VMware Explore 2025 | Paul Turner, Broadcom & Kumaran Siva, AMD

Clips
More from VMware Explore 2025

Paul Turner

Chief Product Officer, VCF Division

Broadcom

Kumaran Siva

Corporate Vice President, Strategic Business Development

AMD

GPU virtualization emerges as the cornerstone of enterprise-scale AI

As advances in graphics processing units accelerate enterprise artificial intelligence adoption, GPU virtualization is emerging as the foundation for where critical workloads ultimately run. While public cloud remains useful for early pilots, production deployments are shifting toward controlled environments that prioritize performance, security and efficiency.This transition is sparking closer collaboration between Broadcom Inc. and Advanced Micro Devices Inc., as the two companies engineer infrastructure purpose-built for private AI with GPU virtualization at its core

play_circle_outline VMware Explore 2025 event highlights featuring Paul Turner and Kumaran Siva.

play_circle_outline Enhancing Product Value: Integrating Private AI Foundation with VCF and Ensuring Compliance in AI Applications Management

play_circle_outline Collaboration between VMware and AMD to enhance AI infrastructure projects.

play_circle_outline Partnership history between Broadcom, VMware, and AMD in virtualization technology.

play_circle_outline Discussions of enhanced performance and efficiency with AMD's GPU offerings and architecture.

Info
Transcript

Paul Turner, Broadcom & Kumaran Siva, AMD

Paul Turner

Chief Product Officer, VCF Division Broadcom

Kumaran Siva

Corporate Vice President, Strategic Business Development AMD

At VMware Explore 2025, theCUBE’s John Furrier and Paul Nashawaty sit down with Paul Turner, VP of Products for VMware Cloud Foundation (VCF) at Broadcom, and Kumaran Siva, Corporate VP of Strategic Business Development at AMD, to unpack the latest innovations driving Private AI and enterprise-scale infrastructure transformation.

Turner details how VCF 9 is embedding Private AI Foundation directly into the core product, enabling customers to run VMs, containers and AI workloads in secure, compliant environments. He emphasizes the importance of open c... Read more

explore Keep Exploring

What event is being discussed in the text, and who are some of the key individuals involved? add

What is the reason for integrating the Private AI Foundation into VCF? add

What is the focus and objective of the partnership discussed in the text? add

What has been the nature and duration of the partnership between Broadcom VMware and the Epyc CPU line, and what future collaborations are planned? add

What is AMD's perspective on the importance of VMware and the VCF package in relation to their GPU offerings and overall market response? add

bolt Powered by CUBE AI

Paul Turner, Broadcom & Kumaran Siva, AMD

search

>> Welcome back, everyone, to theCUBE. We're live here in Las Vegas for VMware Explore 2025. I'm John Furrier, host of theCUBE, with Paul Nashawaty, Lead Analyst, as part of our CUBE Research team. We've got a great lineup. We've got the person who's got the keys to the kingdom at VMware, Paul Turner, Vice President of Products at VCF. VCF 9 really hitting the market in a big way. Of course, we covered the launch and it's continuing to get great press here at the event. And we've got Kumaran Siva, Corporate Vice President, Strategic Business Development at AMD, and a great collaboration. Gentleman, thanks for coming on theCUBE. Paul, great to see you again.

Paul Turner

>> Yeah, wonderful to see you, John.

Kumaran Siva

>> Yeah, thank you.

Paul Turner

>> What an event, right?

>> Looking great. Again, the VCF success has been... We had a lot of customers come on with great news, so congratulations. We know you're under a lot of pressure to keep bringing those great products.

Kumaran Siva

>> That's right.

>> Works as advertised. Congratulations.

Paul Turner

>> As we've talked to you over the last few years, I mean, innovation's everything. And hopefully you're seeing that. Hopefully you're seeing us deliver for customers.

>> And the results on the business side, it's phenomenal. Congratulations on that.

Paul Turner

>> Yeah, good, thank you.

>> Just want to get that out there. What's the big changes in the AI side with VCF? Obviously Private AI's hot. What's the update? Tell us why that's an important part of the news, why it's part of VCF?

Paul Turner

>> Yeah, yeah. Well, part of VCF is part of the news, right? I mean, that is a big change. Why are we doing that? We're actually taking all the Private AI Foundation, building it into VCF, and actually delivering it as just part of the value of the product included in every VCF core that customers have. And the reason for that is that customers need to be able to run any application, VMs, containers, AI applications, and the value of Private AI, if you just think about it, people are worried about compliance, data sovereignty. How do I manage the security of that? They have to be able to run AI and particularly inferencing part of AI applications on their data, in their security areas, in their data center zones, and that's what we're enabling. And we want to keep that as a strong, heterogeneous environment for people to be able to build and run AI applications.

>> It's a really strong feature of VCF. Congratulations. Hock Tan and the team announced the relationship and partnership and the collaboration with AMD. What's the meaning of that? Unpack that for us. Why is it important?

Paul Turner

>> Oh, I think it's... Delighted actually to be here with Kumaran and actually talking about that partnership because we care. If you think about it, we always have looked at AI infrastructure and seen how can we maintain open choice for customers so that they can build it on the infrastructure that they want, be that NVIDIA or AMD systems we previously had announced with Intel? But we really are here to build a stronger open ecosystem, and that's what we're showing with the AMD collaboration. We're taking our capabilities of our Private AI Foundation and all of that is available on the latest processors, latest GPUs from AMD, and that's what it's all about, choice and Private AI delivery.

>> Kumaran, jump in here. Talk about the strength of the collaboration and the history.

Kumaran Siva

>> Oh, yeah, absolutely. Look, so we've been partnering with Broadcom VMware for the last 10 years plus, and we have shown enormous value to our customers on our Epyc CPU line where we partnered together. We have the best virtualized solutions on the market today with VCF and Epyc CPUs, and we're now extending that to a collaboration, strategic collaboration around GPUs. We see, as enterprises truly start adopting, you started going from Snowflakes to actually having broad deployment. That means VCF, that means VMware, and that also we want to be part of that and we want to deploy with you in mainstream enterprise, and that's a really important thing that we'll do together.

>> So the scale's huge?

Kumaran Siva

>> Absolutely.

>> That's a big part of this.

Kumaran Siva

>> Absolutely.

Paul Nashawaty

>> Yeah, scale sounds very huge. Also, there's a lot happening, the excitement on the floor that we're seeing, the announcements that's happening, it's really, really cool. I want to get down to some of the brass tacks though. When we look at this collaboration between the two organizations, we saw a lot in the keynote today, we saw a lot of announcements. What are some of the areas where you can elaborate more about the products and why it works from a technical perspective?

Paul Turner

>> Yeah, sure. So if you think actually about the Private AI Foundation, we actually have to do two things. We have to actually virtualize GPUs properly. We have to be able to take advantage of them, and we can actually go and build a pool, GPU as a Service, a pool of GPUs with AMD. We can actually use our DRS capability to be able to do automatic identification of best and optimal placement of that resource against that pool of GPUs. So you're getting much better efficiency. You're managing this pool, you've got all of your AI applications, you're managing that. Again, much more efficiency of the environment. But beyond that, we are actually able to run our Private AI stack, and our Private AI stack brings a customized vector database. So you can actually do your RAG model and tuning that you need to do because you take a base model, and then you actually use that model and you do RAG optimization of it, retrieval-augmented generation. And then the second thing that we have is AI Blueprints that we build in, so you get much faster delivery. Again, all on top of the AMD ecosystem, and it's quite powerful actually when you deliver that, better operational management of the environment too.

Paul Nashawaty

>> It totally makes sense. I want to talk about the AMD side of it though. So when we look at the announcements Paul was just talking about, a lot of demand hitting the market, what's happening in the market. What's from the AMD's response?

Kumaran Siva

>> From an AMD perspective, exactly what Paul's saying. We really view VMware and the VCF package as being a critical component to making our GPUs easy to use and easy to adopt. Just as he talked about, to be able to do the virtualization of the GPU, the partitions, all of that, we do the software and we tie in to VCF. So we have our GPU operator, we have essential components that we are working very closely together with Broadcom to make sure that we enable the full ease of use. On top of that, we focus on optimizing the inference engines, so things like VLM, SGLang, VLMD for distributed inference. We have all of those capabilities in development and optimized, along with some of the most popular open-weights models out there today. We have, just as you guys bring the components with respect to RAG and all the pieces that can also run on our Epyc CPUs, those then drive into models that are optimized day zero available for the Instinct and for the AMD-VCF architecture for this Private AI system.

Paul Nashawaty

>> No, it sounds like it's a real win for your clients. To join clients that are working on it, it sounds like a real solid win. As an analyst, I have to ask these questions, right? I want to know what's going to happen in the next 24 months. We're sitting here two years from now, what does that look like?

Paul Turner

>> I think I'll start with I think we talked very briefly there about the work that we're actually doing on open models, OpenAI work. We're actually working very closely together on that OpenAI ecosystem. You're going to see more and more development of new open frameworks, models, inferencing servers. We actually have joint projects in with some of the universities. We're really going to open up this ecosystem. Now, all of that can just work as is against infrastructure that we're now providing on top, but we're just a virtual infrastructure and we're virtualizing the GPUs and providing it as a GPU as a Service. So I think that's one big thing, all of that ecosystem of applications that we're building together. The other thing that we're working on is, of course, vMotion and GPU vMotion so that we can do true non-disruptive operation. This is going to be GPU as a Service. I've got to be able to take an application, dynamically move it, no disruption to the application. That's in development right now between the two of us, and it'll be an exciting development. Already we have the ability to partition, monitor the GPUs, do DRS placement. You bring in the ability for us to do dynamic vMotion, you got cool stuff coming.

Paul Nashawaty

>> Super interesting. I mean, we find that in our own research set, in theCUBE Research, that 20% of respondents in our recent survey indicated application portability as critical to them, 67% say it's very important to them. So it's very much in alignment to what you're talking about.

Paul Turner

>> Yeah, yeah. Well, it's important in this case to get more efficiency off a relatively costly resource. Sorry about that.

Kumaran Siva

>> Yeah, yeah, yeah. No worries.

Paul Turner

>> GPUs are costly.

Kumaran Siva

>> No, for sure, for sure.

Paul Turner

>> Hope you don't mind.

Kumaran Siva

>> Yeah, absolutely.

Paul Turner

>> But getting the most utilization out of them-

>> They're valuable.

Paul Nashawaty

>> They're valuable.

Kumaran Siva

>> They're valuable.

Paul Turner

>> They're very valuable, but you-

Kumaran Siva

>> Exactly.

>> Everybody wants them. I mean, the demand is so high because you're streamlining the models. You get choice. I mean, just it's hot. I mean, people want more.

Paul Turner

>> It's one of the things that people have said. If you drive up the utilization and use, you don't actually lower adoption or volume of products. You actually increase the volume of products in the market-

Kumaran Siva

>> That's right....

Paul Turner

>> because people get such value out of it, but sorry.

Kumaran Siva

>> No, no, no, that's great. I think you summarized it well, and by the way, I think it's great. Utilization is super important to enterprises, and we've certainly seen that in the CPU world where we have tens of thousands of applications that enterprises running today on VCF with Epyc CPUs underneath, seamlessly running, live migrating all of those features today. We will bring that to GPUs. Now, what is exciting about GPU as a Service and that layer that VCF is going to bring in place, we see thousands of agents. We see thousands of applications that are going to get deployed across enterprises as they do these different functions, both vertical and horizontal, things like drug discovery, things like mechanical design, things like customer service agents, things that span all enterprises. That's going to become super important over the next few years. The 24 months, I think all of those things are going to start to come to fruition, and VCF and AMD are going to be there to be able to drive that.

>> I love the enablement. I mean, I love that enablement message. When you have a platform, the enablement is the proof point. You start to see more agents, more productivity, more value creation. All right, so I have to ask, customers are also a proof point.

Paul Turner

>> Yeah, of course.

>> Can you guys share some of the use cases right now in the collaboration that's popping out? What are some of the examples can you share? You don't have to name names, we don't have to, but just give us a taste of what's some of the scope, what's some of the value that's being created and extracted?

Paul Turner

>> Yeah, no, I'm actually happy to share a little more on the different customers. So as we look at Private AI, we've getting some interesting use cases. So Loomis, a security company, you probably have seen them, they protect many different enterprises. It's all logistics management and how do we actually help them optimize their logistics management and delivery services? You've got NTT DATA, they're actually using it as GPU as a Service, so actually a managed service for actually GPUs. We see a lot of customers then go down the path of customer service improvements. We're even doing this internally. We're actually using it inside in our own support organization so that we can actually do automatic call-and-response mechanisms, a very standard way to do it. So the use cases are blossoming. If you take Gartner has gone and analyzed all of the use cases that are potential use cases, and they see up to 35% productivity gains across different industry types, different use cases for generative AI use cases. So they're very real. I think what's happening is originally they were built as cloud-first-only kind of services. Because of data privacy, data concerns, sovereignty, Private AI is growing in leaps and bounds. So companies I mentioned we're all actually Private AI deployments of the services, and I expect we're only at the beginning of it. I think you're going to see a lot more deployments of gen AI use cases into enterprises. It'll be part of every deployment of every server infrastructure out there.

>> The on-premises action with the crown jewels, the IPv, the data, is a huge factor.

Paul Turner

>> Yes, yes. Yes, huge.

Kumaran Siva

>> Yes. I've heard so many customers say, "POC and cloud deploy on-prem," right? So private cloud is going to be a real thing. Absolutely the reasons that Paul mentioned. Regulated industries need to keep data secure, and so this is going to be a challenge that the partnership here really goes a long way to addressing. Just like what we did with Epyc where we brought low-power, low-cost virtualization, creating high value to our end customers, we're going to be able to create high value for these use cases in GPU and then with the agentic AI and all these other LLM type of cases coming up as well.

>> All right. You got the customers, you got the collaboration, you got the use cases and features. With deployment being complex for some, you guys are simplifying it, check. Performance is huge. Power, you mentioned. That's a great feature. Can you guys just lay out the architecture of how this works? Explain the secret sauce of how this is working. What's the architecture look like? Because you got speed for deployment, app dev. He covers app dev like a blanket, there's huge demand, building apps on top. That's the thousand flowers will be blooming.

Paul Turner

>> Yep.

>> Talk about the architecture with scale and deployment. They go hand in hand big time.

Paul Turner

>> Yeah, so I think a few kind of things. As we look at the infrastructure, just virtualization of the GPUs we do as base. We build on top of that DRS, vMotion, you can actually deliver GPU as a Service. On top of that, we actually give you the ability to build out a PostgreS kind of vector database so you can actually use that. You have a model manager to actually model your private models and decide which ones you actually want to make available to developers. You have a whole set of effective blueprints that are available that actually give you very rapid deployment. All of that is what's included in Private AI Foundation now part of the core, all enabled of course on AMD infrastructure. But I'll let Kumaran add on all of the extensions that we actually did then from the AMD side.

>> All right. Tell us the secret sauce. Tell us what's going on.

Kumaran Siva

>> Yeah, no, from our point of view, look, so for us, we go back to the models. We have all of these models that are made available day zero. So whether it's Llama, Quinn, DeepSeq, all of these models now available, Mixtral, available day zero. So those things are there and they provide the base infrastructure for these other applications that Paul talked about that sit on top that run on Epyc CPUs, but then drive into our GPUs. We have that capability. Now, the GPUs themselves are super optimized to run these models well, and that's what we focus on from a silicon and from a software enablement standpoint. We bring things like FP4, FP6, FP8, so these quantizations that make the models run well. We have higher memory bandwidth, we have more memory capacity per GPU. That allows you to run more of these models, more concurrent users, run them more efficiently, essentially. So it goes in well with the virtualization capabilities of the GPU and allows better value to the end customer, better performance and meeting of the SLAs.

>> Kumaran, talk about the benefits of the customer real quick because you're bringing speed, you're bringing some efficiencies, you got the power. What are some of the things that they'll see from this?

Kumaran Siva

>> Oh, absolutely. I mean, fundamentally it's going to be about the economics of going, using a GPU for a particular use case will just look so much better. We're using these models for this particular use case. For example, the number of users that you can have running at the same time. Let's say you're a customer service agent and the number of customers that you can service with a single piece of infrastructure will multiply between VCF and AMD and the kind of technologies we're bringing to bear, the responsiveness, the ability to meet SLAs, those will all get better and you'll be able to hit the time to token, the time to... response time's that you can see-

>> Next token out, next token. So it goes beyond the one shot to the-

Kumaran Siva

>> And throughput, yeah, absolutely....

>> reasoning because the inference is going to be big too.

Kumaran Siva

>> Exactly. It's going to just make the AI better, it's going to make it more applicable, and the use cases are just going to blossom.

Paul Nashawaty

>> But it also sounds like it meets the clients where they are in their journey as well, so they can adjust and adapt as they need to go, right? That's really what it comes down to.

Kumaran Siva

>> Yeah, it's a super complex... I mean, this is what we're building here is the foundation. It's just going to be super amazing to see what happens, especially over the next 24 months because you're going to have applications that will optimize different aspects. And so the fact that you are building on such a strong foundation, it allows you to change things and allows those things to evolve.

Paul Nashawaty

>> Dynamic, yeah.

>> Yeah, I mean, I'm smiling. Paul knows I'm a pushover for software and chips because when you put software and chips together, magic happens because you're getting at some of these root level, feels like the '90s again, where talking about memory and the configurations matter because software's going to be very dynamic, programmable. You've got agents in there. This is huge and I think VCF sets the table, Paul. You got the roadmap. We all know the pressure's on you because people are going to want to see more from you.

Paul Turner

>> Of course.

>> They want to see more GPU as a Service. They want to see policy-based stuff happen based upon certain models wanting certain resources. I mean, it's a computer science dream scenario for builders.

Paul Turner

>> Oh, it's the fun.

Kumaran Siva

>> It is.

Paul Turner

>> It's the fun, right? We're building an ability for our inferencing engine so you can actually do shared inferencing of the same model. We already have model sharing that we can actually manage problems, trusted models that you share across your enterprise. Now you're going to be able to deploy them onto your GPUs and actually have multiple application servers actually share those, so multiple application teams. You're getting the IT team actually manage this at scale, which is much better than the little poke fiefdoms-

Kumaran Siva

>> That's right. Definitely....

Paul Turner

>> that would've been run before. All of those things are really going to make a difference. And I want to go back to the openness because I think it's so critical.

>> Choice.

Paul Turner

>> We're building this as an open software stack. We're working with OPEA, which is the open framework for AI applications, and I think we'll continue to expand that jointly. I think we have a phenomenal partnership with AMD that we have built, and I think we're going to continue to strengthen that.

>> Yeah. I mean, the openness is a great call-out because if you look at the success of first principles of VMware and AMD, it's about open systems, which we're building, interoperability, ecosystem. Now you starting to see ecosystem partners come out, I won't say out of the woodwork, but they're coming on the market. Integration. They look like SIs. It's not resellers. It's real technology. It's distributed computing. Any reaction to that ecosystem? Because we're seeing that develop real-time here.

Kumaran Siva

>> Yeah. Open is a big part of what we do. AMD is very committed to making sure that the whole stack that we provide is open-sourced. We have open-sourced our GPU operator, our whole ROCm stack, which is our low level, the programming the GPU itself, completely open. On top of that, the inference engines that we drive and optimize around, completely open. So what we have in conjunction with VCF is being able to bring some of those open technologies and the ability to have that rapid innovation on a solid basis of VCF and on a solid basis of having this reliable substrate that the enterprise is built on.

>> Right. We report on SiliconANGLE. This is something we talk about on theCUBE all the time. The best coders building the systems are getting closer to the hardware because every ounce of optimization has a step-function advantage. Reaction. You agree?

Kumaran Siva

>> Oh, for sure. I mean, I think that's fundamentally what GPU programming... It's easy to program parallel code. It's hard to make parallel code run well, and that is truly been a focus for AMD in terms of our Developer focus, in terms of the Tool Chain. What we've been doing, really focusing on how do you get the most performance out of the GPU that you possibly can?

>> Guys, super exciting. Thanks for coming on and sharing. We could go an hour on this on a podcast. We should do a podcast on this because I think there's going to be an app dev impact because you're going to have new developers coming in with agents and the underlying layers have to be better and will be better. Let's wrap this up. Bottom line, what's next in the collaboration? You get a little teaser there, Paul. Thank you. Where do we go from here? What can we expect?

Paul Turner

>> I think a lot of the focus we're going to do together, of course we're going to make infrastructure super efficient. We're going to make GPU as a Service very efficient so that you can move applications around. We can re-consolidate applications. That's a given. The open-source efforts that we do together, to build to your point, an ecosystem of applications that we can... We don't need to build it all ourselves. We need to encourage an ecosystem to build. If you think about where generative AI came, why has it blossomed in the last few years? It's because Meta made the very smart move to actually open source a model out into the industry and said, "Go at it, innovate." And that's what we're doing together is we're building an infrastructure, but we're telling everyone else, "We've got a baseline. We give you an inferencing server, we gave you a model manager, but innovate, the applications come from you." They're going to come from every enterprise customer. They're going to come from startups. Let's have them innovate. That's the fun.

>> Bring us home. Final word.

Kumaran Siva

>> Oh, I think, look, today you talk to an enterprise, they're doing maybe a dozen possible use cases, and a lot of them are POC today. I think in 24 months, because you think about how we're going to go forward, we're going to have tens of thousands. You're going to have hundreds of thousands of applications. It's going to be pretty damn cool. So I really look forward to that whole layer. It's just as Paul said, the open ecosystem that we're enabling on top of VCF is going to blossom, and we're going to have tons and tons of these great applications across the board.

>> Good. Kumaran, Paul, thank you so much. A tsunami of applications, modern applications, private AI applications.

Paul Nashawaty

>> It's a tsunami of applications.

>> Out in the open, open innovation, of course, we're out in the open here on the floor with theCUBE. I'm John Furrier, Paul Nashawaty with theCUBE. Thanks for watching.

Kumaran Siva

>> Thank you.