Who needs VCs when you have friends like these?

Who needs VCs when you have friends like these? - Stack Overflow

Stack Overflow Business Stack Internal: the knowledge intelligence layer that powers enterprise AI.Stack Data Licensing: decades of verified, technical knowledge to boost AI performance and trust.Stack Ads: engage developers where it matters — in their daily workflow.Runpod is an end-to-end AI cloud that provides developers with GPUs so they can build and run custom AI systems that scale.Connect with Zhen on LinkedIn or email him at zhenlu@runpod.io.Today’s shoutout goes to Famous Question badge winner cigol on, who won the badge for getting 10,000+ views on their question Using JavaScript, is it possible to capture the body payload from an outgoing fetch request?.TRANSCRIPT[Intro Music]Ryan Donovan: What happens when AI relationships evolve from reliable tools to trusted confidants? Raymond Yin of The Tech Between Us and Dr. Marisa Tschopp, researcher with SCIP AG in Zurich, discuss how this evolution will shape our daily interactions with AI. Listen from your favorite podcast platform or visit mouser.com/empowering-innovation.Ryan Donovan: Hello everyone, and welcome to the Stack Overflow Podcast, a place to talk all things software and technology. I am your host, Ryan Donovan, and today we are talking about skipping the VCs and going right to the community, building a company based on community input, and my guest for that today is Zhen Lu, who is the Co-founder and CEO of Runpod. Welcome to the show, Zhen.Zhen Lu: Thanks so much for having me, Ryan.Ryan Donovan: Before we get into the community driven stuff, we want to get to know our guests .Tell us a little bit about how you got into software and technology.Zhen Lu: I took a bit of a winding path as I'm what to do. So, actually, and I'll save you the gory details, but was raised in a pretty traditionally Asian family, was really encouraged to think about higher education, medicine, legal– pretty common stuff, right? I fit pretty squarely into the stereotype, unfortunately. And so, I actually followed that because it was very interesting to me. I ended up chasing hard problems into a PhD in quantum chemistry. Really enjoyed that because it allowed me to dig into not only chemistry, but it was the intersection [of] math, chemistry, physics, [and] biology. Super interesting. Did some research on electronic structure theory of DNA base pairs. But then, eventually found myself wanting something that had a more immediate impact on the world. And so, I felt that software engineering could give me that. So, I ended up swapping into software engineering, working with my Co-founder, Pardeep, and the rest is history, if you will. So, I'm somebody that really likes challenges, and I'm not afraid to change my life around to give myself those challenges.Ryan Donovan: It does sound like a pretty distinct change, there. So, you got into founding a company. You skipped the sort of VC path and went right to community development. We are a company at Stack Overflow that is very community driven, and it can be very insightful, but it can also be very volatile. Why did you choose to go right to the communities?Zhen Lu: Yeah. Ryan, I would love to say that it was all in the master plan, and we knew that it was gonna work out this way, but the truth of the matter is that my Co-founder and I, we identify as software developers, right? So, prior to starting Run Pod, we worked together for about six years in the same software development team. We built the team from about eight people when we started to almost 100 people. And we found a lot of joy not only in the craft, but working with other talented people. So, when it was time to start Run Pod, and we can talk about why we decided to start Run Pod, but when it was time to do that, we really were like, what are we good at, right? We're not expert marketers, we're not expert salespeople, we're not experts in the capital markets, but what we are experts [in] is software. And so, we decided, okay, we're software developers. Let's write some software. We don't need other people's validation, money, et cetera, to just go do that. And we did end up building some servers. We funded them ourselves. We ran them in our basements, and we said, ' we have everything we need.' Let's write the software and let's see if people are actually interested in it, rather than following the more traditional path of raising the money, [and] all of that stuff. So, we were just doing what we love to do.Ryan Donovan: Obviously, when you found a company, there are more than just, writing the software. So, what were the expectations, the theories, hypotheses, that were validated by the community by throwing out to the wolves? And what sort of things were surprises to you?Zhen Lu: First part of the question: the thing that was really validated was that people wanted this. And so, we always set out with a pretty ambitious vision of wanting to define the next level of software development. We had the strong conviction that software development was going to move towards machine learning. At that time, gen AI wasn't really a thing, and we thought that these types of accelerators, GPUs, or otherwise, were going to become a bigger and bigger part of the entire equation. And so, when we looked at what we were doing at our day jobs, we were building some massive distributed systems in the cloud, and we were working on some pretty cool machine learning projects, as well. And so, we were like, this is pretty awful, actually, from a development experience perspective. And we were like, this is gonna just get bigger and bigger. It's just a matter of time. We didn't necessarily know if it was gonna take one year, two years, five years, 10 years, but we were cloud practitioners. What would it take for us to help build a cloud? And we missed the bus the first time around, and we didn't want to do that this time around, so we wanted to do it earlier, even if it took a couple years. Now it turns out that it was the right time to do– there was pain already. Even at that time, this was like researchers – the people really at the cutting edge of ML, they wanted something easier to use that was built with them in mind, rather than them going to a hyperscaler or other cloud, where it's like, 'hey, it's a traditional cloud stuff. If you want these weird GU things, good luck.' I think that's where we got our initial validation from the community, where when we launched it, it was free. We just made a Reddit post. We said, look, we built something for the community. We think it's pretty good. We're not gonna charge you money for this. Honestly, it's running on the GPUs that we have in our basements. So, whatever. All we want you to do is if you wanna use it, tell us what you wanna use it for. If we think it's interesting, we'll get you access to the platform, and all we ask for is that you basically just give us the cold, hard truth, right? And we were incredibly fortunate that we had people that really cared, and they gave us some pretty constructive feedback. But overarchingly, it was like, 'please take my money.' And it doesn't really get better than that. Let's see what we can do to scale from here.Ryan Donovan: Let's step back and clarify what 'this' is we're talking about – what [was] the thing that you built?Zhen Lu: Yeah, so our V0 product that we launched day one was all about development environments that had GPUs enabled. So, what our initial users had a lot of success with was they were used to, let's say, going to an AWS of the world and spinning up a virtual machine, and then needing to install everything on the virtual machine, and then figuring out like how the dependency matrix has all sorts of crap that really no one wants to deal with, and it took a lot of time. And so, when we launched Run Pod, we just went, let's start with the first product, it's just development environments, incredibly fast to spin up, incredibly fast to tear down. The stakes are low for you to just go do the things that you need to do. And that resonated really well with the community, which was great. And then, what we found is then, our customers were becoming more successful. Some of them were actually launching real businesses, and then they were pulling us to offer them more and more abstraction, which we were happy to do because we always knew we wanted to build all of those. So, we added on things like serverless autoscaling for truly custom workloads, really fast cold starts, and just really resonating with the developer in the fact that, they need to iterate quickly, right? It doesn't matter if it's traditional software development or AI software development. The key thing to enable development is to be able to iterate quickly because the faster that you can make your changes, the faster you can see those changes reflected, the better you're able to make progress.Ryan Donovan: When you say you were offering Dev-only environments, does that mean they had to take their production deployments elsewhere?Zhen Lu: At this time, there wasn't actually that much production. If you think about what people were doing with GPUs at that time, a lot of it was research. Now, there were more traditional HPC workloads. These are things like molecular dynamics, protein folding, pharmaceutical drug discovery, those kinds of things. And we definitely did get some companies that were interested in running those types of workloads on Run Pod. But really our bread and butter was to focus on the emerging AI workloads, and our ability to serve those was more important to us than to serve the more classical ML HPC workloads.Ryan Donovan: So, when you're getting this feedback from the community, and they're asking for additional, like you said, the serverless quick start functionality, how much of those specs are entirely community driven? And how much of those are you planning ahead and being like, 'let's validate this with the landscape around us?'Zhen Lu: Yeah, so just to clarify, are you asking where the