Show Notes
In this episode, we cover:
00:00:00 - Reflections on the Episode/Introduction
00:07:30 - The 5 W’s of Servers and their Future
00:30:00 - Investing in Oxide and the Public Cloud
00:36:20 - Oxide’s Offerings to Customers
Links:
Transcript
Jason: Welcome to the Perfectly Boring podcast, a show where we talk to the people transforming the world’s most boring industries. I’m Jason Black, general partner at RRE ventures.
Will: And I’m Will Coffield, general partner at Riot Ventures.
Jason: Today’s boring topic of the day: servers.
Will: Today, we’ve got Steve Tuck, the co-founder and CEO of
Oxide Computer, on the podcast. Oxide is on a mission to fundamentally transform the private cloud and on-premise data center so that companies that are not Google, or Microsoft, or Amazon can have hyper scalable, ultra performant infrastructure at their beck and call. I’ve been an investor in the company for the last two or three years at this point, but Jason, this is your first time hearing the story from Steve and really going deep on Oxide’s mission and place in the market. Curious what your initial thoughts are.
Jason: At first glance, Oxide feels like a faster horse approach to an industry buying cars left and right. But the shift in the cloud will add $140 billion in new spend every year over the next five years. But one of the big things that was really interesting in the conversation was that it’s actually the overarching pie that’s expanding, not just demand for cloud but at the same rate, a demand for on-premise infrastructure that’s largely been stagnant over the years. One of the interesting pivot points was when hardware and software were integrated back in the mainframe era, and then virtual machines kind of divorced hardware and software at the server level. Opening up the opportunity for a public cloud that reunified those two things where your software and hardware ran together, but the on-premises never really recaptured that software layer and have historically struggled to innovate on that domain.
Will: Yeah, it’s an interesting inflection point for the enterprise, and for basically any company that is operating digitally at this point, is that you’re stuck between a rock and a hard place. You can scale infinitely on the public cloud but you make certain sacrifices from a performance security and certainly from an expense standpoint, or you can go to what is available commercially right now and you can cobble together a Frankenstein-esque solution from a bunch of legacy providers like HP, and Dell, and SolarWinds, and VMware into a MacGyvered together on-premise data center that is difficult to operate for companies where infrastructure isn’t, and they don’t want it to be, their core competency. Oxide is looking to step into that void and provide a infinitely scalable, ultra-high-performance, plug-and-play rack-scale server for everybody to be able to own and operate without needing to rent it from Google, or AWS, or Microsoft.
Jason: Well, it doesn’t sound very fun, and it definitely sounds [laugh] very boring. So, before we go too deep, let’s jump into the interview with Steve.
Will: Steve Tuck, founder and CEO of Oxide Computer. Thank you for joining us today.
Steve: Yeah, thanks for having me. Looking forward to it.
Will: And I think maybe a great way to kick things off here for listeners would be to give folks a baseline of your background, sort of your bio, leading up to founding Oxide.
Steve: Sure. Born and raised in the Bay Area. Grew up in a family business that was and has been focused on heating and air conditioning over the last 100-plus years, Atlas. And went to school and then straight out of school, went into the computer space. Joined Dell computer company in 1999, which was a pretty fun and exciting time at Dell.
I think that Dell had just crossed over to being the number one PC manufacturer in the US. I think number two worldwide at Compaq. Really just got to take in and appreciate the direct approach that Dell had taken in a market to stand apart, working directly with customers not pushing everything to the channel, which was customary for a lot of the PC vendors at the time. And while I was there, you had the emergence of—in the enterprise—hardware virtualization company called VMware that at the time, had a product that allowed one to drive a lot more density on their servers by way of virtualizing the hardware that people were running. And watching that become much more pervasive, and working with companies as they began to shift from single system, single app to virtualized environments.
And then at the tail end, just watching large tech companies emerge and demand a lot different style computers than those that we had been customarily making at Dell. And kind of fascinated with just what these companies like Facebook, and Google, and Amazon, and others were doing to reimagine what systems needed to look like in their hyperscale environments. One of the companies that was in the tech space, Joyent, a cloud computing company, is where I went next. Was really drawn in just to velocity and the innovation that was taking place with these companies that were providing abstractions on top of hardware to make it much easier for customers to get access to the compute, and the storage, and the networking that they needed to build and deploy software. So, spent—after ten years at Dell, I was at Joyent for ten years. That is where I met my future co-founders, Bryan Cantrill who was at Joyent, and then also Jess Frazelle who we knew working closely while she was at Docker and other stops.
But spent ten years as a public cloud infrastructure operator, and we built that service out to support workloads that ran the gamut from small game developers up to very large enterprises, and it was really interesting to learn about and appreciate what this infrastructure utility business looked like in public cloud. And that was also kind of where I got my first realization of just how hard it was to run large fleets of the systems that I had been responsible for providing back at Dell for ten years. We were obviously a large customer of Dell, and Supermicro, and a number of switch manufacturers. It was eye-opening just how much was lacking in the remaining software to bind together hundreds or thousands of these machines.
A lot of the operational tooling that I wished had been there and how much we were living at spreadsheets to manage and organize and deploy this infrastructure. While there, also got to kind of see firsthand what happened as customers got really, really big in the public cloud. And one of those was Samsung, who was a very large AWS customer, got so large that they needed to figure out what their path on-premise would look like. And after going through the landscape of all the legacy enterprise solutions, deemed that they had to go buy a cloud company to complete that journey. And they bought Joyent. Spent three years operating the Samsung cloud, and then that brings us to two years ago, when Jess, Bryan, and I started Oxide Computer.
Will: I think maybe for the benefit of our listeners, it would be interesting to have you define—and what we’re talking about today is the server industry—and to maybe take a step back and in your own words, define what a server is. And then it would be really interesting to jump into a high-level history of the server up until today, and maybe within that, where the emergence of the public cloud came from.
Steve: You know, you’ll probably get different definitions of what a server is depending on who you ask, but at the highest level, a server differs from a typical PC that you would have in your home in a couple of ways, and more about what it is being asked to do that drives the requirements of what one would deem a server. But if you think about a basic PC that you’re running in your home, a laptop, a desktop, a server has a lot of the same components: they have CPUs, and DRAM memory that is for non-volatile storage, and disks that are storing things in a persistent way when you shut off your computer that actually store and retain the data, and a network card so that you can connect to either other machines or to the internet. But where servers start to take on a little bit different shape and a little bit different set of responsibilities is the workloads that they’re supporting. Servers, the expectations are that they are going to be running 24/7 in a highly reliable and highly available manner. And so there are technologies that have gone into servers, that ECC memory to ensure that you do not have memory faults that lose data, more robust components internally, ways to manage these things remotely, and ways to connect these to other servers, other computers.
Servers, when running well, are things you don’t really need to think about, are doing that, are running in a resilient, highly available manner. In terms of the arc of the server industry, if you go back—I mean, there’s been servers for many, many, many, many decades. Some of the earlier commercially available servers were called mainframes, and these were big monolithic systems that had a lot of hardware resources at the time, and then were combined with a lot of operational and utilization software to be able to run a variety of tasks. These were giant, giant machines; these were extraordinarily expensive; you would typically find them only running in universities or government projects, maybe some very, very large enterprises in the’60s and’70s. As more and more software was being built and developed and run, the market demand and need for smaller, more accessible servers that were going to be running this common software, were driving machines that were coming out—still hardware plus software—from the likes of IBM and DEC and others.
Then you broke into this period in the ’80s where, with the advent of x86 and the rise of these PC manufacturers—the Dells and Compaqs and others—this transition to more commodity server systems. A focus, really a focus on hardware only, and building these commodity x86 servers that were less expensive, that were more accessible from an economics perspective, and then ultimately that would be able to run arbitrary software, so one could run any operating system or any body of software that they wanted on these commodity servers. When I got to Dell in 1999, this is several years into Dell’s foray into the server market, and you would buy a server from Dell, or from HP, or from Compaq, or IBM, then you would go find your software that you were going to run on top of that to stitch these machines together. That was, kind of, that server virtualization era, in the ’90s, 2000s. As I mentioned, technology companies were looking at building more scalable systems that were aggregating resources together and making it much easier for their customers to access the storage, the networking that they needed, that period of time in which the commodity servers and the software industry diverged, and you had a bunch of different companies that were responsible for either hardware or the software that would bring these computers together, these large hyperscalers said, “Well, we’re building purpose-built infrastructure services for our constituents at, like, a Facebook. That means we really need to bind this hardware and software together in a single product so that our software teams can go very quickly and they can programmatically access the resources that they need to deploy software.”
So, they began to develop systems that looked more monolithic, kind of, rack-level systems that were driving much better efficiency from a power and density perspective, and hydrating it with software to provide infrastructure services to their businesses. And so you saw, what started out in the computer industry is these monolithic hardware plus software products that were not very accessible because they were so expensive and so large, but real products that were much easier to do real work on, to this period where you had a disaggregation of hardware and software where the end-user bore the responsibility of tying these things together and binding these into those infrastructure products, to today, where the largest hyperscalers in the market have come to the realization that building hardware and software together and designing and developing what modern computers should look like, is commonplace, and we all know that well or can access that as public cloud computing.
Jason: And what was the driving force behind that decoupling? Was it the actual hardware vendors that didn’t want to have to deal with the software? Or is that more from a customer-facing perspective where the customers themselves felt that they could eke out the best advantage by developing their own software stack on top of a relatively commodity unopinionated hardware stack that they could buy from a Dell or an HP?
Steve: Yeah, I think probably both, but one thing that was a driver is that these were PC companies. So, coming out of the’80s companies that were considered, quote-unquote, “The IBM clones,” Dell, and Compaq, and HP, and others that were building personal computers and saw an opportunity to build more robust personal computers that could be sold to customers who were running, again, just arbitrary software. There wasn’t the desire nor the DNA to go build that full software stack and provide that out as an opinionated appliance or product. And I think then, part of it was also like, hey, if we just focus on the hardware, then got this high utility artifact that we can go sell into all sorts of arbitrary software use cases. You know, whether this is going to be a single server or three servers that’s going to go run in a closet of cafe, or it’s going to be a thousand servers that are running in one of these large enterprise data centers, we get to build the same box, and that box can run underneath any different type of software. By way of that, what you ultimately get in that scenario is you do have to boil things down to the lowest common denominators to make sure that you’ve got that compatibility across all the different software types.
Will: Who were the primary software vendors that were helping those companies take commodity servers and specialize into particular areas? And what’s their role now and how has that transformed in light of the public cloud and the offerings that are once again generalized, but also reintegrated from a hardware and software perspective, just not maybe in your own server room, but in AWS, or Azure, or GCP?
Steve: Yeah, so you have a couple layers of software that are required in the operation of hardware, and then all the way up through what we would think about as running in a rack, a full rack system today. You’ve first got firmware, and this is the software that runs on the hardware to be able to connect the different hardware components, to boot the system, to make sure that the CPU can talk to its memory, and storage, and the network. That software may be a surprise to some, but that firmware that is essential to the hardware itself is not made by the server manufacturer themselves. That was part of this outsourcing exercise in the ’80s where not only the upstack software that runs on server systems but actually some of the lower-level downstack software was outsourced to these third-party firmware shops that would write that software. And at the time, probably made a lot of sense and made things a lot easier for the entire ecosystem.
You know, the fact that’s the same model today, and given how proprietary that is and, you know, where that can actually lead to some vulnerabilities and security issues is more problematic. You’ve got firmware, then you’ve got the operating system that runs on top of the server. You have a hypervisor, which is the emulation layer that translates that lower-level hardware into a number of virtual machines that applications can run in. You have control plane software that connects multiple systems together so that you can have five or ten or a hundred, or a thousand servers working in a pool, in a fleet. And then you’ve got higher-level software that allows a user to carve up the resources that they need to identify the amount of compute, and memory, and storage that they want to spin up.
And that is exposed to the end-user by way of APIs and/or a user interface. And so you’ve got many layers of software that are running on top of hardware, and the two in conjunction are all there to provide infrastructure services to the end-user. And so when you’re going to the public cloud today, you don’t have to worry about any of that, right? Both of you have probably spun up infrastructure on the public cloud, but they call it 16 digits to freedom because you just swipe a credit card and hit an API, and within seconds, certainly within a minute, you’ve got readily available virtual servers and services that allow you to deploy software quickly and manage a project with team members. And the kinds of things that used to take days, weeks, or even months inside an enterprise can be done now in a matter of minutes, and that’s extraordinarily powerful.
But what you don’t see is all the integration of these different components running, very well stitched together under the hood. Now, for someone who’s deploying their own infrastructure in their own data center today, that sausage-making is very evident. Today, if you’re not a cloud hyperscaler, you are having to go pick a hardware vendor and then figure out your operating system and your control plane and your hypervisor, and you have to bind all those things together to create a rack-level system. And it might have three or four different vendors and three or four different products inside of it, and ultimately, you have to bear the responsibility of knitting all that together.
Will: Because those products were developed in silos from each other?
Steve: Yeah.
Will: They were not co-developed. You’ve got hardware that was designed in a silo separate from oftentimes it sounds like the firmware and all of the software for operating those resources.
Steve: Yeah. The hardware has a certain set of market user requirements, and then if you’re a Red Hat or you’re a VMware, you’re talking to your customers about what they need and you’re thinking at the software layer. And then you yourself are trying to make it such that it can run across ten or twenty different types of hardware, which means that you cannot do things that bind or provide hooks into that underlying hardware which, unfortunately, is where a ton of value comes from. You can see an analog to this in thinking about the Android ecosystem compared to the Apple ecosystem and what that experience is like when all that hardware and software is integrated together, co-designed together, and you have that iPhone experience. Plenty of other analogs in the automotive industry, with Tesla, and health equipment, and Peloton and others, but when hardware and software—we believe certainly—when hardware and software is co-designed together, you get a better artifact and you get a much, much better user experience. Unfortunately, that is just not the case today in on-prem computing.
Jason: So, this is probably a great time to transition to Oxide. Maybe to keep the analogy going, the public cloud is that iPhone experience, but it’s just running in somebody else’s data center, whether that’s AWS, Azure, or one of the other public clouds. You’re developing iOS for on-prem, for the people who want to run their own servers, which seems like kind of a countertrend. Maybe you can talk us through the dynamics in that market as it stands today, and how that’s growing and evolving, and what role Oxide Computer plays in that, going forward.
Steve: You’ve got this what my co-founder Jess affectionately refers to as ‘infrastructure privilege’ in the hyperscalers, where they have been able to apply the money, and the time, and the resources to develop this, kind of, iPhone stack, instead of thinking about a server as a single 1U unit, or single machine, had looked at, well, what does a rack—which is the case that servers are slotted into in these large data centers—what does rack-level computing look like and where can we drive better power efficiency? Where can we drive better density? How can we drive much better security at scale than the commodity server market today? And doing things like implementing hardware Roots of Trust and Chain of Trust, so that you can ensure the software that is running on your machines is what is intended to be running there. The blessing is that we all—the market—gets access to that modern infrastructure, but you can only rent it.
The only way you can access it is to rent, and that means that you need to run in one of the three mega cloud providers’ data centers in those locations, that you are having to operate in a rental fee model, which at scale can become very, very prohibitively expensive. Our fundamental belief is that the way that these hyperscale data centers have been designed and these products have been designed certainly looks a lot more like what modern computers should look like, but the rest of the market should have access to the same thing. You should be able to buy and own and deploy that same product that runs inside a Facebook data center, or Apple data center, or Amazon, or a Google data center, you should be able to take that product with you wherever your business needs to run. A bit intimidating at the top because what we signed up for was building hardware, and taking a clean sheet paper approach to what a modern server could look like. There’s a lot of good hardware innovation that the hyperscalers have helped drive; if you go back to 2010, Facebook pioneered being a lot more open about these modern open hardware systems that they were developing, and the Open Compute Project, OCP, has been a great collection point for these hyperscalers investing in these modern rack-level systems and doing it in the open, thinking about what the software is that is required to operate modern machines, importantly, in a way that does not sink the operations teams of the enterprises that are running them.
Again, I think one of the things that was just so stunning to me, when I was at Joyent—we were running these machines, these commodity machines, and stitching together the software at scale—was how much of the organization’s time was tied up in the deployment, and the integration, and the operation of this. And not just the organization’s time, but actually our most precious resource, our engineering team, was having to spend so much time figuring out where a performance problem was coming from. For example in [clear throat], man, those are the times in which you really are pounding your fist on the table because you will try and go downstack to figure out, is this in the control plane? Is this in the firmware? Is this in the hardware?
And commodity systems of today make it extremely, extremely difficult to figure that out. But what we set out to do was build same rack-level system that you might find in a hyperscaler data center, complete with all the software that you need to operate it with the automation required for high availability and low operational overhead, and then with a CloudFront end, with a set of services on the front end of that rack-level system that delight developers, that look like the cloud experience that developers have come to love and depend on in the public cloud. And that means everything is programmable, API-driven services, all the hardware resources that you need—compute, memory, and storage—are actually a pool of resources that you can carve up and get access to and use in a very developer-friendly way. And the developer tools that your software teams have come to depend on just work and all the tooling that these developers have invested so much time in over the last several years, to be able to automate things, to be able to deploy software faster are resident in that product. And so it is definitely kind of hardware and software co-designed, much like some of the original servers long, long, long ago, but modernized with the hardware innovation and open software approach that the cloud has ushered in.
Jason: And give us a sense of scale; I think we’re so used to seeing the headline numbers of the public cloud, you know, $300-and-some billion dollars today, adding $740-some billion over the next five years in public cloud spend. It’s obviously a massive transformation, huge amount of green space up for grabs. What’s happening in the on-prem market where your Oxide Computer is playing and how do you think about the growth in that market relative to a public cloud?
Steve: It’s funny because as Will can attest, as we were going through and fundraising, the prevalent sentiment was, like, everything’s going to the public cloud. As we’re talking to the folks in the VC community, it was Amazon, Microsoft, and Google are going to own the entirety of compute. We fundamentally disagreed because, A, we’ve lived it, and b, we went out as we were starting out and talked to dozens and dozens of our peers in the enterprise, who said, “Our cloud ambitions are to be able to get 20, 30, 40% of our workloads out there, and then we still have 60, 70% of our infrastructure that is going to continue to run in our own data centers for reasons including regulatory compliance, latency, security, and in a lot of cases, cost.” It’s not possible for these enterprises that are spending half a billion, a billion dollars a year to run all of their infrastructure in the public cloud. What you’ve seen on-premises, and it depends on who you’re turning to, what sort of poll and research you’re turning to, but the on-prem market, one is growing, which I think surprises a lot of folks; the public cloud market, of course, it’s growing like gangbusters, and that does not surprise a lot of folks, but what we see is that the combined market of on-prem and cloud, you can call it—if on-premise on the order of $100 billion and cloud is on the order of $150 billion, you are going to see enormous growth in both places over the next 10, 15 years.
These markets are going to look very, very small compared to where they will be because one of the biggest drivers of whether it’s public cloud or on-prem infrastructure, is everything shifting to digital formats. The digitalization that is just all too commonplace, described everywhere. But we’re still very, very early in that journey. I think that if you look at the global GDP, less than 10% of the global GDP is on the internet, is online. Over the coming 10, 20 years, as that shifts to 20%, 30%, you’re seeing exponential growth. And again, we believe and we have heard from the market that is representative of that $100 billion that investments in the public cloud and on-prem is going to continue to grow much, much more as we look forward.
Will: Steve, I really appreciate you letting listeners know how special a VC I am.
Steve: [laugh].
Will: [laugh]. It was really important point that I wanted to make sure we hit on.
Steve: Yeah, should we come back to that?
Will: Yeah, yeah yeah—
Steve: Yeah, let’s spend another five or ten minutes on that.
Will: —we’ll revisit that. We’ll revisit that later. But when we’re talking about the market here, one of the things that got us so excited about investing in Oxide is looking at the existing ecosystem of on-prem commercial providers. I think if you look at the public cloud, there are fierce competitors there, with unbelievably sophisticated operations and product development. When you look at the on-prem ecosystem and who you would go to if you were going to build your own data center today, it’s a lot of legacy companies that have started to optimize more for, I would say, profitability over the last couple of years than they have for really continuing to drive forward from an R&D and product standpoint.
Would love maybe for you to touch on briefly, what does competition for you look like in the on-prem ecosystem? I think it’s very clear who you’re competing with, from a public cloud perspective, right? It’s Microsoft, Google, Amazon, but who are you going up against in the on-prem ecosystem?
Steve: Yeah. And just one note on that. We don’t view ourselves as competing with Amazon, Google, and Microsoft. In fact, we are ardent supporters of cloud in the format, namely this kind of programmable API-fronted infrastructure as being the path of the future of compute and storage and networking. That is the way that, in the future, most software should be deployed to, and operated on, and run.
We just view the opportunity for, and what customers are really, really excited about is having those same benefits of public cloud, but in a format in which they can own it and being able to have access to that everywhere their business needs to run, so that it’s not, you know, do I get all this velocity, and this innovation, and this simplicity when I rent public cloud, or do I own my infrastructure and have to give up a lot of that? But to the first part of your question, I think the first issue is that it isn’t one vendor that you are talking about what is the collection of vendors that I go to to get servers, software to make my servers talk to each other, switches to network together these servers, and additional software to operate, and manage, and monitor, and update. And there’s a lot of complexity there. And then when you take apart each one of those different sets of vendors in the ecosystem, they’re not designing together, so you’ve got these kind of data boundaries and these product boundaries that start to become really, really real when you’re operating at scale, and when you’re running critical applications to your business on these machines. And you find yourself spending an enormous amount of the company’s time just knitting this stuff together and operating it, which is all time lost that could be spent adding additional features to your own product and better competing with your competitors.
And so I think that you have a couple of things in play that make it hard for customers running infrastructure on-premises, you’ve got that dynamic that it’s a fractured ecosystem, that these things are not designed together, that you have this kit car that you have to assemble yourself and it doesn’t even come with a blueprint of the particular car design that you’re building. I think that you do have some profit-taking in that it is very monopolized, especially on the software side where you’ve only got a couple of large players that know that there are few alternatives for companies. And so you are seeing these ELAs balloon, and you are seeing practices that look a lot like Oracle Enterprise software sales that are really making this on-prem experience not very economically attractive. And so our approach is, hardware should come with all the software required to operate it, it should be tightly integrated, the software should be all open-source. Something we haven’t talked about.
I think open-source is playing an enormous role in accelerating the cloud landscape and the technology landscapes. We are going to be developing our software in an open manner, and truly believe whether it’s from a security view through to the open ecosystem, that it is imperative that software be open. And then we are integrating the switch into that rack-level product so that you’ve got networking baked in. By doing that, it opens up a whole new vector of value to the customer where, for example, you can see for the first time what is the path of traffic from my virtual machine to a switchboard? Or when things are not performing well, being able to look into that path, and the health, and see where things are not performing as well as they should, and being able to mitigate those sorts of issues.
It does turn out if you are able to get rid of a lot of the old, crufty artifacts that have built up inside even these commodity system servers, and you are able to start designing at a rack level where you can drive much better power efficiency and density, and you bake in the software to effectively make this modern rack-level server look like a cloud in a box, and ensure these things can snap together in a grid, where in that larger fleet, operational management is easy because you’ve got the same automation capabilities that the big cloud hyperscalers have today. It can really simplify life. It ends up being an economic win and maybe most importantly, presents the infrastructure in a way that the developers love. And so there’s not this view of the public cloud being the fast, innovative path for developers and on-prem being this, submit a trouble ticket and try and get access to a VM in six days, which sadly is the experience that we hear a lot of companies are still struggling with in on-prem computing.
Jason: Practically, when you’re going out and talking to customers, you’re going to be a heterogeneous environment where presumably they already have their own on-prem infrastructure and they’ll start to plug in—
Steve: Yeah.
Jason: —Oxide Computer alongside of it. And presumably, they’re also to some degree in the public cloud. It’s a fairly complex environment that you’re trying to insert yourself into. How are your customers thinking about building on top of Oxide Computer in that heterogeneous environment? And how do you see Oxide Computer expanding within these enterprises, given that there’s a huge amount of existing capital that’s gone into building out their data centers that are already operating today, and the public cloud deployments that they have?
Steve: As customers are starting to adopt Oxide rack-level computing, they are certainly going to be going into environments where they’ve got multiple generations of multiple different types of infrastructure. First, the discussions that we’re having are around what are the points of data exfiltration, of data access that one needs to operate their broader environment. You can think about handoff points like the network where you want to make sure you’ve got a consistent protocol to, like, BGP or other, to be able to speak from your layer 2 networks to your layer 3 networks; you’ve got operational software that is doing monitoring and alerting and rolling up for service for your SRE teams, your operations teams, and we are making sure that Oxide’s endpoint—the front end of the Oxide product—will integrate well, will provide the data required for those systems to run well. Another thorny issue for a lot of companies is identity and access management, controlling the authentication and the access for users of their infrastructure systems, and that’s another area where we are making sure that the interface from Oxide to the systems they use today, and also resident in the Oxide product such as one wants to use it directly, has a clean cloud-like identity and access management construct for one to use. But at the highest level it is, make sure that you can get out of the Oxide infrastructure, the kind of data and tooling required to incorporate into management of your overall fleet.
Over time, I think customers are going to experience a much simpler and much more automated world inside of the Oxide ecosystem; I think they’re going to find that there are exponentially fewer hours required to manage that environment and that is going to inevitably just lead to wanting to replace a hundred racks of the extant commodity stack with, you know, sixty racks of Oxide that provide much better density, smaller footprint in the data center, and again, software-driven in the way that these folks are looking for.
Jason: And in that answer, you alluded to a lot of the specialization and features that you guys can offer. I’ve always loved Alan Kay’s quote, “People who are really serious about software make their own hardware.”
Steve: Yeah.
Jason: Obviously, you’ve got some things in here that only Oxide Computer can do. What are some of those features that traditional vendors can’t even touch or deliver that you’ll be able to, given your hardware-software integration?
Steve: Maybe not the most exciting example, but I think one that is extremely important to a lot of the large enterprise company that we’re working with, and that is at a station, being able to attest to the software that is running on your hardware. And why is that important? Well, as we’ve talked about, you’ve got a lot of different vendors that are participating in that system that you’re deploying in your data center. And today, a lot of that software is proprietary and opaque and it is very difficult to know what versions of things you are running, or what was qualified inside that package that was delivered in the firmware. We were talking to a large financial institution, and they said their teams are spending two weeks a month just doing, kind of a proof of trust in their infrastructure that their customer’s data is running on, and how cumbersome and hard it is because of how murky and opaque those lower-level system software world is.
What do the hyperscalers do? They have incorporated hardware Root of Trust, which ensures from that first boot instruction, from that first instruction on the microprocessor, that you have a trusted and verifiable path, from the system booting all the way up through the control plane software to, say, a provisioned VM. And so what this does is it allows the rest of the market access to a bunch of security innovation that has gone on where these hyperscalers would never run without this. Again, having the hardware Root of Trust anchored at a station process, the way to attest all that software running is going to be really, really impactful for more than just security-conscious customers, but certainly, those that are investing more in that are really, really excited. If you move upstack a little bit, when you co-design the hardware with the control plane, both the server and the switch hardware with the control plane, it opens up a whole bunch of opportunity to improve performance, improve availability because you now have systems that are designed to work together very, very well.
You can now see from the networking of a system through to the resources that are being allocated on a particular machine, and when things are slow, when things are broken, you are able to identify and drive those fixes, in some cases that you could not do before, in much, much, much faster time, which allows you to start driving infrastructure that looks a lot more like the five nines environment that we expect out of the public cloud.
Jason: A lot of what you just mentioned, actually, once again, ties back to that analogy to the iPhone, and having that kind of secure enclave that powers Touch ID and Face ID—
Steve: Yep.
Jason: —kind of a server equivalent, and once again, optimization around particular workflows, the iPhone knows exactly how many photos every [laugh] iOS user takes, and therefore they have a custom chip dedicated specifically to processing images. I think that tight coupling, just relating it back to that iOS and iPhone integration, is really exciting.
Steve: Well, and the feedback loop is so important because, you know, like iPhone, we’re going to be able to understand where there are rough edges and where things are—where improvements can even can continue to be made. And because this is software-driven hardware, you get an opportunity to continuously improve that artifact over time. It now stops looking like the old, your car loses 30% of the value when you drive it off the lot. Because there’s so much intelligent software baked into the hardware, and there’s an opportunity to update and add features, and take the learnings from that hardware-software interaction and feed that back into an improving product over time, you can start to see the actual hardware itself have a much longer useful life. And that’s one of the things we’re really excited about is that we don’t think servers should be commodities that the vendors are trying to push you to replace every 36 months.
One of the things that is important to keep in mind is as Moore’s laws is starting to slow or starting to hit some of the limitations, you won’t have CPU density and some of these things, driving the need to replace hardware as quickly. So, with software that helps you drive better utilization and create a better-combined product in that rack-level system, we think we’re going to see customers that can start getting five, six, seven years of useful life out of the product, not the typical two, or three, or maybe four that customers are seeing today in the commodity systems.
Will: Steve, that’s one of the challenges for Oxide is that you’re taking on excellence in a bunch of interdisciplinary sciences here, between the hardware, the software, the firmware, the security; this is a monster engineering undertaking. One of the things that I’ve seen as an investor is how dedicated you have got to be to hiring, to build basically the Avengers team here to go after such a big mission. Maybe you could touch on just how you’ve thought about architecting a team here. And it’s certainly very different than what the legacy providers from an on-prem ecosystem perspective have taken on.
Steve: I think one of the things that has been so important is before we even set out on what we were going to build, the three of us spent time and focused on what kind of company we wanted to build, what kind of company that we wanted to work at for the next long chunk of our careers. And it’s certainly drawing on experiences that we had in the past. Plenty of positives, but also making sure to keep in mind the negatives and some of the patterns we did not want to repeat in where we were working next. And so we spent a lot of time just first getting the principles and the values of the company down, which was pretty easy because the three of us shared these values. And thinking about all the headwinds, just all the foot faults that hurt startups and even big companies, all the time, whether it be the subjectivity and obscurity of compensation or how folks in some of these large tech companies doing performance management and things, and just thinking about how we could start from a point of building a company that people really want to work for and work with.
And I think then layering on top of that, setting out on a mission to go build the next great computer company and build computers for the cloud era, for the cloud generation, that is, as you say, it’s a big swing. And it’s ambitious, and exhilarating and terrifying, and I think with that foundation of focusing first on the fundamentals of the business regardless of what the business is, and then layering on top of it the mission that we are taking on, that has been appealing, that’s been exciting for folks. And it has given us the great opportunity of having terrific technologists from all over the world that have come inbound and have wanted to be a part of this. And we, kind of, will joke internally that we’ve got eight or nine startups instead of a startup because we’re building hardware, and we’re taking on developing open-source firmware, and a control plane, and a switch, and hardware Root of Trust, and in all of these elements. And just finding folks that are excited about the mission, that share our values, and that are great technologists, but also have the versatility to work up and down the stack has been really, really key.
So far, so great. We’ve been very fortunate to build a terrific, terrific team. Shameless plug: we are definitely still hiring all over the company. So, from hardware engineering, software engineering, operations, support, sales, we’re continuing to add to the team, and that is definitely what is going to make this company great.
Will: Maybe just kind of a wrap-up question here. One of the things Jason and I always like to ask folks is, if you succeed over the next five years, how have you changed the market that you’re operating in, and what does the company look like in five years? And I want you to know as an investor, I’m holding you to this. Um, so—
Steve: Yeah, get your pen ready. Yeah.
Will: Yeah, yeah. [laugh].
Steve: Definitely. Expect to hear about that in the next board meeting. When we get this product in the market and five years from now, as that has expanded and we’ve done our jobs, then I think one of the most important things is we will see an incredible amount of time given back to these companies, time that is wasted today having to stitch together a fractured ecosystem of products that were not designed to work together, were not designed with each other in mind. And in some cases, this can be 20, 30, 40% of an organization’s time. That is something you can’t get back.
You know, you can get more money, you can—there’s a lot that folks can control, but that loss of time, that inefficiency in DIY your own cloud infrastructure on-premises, will be a big boon. Because that means now you’ve got the ability for these companies to capitalize on digitalizing their businesses, and just the velocity of their ability to go improve their own products, that just will have a flywheel effect. So, that great simplification where you don’t even consider having to go through and do these low-level updates, and having to debug and deal with performance issues that are impossible to sort out, this—aggregation just goes away. This system comes complete and you wouldn’t think anything else, just like an iPhone. I think the other thing that I would hope to see is that we have made a huge dent in the efficiency of computing systems on-premises, that the amount of power required to power your applications today has fallen by a significant amount because of the ability to instrument the system, from a hardware and software perspective, to understand where power is being used, where it is being wasted.
And I think that can have some big implications, both to just economics, to the climate, to a number of things, by building and people using smarter systems that are more efficient. I think generally just making it commonplace that you have a programmable infrastructure that is great for developers everywhere, that is no longer restricted to a rental-only model. Is that enough for five years?
Will: Yeah, I think I think democratizing access to hyperscale infrastructure for everybody else sounds about right.
Steve: All right. I’m glad you wrote that down.
Jason: Well, once again, Steve, thanks for coming on. Really exciting, I think, in this conversation, talking about the server market as being a fairly dynamic market still, that has a great growth path, and we’re really excited to see Oxide Computer succeed, so thanks for coming on and sharing your story with us.
Steve: Yeah, thank you both. It was a lot of fun.
Will: Thank you for listening to
Perfectly Boring. You can keep up the latest on the podcast at
perfectlyboring.com, and follow us on Apple, Spotify, or wherever you listen to podcasts. We’ll see you next time.