Pod Insights a16z Podcast a16z Podcast | Mobile Invades the Data Center About this episode There is a shift in enterprise hardware from expensive, proprietary hardware to cheap components plucked directly from the consumer hardware supply chain. While that trend has been underway for some time thanks to companies like Google, Facebook and Amazon rolling their own data centers, the next wave of enterprise IT is taking its cues from the mobile supply chain’s collection of components -- the stuff that your mobile phone is made of. That includes everything from power-sipping ARM chips to flash-based memory. Andreessen Horowitz’s Peter Levine is joined by Coho Data CEO Ramana Jonnala and Cumulus Networks CEO JR Rivers for a discussion about the consumerization of the datacenter. Where is the hardware coming from? How will software make it all work? What does it mean for incumbent hardware and software vendors? And how will customers buy all this data center muscle?
Transcript 0:00 Speaker 2 Hello, this is Peter Levine from Andreas and Horowitz, and I'm here with Ramana and JR. Ramana is the uh CEO of Coho Data, and JR is the CEO of Cumulus. Welcome guys. What I'd uh what I'd love to chat about is the future of the data center and specifically the consumerization of enterprise hardware and what that means for the data center and how you guys think about that. So I've gone out and I've talked about sort of this uh you know this notion of the mobile supply chain eating the data center. And uh, you know, I have this big belief that mobile, the components of mobile phones will become the basis of the next generation data center, really, really inexpensive, sort of you know, cheap components that require a lot less power, floor space, cooling, but all of that stuff being sort of connected and propagated together with sophisticated software.
1:00 Speaker 2 And so um I'd love to um maybe get your guy, you know, your guys' opinion on this and maybe describe what's happening now in the data center. Um, you know, kind of is it an incremental change? Is it a wholesale change? What are you seeing out there and how are you seeing some of the trends? Ramana, maybe start with you and then JR you can jump in.
1:25 Speaker 1 Absolutely. Thanks, Peter. Great to be here. Uh definitely there's a lot of consumerization. I'd I like to call it sort of the breaking down of the data center. Uh I come from a storage perspective, right? And for the past 30 years, storage purchases have roughly been the same, which is it's like buying a car. Yeah, it's a lifestyle choice. You know, you on performance, then you go buy a high-end uh storage device or a low-end storage device. You silo yourself and then you complain about utilization. I think that's all breaking apart. What's happening is instead of people trying to buy for like three to five years in advance, they are thinking about let me buy what I need right now. You can't do that in the traditional model. That's where first it's breaking down. The refrigerator is full of disks, uh breaking down into smaller components.
2:11 Speaker 1 Uh people are saying, hey, you know, I have visibility into my business for the next year, so I'll buy like storage for the next year. It's up to you. It's incumbent on you as a vendor to make it easier for me to scale from there.
2:24 Speaker 2 Add
2:24 Speaker 1 another box, add another box. This is where the value is not as much in hardware customizations anymore. It's about commodity components and letting the software be, you know, letting the software be capable of orchestrating ease of management and ease of scale. It's incumbent on software. And the other thing, which is catalyzing this is open source. And you know, you have when you have commoditization of hardware, and you know, you're talking about the consumerization. You have a lot of devices, you have a lot of endpoints, and you also have a lot of um, you know, like on the server side, buses and disks and flash and so forth. And it's almost impossible for anybody to have for a single entity to develop device drivers and stacks and everything for that. It's actually a spread out development model, and open source is key to that.
3:13 Speaker 1 Yeah. Every company is contributing to open source and and keeping a platform, a software platform on which you can build more value ads, um, they're contributing to it as a result of which you know innovation is just taking over.
3:28 Speaker 0 Kind of building on what Ramon was saying is I don't know that I 100% agree with you that the components in the cell phone or the mobile the mobile uh you know devices are going to drive the data center of the future, but I think a lot of the tenants that allow those to exist hold true. So like
3:45 Speaker 2 give me some examples. Yeah,
3:46 Speaker 0 exactly. So I mean, some of the really key pieces are as Romana said, looking forward and buying from my business over the next three to five years. What we understand is you know, if you look at cell phone technologies and a lot of the kind of modern technologies around IP IIT, uh there are a lot of industry standard or well known industry mechanisms, an x86 processor or an arm processor or flash memory and disk. These things are well understood. So people can put these together and again in well understood ways. Kind of again to Romano's point around open source. It's it's kind of the whole industry at large moving at different rates with each other, putting things together at this moment in time, recognizing that two years from now it's going to come together a little bit differently. So this allows customers, you know, both you know, web scale type, you know, or web-facing businesses as well as enterprises to acquire and consume that technology on an as needed basis.
4:39 Speaker 0 And recognizing that I can get you know effectively what I need today, it all comes together. What I need in five years is going to be very different.
4:48 Speaker 2 Yeah. Yeah. That's that's interesting. And when I talk about just for clarity here, like ARM chips and flash, to me are innovations that really started in the mobile world and are moving over to the uh enterprise world. It's not unlike uh, you know, we saw many years ago the shift from mainframe to PC, right? Mainframes were in the data center, PCs were on the desk at someone's house. Well, the PC innovations and revolution eventually took over the data center. And I think that it's the innovation is not that you're physically going to have a mobile phone in a data center. You'll have components that have been influenced by the development of mobile phones that I think really it's already transformed. I mean, storage as an example using flash.
5:33 Speaker 2 Like you don't have a spinning disk in your cell phone, right?
5:37 Speaker 0 But what's I think what's kind of slightly counter to that, or not really counter, it's it's aligned but not 100% directly aligned is the fact that I I agree with you around the fact that the proprietary systems that exist in today's data center will open up, much like the the mainframe PC PC took off because it was a grossly open architecture. Everybody understood uh generally how it worked, and people could innovate it on it on it at scale. Yep. Whereas if I was IBM and I did the mainframe, guess who got to innovate on the IBM frame IBM, right? Yeah, well, and not only did it lock customers in, which is bad for the customer, but it also locked up innovation. It didn't allow the industry to move at industry pays. Um, one of the the most interesting things that that I find now is you know, IBM DB2, which was the reason you bought an IBM mainframe, now runs on every processor architecture across every operating system and it performs better on Linux X86 than it does on the IBM mainframe today in
6:34 Speaker 1 terms of transactions per second.
6:35 Speaker 0 Yeah.
6:35 Speaker 1 Yeah. And um the other thing is well, right, which is when you talk about consumerization, right? People say, hey, is I can't trust my cell phone to be my next server because it's going to fail. Exactly. But that's what people
6:47 Speaker 2 said about PCs. Exactly. That
6:49 Speaker 1 they're gonna fail. But the the beauty of innovation is that you will innovate accounting for failure, right? I mean, so if you're gonna build a um, you know, a big system, you break it down into 20 components. No, you will just assume that six of them are going to fail, and you'll try to figure out how best to recover from it. Yeah, but this is a
7:07 Speaker 0 C
7:08 Speaker 1 change. If you cut you off
7:09 Speaker 0 right here, because like the fundamental problem with that thesis is you can't drop the mainframe in the toilet.
7:15 Speaker 2 Yeah. Like your cell phone. The reason your cell phone fails, your cell phone is actually
7:19 Speaker 0 a pretty reliable device. It's got very few moving parts. Um almost everything is integrated. A cell phone operating in a reasonably static, stable environment. I bet you find the MTBF there is better than a mainframe. The reason your cell phone fails is because you throw it at your kid or you know, you drop it in the water or something else like that.
7:37 Speaker 1 Yeah, they I mean, from my point of view, right? I mean, if you see Flash, Flash actually fails more than a disk. Disc has been around for 50 years. Flash actually fails quite a lot now. When you design a system, right, these days, you think about how do you just you just account for it and you account for it in software and you become more innovative, right? I mean, in terms of how to make sure that I can very quickly build something that is super reliant by using scale to your advantage rather than making this this it goes back to tightly coupled with this loosely couple. Loosely couple is just better because you can then you don't get sort of obsessed with a single component failure.
8:12 Speaker 2 So it sounds like um we all believe that open components, both hardware and software will change, kind of are changing the data center. More openness allow enables for more innovation. What do you think this means for incumbent vendors? And like, you know, hey, it's obvious. Hey, we're we're you guys are telling the world here that um, you know, it's open components and open software. So what does this mean for incumbents and like what you know, what's gonna happen in your mind?
8:46 Speaker 0 I think the landscape changes um the way I think about it, which may not necessarily be the way others do is is kind of we'll call them thinner business models. People that are able, you know, like what we do at Cumulus, right? We're a networking company. We have a reasonably thin, you know, up super operationally efficient model where we leverage open source and provide you know great networking solutions to our customers. Um some of our competitors, which are kind of very large incumbents, invest tremendous amounts of money trying to protect their value add. And we kind of go out about the whole problem the other way around. We strip away the only thing only and leave only the things that matter to customers and get rid of the kind of excess value add. And then that leaves a space for you know the supply chain, meaning the you know, resellers, system integrators, and those people to kind of come along and provide value added services that really make all of this new technology consumable by customers.
9:39 Speaker 0 And you know, you might ask yourself, you know, what's different about that as opposed to going to a great big incumbent to build to buy an existing system. The difference is it it keeps everything very loosely coupled. We innovated our pace. All of the other people that come together to make a solution all innovate at their own paces, and the solution providers that the that integrate it and deliver to customers are able. To kind of have this continuum of offerings that make sense again to the customers, creating competition and effectively, you know, focusing all of us on the necessary elements and avoiding all the cruft.
10:11 Speaker 2 Mm-hmm. Mm-hmm. Ramana, maybe um, you know, you can certainly comment on that. But how is this new trend different from what's happened in the past? I mean, I've looked I we've all been around, I've known you guys for a long time. I've, you know, we've all been around, hey, open is good, and like this whole notion of open has been around for a long time. The promises great innovation. Why is this? It feels like there's a real inflection point in data center design and sort of what's happening now into the future relative to sort of what's happened in the past. What do you think the difference is now?
10:47 Speaker 1 Yeah. So I call this that, you know, if you looked at like stacks of applications, this is commoditization sort of moving up. What open source does is that actually takes a lot of your sort of the lower end layers, the things that work with hardware and just makes it commoditized as an open platform for everybody. So it's just available right out. And it's in the interest of hardware vendors to contribute and keep that platform robust and growing. Yeah. That wasn't the case 10 years ago. 10 years ago, proprietary uh systems still held some value. There was this belief, right? Going back to JR's point, which is what has happened is DB2 is just better on a Linux platform on an X86 hardware. And that sort of a commoditization of the core, you know, platform is is what is driving an innovation goes to like the higher and higher levels, which is making this this possible.
11:38 Speaker 1 It also brings costs down too, which is the overall cost of putting the solution together, once it moves from proprietary to open source, not only is it easier to innovate, it also brings in uh like cost pressures and what was you know traditionally proprietary and makes it open and available to a lot of folks. And for example, it was you know, to think of some uh things like open stack, getting the level of um, you know, open stack, open stack summit this year had 4,000 attendees, 5,000 more than doubling from uh the the past year. It's it's going to become the next VMware, to tell you the truth, because that is going to be the platform for you know cloud innovation and an application innovation and so on and so forth. The reason is again, nobody's tied in to anything proprietary there.
12:25 Speaker 1 So, you know, that that's the direction, that's the direction these things are going, anyways. That combined with hardware platforms just makes it, you know, that's that's the data center. Innovation radically increases.
12:39 Speaker 0 And I think there's the inflection point is less dramatic than you think, only because you're looking at it from the outside, not the inside. You know, having made kind of help make the sausage, um, some of the incumbent networking uh providers
12:53 Speaker 1 you
12:53 Speaker 0 know, long, long time ago, people would you know you built the operating system, you built the the all the ASICs, you manufactured it downstairs, you did everything, and then pretty soon you started outsourcing the the manufacturing, you started using industry standard components, right? I don't call them a commodity because they're not necessarily always commodity, meaning there's 65 suppliers, but they're calling industry standard components. Until we love you.
13:14 Speaker 1 Yeah. Yeah, they hate commodities. They like industry standards, right?
13:18 Speaker 0 But I mean, which is
13:19 Speaker 2 means
13:20 Speaker 0 well understood and broadly available to all commers, right? That's kind of the the moral equivalent of that. But the other thing that happened during it's been occurring over a long period of time is these incumbent providers are using a ton of open source. You know, if you look at a Cisco, for instance, the Cisco operating system, and XOS is their their data center class operating system. A phenomenally high percentage of that code is open source. They just don't tell you that it's that. It is Linux, but it's like some super old version of Linux. They don't they don't have a good hygiene around what it means to be a good consumer of open source. So like it's really hard to get access to their patches, you know, it's really super old. So they and they keep it a secret because they want to say, hey, we're the ones that provide the value ad. We control the customer, we can control everything.
14:07 Speaker 0 And and that's the part that's kind of inflecting is rec is customers are starting to recognize that hey, wait a minute, NXOS is Linux, and I'm not getting the benefits of that. Because you know, Heartlead is a great example. Heartbleed came out, Cumulus Linux, we do a you know network operating system focused on data center. We had all the patches and all the the new uh you know binaries up in about three hours from the time that we found out about Heartbleed. Cisco took about three weeks to get it up for an XOS because they had to backport it to their antiquated version of every open source uh package that they were using, and they weren't you know telling people about. So they're kind of in this really weird spot where they're using this stuff, they don't want to tell anybody because it kind of it kind of takes all the the gold off of the the outside of the whole thing, but then they're their customers suffer because that they're trying
14:56 Speaker 1 to maintain that customer mind share. And and guess what, yeah, to your point, right? Which is the next Cisco VMworld, uh, sorry, VMware, that's open source too. I mean, literally, if you looked at how much of ESX is based on open source, it's just that you know, they probably don't come out and claim it right off the bat, right? But it's very well, I mean, we come from uh the Zen world that's our R DNA. So we know exactly how how much of VSX is actually open source, which is good and you know, uh good, and that's the direction that things are going.
15:25 Speaker 2 That's an interesting observation about using open source and sort of hiding behind it as opposed to being transparent about it, and the downstream um uh sort of benefits of being transparent to customers as opposed to hiding behind it because customers now, this Heartblade example is an excellent example of how, you know, hey, like we can fix this in a in a day because like we're very transparent, that it is open source, you can use what's out there and innovation therefore increases and customers kind of get you know, cycle times get reduced relative to new features and adoption curves and all of that.
16:02 Speaker 0 Yeah, it's that that kind of parallel undirected innovation which is kind of the general thesis around open source or or effectively open elements. When you can granimal things together to create solutions, it allows you to you know each piece to innovate at scale. People have really great ideas. You know, the the coolest thing for me, you know, as as a CEO is when a customer gets a hold of of our software, puts it on some hardware, and they they go through and say, well, how would you operationalize this? And we give them this is our set of recipes, and then you know, they call back and say, Hey, guess what? Your recipes were okay, but look what we did. Yeah. And you look at it and you're going, holy God, this is really, really cool. So this customer innovated, took it in a whole different direction we ever would have thought. Yeah. Um that's kind of what makes the whole world go around now.
16:49 Speaker 2 Yeah, interesting. Maybe um, you know, let's shift to some of the business uh challenges and opportunities that you guys are are seeing relative to data center. And I'll characterize that, you know, data in a data center world, it's often as a as a young company, it's hard to crack into the data center stack, right? It's got you know, one incumbent stacked on top of another, you know, all the way up through databases and above. And as a as a young company to get inside of that very controlled stack is a is a non-trivial problem. So I'm just sort of curious. Like you guys are running phenomenal companies right now, and you intend to become part of the next sort of data center of the future. How do you see is it a different now?
17:36 Speaker 2 Do you still have to fit into the existing stack, or is it is there something else going on?
17:42 Speaker 1 Okay, I'll I'll go first. So to quote Ben, you know, Ben Horovitz's blog, right? Enterprise sales is still the same as it was 10 years ago. So you bring the new technologies and you bring everything, but it's still you go, you know, the best chance you have is new apps, new developments and and green field environments, especially for us because we come from the storage side of things, which is probably the most conservative. It's people's data, right? I mean, they don't they don't move things around that easily. But having having said that, I think this notion of uh, you know, like cloud-inspired architectures, everybody's cloud inspired now, which is which is fundamentally changing the data center. There are three elements, I mean, the way when I describe cloud inspiration, right? There are three things to it, which is scale on demand.
18:30 Speaker 1 Don't sell me what you are telling me I'll need for five years. Yeah. Sell me what I need for the next year,
18:36 Speaker 2 right?
18:37 Speaker 1 Then pay as you grow, kind of goes into that, which is make it super simple for me to add one more component 12 months from now when my business is growing and it's up to you to make sure that everything is just work seamlessly. So I want to be able to scale on demand. The second thing is to be able to perform on demand. This is a big deal in storage, which is like as you added capacity, performance tapered off. That's the income. That that's how things are. That's how things have been for a long time. It's to perform on demand. And the third thing is multi-tenancy, uh, which is, you know, in the in the old days, again, going back to my car analogy, which is your important apps, you bought expensive storage connected those two. For home directories, you've got cheap and cheerful storage, and you created these silos.
19:22 Speaker 1 The whole notion of cloud breaks through silos. It's like I want to run both my IOPHungry app as well as my lazy app on the same storage. It's up to you to figure out where each of these apps goes, whether it gets flash or disk. These are so fundamentally, right? People are thinking that way anyway. So for newbies like us who are going into the enterprise, we are able to, you know, that message sticks around a lot of the CIOs in the enterprise are, you know, they use Google Mail, they have Facebook and all that, and they do know that that infrastructure is not built on uh, you know, the way data centers are built. They know that. But still we you know, we Do go into new environments, new apps, that's where we get the most traction. Yeah. And I believe over a period of time, it's not going to be right on, you know, next tomorrow, but over a period of time, you would see this new architecture takeover.
20:12 Speaker 1 Yeah. Jarr, what are you seeing? And sort of
20:14 Speaker 2 on the networking side.
20:16 Speaker 0 Yeah, we're I mean, we don't have people's data. So if someone loses a packet, most on pretty much every protocol these days will recover from it. So we're a slightly better off place. But it's it's kind of weird because the the behavior I noticed not only around networking in storage as well is uh customers have dealt with the existing incumbents and the existing incumbents were not perfect. It's not like they woke up in the morning, they delivered perfect product and have never tripped and fell. It's actually quite the contrary. The customers have had lots and lots of problems with the existing incumbents. And so through that set of problems and resolution, they've not only come to trust the incumbent, which is fair, but they've also become the rely and and fear not having the incumbent. Like, oh my God, what happens if company X goes away? You know, something bad might happen, right?
21:03 Speaker 0 And there's people lurking under my bed. What do I do? Um so what we're seeing now is enterprises, you know, and actually, even the also all the mega scales to you know, actually all the mega skills that I know of right now have made the decision the existing incumbents don't provide very much value. So they're all transitioning away one way or the other. Kind of contrary to what Cisco says in their X number of big data centers in the world use Cisco Gear, you know, BS. Let's let's go. Let's stand up the customers and talk about it, right? Um the and so what's what's happening now is the enterprises are saying, wait a minute, uh you know, a lot of these mega skills are actually moving away and they're going here or there. So maybe the boogeyman isn't as bad, but I st yeah, they're still a boogeyman. I don't understand it, right? And so now they're starting to feel it out and sniff and understand A, you know, maybe there is, maybe their applications warrant that kind of relationship, or maybe not.
21:55 Speaker 0 But now they're trying to explore and figure out what's going on. So
21:59 Speaker 2 you think there's a lot more willingness now to try some new things based on that. I think there's a lot more willingness to look at new
22:07 Speaker 0 things. The
22:07 Speaker 2 the
22:07 Speaker 0 try shows up when you know inevitably someone looks at it and says, you know what, I I need to try. And like in our case, we see you know some of the enterprises trying this just to make the incumbents kind of kick themselves in the head. And the the response is actually the most telling part, right? When when we're up against an incumbent in a in a deal, you know, you know, now we're still pretty young. Oftentimes the incumbent will come in and win. But when they win, they're winning on our message. They're not winning on the message that they want to drive to the customer. They're coming in and saying, hey, you we can do everything cumulus can do, and we can get close to their price points. And you know, you've been using us for years, so just please keep using us.
22:43 Speaker 2 Yeah. Maybe one more uh just quickly uh to wrap up here. The war between public and private cloud. Do you guys have any sort of thoughts on
22:52 Speaker 1 that? Sure. So um, you know, it's actually two years ago that you know, there were a whole bunch of folks. Uh, and this happens with new technology all the time, which is people just go to the extreme and they say it's
23:03 Speaker 2 all public cloud, right? Everybody's moving everything there.
23:06 Speaker 1 Like, why bother ever buying a disk? And um, but the reality, things settle somewhere in the middle, right? And I think the argument now is not as much about whether it's gonna be public, private, what percentage, but the next generation admin. The next generation admin is growing up thinking about cloud. That's how they think. I mean, they grow up using you know iPhones, they grow up pushing their you know, pictures out somewhere to the cloud and being able, you know, all that access. And pay as you grow, pay as you grow economics. I mean, they're they're inspired by these. So the next generation data uh data center is going to look like a cloud, whether it's on-prem or or public, and how what's the percentage of that depends on the industry, how big the customer is, so on and so forth.
23:53 Speaker 1 But every data center is going to look like a cloud. That that's a cloud-inspired data center.
23:58 Speaker 0 Yeah, one of the observations that I have, and it's just an observation, trying to watch and understand is everybody that's been successful in offering up a public cloud, you know, the Amazons, the Microsoft's now, the Googles, um, they actually have a real business that pays for their infrastructure. And I would venture to guess, except for perhaps in the case of Amazon, because they're so big right now, if you got rid of their cloud component, they'd still be profitable. If you just chopped it off tomorrow and said you get no cloud revenue, yeah, they're they're whole. Yeah. Right. So this premise that public cloud is this ever growing, never ceasing resource, I think is a bit flawed. Yeah. There has to be some underlying fundamental that allows a company to build that stuff out. And if that thesis is true, there is probably some math behind it that says the public part of your your infrastructure offering can only get to X percent of your total infrastructure.
24:49 Speaker 0 And let's say for the sake of argument is 50%, right? So if it's 50% and Microsoft says, okay, I have X number of millions of computers that run 50% of my business, my real stuff, then I can only have that same amount that I use for kind of generic public cloud offerings at most. And I think I think there really is a ratio, a golden ratio there that makes sense.
25:08 Speaker 2 Yeah.
25:09 Speaker 1 Interesting.
25:10 Speaker 2 Well, um, thank you both. Uh Ramana and JR, thanks for uh thanks for being part of this. And uh, of course we'll see what happens. I'll let you know in 10 years from now. Okay, thanks,
25:20 Speaker 1 Peter.
Previous Company Building in Crypto: A Conversation between Brian Armstrong and Chris Dixon Next What's Next for Education Startups Page updated August 25, 2026 · transcript and summary are machine-generated