I just think that people would benefit from like working in a factory, going to a mine, looking at a mine, going into like the actual industrial economy that people are talking about. It doesn't work the way you think it does. Just doesn't. And we've got an entire generation that like doesn't have calloused hands. And you'll find like there's a bunch of stuff that's just not connected that you wouldn't connect. People need to spend more time reading Richard Scarry. You know what I mean? Where you're just like, learn how the world works, and then you can learn what the actual perils of automation might be and might not be, and where all those very significant gaps are.
All right, we are back. We are live in the studio with Bryan Cantrill, who is co-founder and CTO of Oxide Computer Company and a longtime systems engineer across a bunch of different companies. Bryan, welcome.
Hey, thanks. It's great to be here.
Absolutely. So, tell us about Oxide.
Yeah. So, we're a computer company. It says it right on the tin. That was very deliberate. You know, when we started the company in 2019, we wanted to make very clear to any VCs that there would be no pivots to software, that we were doing hardware and software co-designed. So, we design and make a rack-scale machine. So, a machine that is designed for cloud computing. So, my co-founder and I were together at a cloud computing pioneer, Joyent. I spent 14 years prior to that at Sun Microsystems, which I have to explain as a computer company.
You know, I've heard of it.
Okay. You know, I just, you know, you've got the Dan Aykroyd thing. I just never know what I got to, I got to go explain. So Sun was a computer company that was consumed by Oracle. I went to this cloud computing pioneer, and we were building out public cloud computing, making that software available for people to run on-prem. And we were bought by Samsung in 2016 because Samsung needed to be able to compete with Apple with cloud computing, and there wasn't a product they could go buy to run cloud computing on-premises. So they bought Joyent. We went out to an explosive level of scale with Samsung, and everything broke. And as we were wondering, like, how do the hyperscalers do this? You realize, because we were doing it on Dell and Supermicro and Force10 switches, and you realize that, like, oh, the hyperscalers don't do it this way, dummy. The hyperscalers are not Dell customers. They're not HP customers. They're not Supermicro customers. They've actually designed their own machines. So they've designed their own rack-scale machines. And we realized that's what we need to go do. We need to go actually do hardware and software together, do what the hyperscalers have built for themselves, make that as a product that could actually be shipped into the enterprise. With the big belief being that on-prem computing deserved modernity, that you should be able to run, you should be able to own cloud computing. You shouldn't be forced to merely just rent it.
So, like, what are the challenges with, like, regular on-prem computing?
Oh god, it's such a, the challenges are so myriad that to anyone that's deployed Dell or HP or Supermicro and tried to get it to work, especially as elastic infrastructure, it's like, where do you even start? It is a cobbled-together mess. It's a kit car. It's not designed to operate coherently. And it starts with racking, stacking, cabling, the software on top of that. You're dealing with multiple different vendors. Any standup on that commodity gear is almost tautologically a snowflake. So when anything goes wrong with that, every vendor is pointing at one another. And so there are, again, it's hard to know where to begin with the problems, and that's why people move to the cloud because there are so many problems. I can just go to the public cloud. It's like, this is so much easier. I can just go to this hotel. Everything is taken care of for me. But then the economics become debilitating. And the public cloud is great, but when you get to a certain size and scale, it doesn't make economic sense. And so that's what we sought out to do. Raised money in 2019. Very hard to raise money, even mentioning hardware in 2019. We had plenty of VC firms that told us that they were only interested in funding SaaS, period. You're like, 'Okay, so if Nvidia or AMD came through here today, you wouldn't fund them.' Okay. But we've fortunately found some terrific investors and were able to capitalize the company, introduce our product into the market in 2023, and it's been off to the races. It's been fun. We've got the, we very much have the 'how fast can you build them' problem right now. It's a good problem to have.
So what kinds of people should be using on-prem instead of cloud? Like, I imagine probably governments and people with sensitive data.
Sensitive data, security, anytime you, and that's kind of like, and actually our first customer was the federal government. They came knocking, and I told my co-founder, it's like, the federal government's not going to be our first customer. But they really beat our door down, and Steve was like, look, I think it's going to be faster if I just take the call and explain to them why they're not a customer for us. And so he gets on the call with, this is a national lab, Idaho National Lab, and he starts to explain why we don't think you're a customer for us. And they're like, can we talk for a second? And they described their problem, which was VMware on Dell and all the problems they had running in a secure facility. And they're like, our understanding is that Oxide would solve that problem. We're like, yes, we're very sorry. Oxide would solve that problem. And one of the things we learned is that when the federal government wants to move, they can move very quickly. So really proud that the federal government was our first customer, but it's not just in those regulated industries. It's also financial services. It's every vertical you can imagine, honestly, needs on-prem compute. And then the one that was the big eye-opener was, as it turns out, in these frontier labs, these workloads, there's a lot of CPU that's needed. The GPU is extremely important, but the death of the CPU has been greatly exaggerated. And so we've found across the neoclouds, across frontier labs, we've got a lot of demand for the product.
Yeah. I mean, we're seeing tremendous demand for CPUs now with all of these agent products.
Totally. Absolutely. Yeah. And I mean, our biggest problem, I mean, we've got shortages of everything right now. And being able to secure the supply chain has been our biggest problem. And getting the DRAM we need, getting the CPUs we need, getting the components we need up and down the stack has been a huge challenge.
Totally. I want to shout out our sponsor, Lovable. You know the app you've been meaning to build, the internal tool, side project, or product you'd otherwise lose a weekend to. With Lovable, you get software you can ship today. That includes an editable codebase you can inspect and change, plus two-way GitHub sync. Lovable also handles the backend and infrastructure, including managed Postgres, object storage, hosting, payments, and more, so you don't have to wire it all up yourself. And through Lovable's MCP server, you can create and deploy your project directly from the agents you already use. Turn ideas into software people love with lovable.dev. Now onto the episode.
So tell us about your article, 'Fool's Expertise,' which you wrote the other day about Jensen's interview on Ezra Klein.
Oh my god. And this is another one of these where it's just like, and I apologize to my wife. You know, it's a Sunday afternoon. I'm like, god damn it, I got another one in me that I got to get out. I just like, and so that in particular was on. So, and obviously most folks, many folks saw this interview that Jensen did with Ezra Klein. And it's interesting, that interview, it's like, I feel that interview reconfirmed your priors, whoever you are. Like, people kind of came out of this being like, and I had a very good friend of mine text me being like, I don't know who this, not a technologist, but someone I've known for most of my life at this point, he's like, I don't know who this Jensen guy is, but he sounds like a bit of a turkey. I'm like, okay, he's not a turkey. Like, he's like, this is Jensen, right? And I went to go listen to the interview, and I thought, I mean, I thought Jensen did admirably, but I thought the questions from Klein, I thought were honestly terrible, because they were forcing, I mean, they were kind of admonishing, he was admonishing Jensen for not deferring to the expertise of AI experts, like pioneers like Jeff Hinton. And like, nothing against Jeff Hinton, like, great pioneering work for sure, but Jeff Hinton is, and I called it out in the piece, not an expert in radiology. Like, Jeff Hinton just shouldn't weigh in on what's going to, I mean, if you want to weigh in on radiology, go spend some time with radiologists. Walk the floor with radiologists. Work a shift with radiologists. See what they do. Learn what a radiologist does, and then maybe you can figure out what's going to happen or not. In 2016, Hinton says we should not be training radiologists. It's like, yo. And this is like the thing that I kind of find upsetting is because I know the kind of effect that that has on young people. And if you're, I mean, you can easily imagine someone who was an aspiring physician in 2016 and was thinking which domain they wanted to go into and may have had a natural calling towards radiology and comes across this article from a putative expert and thinks to themselves, I'm not going to do that. I mean, meanwhile, it's like we need radiologists, and in fact, AI, I think, is going to help radiologists do their job. And I think that this is true across so many of these domains where these guys weigh in on things that are not their domain of expertise. And, you know, this has come to a head certainly with this extinction risk, which is just absolutely very frustrating. And I think many of us got calls from, you know, parents over the weekend and so on when that, you know, a couple weeks ago of like, should I be worried? Like, you're in, you know, you're in technology, should I be worried about extinction? And having to like talk everyone off the roof and explain. Although actually, you know, I should say, people are less on the roof than I think I was worried about because in the larger world, the commonsensical world, these arguments don't carry the weight that they do here in Silicon Valley. I think people have the natural question of like, how would this lead to extinction? You know, I'm a gym teacher in Little Rock, Arkansas, and I'm trying to understand how that would lead to my extinction. It's like, no, you're not wrong. It's like, nope, it does not lead to your extinction. And sorry, on behalf of everyone in Silicon Valley, sorry that we've let this kind of escape containment into mainstream discourse that implying that you should be afraid for your life and the life of your kids and the life of your parents. I mean, that's what extinction means. I think, you know, it's convenient to talk about it about x-risk and divorce it from the underlying humanity. But you're talking about my kids and my, you know, my sibling, my mom. You're talking about like the people that are most important to me, people most important to you, people most important to everyone else. And I just, yeah, sorry. I just, as you can see, gets under my fingernails, obviously.
They do mean it. I do try to explain to people that like, oh, this is like eight-dimensional chest from Dario. It's like, no, no, no. These are true believers. These are true believers. And I think, you know, Bill Gurley has been very outspoken on that. I think he's very right, that these are true believers. And, you know, not for the first time. We've got some very odd beliefs kicking around here.
Yeah, but, you know, obviously with current systems, extinction risk is like not even on the table. Like, it's absurd. You know, Astral is not going to kill everyone in the world. Astral 6.1, which was held back by OpenAI for safety concerns, is not going to kill everyone in the world.
Well, nor is GPT-2, which was held back, which we had safety concerns over GPT-2.
Well, I mean, not AI safety concerns. It was more like propaganda concerns. They thought GPT-2 would...
It was inarticulated. It was safety concerns. It was the same. And it's like very easy in hindsight to be like, 'Oh, no, no, no. The thing I was worried about was propaganda.' It's like, 'No, that's not what you said at the time.' At the time, he said his safety concerns. And so I think this is part of my frustration about this is those safety concerns kind of, they tack into these unspecified fears, and when you drive them to specificity, we contain that fear into something that's much smaller.
Yeah. I think the specific pathway that these people are worried about is something like current AI agents are, you know, starting to be able to improve themselves. Eventually they might be able to improve themselves to a point where the thing that comes out the other end is like very powerful.
What does powerful even mean? I mean, we've already like accelerated ourselves into science fiction as far as I'm concerned.
Yeah, but sometimes science fiction becomes real.
Oh, Jesus Christ, which is fine. Okay, like, so fine, if like, if sometimes science fiction becomes real, if that's like the headline, like let's lead with that headline so the rest of the world can be like, oh, okay, this is like, yeah, sure, sometimes science fiction becomes real. Like this is, but this is science fiction. Let's actually, let's call it what it is.
For now. Yeah, but sometimes science fiction becomes real, like wireless, you know, instant communication across the world.
Lots of times science fiction does not become real.
I mean, like, you know, I grew up, I'm, you know, we were talking before we started recording, I'm very much a bullseye Gen Xer. I've got my space Lego set. You know, we were going to have moon colonization so many decades ago. We were going to have flying cars. We were going to have jetpacks. I mean, there was so much stuff, so much of that futurism in the 80s that felt we were going to have supersonic travel everywhere. Like, none of that has come to fruition.
All of it has from a technical standpoint. It's mostly because of political constraints.
We had supersonic travel in like the 70s.
Yeah. Well, it's not just political constraints, right? I mean, people don't want supersonic booms across the continent.
Yeah. I have not flown in a Concorde, although my grandparents have.
I mean, it is and isn't. You know, my grandfather also flew in a Concorde exactly once. He took the upgraded Concorde class, and it was like, fine, arrived earlier. He said it was very uncomfortable. It's like, it's super narrow. And yeah, it solves one problem, but it creates a bunch of other problems, a bunch of other problems that like we actually, we as a society, like are not that interested in. Like, we don't want to have sonic booms all the time. We just don't. We don't want noise pollution. We don't want, we like, efficiency matters. Like, the cost matters, the economics matter. I mean, that's I think what people get divorced from, the actual economics of these things. And economics is the driver of everything, and sometimes people chalk up to kind of politics what is actually economics.
Right, but going back to the, you know, the doom story path.
