Vultr Builds AI Cloud Beyond Hyperscalers

Hey everyone. We're wrapping up our PlatformCon coverage here. I don't know if we saved the best for last.

I'll let you be the judge. But let me introduce you to my friend, Duncan Eng. Duncan is with Vultr.

It's spelled V-U-L-T-R. You probably know this because we've covered Vultr before. But Duncan, welcome to techstrong TV and our little studio here at PlatformCon.

This is an amazing view. Million dollar view. Yeah.

Yeah, folks don't... Million? I think billion these days.

Yeah. Exactly. This is quite the setup and- It really is ...

thank you so much for having me on. I appreciate it. My pleasure.

So let's start with a little Duncan background, right? You're with Vultr now. Mm-hmm.

What's your role at Vultr? Yeah. So I'm the Vice President of Solutions Engineering at Vultr.

I've been here, I actually just celebrated my fourth anniversary. Really? So, given the times, that's a pretty long tenure at a place.

Yep. I've been in the infrastructure business for almost 30 years now, and have really enjoyed how we've continued to grow and mature this ecosystem. Absolutely.

I know what 30 years feels like. Yeah, we were just talking. But, Duncan, when you say we're in the infra...

Infrastructure means a lot of different things today. We have infrastructure as code. Yep.

We have a lot of people, infrastructure as code name for GPUs. Yeah. Right?

But it wasn't GPUs 30 years ago. It wasn't GPUs three years ago. Or three years ago, three and a half.

You in the data center part of things, or? So, I've kind of done every level of the build-out. Started my career as a network engineer.

As part of my responsibilities there, the team also did some of the data center construction. So, kind of my first, quote unquote, "big data center" was when we were going to build a three megawatt data center, and people were like, "What? " Wow.

Why would you need that much? Exactly. So this was back in dial-up ISP service provider, kind of starting to see- I know those days.

AOL days. Exactly. Starting to see that big pickup on the consumer side.

So started my career there, and then actually moved onto the academic side. I worked at the University of Southern California in their academic computing. Oh, very cool.

So kind of both sides, both on the network side. I rolled out Wi-Fi across their campus, 50,000 students, 50,000 administrators and faculty. Participated in some large technical advisory committees in California covering K through 12, junior college, state universities, and then another network that included things like the San Diego Supercomputing Facility, Stanford Linear Accelerator, the Naval Postgraduate School in Monterey.

Been there. So those types of ESnet, Lawrence Livermore, Lawrence Berkeley. So these Department of Energy kind of projects.

Exactly. So getting to work on very large scale networks, and while at USC, we had started the process to build a top 500 supercomputer, so working in the HPC space, very early on in my career. Absolutely.

I wasn't always on the media side. I've done a lot of startups. One of the startups I helped co-found was big in US government security.

Mm-hmm. We didn't call it cyber then, it was security. But we did a lot of work for the DOE labs there.

Yep. A lot of people don't realize, most of the people working at those DOE labs were University of California employees, actually. Yep, absolutely.

At DOE. They've run Los Alamos for years. White Sands.

Exactly. I was at all of them. Yep.

I was at all of them, but that's all I'm going to say. Anyway, Duncan, let's talk about Vultr. Yep.

As I said, we've covered it before. Some of our audience might be familiar, but I'm going to say a lot of people probably are not. How would you describe the company to them?

So the company is rapidly growing. It's really been a very interesting journey here over the last four years. So let's actually start at our founding.

The company's been in business now that we'll celebrate our 25th anniversary this year. Really? Oh, my God.

I didn't realize it was that old. Absolutely. A lot of folks kind of lump us in with the NeoClouds.

Right. So it's kind of, in terms of just maybe market visibility, I think is the way to think about that. Until fairly recently, it had no sales team, had no marketing team, whereas that's radically changed in the last three, four years, and it's really a case now where because of that long operating history, like you talk about, having worked in security previously, you don't do a SOC 2 Type 2 overnight.

You don't do an ISO overnight. You don't get all of the things ready to do FedRAMP overnight, and FedRAMP is something that we're in the process of right now. We're working on this from a global perspective as well.

Sure. Not just here in the US. We have 33 public locations.

We've got smaller sites that we work with specific customers on, and then a case where even in locations like India where we have three data centers, we're working on their version called Mighty, in terms of how we work with their federal government and states as well. Sure. So, first of all, I will tell you, having done a bunch of startups- Mm-hmm ...

it's not a bad thing to always be grouped into the latest fad. It's not even a fad, but to the latest sort of market. And this whole NeoCloud thing is blowing up, right?

Because putting all your eggs in three baskets may not be the wisest strategy. The whole issue of digital sovereignty- Absolutely ... sovereign AI, data sovereignty.

Mm-hmm. People are looking for alternatives- Absolutely ... that the NeoClouds offer.

So I don't think it's a bad thing, right? Not at all. You're hanging with the kids in the NeoCloud category.

But now you're operating full-fledged Your own data centers kind of thing, or? Yeah. So for us currently, don't own the walls and don't own the roof.

Everything else is ours. The servers, the network, the storage, all of those are ours. We, at this point, are still working with data center providers to provide the physical facility, the MEP facilities.

For us, it's really a case where if you just really look at this as a return on investment perspective, I can spend hundreds of millions of dollars building the building, or I can take that same money and buy GPUs, and which one delivers me a greater return on investment? Right. So, not a whole lot of sense to buy concrete.

No. Well, and we see that going on, right? Mm-hmm.

Even Google. Yes. They just made that deal with SpaceX.

Mm-hmm. For, what is it, 110,000 GPUs a month. Yeah.

I think it's almost a billion dollars a month. Yeah. At the same time, SpaceX sold to, is it Anthropic or OpenAI?

Anthropic, yeah, as well. 220,000- Yeah ... GPUs.

They've already committed 330,000 of their GPUs, though they're cancelable on 90-day notice- Mm-hmm ... to these two companies. Then they announced yet a third deal with another neo cloudish kind of provider.

So look, we see the problems with data centers being built. You can't get enough electricians. Mm-hmm.

Can't get enough concrete, can't get enough electricity, can't get enough water. Local opposition. Yeah.

The NIMBY crowd. Quite frankly, who the heck wants to be the guy building the data centers anyway at this point? I mean, they're a lot of trouble.

Yeah. I also think it's a case where that market and those builders are starting to be a little bit more aware about how do they need to reach out to their local communities. Yeah.

How do they frankly need to explain how this campus here uses less water than the local golf course does? If you live in an area that has 18 holes or 36 holes, that this is a completely different scale where a closed loop data center significantly uses less water than some of those- Absolutely ... very kind of common use cases.

There's a lot of that, but you also, Northern Virginia's the data center capital of the world. Yeah. People there are complaining about noise- Yep ...

and this kind of thing. But here's the bottom line. We're committing $8 trillion worldwide to new data centers.

Mm-hmm. Talking about putting data centers from the bottom of the sea, which China's already done- Right ... with a wind farm on top of it, into space- Mm-hmm ...

with Elon and them. Data centers are coming. They're already a part of our world.

Yeah. But as we scale out this AI era, we're going to need those data centers. Now, I wrote a piece a couple weeks ago, Dustin, where I analogize the AI industry to the airline industry.

Mm-hmm. Most airlines don't own their planes. Mm-hmm.

Right? They lease those planes. Exactly.

They make long-term commitments for fuel. Mm-hmm. There's a stratosphere, if you will, of partners that make up the modern airline stack.

Mm-hmm. It's going to be the same thing- Absolutely ... with AI.

We're going to be a stack. The question I have for you is, is Vultr just sort of we give you the GPUs, use our GPUs and now everything that goes around the GPUs in terms of power ping pipe or what have you, or do you add value above... Not that there's nothing wrong with just doing GPUs- Mm-hmm ...

but do you add value above that? Yeah. So I think that's one of the big differentiators between us and some of the other- Neo clouds ...

more recent entrants is that because we started as a cloud provider, and a case where even going into this, we always saw this as the GPU in and of itself is not the end goal here. Data's got to come in and get ingested, somehow fed to the GPU, the GPU will do its work, and then you've got an output that then needs to either be presented or needs to be manipulated again. So even before the last kind of agentic boom of the last six, nine months, we always saw CPUs as an important piece of that entire ecosystem.

So for us, it's really a case of bringing on all of the experience that we have orchestrating cloud workloads, all of the experience that we have dealing with networking to be able to tie into someone's on-prem or colo solution into their multi-cloud existence already, and how do we add to that? How do we bring the compute, whether it's CPU, whether it's GPU, to solve the business problems at the end of the day? " Besides the application here, the example I use all the time is I want to generate a cat wearing pink pajamas versus a cat wearing blue pajamas, right?

I'm sure someone will pay something for that. I'd rather spend my time and energy on working with companies that do interesting problems, solving how thermodynamics in a ceramic chip packaging for a chip, using diffusion models or a company that we work with doing precision medicine. There are some very amazing horizons that we can get to using these tools that we could never really consider doing five or 10 years ago.

Love it. Duncan, I want to pivot a little bit and talk about platform engineering. We are here at PlatformCon.

Yep. Absolutely. Look, I've watched this platform engineering community grow, right?

com. com. Obviously, like everything else in our world, AI has had a huge disruptive...

I don't know if I want to use the word disruptive, but it's had a huge impact. I think it's another tool, and we're learning how to use that tool in better ways. Yep.

Yeah. Specifically for our platform engineers out here How's the game changing? Yeah.

I think the evolution in the last two or three years has been absolutely stunning. We've essentially gone from the concept of a one-shot, a prompt, "Please do this," and have the system shoot back something for you. If you think about the evolution of a software engineer, a DevOps engineer over their career, when we start at the beginning of our careers, we have limited context, we have limited ability to execute, and that's essentially what that was.

Limited context, limited ability to execute. You'd figure out what it was good for, and you're like, "Okay, I can now do this that I would have to spend time. I need to rewrite this array.

" As opposed to doing it by hand, that could do it for you in two minutes. You're like, "Great. " I think we continue to see that grow and continue to mature.

You've now worked a couple of years in platform engineering as a DevOps. All of a sudden, you've got broader context, you have broader ability to execute. Now it's almost like that deep reasoning that we saw a year or two ago.

Right. Where once again, I now am scoping something in a much more kind of way that maybe a product might be scoped, a PRD. I need you to execute or I need you to build this function that does this and does that and interacts with this.

It has the ability to pull resources from outside, has the ability to architect something. Might not do everything correctly, but starting to be a little bit more extensive in terms of how it can get rolled out. And I think the most exciting part of the last six months is kind of this evolution of loop engineering, right?

No longer is the senior engineer, lead engineer sitting here writing the code themselves manually. They're now orchestrating, first of all, the planning piece. What are we trying to build?

Okay, now that we've identified that, how are we going to validate that? And then from there, okay, we've built it, we validated it. How do we then stage it to go into production?

And then everyone's got their own answers on whether you're going to let your AI systems push into production on your own before, still have human in the loop. Oh, that's what I was going to ask. Is that what loop stands for, human in the loop?

No, it's really that kind of- I hear you ... iterative process, but having the human in the loop there where, especially for mission-critical type systems, financial services systems where from a regulatory perspective or just from a, at the end of the day, an engineer hit the button, congratulations, you didn't do your job, you can get fired. Right.

Firing an agent doesn't get you a lot. Well, every time you fire an agent, it's kind of a death penalty, right? Yeah.

But it does bring up the issue of the different roles of the agents are going to have. Yeah. This agent may have this role, this- Yep ...

But that's also tied into another issue that we hear a lot of. Well, look, my senior engineer should not be threatened by AI because he's going to be a loop engineer. He's going to be more- Yeah ...

valuable than ever. Yep. But you know what?

The junior guys, yeah, I don't know if we're going to need them because the agent may do a lot of what they're doing. What's your take on that? I think this is just changing the onboarding for an early stage career.

You still need to fundamentally understand these ideas because if you don't understand the ideas, you can't fix it when it's broken. The system is going to output something. You still need to be able to analyze whether the system delivered it properly or not, and it's a case where the expectations of someone early in their career are going to continue to modify and change, where they will still learn the, let's call it the architectural, the context of how do we want to do this so that they can move forward in their career.

I think companies that take a very short-term analysis of the AI can do items one, two, and three that my junior engineers could do, don't understand that they will need someone with that context of how it got built, why these decisions were made, so that down the line, when their senior engineers move on, that they will have someone with the capability and with the context to continue maintaining and extending the system. Duncan, I'd like to talk to you, but we're going on here a long time already as these things go. You know what we didn't talk about that we'll wrap with it?

Mm-hmm. People want to find out more about Vulture? Yeah, absolutely.

Come to our website. And we've got a booth here at PlatformCon. We're at most of the large trade shows, whether that's KubeCon, whether...

We're really invested in being active parts of the community and please come and reach out and more than happy to do that. If I can even throw in a plug, I'm hiring solutions engineers on my team, both on the CPU side and on the GPU side. There's enough people looking to judge, please.

So absolutely. Thank you for that. I will have some post things that you'll find on my LinkedIn, and please don't hesitate to reach out.

Love it. Duncan- Thank you so much ... pleasure.

Duncan Eang, Vulture, here on techstrong TV at PlatformCon. Hope you've enjoyed today's coverage. We've spoken to some interesting people.

com, or any of the techstrong sites you hear on techstrong TV. But for now, this is Alan Schmel. Thanks for joining us.

Bye-bye.