VMware Explore 2025 | Chris Wolf, Broadcom & Gavin Jolliffe, Xtravirt

Clips
More from VMware Explore 2025

Chris Wolf

Global Head of AI and Advanced Services, VMware Cloud Foundation Division

Broadcom

Gavin Jolliffe

CEO

Xtravirt

Broadcom transforms VMware Cloud Foundation into a springboard for private AI

A series of announcements from Broadcom Inc. this week underscored its plan to transform VMware Cloud Foundation 9.0 into an AI-native platform. At the heart of that vision is artificial intelligence, with private AI serving as a key element that makes VCF a secure, scalable private cloud foundation.Chris Wolf, global head at Broadcom, and Gavin Jolliffe, CEO at Xtravirt, talk with theCUBE about private AI during VMware Explore.On Tuesday, Broadcom announced the addition of VMware Private AI Services as a standard solution of VCF 9.0. The offering will enable organizations to design, deploy and govern AI models on private infrastructure

play_circle_outline Introduction of the discussion at VMware Explore 2025 featuring Chris Wolf and Gavin Jolliffe.

play_circle_outline Private AI's progression in the last three years and its growing validation among customers.

play_circle_outline Necessity of identifying business value to justify AI investments and avoid unnecessary spending.

play_circle_outline Deployment success at the University of Bristol and collaboration under the Private AI Foundation.

play_circle_outline Navigating Data Compliance: The Intersection of Sovereign Cloud, Private AI, and the Digital Operational Resilience Act (DORA) Impact

Info
Transcript

Chris Wolf, Broadcom & Gavin Jolliffe, Xtravirt

Chris Wolf

Global Head of AI and Advanced Services, VMware Cloud Foundation Division Broadcom

Gavin Jolliffe

CEO Xtravirt

At VMware Explore 2025, theCUBE’s John Furrier and Paul Nashawaty sat down with Chris Wolf, head of AI and advanced services at Broadcom, VMware, and Gavin Jolliffe, CEO and founder of Xtravirt, for a deep dive into the realities of enterprise AI adoption. Wolf, who first coined the term ‘private AI’ on theCUBE, details its evolution from a provocative concept to a validated market with more than 80 customers today. He explains how VMware, under Broadcom, is delivering a full-stack AI platform on-premises - one that spans compliance, audit, disaster recovery ... Read more

explore Keep Exploring

What event is being discussed, and who are some of the notable individuals present? add

What has changed in the market for AI solutions over the past three years? add

What considerations and strategies are taken into account when implementing AI use cases in a business context? add

What is the Private AI Foundation with NVIDIA and its significance to the University of Bristol? add

What are the implications of the Digital Operational Resilience Act (DORA) for organizations in terms of regulation and governance of private AI? add

bolt Powered by CUBE AI

Chris Wolf, Broadcom & Gavin Jolliffe, Xtravirt

search

>> Welcome back everyone to theCUBE. We are live in Las Vegas for VMware Explore 2025. I'm John Furrier, host of theCUBE. Paul Nashawaty, lead research analyst at theCUBE Research. We've got a great lineup here. We got a bunch of OGs and virtualization, and the person who coined the term private AI, Chris Wolf, Head of AI and Advanced Services at Broadcom, VMware. Chris, great to see you.

Chris Wolf

>> Always a pleasure, John.

>> Gavin Jolliffe, CEO and founder of Xtravirt doing virtualization before it was cool, writing white papers now leading the charge. Great roster and doing great service. Great to see you. Thanks for coming on.

Gavin Jolliffe

>> Thank you, John. Thanks for having us.

>> Paul and I are very interested in the private AI conversation, but more interested in how that's going to become about. So Chris, I have to ask you first, since you coined the term and introduce it on theCUBE, I won't say you've got a lot of stares, but people are like, "Hmm, very provocative concept." Sovereign AI is talked about. Sovereign cloud is the hottest AI infrastructure conversation we're having on theCUBE. How has private AI progressed for you? Give us the lay of the land now and just add some color to kind of how it's evolved.

Chris Wolf

>> Yeah, it's really interesting because three years ago, we had enough validation from our customers that the notion of being able to bring an AI model adjacent to your data that you're owning and control, keeping your data in its existing repos was going to be a really important space, and it was a space that nobody was talking about. And there was a lot of exaggeration that you were going to need hundreds to thousands of GPUs to run any kind of AI use case. Right? So you were kind of dumb if you said AI wasn't a cloud use case back then. But today, what do you see? You see hyperscalers offering solutions that go on-prem. You hear them talking about a private AI, like there's true validation of the market three years later. Today, we shared we have more than 80 customers now. We have a massive in terms of just a pipeline of new customers that we're working on board, so it's absolutely taken off. I think what's happened since then is we have a full solution now. So all of the services that you would see in terms of a cloud AI solution, I can deliver as a platform now on-premises. And then what you're seeing evolve now is really the choice of models and services. So even our partnership with NVIDIA, you have over 100 AI services that you can deploy as NIMS today, on-premises or in a sovereign cloud anywhere you're looking to go. So there's really a maturation, I would say, of the market that's been pretty exciting.

>> What's interesting too is the enterprise cloud business that VCF is going after also aligns with the enterprise AI. I will say stall factor, but there's a lot of complexity involved, and so a lot of the top enterprise want their jewels, the data on-premises, and they'll stack up a bunch of. Now, they need the cloud. That's another factor.

Chris Wolf

>> Yeah, it is. And part of this journey too for the enterprise has been you've had to figure out what are the use cases that add business value. We can't go back and do AI for the sake of AI. People did that with blockchain. It didn't go well. You spend a lot of money on things, right? So AI has been about, in some cases, starting with GPU as a service, figuring out for generative AI where is the real business value. Where does it make sense? And being very fine-grained. An organization that's been running our platform, I met with their CIO this morning and they're using it for one of their use cases is financial analytics, and they had to find not just the use case, but the right model that is the most optimized for that use case, and then they're lighting that up and are really seeing the business value.

>> Gavin, talk about Xtravirt. I mentioned you're an OG. You're a premier partner, you have a lot of great customers. We've highlighted some of them here on theCUBE. What are you up to now? Explain what you do, the founder, CEO.

Gavin Jolliffe

>> Start at the top?

>> Where are we today? Because you have a template you're executing.

Gavin Jolliffe

>> Yeah, absolutely. So thanks for the backdrop and thanks, Chris, for your support over the years as well. So Xtravirt is a UK-headquartered specialist integrator for VMware private cloud. So we're all in VMware, always have been. We deliver advisory, professional, and managed services to large enterprise, mid-market, across all sectors. For us, our core strategy outside of being now 20 years old, which I can only believe when I'm not looking in the mirror, is around something we call VCF anywhere. That's our core strategy. And what that means is enabling customers to be able to deliver private cloud, private AI, and other strategic capabilities, whether it's in their data center through a cloud service provider, a hyperscaler, a mix of all of those. Really, critically for us, deployment of a great platform is not where we stop. For us, it's about embedding the solution into the heart of a customer's operating model. For us, it's about making sure that we're delivering critical lifecycle and managed services expertise, but also enabling the business and organizational change that will help them unlock it. Critically, we make sure that the customer always retains control, so we act as a strategic extension in that regard.

Paul Nashawaty

>> So I have a question around for both of you, Chris and Gavin. This is actually really interesting, because when I look at my research and I look at what we're talking about, there's many organizations that are in various stages of their journey, whether they're immature in their journey or very mature in their journey. There's a lot of discussion around where the budgets are going, where the money is coming from. And when I talk to CIOs, 25% of their budget is allocated to AI projects. Now, they don't necessarily know what those projects are being spent on as long as it says AI in front of something, right? Now, I think if we just start in the context of for solving cloud, for solving AI solutions, how do you bridge the gap between an organization that's just trying to start and where they want to go in their journey? Now, it sounds like from Gavin's perspective there's a blueprint, but Chris, from the perspective of the tech stack, how does it work?

Chris Wolf

>> Yeah. We try to guide customers to be very precise on the use case first so they can demonstrate the business value. Even in Broadcom, if a division in Broadcom is committing to an AI use case, the general manager is required to also commit to a specific ROI statement to make sure that we're not doing AI for the sake of AI, that there is business value in what we're trying to do. And a lot of these are efficiency gains and allowing us to move faster as a company because R&D is so important to us. But if you step back for a customer, like let's take a very cookie cutter use case, very pragmatic, which would be their contact center. Everybody has a contact center. They have a lot of private data, they have a lot of their own private KB articles that they've developed over the years. That's an internal knowledge base with a lot of proprietary IP. They want to keep that protected, so they're not going to put a chatbot in front of it where they might have concerns over guardrails and expose it to the customer. So you put that in front of your support agents, so you have human in the loop for the generative AI tool. And then say even if you're on the low end of a benchmark that we've seen industry-wide for efficiency gains, maybe you're seeing a 10% productivity gain. That's at the low end. It could be as high as 35%. So now, you're seeing in terms of support tickets, I'm closing maybe another 30, 40 tickets a week depending on the type of use case. That's a huge efficiency gain that I can bring that service in to the organization and start. And then maybe after that, I'm looking at legal, I'm looking at sales contracts. There's other things that I can start to pick up after that.

Paul Nashawaty

>> I like where you're going with this. And the fact that there's a lot of ... How do I say it? Not everything needs to be refactored, right? Some stuff works the way it works. Because when I talk to CIOs and CIOs tell me that modernization is their number one goal, they're trying to modernize these AI solutions that they have in place. They run into complexity and skill gap issues. Gavin, this is where you come in. When we look at the skill gap issues, a lot of times, I talk to all these organizations, they say, "Well, what do you recommend Paul on how we should move forward? We can't find people to do the job. How do we move forward?" So I tell them to work with the service delivery partner, right? So the ecosystem in that really kind of helps. What are your thoughts on that?

Gavin Jolliffe

>> Yeah, absolutely. So to Chris' point, you need a strategy, you need a vision, you need to understand where you're trying to end up, and then it's about understanding what capabilities you want to bring forward. It's where you're going to invest. How do you want to leverage your organization? Then how do you want to fill the gap? So our role really is to ensure we understand where the customer is going, what capabilities they have today, what they need at the end of that outcome, and then provide the right collection of capabilities all the way through that journey. Which means that it's not one and done, it's not monolithic, it's not a blueprint that is one size fits all. You're constantly adapting to wherever the customer is.

>> You guys did a deployment at University of Bristol.

Gavin Jolliffe

>> Yes.

>> Keith came on theCUBE twice. He's very happy with you guys.

Gavin Jolliffe

>> You can't keep him away.

>> He's a walking testimonial for you guys. He said to me, "We can focus on what we want to do." They do a lot of research, they need to protect. Also, security was a huge concern in UK. As you know, a lot of hacking has been going on. So he's very happy with VCF, so check. You also helped them with the Private AI Foundation with NVIDIA. Explain that piece. What is that about? I've heard that, I don't know much about it. What is the Private AI Foundation with NVIDIA?

Gavin Jolliffe

>> All right, yeah, brilliant. So I'll stand back and I'll unpack it at slightly higher level. So this has been a really exciting journey for not only university, but it's a UK national endeavor. So University of Bristol, as we know, is a global top-50 institution, right? They are at the forefront of advanced research and development within the UK. The UK government invested around $300 million in their Isambard supercompute platform. Now, that supercompute platform is today one of the world's fastest AI supercomputers. So you can imagine the immense cost of managing that density of resource. So by bringing VMware VCF and Private AI alongside to complement it, what we ended up doing was enabling the university to have a tiered AI capability, which meant that they could apply the right platform to the right cost to the right resource groups. By being able to extend VCF with private AI and NVIDIA GPUs, it also gives them a secure and governed way of being able to allocate that resource whilst being able to keep their operational costs in check. And, yeah, it's been a-

>> Just not interrupt. Was that resource tiering, was that for like workloads, or for the AI resource? What was that specifically?

Gavin Jolliffe

>> Because Private AI is a scalable platform within private cloud, it means whatever the resource slices or requirements are from each of those resource groups, they've got full control.

>> So they can flex it?

Gavin Jolliffe

>> They can flex it, they can make sure that it marries to their costs. Yeah, they have full strategic control on governance over that. I think going back to that, what does it take to stand up a platform like that? It's been a seven-year journey, ongoing journey as a collaboration between the university, VMware and Xtravirt. It's been obviously underpinned by Keith's team's great vision. A lot of trust and a lot of deep collaboration over that time. But what really started as foundation laying for all of us has now transformed into a real great launchpad for transformative development and innovation, not only for the university, but in the wider UK context now.

>> He mentions to me they got a quantum computer going too over there.

Gavin Jolliffe

>> Yeah. Sometimes pulls it out of his bag.

>> I mean, that's public information.

Gavin Jolliffe

>> I didn't say that.

>> Chris, what's the Private AI Foundation? What does it mean? I see NVIDIA is involved. They love this physical AI construct. This kind of feels a lot like that. Is this kind of marrying AI with the reality of the IT piece?

Chris Wolf

>> It really is, because I think there's multiple sides of this. First, you look at ... You could say, "Well, why don't I do AI in cloud? Why don't I do AI on-prem, on bare metal?" There's all of these options out there and we get these questions all the time. If you're just trying to do some research and experimentation, then maybe some of this bare metal stuff is okay for your needs. But when you want to go into production, you have compliance, you have audit, you have backup, you have disaster recovery, you have security scans you have to do, right? There's a whole list in terms of operational mandates and constructs that you need. And then if you want to really have the resource efficiencies, you're trying to maximize the use of all of your cluster capacity, and it's not just the GPU. Everybody focuses on the GPU, but how am I making sure my network is properly provisioned for a 70 billion parameter model versus an 8 billion parameter model? It's entirely different math. How do you load balance that at scale across your clusters? Nobody has a distributed resource scheduler like us. It doesn't exist outside of our core technology, and this makes a huge difference. So if you want to buy a bunch of hardware and utilize it 35%, there's lots of choices out there. If you want to buy a bunch of hardware and utilize it to 80% to 95%, 100% of your capacity and be able to maintain your SLAs, nobody can do what we do. And this matters, because you're not just needing fewer servers, but now that I need less power into my data centers to drive this. And power, especially in parts of Europe, that's a real concern.

>> Well, to Gavin's point, I mean this is where they're buying gear. So it's not like ... It's expensive. So when it's not being used for this group, I can flex it over here, and that's huge. That's basically cloud. It's enterprise cloud.

Chris Wolf

>> Yeah.

Paul Nashawaty

>> Yeah. Time-sharing of the gear is one thing. So Chris, what you're talking about is an acceleration. I mean, this is basically the core VMware kind of principle, utilization of the resource that you have, get maximum utilization. Why dare I ask? Why don't people don't know about this? This is something that's been around for a long time.

Chris Wolf

>> Yeah. It's a fair question. We've had to do a lot of education because if you go back to the early days of HPC and ML, it was a bare metal use case. So the groups that have run these clusters, that's what they know. And what happens with these use cases is even in the old HPC days, we started picking up customers for that as well because the churn and the time it would take to churn over a cluster and deploy a new type of test workload or whatever it might be, it might take days to weeks. We even see that today with AI models that you're trying to bring into production. I got to get the right device driver for the particular accelerators it's going to run on, right? I need to have the optimized kernels based on the model and the accelerator. There's a whole software stack around it. The really efficient IT organizations we talk to, that alone might take them two weeks and we're saying we can do this in under an hour. So some of this is education. We've done a lot of performance benchmarking where we can prove that under the worst of circumstances, if you're looking at ... I'm using VGPU from NVIDIA with our stack, you might see a 2% performance tax, but if I get all the consolidation benefits, 2% is really nothing. That all said, we also shared today we have full support for pass-through with NVIDIA. So if you're really that worried about the 2%, pass through your GPU. We're going to still virtualize all your networking storage, memory, everything else, give you all of those benefits, and we're going to give you direct access to a GPU at the same time. So there is no reason, no excuse. Seriously, if you are not looking at this, you are doing a disservice to your company because we're going to probably cut your cost in half. You're wasting your company's money if you're not talking to us.

>> All right, so that's a bold statement, so I'll put my skeptic hat on.

Chris Wolf

>> Go ahead, prove it.

>> Well, there'll be skeptics out there. So how do you get through the knothole on that, because that's a bold claim? And show me the money. Every CFO will be like, "Okay, tell me more."

Chris Wolf

>> Yeah, yeah. We have we a spreadsheet that us and our partners like Xtravirt will bring to our customers. We can do, if you look at X number of tokens on any cloud provider, if you compare us to the on-prem solutions, it's math. It is fully customizable. We can show it to you. And then normally, what happens though, John, to your point is when we go through all of this and we say, "Look, you get choice of accelerators, choice of models, you get common infrastructure, we're going to cut your TCO probably in half if not more." What do you do in that circumstance? You start to do a POC, you stand up a use case, you work with a great partner like Xtravirt, you get that first app running and then the proof is right there in front of you. So we want to guide the organization through the financial analytics, to that proof point. And now, the fact that we're just bundled with VCF, right? It's like you can turn it on when you're ready. You use some GPUs.

>> Your point earlier was be intentional about the use cases. That's the reason why. Get the proof, math, and the outcome.

Chris Wolf

>> Yeah, we are not being hand-wavy. I think that's a disservice to your customers. I've had so many CXOs that I met with where they say, "You know what? Thank you. Because every other vendor that comes in, they say they can do AI for anything."

>> High in the sky, nonsense.

Chris Wolf

>> Yeah, it doesn't help. It doesn't help them to solve a business problem.

>> Gavin, I have to ask you a question because this come up here this year at VMware Explorer. An old timer like me, actually longer than me, probably more like you been, 20 something. How many years, 20?

Gavin Jolliffe

>> Shh.

>> Says to me he works for a big bank. I can't say his name, because people know who he is. He says, "The young kids doing the AI stuff, they don't know what hardware is." And I'm like, "Oh, that's good, isn't it? You abstract out of the complexities." He goes, "No, no. But when you start making these calls around scheduling and resource management, you got to know things like VP and L2. There's calls you got to make in a systems mindset. They have to train the young guns on how hardware works." And I'm not kidding, this actually was a real conversation. He go, "What do you do?" I'm like, "We teach them about entropy and heat." What's your reaction to that? Do you agree that there's kind of almost a spoiled nature of IT? I see that as-

Gavin Jolliffe

>> Oh, man.

Chris Wolf

>> Get off my lawn.

>> That's .

Chris Wolf

>> Please, get off my lawn. Exactly, John.

>> Get off my lawn. I walk to school bare feet in the snow. No, I'm serious, because it's easy, great, thank you. But is there a problem?

Gavin Jolliffe

>> Let me pivot quickly on that one. Can I give you an analogy?

>> Yes, please.

Gavin Jolliffe

>> All right. And it's a bloke analogy because it's about cars. So I think the journey that ultimately we're on is a little bit like your first car, right? You could lift the hood, you could tinker with it, you could break it. I broke many of my cars and my friend's cars. What do you do with an electric car today? You lift the hood looking for a space for your golf clubs. So we are talking less about the mechanics, and we're talking more about the journey these days. So whilst there are going to be a high level of elements, these are all being shifted up the stack. Because what we're seeing with VCF as a fully integrated platform these days is shifting that story on. It's worrying less about being a mechanic and more about how do I get to my journey?

>> So in this case, the abstraction to your point about having the resource?

Gavin Jolliffe

>> Exactly better. So your analogy around the younger generation, I sympathize and I get it. I got two kids too.

>> And they were loaded software in a server.

Chris Wolf

>> Sometimes you want to stay in your lane, because we've seen customers where it's like literally the data science team is trying to build and manage infrastructure. Focus on what you're good at. Data scientists and ML engineers, they're so specialized, they shouldn't be doing that work. Let the IT ops teams with the expertise do all that hard stuff because you don't know what you don't know and you don't want to have some kind of data or compliance risk that you create inadvertently.

Gavin Jolliffe

>> Good point.

>> All right, so let's talk about the sovereign cloud impact, because it's a really important area we're covering now and much aggressively going into Mobile World Congress. Sovereign cloud I think is the prerequisite to sovereign AI, because it's already in VCF, great on your side. But as you go broader across environments, it's a lot of formulations, policy involved, there's governance, all that good stuff. What's the relationship between sovereign cloud and sovereign AI? Are they together, are they hand in glove? How do you guys think about that piece? Because that is private AI. It's going to come right down the street for us.

Gavin Jolliffe

>> Yeah. I mean, private AI solves a lot of the challenges that sovereign cloud is trying to address. So if we think about organizations that have data sensitivity that requires it to stay within national boundaries or local jurisdictions or away from foreign influence, those are all going to be factors that private AI can address. I think more broadly, there are going to be pools. Some of those pools, we know we talked about cost. So you can iterate and test models as many times as you like on a private AI world without additional incremental cost. So your acceleration can move. You've got encryption and access under your control. You've got an element of future proofing and hedging against macro events as well with that view. But I think there's more interesting dynamics, and you touched on regulation. So a real interesting one that we are starting to tackle is DORA. So it's a new piece of legislation, the Digital Operational Resilience Act. I said that first time correct today. What that is, is it requires organizations to underwrite their ability to recover from a cyber attack or a system failure. Now, the penalties for that are the most eye watering that we've seen yet. So up to 2% of global revenue, up to a million euro individual fines, and up to imprisonment for senior execs and board members. Now, quite rightly, that is making a lot of board leaders stand up and recalibrate. What we're seeing is not just private AI in terms of where does data sit, but now who's governing it? How is it being used? We're getting new dimensions that are drawing the why and what's important here.

>> Yeah. I mean, I think it's a huge ... That's what's going to scare people off the board, for sure. Chris, it sounds like private AI is sovereign. Sovereign AI and private AI is sovereign cloud. With VCF, is it one thing?

Chris Wolf

>> It is. It's a feature of sovereign cloud. And the key thing with VCF that I think kind of gets missed sometimes in the conversations is everybody is claiming they can do sovereign cloud, but the details matter here. They really matter depending on what the organization or entity or jurisdiction is looking for. And the difference with VCF is we run in a fully air-gapped environment. The organization owns the control plane. It's yours, right? It's your software, you own it, you operate it. That is fundamentally different than what you're seeing with any of these other solutions, and that really matters. So that is absolute control that you have. No matter what happens in the world, you have ownership of the software stack and your intellectual property.

>> Well, guys, it's great to have you both on. Thanks for the commentary. Final question. What are you guys most excited about right now? Chris, we'll start with you. Obviously, private AI is booming. You've found or coined the term. What are you excited about?

Chris Wolf

>> What I'm excited about is that we've had this conversation around generative AI for some time now, and we're seeing so much energy in people taking next steps. We've seen this convergence of an ecosystem around what we're doing with our product and technology. I think people are really understanding from a Broadcom perspective what we do. When you're talking about open ecosystems, when you're talking about interoperability and choice, there's just no better AI partner today than us. And of course, enabling our partners to do all of this the same. So really excited for the future based on our core principles, and we've had really good validation of that.

Gavin Jolliffe

>> Gavin, take us home.

>> Take us home. What are you excited about?

Gavin Jolliffe

>> I'm going to echo a little bit of what Chris has said. What we're getting now is absolute clarity in terms of where we're moving. We've moved from a selective product-centric model into a fully integrated R&D private cloud. And that's really exciting because customers know which direction they're heading in and it's an absolute enabler now. So our drivers to really support customers on accelerating that journey, and we are right at the heart of that. So for us, every day is exciting, right?

>> Gentlemen, thank you so much. Appreciate it. I'm John Furrier, the host, with Paul Nashawaty here live in Las Vegas for VMware Explorer. Thanks for watching.