A Product Market Fit Show | Startup Podcast for Founders
00:12:58 · Steve Tuck
[00:12:58] S0: This is 2009. Okay. So cloud computing is super nascent. If you were running an AWS in 2009, 2010, it was kind of wild that there was this new mechanism where you could hit an API and kind of throw code at it and run applications. And like everything else just kind of worked. Also, sometimes things didn't work and you would have an outage that would be like three days long. And so the first era of cloud computing were all of these gaming startups and apps and fintech companies. They were the ones that were willing to go launch in the public cloud and operate there. And it would be a couple of years until the mainstream enterprise was like, wait a minute, we actually could use this, should use this and began to adopt it in scale. And then obviously the last chapter, most recent chapter of cloud computing has been AI. This is where all of the major labs grew up. This is where all of kind of the AI building has occurred is in cloud computing. And the thing that we learned over the 10 years at Joyent was sort of threefold. Number one, the way that Joyent was doing this was writing all the software portion, the cloud control plane and the update engine to be able to deliver cloud services, but buying off the shelf hardware. And short version, it turns out when you take software and try and pack it onto off the shelf PCs and rack form, things don't go well. All the problems around performance, stability, availability, they kind of stem from that hardware software interface. And, you know, like Apple pioneered and learned and like Tesla pioneered and learned the hyperscalers, Amazon, Microsoft, and Google realized we've got to blow this up. We have to redesign the modern computer from the ground up and co-design software in here for the cloud era. And so the machines that power the world's biggest cloud computing hyperscalers look nothing like what you can buy off the shelf from Dell, HP, Lenovo's, micro, you know, storage providers, network providers. And when we were kind of evolving at Joyent, we then got acquired by Samsung. And that's when I moved into kind of an operating role and then eventually would run the entirety of that operation. At the scale we were running, which was powering the Samsung cloud globally, that hardware software disparity just completely broke down.
00:15:16 · Steve Tuck, Pablo Srugo
And it was at one of our lowest moments where I was at OCP Summit, which is a place where all the big hyperscalers come to show off their wares, that we would first witness these incredible rack scale machines that Facebook had for themselves that were transformative. They were so much more dense and energy efficient and operable than anything that you could kind of compile off the shelf in hardware. And I was desperate. You know, I had my checkbook out from Samsung, like, how do I buy these? And, you know, how fast can you ship them? And the Facebook team kind of looked at each other and like, sorry, do you work for Facebook? [00:15:55] S1: These are not for sale, right?
00:15:57 · Steve Tuck
[00:15:57] S0: No, they're not for sale. These amazing machines are not for sale. And it would have truly been the hand of God reaching down in the midst of this five country build out we were doing if we could get a hold of these modern appliances that apparently the best companies in the world have been running on for seven years. And that was where kind of the first deep seeds, we wouldn't know it at the time, but like looking back on it, I think that painful realization coupled with the fact that it's so hard to build high quality, reliable, automated systems when you've got this kind of kit car approach of hardware and software, that we would land on this realization that the same machines that power the cloud need to be available for those that want to operate on-prem, that want to own and control their own infrastructure, that want economics that are massively different than the rental economy of cloud computing, that want, you know, systems and data that no one can access. And yet to do that today for the nearly 60% of IT infrastructure that runs outside the public cloud, you have to go back in time like 20 years and you have to absorb all this overhead in your business to buy these disparate pieces and weave them together. And then, you know, if anything breaks, by the way, you're the only one running that combination of things on the planet. You're the only one responsible for it. And it was in these years that we really realized like cloud computing is the future of computing. Cloud computing is an architectural concept, not a rental concept. And that the machines that power the cloud need to be available in product form to kind of transform the on-prem IT economy, which is huge. $100, $150 billion growing as the economy moves online.
00:17:40 · Pablo Srugo
[00:17:40] S1: What is, by the way, on the, for the on-prem, like there's a security reason to stay on-prem. There's a cost reason. Like what are the top, you know, two or three reasons why some of the big, big enterprises are going to keep some of that payload for always, you know, on-prem.
00:17:54 · Steve Tuck
[00:17:54] S0: Security is one. Just, just that governance and understanding that you can verify no one has touched these systems. So you can put subsets of data there and know exactly, you know, how it's being used and who is accessing it. Economics is a big one. Like there's a reason that people lease apartments and buy homes instead of everyone living in a hotel room. You know, when you have a certainty around a subset of usage over years, then owning is not just a little bit more cost effective. It's orders of magnitude. And then there is latency reasons. You know, we have customers that running infrastructure that's collecting data from their own factories and data centers and systems that they need to aggregate locally and make decisions on in real time. And so latency can be a key contributor. And then for software companies, you know, cloud SaaS companies, AI labs, it is the ability to sell the new markets. You know, if you're trying to provide services to a bank and you run explicitly, you run your SaaS platform in the public cloud, you're only going to get a percent of that bank's data. But if you can bring that into a single tenant on-prem model, you now can service the entirety of that bank's data. And so in regulated industries, there's actually an interesting emergent kind of angle that we have heard from customers, which is business growth, is expansion to new markets, being able to grow the top lineup.
00:19:24 · Pablo Srugo
[00:19:24] S1: But who pays for the rack in that example? Like if Salesforce is selling into a bank, maybe that's a bad example or not. You can tell me, are they buying that on-prem rack and putting it inside? Like how does that work? Where are the economics?