Prashanth Shenoy, Broadcom | VMware Explore 2026

[music] Hello, I'm John Furry, your host of the Cube. This is the Cub's VMware Explore coverage. Of course, we have two days of live coverage from the show floor. I'm here with Broadcom executives to unpack the innovation at and news at the show. I'm here with Pashant Shenoi, VP of product marketing with VMware Cloud Foundation division. We're going to talk about private AI cloud and the VCF AI factory. Pashant, great to see you. The world is interesting times we're living in right now with neo clouds and private cloud and AI intelligence getting injected into one the infrastructure and also enterprises. >> What's the news about with VCF? I mean you must be loving life these days. >> I know it's been exciting, hectic but but pretty cool. So uh the main uh news at the show here is all about VCF AI factory and I think we've reached a very critical juncture in in the world of AI right like a lot of our organizations are moving from pilot to production at scale. So there are big concerns around cost tokconomics security privacy concerns of their data. So a lot of our organizations are looking towards private cloud as the preferred platform for deploying their production AI workloads and that's where the VCF AI factory really comes through right it's a very turnkey unified integrated solution from metal to model right so it brings all of the underlying ODM OEM servers that we have worked with to certify and validate and test the VCF AI software solution and the model as a service platforms with the choice of GPUs whether it's from Nvidia and AMD all packaged integrated so it provides really a faster time to fast AI model deployment provides a very unified simplified day 0 day2 operations with the right security and privacy requirement that a lot of our customers have. So we are pretty excited to help bring that to the market. You know, it's been interesting the past three years watch AI factories kind of grow from a concept before the name was coined by Jensen uh at NVIDIA. We were calling it large scale clusters. So in IT, you know, you had big data centers, hypers scales were doing their things, large enterprises had their IT with virtualization and you guys were a big part of that with VMware. But then the world started changing. You saw the density of the servers. Rack scales back, the data centers back, not that it ever left, but it's much more connected in the hybrid way. So it's not like clouds going away. That's growing. >> Data centers growing. The rise of the Neo clouds or Neolabs is also is called where you have this kind of new intelligence service provider capability operating at hypers scale >> is proven that this new model's here that AI intelligence >> is now being injected into the market. So the demand's clear everyone talks about that. So having a VCF AI factory actually is the timing of all the work that you guys are doing we've been covering is packaging VMware putting everything into simple SKUs turnkey now with AI factory take me through how that works you just like here's your factory and then pumps out tokens take me through because the the demand is stand up the factories and then operate it >> just what sounds like what VMware used to do for virtualization but a different scale how does it work >> yeah absolutely because the main problem you're dead on right Like the the concept of factory is like without the factory our customers have to manually build and stick together things right from the GPU accelerated servers to virtualization to storage requirements to data to Kubernetes services all packaged and then with the model gallery that they need to pick and choose right so they have to manually build it so there's a lot of silos created it's not performance validated it's not tested so it takes a lot of time from day zero to day two operating. So we have solved that for our customers, right? And the good part about this is the existing VCF software that a lot of our customers have is ready for an AI factory. So what we have done is really work with these choice of OEM and ODM vendors including Cisco, Dell, Lenovo, Super Micro, work with the chip vendors and the AI software stack vendors like AMD and Nvidia and worked with over 150 models that we have tested, validated and optimized to run on VCF and brought it to our customers. So they get that flexibility and choice but they don't need to be involved in doing the bring up themselves. Right? So that's the key aspect and add to that we are also announcing our partnership with metalsoft which helps you do this integrated heterogeneous firmware and hardware bring up right so you don't have the siloed hardware management and then the software thing all of this is integrated into the VCF ops console so the operations become drastically easy and the provisioning of this hardware and management gets reduced from like months weeks to now minutes >> so the standing up the factories that's what you mean by turning it on >> okay I load it in like like the old school days of here's some Linux turn on the server it's up and running. >> Yeah, it's validated and tested so that you can go with confidence that the day zero server certification is done for these GPU accelerated servers. Uh the model gallery, the model runtime, the secure agent framework, all of that AI services that we have built into the VCF is already there for you to get up and running. So you get that observability and visibility that you need but you also take out the pain from the day 0 day2 provisioning operations management and optimization of your infrastructure and that's a big deal because they get a lot of choice and control to manage their tokconomics >> talk about the uh the VCF software versus the private AI services you just mentioned what's different I know VCF is well understood package people can turn stuff on and that was the goal of you guys is to get customers doing more things so this is a good opportunity But the private AI services what what's in there? >> Yeah, explain. >> Private AI services is mainly to bring enterprises AI applications to their secure private data that they have on premise and run that in a very secure scalable manner on our VCF private cloud. Right? So what that means is you have things like model gallery, model runtime, model governance, API AI gateway all built in as part of the private AI services. now as part of VCF. So those are some of the key capabilities, right? So multi-tenant model sharing is one of the new private AI service that we are unveiling which means that you can share the models across your tenants or across your line of business but your data is governed and private for that particular tenant and model right so really helps you with the GPU resource allocation as well as ease and efficiency. So that's number one. uh AI gateway is another key thing where you have a unified governance and a single consumption interface for local models to interact with cloud hosted models. So we work with over 40 cloud model providers including Google, Anthropic, OpenAI etc. uh so there's still this hybrid environment but with the choice and flexibility and finally as the world moves to agentic so there is an AI agent secure framework which means that this dynamically generated code by agents you have containers that they can run in but we govern who these models and codes can talk to what tools can they access how do we validate the output before they can actually go into the real world so providing an AI sandbox for example is a very very creep part. So these are some of the V private AI services that we built in on top of the existing VCF that you said customers already choose and use to run. >> What's unique about um you guys is you have a unique view with the customer because a lot of enterprises have VMware >> and they want that domain specific data to be unlocked get value out of it. Um the thing that we're seeing in the market I want to get your reaction to this is that >> enterprises they're good at data platforms they've been doing data platforms and so certainly evolving they love the models and you mentioned frontier general intelligence internet and then their their maybe their internal models or stuff for their their [clears throat] data >> but a new layer is emerging that we're seeing that that we're seeing is >> it's a context layer it's not just a semantic layer it's like okay I got data platforms everywhere and I got these models >> so I got to connect them and bring context to them. Yeah. >> So, you're starting to see conversations here with you guys and others where it's like, okay, here's the data that we're going to put in for this agent or we're going to use a graph database here. So, the the data and models get separated >> and this new layers emerge. I don't know if call it a context layer, but it's a layer to manage >> the context and domain specific. It might be a big model, yeah, >> small model, >> a fusion of two. >> Yeah. general intelligence from open AI and CHPT versus my finance model or my manufacturing model. Yeah. What is that? How does that how does that resonate with this service? Because >> yeah, >> a factory has to deal with this. >> Yeah, absolutely. You're spot on. So, one of the things that we have realized and our customers have realized is not every use case that enterprises have require a frontier LLM model, right? So, there's a lot of SLMs emerging in the market. So it's all about purpose, fit, governance and cost which is very very critical. So our job is to provide a choice of these models SLMs, LLMs, openw weight open-source models as well as commercial models to be available for our customers for their fit and purpose. So that's point number one which is what we are providing right. We have around 150 uh models that we have tested, validated and optimized to run on VCF. That includes everything from Google Gemma to Microsoft 5 open source and open weight, NVDS, Neotron, ZAI and pretty much NEC, Cottomy for some of the regional models requirement. So our cho our job is to provide the choice and flexibility of SLMs, LLM's openweight, open- source model to run on the same VCF platform that they have used to run their VMs, their Kubernetes, their container workloads. Then comes the context which is very very critical right like who gets access to what models. How do we create this AI sandbox? How do we create the model governance? How do we provide the observability and intelligence of how your capacity utilization of your GPUs your memory your storage is functioning so that the infrastructure admins the VI admins who are now AI admins as I call them are now responsible for managing that right right managing the AI tokconomics. So that's I would call the context layer which is part of >> and you guys do that. So you provide a capability to for that I'm an enterprise I can I can manage my data and I can I don't have to bundle my data with the model I separate them and you guys have a little ability to do that. So we have the data services built in and and Purima is also going to talk about the data foundation as part of Tanzoo that provides the key set of data services that is required right like things like object storage which now we introduced as part of 9 uh 9.1 as part of VSAN is very critical because AI is running on unstructured data right so how do we manage that how do you provision this how do you provide high availability all of that capabilities is part of the VCF AI factory and the VCF private AI storage >> object stores is hot right now and it's really compatible with the AI. Um, I'm glad you explained that. I really wanted to get that out because this is the big challenge. Okay, I got the models, got the data. Great. Now, h how does the factory VCF factory compare? Uh, let me rephrase. What's the relationship with AMD and Nvidia? Yes, they are suppliers on the semi side. Of course, Broadcom has a lot of chips too in-house for that. um semis and the density of these systems that are going to run ECF or run software at scale, >> they're getting denser. So the HBM memory, you got now I see solid state has got some HBM kind of features too. Yeah. >> Um and you're starting to see the role of storage change. That's why object store seems to work well. So you got storage closer to the GPUs. You got CPUs now with pre-filled decode. So the architecture is starting to form. Yeah. With all these big semiconductor systems. >> Yeah. What's the relationship with AMD and Nvidia? How does that tie in? >> Yeah, absolutely. I mean, AMD and Nvidia have been long-standing partners for us as part of VCF, right? With the VCF AI factory, what we made sure is all of the GPU accelerated servers from the OEM and ODM vendors, they run typically Nvidia or AMD. So, especially with AMD, we are now working to have like a VCF AI factory with AMD edition there. Right? So, that's that turnkey tested, validated. So it includes the VCF software. It includes the AMD MI350 GPU instance series that we have tested and validated on across the OEM and ODM servers. It includes the the Rockm Enterprise AI software from AMD as well as their DVX drivers, right? That helps you allocate the GPU to run on large VMs across a single V VKS cluster for example, right? So it's fine-tuned, tested, and validated all packaged with the private AI services that I talked about into one software that's out and running out the gate. Right. So that's >> so validations all 100% validate with Nvidia and AMD. >> Yeah, absolutely. And goes with the same OEM. So the day zero server certification becomes very very critical, right? Otherwise the customers don't have the confidence in terms of the scale in terms of the performance in terms of like are they certified to run this models etc. So we make sure that out of the gate we have day zero server certification with customers with vendors like Dell with with Cisco etc which becomes very very crucial to take the friction out of these fast first AI model deployment. So if I'm an enterprise large enterprise I have a lot of Dell for instance or super micro >> or maybe even Lenovo okay but Dell has a lot of AF factories going out can I run VCF on Dell >> yeah so that's one of the key things right it again goes back to as the server costs have gone through the roof our customers are looking at how do I get that how do I sweat my existing hardware number one uh when I get new hardware from the factory how do I make sure it's AI ready. So we worked with Super Micro, Lenovo, Cisco and Dell to create the VCF AI ready node, right? So what that means is VCF with these GPU accelerated servers from all these partners that are tested and validated to run the AI factory and that makes it very very crucial for our customers to get that choice and flexibility but with the cost economics and the server vendor of choice that they want. Right? So that's a key part. So how has this impacted the ecosystem because I mean we've been covering VMware I mean this is our 17th event years put the gray hairs long been a long time but the ecosystem you always had great ODM and OEMs okay you've had a robust ecosystem it seems to me I'm getting tons of inbound emails and messages around this year around working with you guys you mentioned the big names they're all the biggest hardware names hardware systems >> players >> how has the ecosystem changed because now the game is on again in a big way there's a lot growth. I mentioned the Neo clouds are growing. That shows the demand curve. The enterprise is opening up. >> Yeah. >> For enterprise AI because it's about a year into pilots. We're starting to see progressions with coding and now Agentic >> still not penetrated to the level we had hoped, but we think next year is going to be huge. >> Yeah. Yeah. >> What's the ecosystem like now um with VCF AI factory? You're just turnkeying large scale software system. Yeah. Yeah. >> To run on the large rack scale and new AI dense same differently configured but high performance systems. >> Yeah, absolutely. One of the things that I think every vendor and every AI player has realized that it's not a one vendor game at all. Right. From from the chipset to the systems to the servers to the AI stack to the models, right? So it's going to be a very the the vendor that truly will be successful in my opinion is a vendor that has a very strong ecosystem play that will take the friction out of deploying uh the AI use cases that our customers want but in a very costefficient secure manner right and that's what we are trying to be for our customers right how do we take a set of curated ecosystem partners from the server side from the AI I chipset side from the system side, the software side and the model side and help package it all together to provide as a turnkey solution offering to the customer. So the ecosystem is going to be pretty huge. >> Let's pivot to frontier AI security. This is coming up. Um in private cloud, private AI cloud, I like the name, but you can just it's secure cloud. I mean it's private is essentially my workloads running >> private and private cloud coming together. >> It's hybrid cloud. It's distributed computing, but you can get the security. So it's private to the data. I get that. But security has to be built in from day one. Talk about the the Frontier [clears throat] AI security story with VCF right now. >> Yeah. One of the interesting things that we saw this year after some of the frontier AI security models uh being launched in the market is the 3Vs that I call of the security landscape. The volume, the velocity and the type of variety of these AIdriven threats have just exploded. Right? So uh attackers don't take weeks or months, they take hours or minutes to to get into the system. That's point number one, right? So you can't have patches now done on a six-month yearly basis. You got to even weeks are not enough, right? So you need to have a continuous patching. So the concept of what I call intrinsic security or built-in security at your infrastructure layer, workload and application layer is extremely crucial in the frontier AI model. So one of the key things that we have done in VCF 9.1 is to harden the infrastructure from the ground up. So 9.1 is our marquee uh release that comes with all of the vulnerability fixes that we've already identified based on testing of the frontier AI model and we now have monthly patches released out into the market. Right? So that really gives the confidence that it's a highly resilient AI infrastructure. So that's some of the that's how we envision the threat landscape. So having a very hardened infrastructure with vulnerability discoveries, virtual patching requirements, monthly patch released with lateral security is very very crucial to help drive this frontier AI security landscape. >> The we all saw the black hat had a great postmortem on the open AI hugging face >> agent on road going. Um that speaks to some of the data modeling >> having make sure the open weights are verified. Yeah. um mythosis of identifying like so many holes in the system. It's like Swiss G's they got to get filled. But you guys are using AI on your side to actually not only identify but to move fast on patching. What is the current turnaround on patching now? Of course you upgrade the commercial software and you got the open source software from VMware. >> Yeah. >> What's the just give us some anecdotes on like what's the frequency and what's the velocity look like on your side? Yeah, we are uh pretty much releasing monthly patches for for VCF9.x, right? So that's one of the key shifts that we have done. So customers don't really need to wait on a quarterly, six-month time frame. We have monthly releases and all of these releases come with all the vulnerability testing, the discoveries that we have identified on an ongoing basis and we patch that and we release an EP express patch for our customers on a monthly basis and then there's a quarterly big patch release too. So we are making sure that we are moving in with the times providing a lot more faster patching that's one but also to make the patching experience itself easy and do in place upgrades we have ESX live patching that we introduce right so now it's there across the stack so you can do in place upgrades without bringing your servers your workspace uh your work workloads down so it's very very critical to have that resilient infrastructure that aids harden and then it's easier to patch through virtual patching and and live hatching. >> You know, the agents become great uh service in this area. It's like uh you know what keeps you up at night or security while you sleep. Agents could do all this. Oh yeah. I mean you see a future or now >> seeing some real time agentic work being done once you baseline >> the security layer. >> Yeah. >> It could actually go in and actually while you're sleeping actually take care of a business. >> Yeah. So we have a lot of AI assisted operations built into the system that I haven't quite talked about. Right. It's it's VCF for AI and how AI is used within V VCF for what we call AI assistant for VCF. How do you do your day 0, day 1, day2 operations a lot more from a monitoring and troubleshooting perspective from proactive diagnosis and triaging systems when it happens and also the upgrades the life cycle management capabilities etc. So using AI agents at the vsenter level for storage for networking VCF operations are all the things that we are building um as we head into 9.1.x X >> you guys participated in glass swing. Um what were some of the results that come out of that? This is you know people looking at to get security people coming together. What assessments, what data, what blueprints came out of that? What's going on on these this in these security groups that are coming together? >> Yeah. Yeah. One of the key things that we figured out is while patching and all of that is good, our customers want some basic questions answered, right? How secure is my infrastructure? What is my risk analysis look like? Uh how do I get to have a very reduced business risk as more and more of these frontier AIdriven threat comes in and how do I continuously keep it that way? Right? So what we're releasing here at the show is the frontier AI security readiness program. So it has really four key areas. One is the assessment area, one is the architect area, the third one is implement and finally it's really upskilling and retraining our VI admins to be AI resilient admins. Right? So these are the four in the assessment we really work with our customers and say we assess their underlying infrastructure and we have in ANS our security vend uh a built-in system that can go and look at the entire threat landscape of the customer and give them a threat score. Right? So doing an assessment giving a threat score to say and then doing a gap analysis of here's where you want to be and here's where you are. So that's assessment. >> Second is the architect right which is how do I create and give you blueprints and best practices for architecting uh resilient private cloud infrastructure. Three is the implement working with our service delivery partners and our professional services so they can provide patching services upgrade security observability triaging and recommendations etc. And finally it's the upskill. So we are releasing this AI resilient infrastructure expert certification. So our VI admins can be security experts because everybody is going to be called to be this AI security expert and make sure the infrastructure is resilient. So all of these four packaged into a program that is available through our partners and through broadcast. I think that's a really important point uh and I wanted to bring it up because um every conversation that I'm in the past couple months around mythos has been scared >> people were scared out of their out of their minds. >> Yeah. >> Every CISO, every executive because it was so powerful, widely reported um certainly on the mainstream but when you get into the circles of technologists >> Yeah. >> this was a big deal because it was a seminal moment. It was almost like a chat GPT moment for security. Glass wing was important. So explain to people what's going on behind the scenes because this wasn't this was a call to arms if you will for the industry. Why is Glass Wing important? I know Anthropic kind of was behind it but it's really not just Anthropic. They just happen to be the stewards of Mythos. So yeah take us through the importance of that moment. What it means for customers because they're in the same boat. >> Yeah. Yeah. It was a watershed moment because it told both the customers as well as the vendors, the security vendors and the infrastructure vendors on [clears throat] how AI will shift the security landscape. Right. The three Vs that I talked about is what Methos Glasswing and other frontier AI models really brought in. Right? The level at which the software the AI models can detect vulnerability in the software right from the kernel all the way to the infrastructure and layer 7 layer is is mind-boggling. Right? So it gave us a fresh view into we as vendors have the obligation and the responsibility and accountability for our customers to really build highquality secure software releases and the way in which we show up the way in which we release these is going to be very very critical for us and and and there are always two sets of customers right one who understand the risk and they're like hey how do I do this that's why the security assessment program that I talked about is really critical they're already brought in. The others are like, "Hey, I'm going to wait and watch and see how real the threat is." And that's going to be that's dangerous. >> That's that's a very dangerous. So, we are truly sitting with our customers to say, >> "Hey, let's run and and find out your risk score, right? Maybe you're secure as you think you are or maybe you'll find some gap and then you'll realize the severity of the situation." So glasswing and and programs like that have really given us some early more advantage as one of the few vendors who participated in the early release of the glass wing to strengthen our own software and change our software development life cycle completely. So that's been very very crucial >> and that's a that's a recent initi recent thing that for you guys and also you can bring better value to the customers but also they're in the same boat so you are aligning with the customer because it's like a opportunity >> to walk the talk and make the product better. >> Ah absolutely >> which we've talked about last year. >> Yeah absolutely. All right, final question. As you know, we've talked many times covering the VCF evolution, watching it go from getting repackaged in the and the initiative with with Chris, yourself, Chris, Paul Turner this year. Talk about the significance of the private AI cloud, specifically the VCF AI factory because I think this is a moment that's almost like a >> a reward slash market opportunity for the work that you guys were doing because there's been a lot of people saying, "Oh, VCF is why is all this change happening?" But there was an intentional reason again we've documented on previous cube videos but this was to be one package less SKUs easier business model but functionally from a technology perspective it should work in the future state. Yeah. Not maybe everything's turned on but now with factory there's demand with AI that's a forcing function. >> Yeah. >> Talk about the importance of this moment for why VCF was designed the way it was >> why this is relevant and what comes next. >> Yeah. In a way it's we we had the crystal ball uh I would say when we built the VCF factory right like what I call the innovate forward things like memory taring that we built in because when we did that two years back the server crisis wasn't there the hardware supply crisis but we knew that as AI yeah it's it's there big time right now but we built in so our customers can use their existing server and even the new servers and get the maximum value and efficiency out of that things that VMware has done for the last 25 pretty much. So the same thing with things like VSAN global DDUP and compression. So packaging all of that into a single private cloud infrastructure was very critical. So that is the journey that we launched into and really accelerated as we became part of Broadcom and that has given us a very solid secure foundation to build an AI factory on and that's why last year all the private AI services became part of VCF right so customers don't need to buy two things because we realize that AI is moving at a very fast rate people are moving from pilot to production a lot of inferencing fine-tuning rag use case are happening on premise so how can we help our customers create that turnkey solution That is the genesis of VCF AI factory which is where we feel next year we're going to sit here and have the same conversation to see how it is truly accelerated deployment of these AI workloads and the good part is you don't need to create another siloed infrastructure right the same infrastructure that you've tried and tested for running your VMs for running your containers can be used running your agents and AI workloads right with the same unified operations management and the security and data privacy so that's what I I think we will work towards and and make sure our customers adopt and deploy these so they can get the value. >> It is a new generation. It's a next level for sure. I mean back in the old days was like get a server and you got to load something. It's got to run an operating system. So you proprietary moves to Linux then you run applications on it you get virtualization. The AI game is a large systems game. So it's not a server it's servers and storage rack scale. You got to run something on it. So people have been asking me, what do I run on my AI factory? So if if it's just hardware, >> you got to run something on it. >> Yeah. >> You want it to be VCF, right? >> Yeah, absolutely. And we want to make sure it's run in a way that provides the cost benefits, >> overcomes the AI tokconomics challenges that our customers have and provide the scale and efficiency, the rack scale efficiency that our customers want when they want to run these large scale distributed systems. So we are pretty excited to unveil this to the market and and work with our >> partner. The AI infrastructure is booming for intelligence tokens. Um you know in every every shift it's ease of use. How fast can I stand it up? >> How do I operate it day one and then what happens after day one? [laughter] >> Yeah this we've seen this movie before. >> Yeah absolutely it's exciting times as you said. >> Pashant great to have you on. Appreciate it. I'm here with Pashant Shoy. He's the VP of product marketing at VMware cloud foundation division unpacking the VCF AI factory. Again, we are in a whole systems game rack scale, distributed computing, private AI cloud, really, really, really big hypers scale activity in the enterprise happening as well. This is the cub's coverage of VMware Explorer. I'm John Furry, the host. Thanks for watching.