Chris Wolf, Broadcom | VMware Explore 2026

[music] Hello, I'm John Furry, your host of the Cube here with VMware Explore coverage. We have two days of Cube live coverage from the show floor. I'm here with the Broadcom executives to unpack the innovation and news at the show. Here now is Chris Wolf, cube alumni, global head of AI and advanced services at VMware Cloud Foundation division to talk about the evolution of AI in the enterprise as we market continues to move from private cloud to private AI to private AI cloud and the impact on operations teams as AI transitions into the enterprise. Intelligence is going there and we're going to talk about Broadcom's unique position to help make that transition complete. Chris, great to see you. Thanks for coming in. >> Yeah, always a pleasure, John. You know, we talked I mean years ago when when uh private AI was coming onto the scene, you were kind of calling it AI is coming into the enterprise. Cloud AI, private AI was just started to get going. Now, three three plus years later, >> yeah, >> it's not only happened, it's happening inside the enterprise as everyone realizes that I have to inject intelligence into the enterprise. So, guess what? Cloud native really brought in the AI era and private cloud is what everyone's kind of doing if you look at the vertical integrations the domain expertise give us a state of the union on enterprise AI where is it today from VMware perspective and how how's Broadcom look at it >> yeah I think it's exciting I mean as far as we've come we're still in the early days if you really think about it so you know you've seen this experimentation you've seen now these use cases emerge with some of the frontier models in the clouds right so that that's one, but then the the counterbalance to that is like, hey, wait a minute. I'm concerned about data leakage. I'm concerned about, you know, somebody else leveraging an AI model's expertise to disrupt my business. So, you have this pressure about bringing models to your data, making sure you have control of your data. That's one. You have sovereignty considerations. You have tokconomics considerations as well. So, this doesn't mean don't use frontier models. It means be practical, right? So use frontier models where it makes sense, where I need deep reasoning, use specialized models, local SLMs where where they make sense as well. So you really seeing this breath of coverage happening in the industry and now it operations is caught in the middle of all of this. >> You know, we uh been following the cloud obviously go back to go say 2013 AWS rises up, they get all the startups, then they start crossing over the enterprise. Security was the one of the main reasons they saw the enterprise adoption. Then I never thought you'd see a another hyperscaler. But with the rise of on premises and now hybrid, you're starting to see cloud scale hit with new entrance come in. We see corewave out there, these neoclouds and the enterprises are now moving that way. So private cloud, okay, is what people are doing as I mentioned, but AI is involved. What's different between kind of the private cloud of old if you want to say cloud native only and then as AI comes in how does that change what private cloud is and what is private AI cloud so because that convergence is happening what is it how do you describe it >> yeah it's it's a good point the convergence is happening so you you have a mix of workloads right so you do have in the modern app space you do have containerbased workloads that are CPUbound that are also driving AI forward you start to look at the the rise of AI agents and these these agentic loops happening right uh and then also uh ephemeral workloads like we've seen for a long time now in the container space so that's on one side of it and it creates a new set of challenges right like um how do I have a pool of sandbox ready infrastructure available so that I can uh bring up these AI agents but do it securely and prevent any type of escapes right or privilege escalations and things of that nature and then you have the GPU side as well where I'm bringing in a a range of frontier models you have uh gateways that are emerging where I need to start to make decisions between do I uh route traffic a particular prompt to a cloud-based model can I use onrem there's there's a new set of governance challenge that's coming around it but at the end of the day when you have a modular IAS layer that can scale that can make the most efficient use of your resources right you're setting yourself up for success uh hardware costs are continuing to rise you have energy costs that are concerns so What's exciting for us is this is creating new uh workloads and new use cases for virtualization because I need to pull and consolidate my capacity and we have the best technology to do it. >> You mentioned uh the prices but also the nature of the infrastructure AI infrastructure is changing. If you look at um how AI has disrupted the stack we're hearing even things like you know I got GPUs here I'm using more compute for prefill decode and prefill starting to be unbundled. So that's happening because they want to be efficient use of the resource. Y >> um you're hearing words like control plane, resource management. I mean a lot of the cloudnative things like Kubernetes and containers, these things are in the infrastructure, but the nature of the infrastructure has changed and people are now get visibility onto this and so they want full control of the resources up and down the stack. That's going to require software. Talk about that dynamic because that makes to me private cloud the use case of the hybrid environment. So you got hybrid cloud but the private is I need security. I need to know the resources. I got to make efficient use of the compute XPUs and GPUs. So people are starting to line up behind this. What's your view on this as AI is a big part of it? >> Yeah and and this it's a it's a great astute point John because if we go back you know 2030 years the complaint around IT and IT operations was well hey wait a minute security is the afterthought right? And that doesn't happen anymore. Security is really ingrained into our processes and it's really at the forefront. But what we see time and time again from the organizations we work with is what they're being told is either a, you know, go to cloud and then they run into budget and economic issues or control issues with that or b, hey, it's a hardware problem. Buy your hardware first, figure out the software later. No, that's a horrible idea because you have to make sure that your software choices are compatible, right, with the hardware you bought. So you have to bring software to the forefront. software is what's giving you the ability to have autonomy in terms of the accelerators you use to have the flexibility to ensure that I can use uh cloud models when I need to use local models when I need to do all of this in aggregate and then also give you the ability to schedule and balance your workloads right across your private cloud so you need a secure modular by design software solution to do all this and I just want to make one other point it's not just about grab all of your software pieces and parts We we can't operate that way anymore because what uh new models like Mythos have shown us >> is that the threat landscape is continually evolving. You need to have vendors that have the capability to be able to keep up with this pace of change to be able to help you to patch rapidly, stay on top of these things. So my view is you're going to have modular designs, right, that give you this high degree of flexibility, but then you still have these open interfaces above it. So from a customer perspective, they're not concerned about lock in. We can still engage with the ecosystem and things of that nature. >> You mentioned virtualization earlier about, you know, how that's going to be a potential benefit. What is Broadcom's um innovation around helping around memory for example, memory shortages are clear. Um I think last year we talked about taring that was a big thing and I think we were coming, hey, that's a pretty big deal. That seems like it's an opportunity. So as you have this kind of virtualization like dynamic in the market, innovation is going to come from managing the constraint, supply scarcity, memory scarcity, but how memory is configured. How do you guys look at this? Because this is kind of in your wheelhouse because AI needs to have these resources up and running, but from an IT shop, I want to make sure I'm running it because I got to stand up agent sandboxes. I got to let agents get going, but I don't want to see things like the open AAI hugging face problem that happened that was detailed in black hat. So, you got a security issue around agents. >> Again, it's kind of like microservices meets AI because you got to see everything. You need telemetry. >> You you you need Yeah, there's a lot to a lot to unpack here. So, if you go back like and this is going to date us a little bit. So think about 15 years ago we were talking about virtual desktops and boottorms like Monday morning people come in right financial trading firm and all of a sudden you got 600 concurrent loginins trying to grab virtual desktops. So then it it was a new approach to things like prefill right where we're saying hey I need to have a warm pool of desktops that's ready so I can just turn them on at a moment's notice. Now with AI agents it's a similar type of problem. So these IT infrastructure teams, they know they need to keep a warm capacity of VMs that can start to isolate agents, prevent escapes. Like that's part of the reality there. Then you have to start thinking about well, how do I manage memory? How do I manage the KV cache across my different GPU memory uh seg uh you know availability and then how do I start to segment that out between different classes of storage. So there's a lot to to manage in this and this is where you need to look at solutions that have the capability to really simplify all of this and then work with folks like us that have the expertise to really help you navigate this space. >> You know one observation I would say from this year at explore is the um activation and engagement of the ecosystem. give us an update on the AI ecosystem for Broadcom and VMware because you're starting I guess I guess said people starting to see visibility into what the tech is going to be some of the architecture and unit economics start to re rear their head because revenues now involved this is not just IT projects this is business model money is on the table this is a change a little bit so what's going on in the ecosystem >> um like like you saw my reaction I'm really excited about this uh we'll we'll start from the bottom and kind of move our way up. So you start to look uh at the base layer. You know, the the struggle that organizations often have is how do I start to piece all the all of the hardware together and then be able to bring time to value really quickly, right, to these different AI initiatives that I have. So starting with AI factory, we're able to give you from bare metal top to bottom provision across all of these different uh hardware OEMs so that I can uh support uh not just from day zero and getting stood up but ongoing operations in terms of patching and management. So that's big and we're working with lots of folks whether it's Dell or Cisco or uh Lenovo Super Micro to really bring this to market. As you start to go further up the stack, there's a lot more innovations we're doing here as well in the ecosystem. So uh around model providers like one of the the things I'm really proud of is we talk about how we have more than 150 different models just ready to go on VCF. When you extend that even further into the cloud there's another 40 model providers that we support as cloud-based resources. And now from end to end I can support this from IT operations perspective uh through VCF through our AI gateway. And this is everything from uh Alibaba to Google to open AI to Anthropic uh you name it. We have this great amount of coverage now for whether it's cloud whether it's on prem and a huge ecosystem around everything we're doing. >> You know I was um I noticed looking at some of OpenAI's numbers 50% is enterprise revenues that's up. Um they're obviously growing anthropic as is as well as those models come in. There's a big discussion around uh enterprise unlock of the data that's specific to the enterprise again securely data but like most of the public general intelligence from the frontier models is is the internet they got a lot of general intelligence but there's a there's a move towards specialized intelligence where I want to have my data specialize on workloads I might not need the big models I might want to use the open weights I might want to use open source there it's a good use case we're seeing great reviews there but as people start to realize I have domain expertise I need to build my company brain for my company as a competitive advantage. This is where the private cloud works and we're seeing this right in real time play out that I want to protect my data. I don't need I want an on-prem. I I'll buy an AI factory but this is now a focus. How does that change u some of the things that you're looking at and and your relationships like with Google for instance and other other big partners because they're moving super fast. they want that long tail or mid torso of the of the market. >> Yeah, it's it's it's been interesting. I think first of all, there's this uh reconciliation of uh expertise in the space. >> So when you look at some of the frontier model providers, if they want to get access to on premises and edge environments, there's no better partner than Broadcom for for all of this. So they're coming to us because it provides rapid access. they they can't work on building their own hardened appliances and take years to develop something and then try to bring it into enterprise IT, right, when we're already here. So that's been a I think a key space for us where we have the ecosystem now coming to us like some of the names that I've mentioned because they want access to the customer base and it's an opportunity for us to really work closely together to solve these common problems >> and you have a relation with Google and share what's happening with Google specifically. Yeah. So, Google's one of the we have a uh session this week on the different model providers we're working with. I'm really excited to moderate that and Google's one of the participants. So, we've been doing a lot with Google initially around some of their Gemma models because what they're seeing is even for these sovereign AI use cases. It's not just about a local model, but it's about having an airgapped control plane. It's about the customer being in complete control of that environment top to bottom. And Google's done a phenomenal job in terms of to your point, you know, not just releasing models, but really working to customize these to provide this the right solution for the customer. And when you look at these edge environments, most of these models I can run on a single node today. >> So that's exciting. Even some of these more complex mixture of experts models, I can get to a single node. It's not every use case, but for a lot of them that the businesses need, we have that. So you're looking at model support and growing that part of the relationship with Google. And then you're looking at the other side of this which is I we talked a little bit about agent sandboxes. Google's done a nice job with agent executor and starting to help you with these hybrid architectures as well. So you have some local models, you have agent exeutor and then through our gateway we can bring all of this together. So you can have local agents, local models as well as Gemini models in the cloud as well. All working as a unified solution. And we're not just testing this in terms of what we bring to customers, but we're also embedding these solutions into VCF itself. I want to get into the news here. Um, you guys, uh, just put out a release. I want to get your reaction because this is super cool. Um, Broadcom introduces VMware private AI cloud. Okay. Essentially, this is an AI factory announcement. So, you guys have an AI factory. So, talk about the AI factory uh, solution from Broadcom because this is what people want are building now and standing up and operating and investing in right now is AI factories. They want more tokens. time token to revenue is starting to see a metric. You're starting to see outcomes be directly related to these factories. It's a new infrastructure node in a hybrid environment. Kind of sits perfectly into the private AI and private AI cloud focus. Talk about the Broadcom AI factory. Yeah, I think this is a coming together of the ecosystem because I I certainly give credit to our server OEM partners for the work that they've done in shipping their AI factories where there's been the opportunity for us to engage with them has really been around completing the vision of AI factory through software which we're the best at right given our customers a common lens where I can just hit a button and I can go soup the nuts from bare metal to uh standing up uh all of my software components So that in a matter of minutes I'm going from again I've laid everything down and now I'm serving models right I can stand up this capacity uh from a day two perspective I can patch and update all of those components end to end and be able to maintain all of that and this is really complex you're talking about uh you know VMs container environments you're talking about model runtimes you're talking about the AI software stack you're talking about frameworks you're talking about GPU operators there's a whole lot that goes into this and when you start to uh even step back, I'm starting to have to look at how do I distribute models across my capacity, right? How do I uh deliver more efficient capacity for these models? How do I decide uh you know what model to place where? How do I intelligently pack models across a pool of GPU resources? So, I'm using all of the available memory. These are really hard problems to solve. And when done wrong, you wind up needing a lot more servers than you really need to to solve the problem. uh that drives up costs and it's not just people focus on the server and the energy costs but it's there's incremental software licensing for all your software components right that's additional cores you're paying for that maybe you didn't need >> so this is what we're able to do end to end with partners such as Cisco and Dell uh giving you these AI ready nodes right from the OEMs we could then stand up our software stack get going and then you can start to layer these other pieces on top >> so the VCF AI factory is an interesting approach and I want to find out what's different because if I go back and look at VCF from last year and the year before we were talking about as it was coming together and being you know mobilized around as one private cloud opportunity adding AI into it is going to have a lot of new things what's new about the VCF AI factory specifically is it the agent piece is it the software stack is it is it the is it the configuration because people want the software stack in these AI factors not just buy a Rex scale set of hardware I mean I got to run something on. It's not like you're loading Linux on a server. This is like you have to purpose-built rack scale >> um intelligence. >> This is end to end everything from compute, networking, storage, memory uh to your uh your AI software stack to our model runtime giving you all of that as an integrated automated appliance is just tremendous value for our customers. And you're saying what's different about it? Well, the way we've been able to build this thing is you can go through and set all of your configuration parameters, get your first factory set up. Once you're done with that, we're going to output a YAML file. So now from that YAML file, you can make whatever edits you need to and then boom, boom, boom, boom, boom, you can spin up additional uh appliances, additional clusters, right, as your needs require. And this is fully automated and supported end to end by us. So you have one source for your from your AI infrastructure software all the way up the stack for when you run into any kind of issues. We're here for you and it's you know obviously our our customer support and service has been worldrenowned uh from our our customers. We continue to hear great feedback. >> What are they excited about? What's the big excitement from the customers? Because we know people want to run fast with the factories. They want to stand them up. The edge is coming to around the corner. What is the customer reaction to this at VCF AI factory? the the the customers are really excited about this because what we've seen a lot over the past like probably since you and I last talked I don't know how long's it been John a year or whatever >> about a year ago. >> Okay. So when we last talked a year ago uh one of the things I wasn't talking about enough was people were running into buyer remorse. They bought what they thought was this full turnkey solution and as it turns out it wasn't because they have to figure out how do I do agent tracing? How do I do observability? How am I doing high availability? How am I managing and optimizing storage? Can I uh tar my memory? Uh how am I doing configuration and change management? How am I doing disaster recovery? For production inference, you have a huge set of problems that you have to navigate. So with us, with our AI factory solution, you get all of this out of the box. You don't have to be like, let me stand up the factory and then peacemeal the things that are missing, right? And then have to navigate change management across that. It's the most common question I've asked these data science teams and chief AI officers when they've when they've deployed these other factory solutions. I first thing I ask them is how's change management working out for you? And then they'll go on a rant about how difficult it is. >> It's a major hassle. Change management is the number one thing with AI right now we're seeing. >> So life cycle management for us getting all of that out of the box and making it simple for customers is a huge huge deal. >> Yeah, I love I mean I think this is what the demand is. Okay, it's got to stand up and I got to run. I got the software stack. Um, another thing I want to get to is sovereignty because sovereign cloud we've heard and discussed many times. Now you got sovereign AI. Sovereignty is many issues. It's an on premises. It's also a geography issue. Uh, it's also a money thing too because if I'm going to run it in a country that's I want to keep those AI dollars inside inside my factory uh environment or the country. So what is the latest uh update on the innovation around sovereignty? How are you guys thinking about it? What are some of the conversations? Yeah, it's been uh a huge area for us. And now I do remember the last time we talked because uh we were talking about one of our customers which was shared services Canada and and how the Canadian government is running their canhat app on VCF today. Uh we see this in the US, we see this in Europe, we see this in Australia and other parts of a the APJ region where the benefit that they see from us is the fact that I can localize models. I can localize my data plane. I can localize my control plane. You have full ownership of everything end to end. identity, your encryption keys. There's no way that an external entity can shut off your private AI cloud with our solution. We're one of the only vendors in the world that can make that claim. And that's been huge for us. It's been huge for business. So when organizations are looking at like, well, how do I handle these sovereign issues? How do I make sure that some uh foreign government isn't disrupting AI? >> We're the partner for that. So it's nice guardrails, too. They they get the quality, get the security, and also the control plane, resource management, all kind of turnkey. All right. Uh to close out, how should customers think about their AI strategy in the next 12 to 18 months as this is moving super fast, we're going to see new changes? Again, mentioned some of the the stats we're seeing like with OpenAI and Enthropic, 50% market share in the enterprise for OpenAI. Enthropic is doing really well. You're starting to see AI and integrating into these environments. What's on the mind of the customer now for the next 12 to 18 months? How should they be thinking about it? >> More than ever, they have to architect for the expectation of change. They can't architect based on what looks good today because the space is moving too fast. That should be first and foremost. How can I have common governance for agentic AI? How can I have endto-end tracing? How can I have the flexibility to onboard all of these different models or even different AI path services as my needs change? So, talk to us. Have a conversation first. We all don't know what we don't know, right? So, level set yourself. Make informed decisions. Don't just start running at something because you know you need to make a decision. Take a step back. Make sure software is at the forefront of of your architecture and decision-m and then go from there. And we're here to help. >> Well, there's no doubt that it is now infiltrated all aspects of the business. It's not an IT project anymore. It's a business project and the world is getting waking up to the AI intelligence value. Chris, great to have you back on the cube here at Explore Coverage. Thanks for coming on. >> Always a pleasure, John. Thank you. >> I'm John Furry with the Cube VMware Explore coverage. We have two days of live coverage. Look out for the Cube. Thanks for watching.