 Welcome back everyone. We're live in Las Vegas for VMware Explorer. I'm John Furrier, Dave Vellante. Day three, we're grinding it out. We've got a couple more interviews to do half a day today. And then by day three kickoff, Chris Wolf here, Vice President of VMware runs the AI Labs. He's been the star of the show. His voice is probably getting a little bit sore too. He's been given presentations, going to dinners. Chris, you've been doing a ton of activities. You've briefed lots of multiple times with the analysts and the press. I've seen you on stage multiple times. You're talking to everyone about the AI action here at VMware Explorer. Yes, indeed. Yes, indeed. It's been a busy week for us, but it's been a celebration, I would say, because it's been a culmination of a cross-business unit effort. What we'd like to say, executing is one VMware to land everything that we did this week. I got to say, I like the stories getting better and better. I like our super cloud and there's a runtime, like a multi-cloud element of clouds coming together. And all the Aria, Tanzu kind of connecting, I think that's smart. Just a lot of smart things going on with the story and the portfolio. But the big story is the gift of AI that's just dropped in the industry of the past half a year or so and longer. The generative AI movement in Ragu on stage was really kind of laid out. This is legit. This is basically what he said. He didn't say that exactly. He was very bullish and he's smart and you got this tailwind with the generative AI. It's just beginning. It gives a little lift and also shines a light on multi-cloud too about what could be possible in the modern app era we're getting into. These AI apps that are soon to be coming out in droves. Yeah, and I think it really emphasizes multi-cloud because people want the benefits of AI but they also want privacy and control of their data at the same time. What was the big takeaway from you this week? Obviously you had a key message. Let's go through the key message that you had from a product standpoint and the positioning. And what were some of the conversations you had and the reaction? The reaction has been enormously positive between customers, partners. They've been telling us this is exactly what I'm looking for. These are the use cases we have. They appreciate the fact that where VMware is focused is aligned to their use cases such as AI-assisted software development. Organizations want to maintain control of their source code. It's important to them. They like the idea that they can bring these technologies on-prem. And I think the aha moment for most has been the low cost that they can gain access to these. There's this perception that you need gobs or thousands of GPUs to train mega-LLMs. And when we told folks we trained the SafeCoder model or tuned it against our source code on 1A100 in four hours, they were shocked. But just with that sample, we saw a greater than 90% acceptance rate among our software engineers. So that's a good setup from this question. So Nvidia's results Wednesday night, their revenue doubled. This is amazing. They beat by $2 billion in the top line and their earnings increased by 400% so they're showing operating leverage. In other words, their earnings are growing faster than their revenues. So it's very clear that AI is having an impact and a meaningful impact on Nvidia, obviously. So the chip level. When do you think from a software company standpoint you're going to see that kind of uptake, maybe not that performance, that's insane. But start to have a meaningful consumption from customers, whether it's software infrastructure, other associated tooling that will impact your business. Is the lag, you think six months? Is it a year? You used to install a server and it was like six months before you actually got it up and running. How do you think AI will play out? I think you'll start to see meaningful delta within the next six to 12 months for sure. And we'll start to see some of those early first movers within the next three months. Because for VMware, it's now enabling our channel. It's enabling the SI partners to deliver these solutions on-premises. Customers are still in learn mode, but we're definitely seeing the uptake here. And also with the scarcity of GPUs, it's one of the reasons people are coming to us is because we can give them a platform that provides full awareness of their CPU and GPU inventory and can intelligently share those between workloads, which is really key just to gain access to these compute devices, but then also to save money at the same time. Chris, do you think that'll manifest itself as direct AI spend or more sort of buried in the normal infrastructure, the vSphere, the vSAN, and the NSX and the sort of related core that you guys have? How do you think about that mix? Yeah, certainly the core because AI workloads are going to drive compute density and compute capacity, which is going to benefit for us, Cloud Foundation, in particular because we really see this as an important platform for customers to be able to run these workloads in proximity to their data and to also maintain control of their data. And then you are going to see some ISVs start to benefit as well from this. You know, certainly our Tanzu business at VMware, but then in addition, you know, folks like Huggingface, they just announced a new round of financing today and a greater than $4 billion valuation, right? So the future is bright. So my third question related to that is two years ago in a breaking analysis, we published a post something like how NVIDIA plans to own the data center with AI. In there, the premise was that more and more workloads were going to go to AI. What Jensen on the earnings call was called accelerated computing. It basically put forth that it's very self-serving that there's a big shift going on from general purpose to what he calls accelerated computing. Obviously a lot of the workloads in VMware are general purpose. So first of all, do you buy that premise and what impact do you think that has on VMware? Yeah, there's plenty of general purpose workloads. And I think what sometimes people lose sight of is all of the focus for AI tends to go to training. But it's inferencing, it's day-to-day execution of drawing inferences against the model. That's the 24-7 workload, right? And that works very well with a platform like VMware where we can run those inferences in VMs. We can have a common management plane. And then when you have a lull in compute utilization say in the evenings, that's the optimal time to start running more training workloads as well. One of the things I want to ask you about this private AI that was a big talk on the keynote, Ragu said it was an industry imperative, not just a VMware thing. So I want to get your clarification. You also said to the analysts when we were on the briefing before the show started that private AI is not private cloud. Can you explain that real quick to clarify that private AI positioning? Yeah, it's an architectural approach that ensures privacy and control of your data. At foremost, that's where it's at. There's other parts in terms of understanding access rights, auditability to ensure that I have limitation in terms of how folks are able to access even my data sets used for training. But that could be done in a virtual private cloud, right? That could be done in a public cloud. There's ways that I can approach this architecture without it having to be say in a private cloud on prem. And that's our point. This is not just about VMware. This is not just about private cloud. This is about how customers should be thinking about bringing AI to their data. What about the difference between the LLMs that are coming out? We've seen that first batch come out, the long tail open source. People are kind of figuring out it's kind of costly to train these things on the bigger models. Smaller ones, maybe smaller. Where is the cost consideration? What have you guys learned on how to approach some of these data sets? You mentioned four hours training of your data set. What's the training? Because it seems like if people don't get ahead of that, they could get over their skis, if you will, on the cost side. Whoa, it could be expensive. Yeah, yeah. Or not. So how do you look at the cost? I mean, how do you scope it? How do you scope it? Yeah, so what we're working to do now is to publish both cost data, as well as sizing guides to help our customers and our partners so that they can understand what the potential costs will be up front. Because people, you know, there is quite a bit of unknown here. And people need that type of guidance and experience. So what we're seeing, especially with the open source models, is that we have a fairly predictive cost, especially when you're deploying on-premises, right? Because it's a fixed capital. I'm not paying per token, as I would in a cloud type of model. And that's where it can add up quite quickly for, say, a large organization. Even if it's just like five tokens a day, that adds up to millions of dollars quite quickly. And that definitely has to be a consideration. I want to ask you a question. Maybe put your analyst hat on. When you do these TCO analyses, one of the big assumptions you have to make is server utilization. And it seems like the server utilization in the cloud is very high. Significantly higher than on-prem. So to the extent that you can attack that. Now, maybe that's a premise that's not correct. I'd love to get your feedback on that. Will GenAI, specifically, an AI generally be different where the utilization could be higher relative to what's happening in the public cloud because they have such a multi-tenant model? Yeah, 100%. And that's the first area of tooling we're looking to bring to customers. And that was per my earlier point. We want to give customers a resource pool that they can immediately take advantage of. So when they have any of that idle compute, they can put it to work. And that is absolutely true. We have a lot of organizations that they might have 30% plus spare CPU capacity right now in their VMware clusters. So in essence, you're almost getting that for free. And then when you're tuning smaller-sized models, because you can do it so much more quickly, you can do it more frequently, and you can lower your carbon footprint, which is also something that I know a lot of folks are concerned about, is just a raw energy consumption that is associated with AI. So when you think about something, and we haven't heard much about Project Monterey this week, we asked some questions in the sort of private meetings, and we got some basic answers. But it would seem that that gives you silicon optionality where you can bring in, you know, whether it's a lower cost arm capability, maybe bringing in NPUs in certain use cases, and then potentially an equivalent to the extent that the merchant silicon business delivers it, an equivalent of like a tranium or an inferential, which is using the Amazon model. Do you see that as something that could potentially drive down the cost of your hybrid and on-prem model? Yeah, and even like you mentioned Monterey, if you think about performance nicks and DPUs, it's been great technology really waiting around for the killer use case. And I would say AI is absolutely the killer use case, so you're going to see that. And that's really where we're focused too, is modularity, because we don't want to just force one stack on customers. We want our customers and partners to have some flexibility to architect against what their particular business needs. Chris, I want to ask you about what I'm most excited about was the news on the AI private foundation, private AI foundation with NVIDIA. Obviously they were on stage. Love that map, that ragu show. If you're watching this, check out the keynote. There's a nice section in there where it lays out the NVIDIA piece, and then all the apps are going to be impacted. But you talked about safe coder. Can you explain a little bit more about the safe coder role? Not the star coder, but the safe coder on VMware, because you guys had a great example. And I think this is where the IT action is going to be, kind of coding, configuration, some of those automation stuff DevOps things that I think platform engineers are going to really get jacked up about. So can you unpack a little bit of this star coder example that you guys were talking about? Yeah, and you mentioned private AI foundation with NVIDIA. We're excited about that. We have the major OEMs lined up to help us ship that. And that's going to be just your simple turnkey solution just to get going with AI right away. And then when you take something like safe coder and bring it to this type of solution, really, if I go back to our own experience with it, so we've been running this the last couple of months. We downloaded the model, the base model from Huggingface. We tuned it. And the way we approached the tuning it was to say, let's look at the commits of our top software engineers in VMware. Let's tune it against that data set, which happened to just be 70 megabytes, which is kind of funny. This is source code from your VMware developers. Yes, right. So this is running on-prem. We tuned it against our source code. We optimized the model against that. That took about four hours of time. Then we started onboarding software engineers. And so what we found was we can reach a density of about 400 software engineers per A100. So if you think about, you know, beefier hardware like an H100, it gets even better. Right? So that was like, wow, like this is going to be extremely low cost for us. That was exciting. And then the other part is with the support for 80 languages, we tried other things like we tried Pyvee Mommy and let's see if we can do some Python scripting with this as well to help the typical VI admin. And really amazing. Like I can write my comment and it's like, oh, okay, I see what you want to do here. And here's your code. So even somebody without coding experience can start to get further involved in the automation game here, which is really exciting. So essentially you train. You train. That's his premise. So you train. You train on the code. So the code is fully there. And then it's just accessed. Yeah. And for any organization, you can deploy this on-premises. So you don't have to have a connection to the cloud. If you want to have everything firewalled and isolated, you can do that, run it on our platform, and you can get this value pretty quickly. And that's our next step now is we wrote a technical blog on all of our learnings. Certainly I'd encourage folks to go check that out on this Office of CTO blog at VMware. And we're now enabling our partners with that information as well. Take this out to the edge, if you could. Take this out to the edge. What does that look like? Have you talked about inference before? And I would argue there's actually going to be even more data at the edge, inference, AI inference in real time. How do you play there? Yeah, there will be. And I think the thing to think about too is you have multiple lines of business that want to access that data at the edge. And those multiple lines of business might have different preferences for what are the AI services that they want to use. They might come from different clouds or different ISVs. So our value is we're providing one platform and then optionality via software. So when you have different lines of businesses or business partners that need access to the data, I don't have to spin up another silo. I have a software update. I push this out to that edge site and now I'm off and running. And that's exciting. We also announced our Keswick project and VMware Edge Cloud Orchestrator. That's allowing that really fluid and massive scale management of ESXi nodes at the edge. So that's the fleet management at scale piece. What does that actually mean? I mean, what does this orchestrator do? Yeah, we put a Kubernetes API above ESXi and we implemented a full GitOps management model. So I can now just do updates via Git repo. I can do pole-based updates for those loosely connected edge sites as well, which we think is really important for edge orchestration. And I negate the need to have vCenter instances at every edge site. So vCenter is still a very important part of the architecture, but this is providing that massive scale. And then through GitOps, I have a real interesting way to do further extensibility now around management of these solutions. I think that this kind of points to my previous question on Dave's Edge question. This comes down to this code system, the StarCoder example you mentioned, kind of teases out the idea that the cloud administrator role will be automated. I mean, if you think about it with the orchestrator, things like that, and you've got this code assistance with AI, a lot of the platform engineering tasks are going to be run, assisted with AI on behalf of the platform engineer or the DevOps engineer or the coder, that means the architectural role and coding will be the key. Do you see that admin piece getting helped out here a lot with the AI, or am I over jumping to conclusions? No, I like how you phrase it, helped out. And that's what's really important. And that's why VMware, we were very deliberate in calling our AI capabilities in our products Intelligent Assist. The AI is assisting you, it's not replacing you. It's allowing you to do more faster, and then for folks that are going to say, oh, well the AI isn't perfect, that's the point. We want the human to go in and put the final touches on generated code, and then we can go ahead and package and execute. We wrote an article actually on silkenangle.com on the panel, I think you were on yesterday about the biases in AI. Everyone's freaking out about it, and we've been saying on theCUBE, there's well-documented online activity in the chess world because chess is at computer, it's against humans. It's very notable, and they're all active, as you know, in the forums. But the data from the chess community has shown that humans with AI against other humans with AI is great, but humans plus AI beat just AI, so the machine. So the humans plus AI is better, and actually changes the game on who's the grandmaster. So people who couldn't make grandmaster status with AI can be grandmasters. So it's a unique human aspect of it. So I think the creativity piece is going to be one of those new things that hasn't been talked about much in the tech world. I mean, what does the creative class look like in technology? Have we ever seen like true creativity? If you go back even 20 years, right, even people in our field, folks that have started to learn how to more effectively use Google, and I mean we're really dating ourselves here, became more effective at work, right? So it's the same thing with AI, right? It's like it's a new tool that's going to make you more effective in your job, and the people that are learning how to use that tool should. And that's why we shouldn't be running away from these things. We should be embracing them, but of course you just have to have some guardrails around how those tools are used, which is what we do inside VMware. You guys done a great job, and you got a good view on it. I think that's the right word, guardrails, keeping an eye on it, let it run a little bit, don't let it go out of control. When it jumps over the guardrail, you're going to rain it back in. Chris, thanks for coming. I know you've got to catch a flight. I really appreciate taking the time to come on theCUBE and to make congratulations on all the great work and looking forward to continuing the conversation on the AI labs. Thanks for coming on. Always a pleasure, John. Thanks so much. Chris Wolf here, star of the show here at VMware Explorer. I'm John Furrier, Dave Vellante. Day three, we've got a half a day. We're going to go all the way to the end. So they kick us out. You'll hear the bell. Now it's closed. We've got a lot more. Stay with us. We'll be right back.