This little server startup wants to take on a horde of tech giants
Oxide Computer is a bold bet by three prominent engineers that some day, more and more companies will want cloudlike hardware under their own control.
Jessie Frazelle rolled her eyes at fellow Oxide Computer co-founder Bryan Cantrill as he recounted a pitch meeting during which he compared their new company's product strategy to IBM's fabled AS/400, a "midrange computer" the size of a washer/dryer introduced in the late 1980s.
"The AS/400 is the most successful least-understood product in the history of the industry," he insisted, acknowledging (for the millionth time, judging by Frazelle's expression) that while the comparison might be apt, it's maybe not the best idea to link your plans for venture capital money to a 30-year-old product. Still, IBM's product "was very much in this same zeitgeist of fully integrating hardware and software and delivering true value," and Big Blue sold a ton of them, he said.
Get what matters in tech, in your inbox every morning. Sign up for Source Code.
Frazelle, Cantrill and their third co-founder, Steve Tuck, are attempting something rarely seen in modern Silicon Valley: They're building a new computer company. Oxide Computer has vowed to deliver a new type of server in which the hardware and the software were designed together — an approach used by the "hyperscalers" (AWS, Microsoft, Google and Facebook) to power the most advanced data centers on the planet — to businesses that want or need to manage their own fleets of servers.
Oxide, which allowed Protocol behind the scenes for its first interview since the company launched in December, has raised $20 million to compete directly against industry giants like Dell Technologies and HPE for the tech budgets of server huggers. The company is helmed by a trio of experienced tech infrastructure veterans with backgrounds at Sun Microsystems and Docker. And it doesn't expect to have a product ready for its initial customers until early 2022: It will need way more people than the 15 employees spread out across a cavernous garage in Emeryville, right across the street from Pixar's huge campus, in order to make that happen.
The AS/400 might be an example of a design philosophy that only techies of a certain age will remember, but everyone has heard of the iPhone: Apple's current trillion-dollar valuation is due in large part to its decision to design the iPhone's hardware and software in concert, creating a product that changed the course of computing history.
Oxide is a bet that server buyers are ready for their iPhone moment.
"We know the market is there," Cantrill said. "This is a market that has been sorely underserved."
Cloud for the crowd
Cloud computing might be a runaway train, but one of the most visible trends in this space over the last few years has been the rise of hybrid cloud computing, in which companies run some of their software on the cloud and some of their software the traditional way, on servers they own and manage themselves. Even AWS now ships hardware, with the launch of its AWS Outposts product late last year.
Yet amid a surge in innovation across other areas of enterprise technology over the last 20 years, the standard retail server hasn't changed that much. Oxide's mission is to create a server using standard x86 processors from companies like Intel or AMD that strips away a lot of the extraneous software that ships with a mainstream server, giving customers a "clean sheet" designed for their operating system, infrastructure, and application requirements.
Companies like AWS don't buy their servers from traditional vendors like Dell and HPE; they design and build their own machines to exact specifications, down to the low-level firmware code that controls the basic interaction between the server's hardware and their customers' application software. The cloud leaders can afford to spend billions each year on the equipment and talent needed to make this happen, because as Cantrill pointed out, computers for which the software and hardware were designed in concert tend to deliver better performance and reliability than all-purpose off-the-rack servers.
"All three of us have spent a lot of time working in hardware and software kind of up and down the stack, and as we've gotten to a point in our career where we're running things at scale, I think we all found a significant divergence between the software that is being written and run and the underlying hardware that supports it," said Tuck, Oxide's CEO. "That is often an area where problems emerge."
Still, most companies managing their own infrastructure have little choice beyond the usual suspects when the three-year lifespan of their servers runs its course. Those retail servers are still designed as one-size-fits-all boxes that must accommodate all of the ways a potential customer might use those servers, which means they ship with a lot of extraneous firmware.
"What we're cleaning up throughout the bottom of the stack just makes it all run better and [makes it] more debuggable," said Frazelle, the startup's chief product officer.
This approach could also head off potential security threats, especially in the era of seismic chip bugs like the Meltdown and Spectre issues that floored the enterprise computing community two years ago by exposing how vulnerable systems were to attacks exploiting this code, once thought to be a safe space.
"All their features have the highest level of privilege that you should ever have on a computer, which is nuts. It's like, why are you doing that?" she said. But the traditional vendors are more or less tied to this approach because they can't anticipate how their customers are going to use their servers, such as which operating systems they plan to install on those machines, she said.
Bucking the trends
It might seem a little crazy to launch a server company in the middle of a historic transition between enterprise computing models.
After all, Gartner expects worldwide spending on cloud infrastructure services to almost double by 2022 compared to 2019, while reporting that worldwide sales of servers grew less than 7% in the third quarter of 2019. HPE, the second-place server vendor, reported disappointing earnings last week thanks in large part to a 16% decrease in server shipments, which wasn't helped by the coronavirus-caused manufacturing chaos in China over the last few months.
Cloud providers have successfully captured the segment of the market that was managing their own servers because they didn't believe there was an alternative, Cantrill said. But he and Oxide think there are a lot of companies that need to manage their own servers for regulatory or control reasons, and they simply won't move to the cloud.
And there's another potential scenario that Oxide is watching very closely.
"The big one that we see coming, and honestly is arguably happening even faster than we thought was going to happen, [is] the economic reason: folks that grew up on the cloud and grew their service on the cloud, and they've now become important, and have begun to focus on their margins," said Cantrill, Oxide's chief technology officer.
Dropbox turned a lot of heads when it revealed during its 2018 IPO that it saved $75 million a year by moving off AWS and onto its own network of servers, and the Oxide co-founders believe that a generation of enterprise software companies built on cloud services will start to think about managing their own servers as their application requirements become more predictable and their coffers swell.
"This all happened because [cloud is] a great product that allows you to move fast and be able to build products and services quickly and not waste a bunch of valuable company people and time doing things that can be automated," Tuck said. "But when you kind of realize that this rental utility model is the only way to consume it, it does start looking a lot like a hotel being the only way you can live."
"People think this market is dead," Frazelle said. "They don't see the fact that people are literally running [their own servers] for good reasons. It's crazy."
Still, a lot can change between now and 2022, when the first Oxide servers are expected to become available, and server vendors desperate for growth will surely monitor Oxide's strategy.
"They're building a business model that basically has limited defensive value," said Kevin Krewell, principal analyst with Tirias Research. "If they don't know why the other [server vendors] don't do it, it's because (A) nobody's asked them for it, or (B) they haven't felt compelled to do it."
Getting past the past
Cantrill, Frazelle and Tuck are longtime friends who have worked for some of the largest and most influential infrastructure tech companies of the internet era.
Tuck and Cantrill worked together for years at Joyent, a smaller cloud provider bought by Samsung in 2016, and Cantrill spent years at Sun Microsystems, a company that also pursued an integrated hardware/software strategy. Frazelle was an early and important engineer at Docker, which ushered in the modern software container movement, and has also spent time at Google and Microsoft.
Their plan for Oxide is to take the best ideas those companies advanced and avoid the worst. When you apply for a job at Oxide, don't expect to be subjected to the traditional whiteboard coding interview; instead, you'll be asked to provide written answers to a series of questions about your work experience. And during your interview, you'll be able to see how the three co-founders answered the exact same questions.
"Each one of us has applied for our jobs," Tuck said.
Get in touch with us: Share information securely with Protocol via encrypted Signal or WhatsApp message, at 415-214-4715 or through our anonymous SecureDrop.
And those are big jobs. Oxide is trying to build a new type of computer that will appeal to companies that want to control their own destiny using the best of what the cloud has to offer. It's an interesting idea but an unproven market, and if Oxide is successful, it will be the sign of another important shift in enterprise computing.
"You should not have this false dichotomy where you lose all the rest of what that cloud infrastructure experience is just because you need a massively different economic footprint," Tuck said. "How many [big companies] want an experience similar to what drew them to the cloud in the first place, for that infrastructure running on premises? Again, the answer is a lot."
Correction: An earlier version of this story misspelled Steve Tuck's name and erroneously stated his affiliation with Sun Microsystems. This story was updated March 10, 2020.