just getting started today. And so as I alluded to previous in the talk, I think that not only are we going to see a whole new set of challengers, which is really exciting, but I don't think it's ever been, like in the history of computation, I don't think it's ever been easier for new entrants, especially small technical startups to come and get a foothold. And I'd like to describe why that is. Okay, so I'm going to be talking about broadly three trends that are causing this disruption and creating a lot of the dynamic. And this is the move from hardware to software, the move from software to services and the rise of developer.
3:54 Speaker 0 So I'm going to start by talking about each one of these in turn. So the first trend I want to talk about is the software-defined movement. So there's been many definitions of software defined, so many so, and it's been so diluted by marketing literature, I've kind of lost track of what they mean. So I'm going to come up with my own working definition. So the working definition that I have is if a bit of functionality can be implemented and delivered entirely in software, then it's software defined. That means it's not encased in sheet metal. That means it's not bound to an ASIC. That means entirely in software. It's software defined. Now, infrastructure is going to a similar transformation to what consumer tools went through in the early 2000s. So do you remember uh GPS devices, right? These are these fixed function things. If you rented a car, you're gonna go stick it in because you didn't know where you're gonna go, and then it would show you where to go, right?
4:43 Speaker 0 So back in the early 2000s, most digital devices in our lives were fixed functions, right? You'd have GPSs for, you know, to find things. You would have devices that are special purpose for games, you had special purposes for healthcare, you had special purposes for learning. Every device was fixed function. And all of that changed and changed dramatically so around, you know, 2005, 2006, 2007 when smartphones started to take off. So I think one of the good examples of this is Waze. So previously, like I mentioned, GPS devices were hardware, they were fixed function. And Waze started in about, I think it was 2007. And it went through three rounds of funding, probably got 70 million dollars in total. And within five years, it had amassed, I think 50 million users and was sold to Google for a billion dollars. And if you look about what happened there, is you had a small team of software developers that just focused just on innovation, just on delivering software.
5:32 Speaker 0 They didn't have to deal with, you know, hardware supply chain, dealing with all of the difficulty of building a hardware device and focus just on what they were doing best. And so you create billion dollars of value very quickly. And of course, this isn't just ways, this is many different areas of the consumer device. I mean, all of games have been disrupted. A lot of healthcare is starting to be disrupted. And so we've seen this play out in the consumer era. And if you look at like what was the single innovation or the single trend that made this possible, is that for the first time, everybody or a large portion of people had a ubiquitous platform that you could install this software on. I mean, previously it just wasn't capable. So you had to not only give them the device, but you had to give them the device and the software on top of it. And now the device had already been solved, and it was sufficiently general that it could run all of these applications.
6:21 Speaker 0 And that was the inflection point that caused this massive destabilizing in the consumer industry. Okay, so what's really fascinating is infrastructure is following a very similar pattern. Traditionally, there was no standard interface for inserting software into the enterprise. I mean, the only standard interfaces really were at the network level, right? Stuff like Ethernet and IP. And as a result, if I wanted to build some functionality, say security and bring it to a company, the only approach I could take that was reasonable was to take that functionality, put it in a box, and then have that box connect to the network. I mean, it's the only insertion point that I had. So in the last, say, five, six years, this has changed dramatically so that we've seen a massive unification and a maturation of abstractions in the software stacks at companies. And this has happened at every layer, right?
7:07 Speaker 0 We're seeing maturation of insertion points for infrastructure at the hypervisor. Um, we're seeing it at the operating system. Uh, we're seeing it in many platforms as a service and application platforms and the cloud management platforms. So, what does that all mean? I mean, that sounds like a lot of technical words, but what that means is if I want to build functionality just in software, I actually now have insertion points that I can reasonably expect to be there in any company. For example, I can assume they're gonna be running Linux, or I can assume that they're gonna have adopted some open source that I know. So now I don't necessarily have to take my functionality and put it in a box. And that really reduces the amount of investment I have to do in building the product, the amount of investment I need to do into distributing that product, but it also allows me to really focus on my core competency, which would be like software development for whatever this is.
7:57 Speaker 0 I have a personal experience with this. So I uh founded a company in 2007 called Nasera. And the idea at the time was to take network functionality, L2, L3, L4 through seven services of traditional networking functionality and deliver it entirely in software and run it within the hypervisor. And at the time, people thought this was nutty, actually. I mean, I remember very well. So, you know, I was raising money. I talked to one of the investors. He was actually very supportive, ended up being a very supportive investor. Um, but we were having drinks over at the Rosewood by Sand Hill, and I was saying, listen, I think that the right thing to do is not to worry about building hardware supply chains to implement functionality and software. And I remember he looked at me and said, you know, Martin, like I know you're a super excitable kid, and I know you're a really smart kid, but like in my heart of hearts, the only way you'll ever be able to do this is by putting it in an appliance.
8:47 Speaker 0 Because that's the only way that customers know how to consume it. It's the only way that you can actually have unification of hardware. Otherwise, you have to deal with all sorts of different environments. It's kind of the only way that customers know how to buy it. And at the time, I thought, well, okay, you know, there's a new way to do it. And over the course of Nasira, which was five years, I actually got in an appreciation for that. It is true that that is the state of the art or was the state of the art, but you've had so many changes with things like say OpenStack or the adoption of Linux, the rise of hypervisors that the environment actually changed. And what we were able to do over the next, you know, five years and only 40 million dollars in funding was to build a billion dollar company who, you know, at say six months prior to this recording, that product that we built all in software was at a 600 million dollar run rate.
9:35 Speaker 0 And so this was this massive shift from people thinking that you can only do things in hardware to now this very, very large product that's generated, you know, billions of dollars of value. And this isn't limited to Nasera. Now, this year wasn't a one-off. You actually see this disruption happening at all layers of the stack, whether, you know, it's security in the operating system, whereas storage, whether it's the actual network infrastructure itself, compute databases, all of these verticals that have traditionally been tied to a box, traditionally been tied to hardware, are now being disrupted by software only players. And that to me is very Exciting because that means to me that if you're gonna do a company and you believe that you understand a disruptive technology that can be delivered in software, it's actually practical for you to do it and you have a major advantage because you can take less funding to get started.
10:23 Speaker 0 And you can focus all of your effort on just that software project. Okay, so as I alluded to, you know, of course, software reduces RD and it helps you to distribute it. I think it's worth just focusing a little bit on how much easier it is to deliver. So let's say that you were gonna do a startup or you wanted to develop a product and you want to tie it to hardware. I mean, the reality is the hardest supply chain has a lot of components and takes a long time to get from, you know, where you originally create it to the customer. I mean, let's say I was gonna build a new type of security functionality and it was gonna be delivered in an appliance. Well, most likely there's an ODM in Taiwan that would, you know, take the software that I built and maybe put it on a special board, wrap some sheet metal around it, that will go in shipping containers, those shipping containers will go in boats, those goats across the ocean, you would unload it, those boxes would then go into a warehouse, that warehouse would probably unload it onto trucks, and those trucks go to other warehouses around the country, and then eventually it would get to the customer.
11:21 Speaker 0 And anytime I wanted to change something that was fundamental to that, unless I could do software only upgrade, which is very rare, but something fundamental, you've got to refresh that entire thing. So it's taking months and a lot of money to build out all of this inventory and to deliver that functionality to the customer. But when it comes to software, I mean you can literally send an email with a link where they can download that software and try it locally without having to do all of these things. And this is why this disruption is so powerful, just like it was in the consumer device, just like it is like ways compared to GPS's, because functionality can be tried incrementally with low cost without having to deal with the hardware at all. So moving things to software is a huge advantage for distribution and for supply chain, but actually supporting software in the field is still very difficult.
12:08 Speaker 0 Uh, you don't have control of, say, the customer environment. These can be any type of hardware and any type of configuration. They may or may not have skilled administrators, debugging remotely can be very difficult. Customers upgrade at very different cycles. So you may have to support many versions in the field. And so companies that shift software, often over time, a large part of the RD is just focusing on this issue of having software in the field, and they build these large post-sales operations teams as a result. And so the next trend I'm gonna talk about is tackling that issue. So the next trend I want to talk about is the movement from software to services. Okay, so in the previous trend, I talked about how consumers moved from hardware to software and infrastructure is doing the same thing. I'm gonna do a very similar analogy, but in in enterprise applications. So in the late 90s, uh, because of all of the problems I just mentioned, applications started to move from on premise installed to as a service, right?
13:00 Speaker 0 And again, just to reiterate, you know, if you do something on premise, you don't have control of the environment, you may or may not have skilled administrators, remote debugging is hard, you have multiple versions, et cetera. And this is changed the way that we think about applications and clearly as a broad trend now. I mean, the vast majority of applications that are created today are as an app or as a web service, for example. And so again, we're seeing the same type of transformation happen in infrastructure. So as I mentioned, at every layer in the stack, we've got the standardization, we've got this maturation of abstractions. These abstractions allow for insertion points. And these abstractions and insertion points can be used to provide core infrastructure, things like security, things like network services, things like identity as a service. And that can either plug into, say, on prem installments or SaaS installments.
13:48 Speaker 0 And from a startup standpoint, this is fantastic because it reduces the amount that you have to invest in post sales and managing multiple versions. It also makes it much easier to say upgrade and push new features out to the customer. So we believe this is a broad trend for infrastructure. And that is providing the infrastructure as a service. And yes, you may have an on-prem installation, but often even that is treated like a managed service. Of course, the canonical example for this is Amazon Web Services. But what I think fewer people understand is how many areas and how many verticals have companies behind them that are going after large areas, delivering them entirely as a service. And this again is core infrastructure. Okay, so the last trend I want to talk about is the rise of the developer as a buying center. Now, this is an often talked about trend, but I think it's not often really well understood.
14:34 Speaker 0 So I want to start kind of very basic and work our way up to what that means. Okay, so as I've mentioned, core functionality is moving from having been part of hardware, being attached to hardware to software. And as a result, it's being installed in N stacks, right? It's being installed in the operating system and the hypervisor, part of the cloud management stack, or maybe it is an application library. And as a result, it's moved into the domain of the developer, right? Now now developers are starting to get familiar with these tools because it's being involved in the applications that they're developing, for example. So as a result, developers are now having a lot of influence on choosing infrastructure components on influencing budgets. So this is certainly not always been the case. So take me for example. My first job out of college was working in computational physics at a national lab.
15:20 Speaker 0 I worked in large simulation codes, I was an application developer, uh, and I didn't have budget for a pencil, right? I mean, you know, I got whatever central IT gave me. I didn't even have root on my own machines. I wrote code. But now, because functionality is moving from hardware into software and deployed in the endpoint, it's very relevant to developers. And more and more they're able to determine what infrastructure components are implemented and how they're used and what is procured. And this changes everything. All right, so now I want to describe why it changes everything. So let's say you're listening to this recording and you're like, you know, Martin, this is fantastic. This is the time. I'm gonna do my startup and I'm gonna listen to everything you've said. I'm gonna go implement some infrastructure technology, and I'm gonna implement it all in software and I'm gonna deliver it as a service, and then I'm done.
16:07 Speaker 0 Unfortunately, you're not. Unfortunately, the most difficult thing with starting a company in the enterprise space is actually bringing whatever you're building to market. And the reason is is the incumbents have pretty much the entire go-to-market side locked up, all aspects of it. You know, many of the incumbents have many decade longs or even century-long relationships with customers. And this transcends any single salesperson. The procurement processes are very complicated, often very difficult to navigate. Often incumbents' tools come with certifications. These certifications are adopted by operators, you know, say, for example, uh CCIE operators have learned these certifications, it's give them job promotions and they are very, very tied to the certain product lines as a result. For example, they'll put the certification number on their business card. Incumbents have deep relationships with analysts that go back for decades.
16:55 Speaker 0 The channel, the channels that take the products and resell them to the enterprise, they are already carrying tons of product or a lot of revenue from any particular incumbent. And so the company has leverage of that channel. And then, of course, all of the products that have been produced and have been around for a while have all of the interfaces you need for operations. And so it's very, very difficult for a startup to come in and enter that, right? You don't have the relationships. You're not probably not even in the procurement system. And if you are, you probably don't know how to navigate it effectively. Uh, it's difficult to go to somebody that works very hard for certification and say that that's no longer valid. Um, the analysts don't know who you are. The channels have no incentive to carry your products. They're already carrying billions of dollars and making tons of money off of an incumbent product.
17:42 Speaker 0 And you probably don't have any of the interfaces you need for operations because you focus so much on getting the core functionality done. And this has been one of the most expensive and most trying parts of bringing a product to market in the enterprise. So developers having control or influence over budget changes all of this. And I think this is probably of all of the things that I'm speaking of, the most significant and the most significant shift. And I'd like to explain why. So developers have very different aesthetic, a very different procurement aesthetic than an operator does. They aren't generally influenced by analysts. They don't have specific certifications for specific products. I mean, developers normally have a computer science degree, for example. So they're much more about fundamentals and primitives than I've learned how to use incumbent product X. They actually don't care about long-standing relationships with sales reps.
18:30 Speaker 0 I mean, what is the idea that a developer has when it comes to buying something, right? Well, normally a developer, say a developer is driving home, they've got a great idea. Well, they swipe a credit card, they go on Amazon and they implement it within Amazon. So to them, procurement is something that should be very quick. And they certainly don't want to deal with like the slow procurement process, right? So developers are very different buying center. So it's probably good to ask what do developers care about. Well, they care about traction among other developers. They care about community, they care about open source, um, they care about technical elegance. Actually, they care about low friction to adoption. Like developers have kind of been bred in this era of like infrastructure on demand, largely because of these big IaaS players. And so if they want something, they want it immediately, which almost suggests that you want to provide it in software or as a service.
19:19 Speaker 0 And so it's just a very, very different procurement side. And because of that, the old go-to-market machines that the incumbents have relied on for so long are starting to become obviated. So as a result, with developers pushing and influencing purchasing decisions, we've now had these like very fast sales cycles where you can really get access to many customers, the top of the funnel through things like open source and reaching the developer. But here's where I want to make a very important distinction, which often gets lost in this conversation. That's the following. Developers influence budget. They often don't control a lot of budget. And so you can almost think of attracting the developer and getting them to use your product almost like a replacement for product marketing. It's like the top of the funnel. So you can get them excited, you can get them engaged, you get account credibility, you get account traction, you can create a phenomenon.
20:08 Speaker 0 But that doesn't necessarily result in a in a sale. Actually, you know, the enterprise sales process still requires direct sales and feed on the street. And so our belief is that startups can get a tremendous amount of account traction through things like open source, the things like targeting the developer, but they still want to build out a direct enterprise sales force if they want to go ahead and generate a lot of revenue and monetize that way. And you know, this actually aligns with the philosophy of the firm here at Andreas and Horowitz. Our focus has largely been looking for the technical phenomenon, the technical founder and then building around kind of a strong go-to-market engine and go-to-market know-how to help drive things like direct sales. And so just to reiterate the point, yes, developer buying power is very important. Yes, that can get you into accounts and account traction, but you still have to build out an enterprise sales in most cases in order to monetize that.
20:59 Speaker 0 So as a result, we've seen lately this Cambrian explosion in developer-centric startups, right? I mean, I've got here a list of a number of companies on our portfolio. There are many more out there also in the industry. And these companies are generating a lot of revenue and lots of valuation by focusing at different buying centers. And I think we're seeing growth faster than we've traditionally seen growth in the infrastructure space. So listen, everybody has seen virtual cycle graphs. Um, but you know, we believe that this really is a set of trends that feed on each other. So as more functionality moves to software and gets delivered as a service, it moves into the domain of the developers who influence buying decisions who create more markets for things in software and so forth. And again, to reiterate the point that I've been making all along, every one of these gives an advantage to small technical startups, if you think about it, right?
21:47 Speaker 0 You know, small technical startups can just focus just on software development. They don't have to build huge back end post-sales operations centers. And then all of the buying aesthetic of developers aligns with, you know, a technical startup, things like open source, things like technical elegance, things like the ability to build community. And so we really feel like we're in this kind of epoch where infrastructure is shifting, like all of it is shifting towards this new development paradigm, distribution paradigm, service paradigm, and uh and targeting kind of a different uh customer constituency as well. Okay, so to make a final point on this, you know, listen, we're talking in the abstract here. I do think that this is a very, very broad reaching shift. I don't think that there's any silo in infrastructure that's safe from this. I think we're talking trillions of dollars of market cap and hundreds of billions of dollars of value that are going to be disrupted.
22:37 Speaker 0 I think this is gonna usher in an entirely new era of companies that are providing services and different technologies. And of course, listen, we we don't know the future. We don't know exactly which companies or what types of companies are going to exist. But if you have a good idea about what those might be, please come talk to us. Thank you.
Previous a16z Podcast: Lobbying Tech Next The Promise of AI Page updated August 25, 2026 · transcript and summary are machine-generated