 Hey, good morning everyone and welcome back to theCUBE, the leader in live tech coverage, covering SC23. This is day three of theCUBE's four days of wall-to-wall coverage of this event. Lisa Martin with John Furrier. I love talking with companies that are new to theCUBE so we really get a great opportunity to understand what they're doing in this context, HPC, AI and ML. Kevin Cochran is here, the CMO of Vulture. Welcome Kevin, great to have you. It's great to be here. Thank you so much for having me. Love the name. Vulture was founded back in 2014. Tell us what the company does, your mission, your vision. Great. Vulture is the world's largest privately held cloud compute platform. We operate in 32 global data centers around the globe and we offer full stack cloud compute including all of our top of the line NVIDIA GPU plans. We were started in 2014 as a platform built by developers for developers. And the pain point that we were looking to solve is how to give developers access to unrivaled performance at an affordable cost with unparalleled global reach. And so the platform simply grew organically. Developers found us, loved us, and we scaled very rapidly. And then about a year and a half ago we started investing in enterprise sales and enterprise marketing to start tapping in a broader opportunity with the coming generative AI and HBC revolution. What was the reason why the developers adopted you? As people know in theCUBE is developers are the new, set the new standards. When they're doing stuff, things are happening. So what was the core reason they were jumping on board? Just a few simple reasons. We're simple. We're easy to use. Our pricing is transparent. It's predictable. And as a group of network engineers that simply geek out every single day running performance tests to eke out every single millisecond of latency, the performance was just unparalleled. So developers just fell in love because it's a platform that developers understand and just love to use. And it's fast. And it's fast as hell. No one ever cried when he had fast packets moving around. That's correctly. And no one ever cried when your bill was between 50 and 90% lower than what you would get from Amazon. It sounds like you've had some, an acceleration of growth. You talked about enterprise sales in the last year and a half. How has the last year of generative, the boom of generative AI, how has it been an accelerance of Vultures Graph? Yeah. So I think it actually started even before the debut of the H100 and ChatGBT where enterprises are starting to rethink their enterprise application architecture to push cloud native applications and serverless functions ever closer to the edge. And what happened in 2023 is what we call the art of the possible. With the debut of H100, with the debut of ChatGBT, enterprises started rushing en masse to start building training clusters, to start experimenting what new power could they put behind all of their new cloud native applications to deliver better outcomes for their customers and their employees. But what we think is very exciting is what's to come in 2024 because it's trying to move beyond experimentation and single large scale training clusters and moving beyond the art of the possible to the realization of actual outcomes. And this is where it gets exciting because this is where enterprises are going to start architecting for the future. Let's unpack this because I think this is the heart of the conversation Lisa, we've been having all week and coming in from the event. So every enterprise saw ChatGBT, the consumerization of that feature kind of woke up the market. Everyone's educated. It's become a user expectation. Every company is looking at putting their AI clothes on, it's what I call, and they're going to go to the AI party. It's a gift. Everyone's happy. Great. So more GPUs are sold. People are starting to stand up infrastructure. They have to get their training wheels on to figure out how to approach the AI infrastructure. Not hard concept to grasp. They got to just do it. So we've seen a lot of activity. Take us through the progression of a customer because they have to train their data and then the values coming out on the inference side. So we're seeing this training versus inference conversation. And as context that Tim Hocken at KubeCon said last week on the stage, he's from Google said inference is the new web app. Implying that this is going to be a future thing that everyone will have to be doing and using. What does that mean? Take us through training versus inference. What's important? What's the sequence? What's the progression? Well, like I said, 2024 is going to be about the realization of outcomes, which means we need to industrialize the continuous innovation and delivery of innovation with generative AI. So in the cloud data application world, we spent the better part of 10 years optimizing CI CD pipelines and putting in place proper infrastructure to support build, test and production deployment of all of our application code. And now we need to look at our generative AI, our actual models, and we need to apply the same principles, which basically means we not only need to stand up training clusters, we need to actually optimize how those trained models are then deployed and then fine tuned in region against localized data sets. And then once fine tuned how they can then be deployed efficiently to large scale inference clusters running the new GH 200 super chip from NVIDIA powered by Vulture and actually deliver predictions for incoming user requests. Think about this. We just spent the past 10, 20 years pushing compute power as close to the edge as possible so that people have the fastest performance response when they're on their mobile phone or on their desktop looking to access our services or information about our company. All of that is now going to be truly hyper-personalized based on the context of the user. True personalization finally becomes a reality with inference, but what happens if all of a sudden every single user request on your mobile phone if you're in Mumbai has to round trip to a server machine in Santa Clara? The personalization angle seems to be... That makes no sense. So is that an infrastructure problem or is that a data problem? What is the situation on the ground to solve this problem? But the situation on the ground is we need to unlock the potential of innovation in open source LLMs so that enterprises can take these things down, actually use their proprietary data sets to train these models, tune these models and then ultimately scale out all of their inference clusters that across the world, close to where people live and work. Inference is the new web app because in the web application world, you actually are running your actual raw compute power at the edge. So too should your model be running at the edge. This is where Vulture has been architecting a new global architecture for generated AI. Your Kubernetes clusters running your core web application code, co-resident with your Kubernetes clusters running your inference nodes and models, all scaling up and down seamlessly around the globe. So just to close this out, I know Lisa wants to get into the question, but I just want to close this thread out because you're right at the end of it here. You got the training your data at step one, and that's just a setup, almost like a set up a sandbox. That's correct. Set your training data, you're under data. The inference is where the application is. That's going to be the ongoing data iteration, whether using synthetic data to train something or using existing data to have generative results. That's right. That application is, that's the new game. So from an infrastructure standpoint, train provision stuff for training, and then the scale and the app is the inference aspect of it. Is that right? That is exactly right. And what we are going to see over the next year is the first new web applications going live that are powered by generative AI models and the first build out and scale out inference clusters to actually support real time delivery of predictions to personalize each and every user request and experience based on their context. And over the course of the next three to five years, we're going to see a wholesale redevelopment of every single web application. And every single web application will actually be delivered by an inference cluster. Inference is the new web app. That is our mission here at Vulture. That is our vision. And that's what we're going to make a reality with GH 200 Superchip. So when you're in customer conversations, Kevin, what do you articulate as like the top three differentiators of what Vulture is doing in this space that nobody else can touch? Absolutely, great question. Number one is we have Unrival Global Reads. So we are the only independent cloud provider that can provision top of the line NVIDIA GPUs on Dell hardware across 32 regions around the globe, the same scale as Amazon, all six continents. We're also the most secure and compliant independent cloud. So when you're dealing with your proprietary data sets and you're dealing with rebuilding your enterprise application, security, compliance, governance, these things do matter. It is part of the global scale. Third is we have a unique set of services to enable the entire pipeline of distribution of both your application artifacts, your code, co-resident with your generative AI artifacts, your models, and we're the only vendor that basically provisions Kubernetes clusters for your application models, your application code, so that you could actually truly reinvent your enterprise. Why is this different? Because you mentioned Amazon earlier from a cost perspective, what's different about this? Is this because AI now is in demand so that you basically have an AI cloud for that kind of demand? Or is this just a cloud environment for just developers? Because these are the clouds out there, you get the hyperscalers. So where does this fit in on the, if I had to kind of put a power law together, are you right up against AWS or you just like, no, we're the AI cloud, period, full stop? It's actually really interesting because of our unique position of being able to scale out today across 32 inference clouds of the new GH200 Super Chat, everyone keeps saying you're the inference cloud, you're the inference cloud. No, we're just the future of the cloud. Because the future of the cloud is large scale, globally distributed, Kubernetes clusters at the edge, and as many points of presence as possible, running all of your serverless functions and all of your generative AI models. We just are at the cutting edge, the bleeding edge, some might say, of putting in place the application and infrastructure architectural model to unlock 2024 potential. We had Dell on earlier, I know Dell's a big partner, we'll get into that in a second, I'd love to get your perspective on that, but the Dell folks were talking to us from a couple of different perspectives, from a design perspective of green sustainability, what not, but from a storage perspective, it's not about storage anymore, it's about compute computation. We're at high performance computing show. So it's not compute, it's a computation platform. Right, correct. That's essentially what you got. That's exactly what we have. And the scale out piece, inference just happens to be the element that will be lingua-franqua app for powering the experiences from broad horizontally scalable data sets, and then offering precision personalization. That's 100% correct. This is the future. This is the future. It's old. How does Dell play involved? I mean, we believe it, we've been saying this, it's cloud's horizontally scalable. That's the goodness of the cloud. Correct. The problem is the vertical specialization of the apps that's set on top of it, but then with AI you can now have both, best of both worlds. You can get domain expertise in your app or your environment or your application, taking advantage of the broad available data. So you're going to have breadth and depth with AI that you couldn't do before. And isn't this incredibly exciting time? I said AI is the fountain of technology youth because at my age, I want to be 25 again. That's right. It's like, how could you not be drunk on AI right now? I mean, this is real. The hype matches reality. And we're just getting started in 2023. I don't even think we understand what happens next in 2024 when enterprises actually start realizing outcomes and moving from the art of the possible to actually achieve new things. So I was talking to Dell and Savannah, at least in the opening set. Dell as an interesting guy, Michael Dell's my age, so we kind of grew up together with him in the industry. They had a mail order company. They sold machines through this dorm room, through the mail. The internet came. So technically the mail was going away. Dell wasn't disrupted. They transitioned to the web. It became the number one PC manufacturer server in the world through supply chain, direct selling mail, now web. They're kind of transitioning to an AI company because they have the goods. How would you see that described to Dell the today's modern AI Dell? What is, what's their role? What do you see them doing? Well, we have two large strategic partners, obviously NVIDIA with the new grades, Copper 202% and also Dell. So we are standardized Dell shop. So our entire Vulture cloud is powered by Dell solutions and hardware. A lot of machines. A lot of machines. A big customer then. Big customer. And what wasn't about from a partnership perspective, those are two big powerhouses, NVIDIA and Dell Technologies. What did they see in Vulture that made them think this is the right solution together and we can be on the frontier of the outcomes AI can produce. I think there's just an incredible alignment on vision for where the world is headed. So this exact conversation that we're having, this is the exact conversation that we have with Dell in terms of where we're aligning our resources for 2024 and 2025. So whenever you have like-minded people that have a big bold vision to transition the tech sector in a new way, it's an unbelievably energetic, exciting time. Just as you mentioned Dell transition from kind of the mail order to the internet, they're doing that again. And we couldn't be more pleased, honored and proud to be with them on that journey. You know, Dave Vellante and I have been anti-patriots, ex-patriots, I wouldn't even call it repatriots. Repatriation has been a conversation that the cloud people are dealing with, which is, oh, people are moving their workloads off the clouds back on-prem. Yeah, some people have done that for whatever business reasons, compliance, data, but the numbers haven't been massive. However, the growth of on-premise and edge is growing significantly. So that's not coming from repatriation, it's coming from cloud operations. You mentioned Kubernetes, cloud-native microservices. So you're seeing this whole another wave coming that's under the covers. It's fueling the AI revolution, which is cloud-native services and net-new workloads. Right, net-new workloads. Net-new build-out. Net-new build-out. Yeah, I mean it's interesting, but let us also talk about the repatriation for just a moment for one reason, is yes, people have actually realized in the initial move to the cloud for their web applications, it was, you know, this is the fastest, cheapest way to quickly build and scale out core operations. And the dirty little secret was that that was not necessarily true. At the end of the day, what we wound up seeing is that because of unpredictable, you know, highly expensive charges from the hyperscalers specifically, suddenly scaling your business profitably became almost impossible because when you looked at the impact of margins and you looked at your cost of operations, they were too great, so much so, you know, Andries and Herb, what's famously called the trillion dollar paradox, which is there's a certain scale you get. Not Martin Casado, he's a rebel. It's sort of size and scale, it just simply made sense to actually build out your own data centers and you would actually save incredible amounts of money. So what we have to do, the reason why we have to worry about this is let us not repeat the sins of the past. What we don't want to wind up with is the 10 trillion dollar paradox, because right now you see the hyperscalers building an entire stove pack stack of services. So you're 100% dependent, all your API calls are going to go to open API, which is horrendously expensive. Soap pipes don't win in this new model. You have to go with an open ecosystem, you have to leverage the best of open source and you need to be absolutely committed to the fact that new workloads will constantly need to be deployed, a rising tide lifts off both and the goal of cloud providers like Avolcher need to be as efficiency minded as possible, more compute power at lower cost, every single year to help you unlock greater potential and more outcomes. We cannot, we don't want the 10 trillion dollar paradox coming in three years. So first of all, I appreciate that and I want to just push back a little bit on the injuries because we know Martini wouldn't even let us on the stage when he had his tweet chat, when he had his spaces. The issue here though is architecture. So cloud was good when you wanted to lift and shift, push up in the cloud, you get agility. As you grew in the cloud as cloud native or if you were born in the cloud, you're relying on one cloud, I get that. But enterprises weren't born in the cloud. They did lift and shift and then they went in the clouds and wait a minute, that's not optimized for our architecture. So I think what I want to get your reaction to is what AI points out and what the pandemic pointed out is that a lot of enterprise weren't modernized. They're architecture, they're platform, engineering or they're architecture needed to be rethought. And I think everyone's kind of realizing that like probably five years ago and that's like now AI's, they have to move. They got to renew it now. So I think this is where I see you guys killing it is because okay, I have to host and control. I need to command and control, I need to control plane here. But I also I want the cloud too. There's benefits to the cloud. So I think this super cloud environment we've been calling it or multi cloud as some other people call it is not just running workloads across clouds, it's using all the resources available in a distributed computing way. So how does that, so where are the enterprises in that conversation? I mean, you make such a wonderful point which is in the early adoption of cloud, net new workloads for new cloud native applications were being built. Yes, but it was also a lot of lift and chip. But what happens with AI is suddenly all of those legacy applications, you realize that through the power of AI you can completely transform each and every business process. You can improve every single point of decision making and the entire supply chain, the entire manufacturing for everything needs to get rebuilt. So everything that was not modernized now needs to get modernized or entirely rebuilt. So it's not a question of lift and chip anymore. People really do need to rebuild and rethink their entire enterprise applications. I remember back in the early days of the internet. So I was working with the company and we were pioneering kind of a new platform and architecture for helping people build and scale global web applications worldwide. And there was a shift from client server to web application, right? We're going exactly. So it wasn't about, originally people tried to just put a web front end onto their client server application and how well did that work for people? It didn't work. And eventually people wound up actually re-architecting and rebuilding and built all their current web application. We have to go through that same process again. Great example. We need more time, but this is a masterclass. Definitely let's keep the conversation going. We love to appreciate your plug for working with Dell. Obviously they're a great partner of ours and they help support the sponsor theCUBE to come here and we really appreciate that. And I love the vision you have. And I think you're right on, inference is the new scale out application. What's going to power and that's data driven. Everything's involved, full re-architecture. Great stuff. Last question, you talked about 2024 being the realization of outcomes. What do you hope are some of the outcomes that we start to see as we enter 2024? So I think there's the traditional outcomes that people talk about, but I'm going to throw something a little divergent. I think one of the outcomes that I hope that we achieve in 24 is kind of a policy framework that all enterprises can adhere to so that as we're actually building our first AI powered applications that we're doing so in a thoughtful way, leveraging this standards for things like explainability and accuracy. So we don't wind up having unintended consequences when we're trying to do good. So there are standards out there with NIST. I encourage everyone, read those standards and be thoughtful about what you're doing. We're all achieving to do good. We just have to be cognizant that we might unleash unintended outcomes. Right, that thoughtfulness is key. Kevin, it's been a blast having you on theCUBE. Thank you so much for unpacking. Thank you so much for having me. Vulture, your mission, vision, differentiators, what the technology is doing, your partner ecosystem, we'll have to have you back as I think we're just scratching the surface. I love that. Thank you so much for the time. Our pleasure. Bye-bye. For our guests and for John Furrier, I'm Lisa Martin and you're watching theCUBE live from SC23, day three of our coverage continues in just a minute.