VMware Explore 2024 | Krish Prasad | Broadcom & Manuvir Das | NVIDIA

Clips
News
More from VMware Explore 2024

Krish Prasad

Senior Vice President and General Manager, Broadcom's VMware Cloud Foundation Division

Broadcom

Manuvir Das

VP, Enterprise Computing

Nvidia Corp.

The hardware/software combo: Nvidia and VMware partner to power next-gen enterprise AI

The VMware Cloud Foundation Division at Broadcom Inc. and Nvidia Corp. have partnered to make AI more accessible, flexible and powerful than ever before, ensuring both variables expand in commensurate degrees. Nvidia and VCF are doing this by pushing the boundaries of computing through the seamless integration of hardware and software innovation.Unwrapping next-gen AI infrastructure with Nvidia’s Manuvir Das and VMware’s Krish Prasad.“A year ago, these models that are the basis of AI were just not good enough for enterprises to actually adopt in use cases,” said Manuvir Das (pictured, left), vice president of enterprise computing at Nvidia. “That’s fundamentally changed in the last year

play_circle_outline GenAI solutions for enterprise AI adoption

play_circle_outline Partnership between Nvidia and VMware Cloud Foundation Division (VCF at Broadcom)

play_circle_outline Private AI for data privacy and AI processing proximity

play_circle_outline Revolutionizing Platform Networking: The Future of Data Engineering Resiliency

play_circle_outline Collaboration between Nvidia and VMware for AI democratization

Info
Transcript

Krish Prasad | Broadcom & Manuvir Das | NVIDIA

Krish Prasad

Senior Vice President and General Manager, Broadcom's VMware Cloud Foundation Division Broadcom

Manuvir Das

VP, Enterprise Computing Nvidia Corp.

Coverage of VMware Explore 2024 in Las Vegas continues with discussions on GenAI by Krish Prasad from VMware Cloud Foundation Division at Broadcom and Manuvir Das from Nvidia. They emphasize the enterprise adoption of AI and the importance of a platform like VMware for running AI applications alongside others. Specific use cases like intelligent data extraction and healthcare applications are highlighted. Companies like Amazon are using AI agents for future planning. Networking and resiliency are crucial for the success of GenAI. The partnership between Nvidi... Read more

explore Keep Exploring

What are some key milestones and accomplishments in the past year for the GenAI solution? add

What pieces are provided for developers to write their apps on top of OEM infrastructure with GPUs and VCF as the operating system? add

What does adoption of private and private AI look like? add

What kind of work is being done in order for GenAI to fly the hard way? add

What are the goals of the partnership between VCF, VMware, and Nvidia in relation to democratizing AI for enterprise companies? add

bolt Powered by CUBE AI

Krish Prasad | Broadcom & Manuvir Das | NVIDIA

search

>> Welcome back, everyone. theCUBE's live coverage here in Las Vegas for VMware Explore 2024. I'm John Furrier, host of theCUBE with my cohost, Dave Vellante. Our 15th year covering this event, seeing the transformation of VMware, the technology, and ultimately the customers. We got two great guests here to break it down. Krish Prasad, Senior Vice President, General Manager of the VMware Cloud Foundation Division, also known as VCF at Broadcom. Manuvir Das, VP of Enterprise Compute at Nvidia. Manuvir, good to see you again. Welcome back to theCUBE, both CUBE alumnis.

Manuvir Das

>> Thank you.

Dave Vellante

>> Good to see you, guys.

Krish Prasad

>> Thank you. Great to be back.

>> Enterprise Compute at Nvidia. Okay, you're the man of the hour. Everyone wants to Nvidia. Can we get some GPUs for theCUBE?

Krish Prasad

>> They're right down here.

>> Only kidding. I'm sure you get that question all the time. Obviously the success, and congratulations on all the great stuff. Love GTC this year continue to thundering away on the momentum. The customer base, the market is evolving, products and transitions as well. All the platforms are there. You guys were on stage, Jensen on stage last year with the partnership with Broadcom. Now VMware bought Broadcom. Give us an update. How's that going? Because that's been a key partnership, and with VCF kind of hitting its stride, Krish.

Krish Prasad

>> Yep. So last year, Ragu and Jensen were on the stage together, and we announced the new solution for GenAI, which is a well-integrated stack, which has Nvidia software in it, the virtualization of the GPU all put together. And at that time it was a conceptual announcement. "Hey, we are working on it together." And now, what, a year later, we have already released it. It is out in the market. Customers are using it, and it's one of the hottest areas that we have in our portfolio.

>> Manuvir, what are some of the key milestones that you can share over the past year? What are some of the accomplishments this past year?

Manuvir Das

>> Yeah, I think if I may put it this way, the world has changed in the last year for enterprise companies, which is the common focus for us. Because a year ago, these models that are the basis of AI were just not good enough for enterprises to actually adopt in use cases. OpenAI had gone a long way, but the ecosystem of models from Meta and Mistral, et cetera, were just not there. And that's fundamentally changed in the last year. These models have become really, really good, like if you look at Llama 3.1. And so the milestone you're talking about, what we realized was the point of inflection is arriving where enterprise companies can deploy GenAI for themselves. And so now the challenge is: how are they going to do that? What are they going to put in their building to make that happen? And so we've come to the point where enterprise IT needs to deploy a new platform. They deployed database, they deployed virtualization, they made the move to the cloud. Now the next platform is: how do I get my own GenAI infrastructure for my company so that all the people building use cases across my company, building chatbots across my company have this one platform that they can depend on? And that was actually why we did the work together because we said that platform is you pick your OEM infrastructure with GPUs in it, and then you use VCF as the operating system that's on top of that infrastructure. And between us, we've given you all the pieces that you can just, as a developer, write your app, and everything else is provided for you.

Dave Vellante

>> Well, in the past, you do that on top of-

Krish Prasad

>> Yeah, I was going to say, the interesting thing is we started this partnership around five years ago, and at that time GenAI was not in the picture. But we were saying, "Hey, we want to bring AI to the Enterprise." And that was the vision. And both of us got our engineering teams together and they have been working very hard to really virtualize the GPU, the software stack that Nvidia produced. And we integrated all of that and that was the announcement we made last year.

>> If you look at the success of CUDA and what's happening with the hardware, the price performance of the models, like Lamy mentioned, tokens of prices are dropping, throughputs going through the roof, more horsepower, democratized, supercomputing for the masses. You guys were years ahead. What's that mean for the customers? How did that translate into the partnership specifically?

Manuvir Das

>> Yeah, I would say, so they now have choice, because when GenAI began really, the only way to really experience it was somebody's got a canned service in the cloud and I'm going to use that. OpenAI started this whole trend, and they've been brilliant in what they've done. And they still are brilliant, and they're expanding their capabilities all the time. But I think customers want choice. For some people, it's okay to use somebody else's service. For other customers, I need to have my own sort of private way of doing my AI. Whether I'm on-prem or on the cloud, it doesn't matter. And so the solution we've built allows a customer to say, "I'll choose my environment." Maybe I'm an AWS shop, maybe I'm an Azure shop, maybe I'm an HPE on-prem shop, I'm a Dell on-prem shop, it doesn't matter what choice you make, we give you a way to build your GenAI.

Dave Vellante

>> So, what does that adoption look like, Krish? Because most companies just haven't been able to get their hands on these GPUs, so they've had to go to the cloud or these specialty clouds. It's starting to free up. We hear the earnings calls and the timelines. We'll hear more tonight actually. But what's it like on-prem? Customers clearly, the ones we talk to, want to do AI on-prem. What's that adoption look like?

Krish Prasad

>> Yeah, so the private and private AI really is about not on-prem, it's really about keeping the data private and bringing the AI processing to the data. And so VCF is an ideal solution because it not only runs on-prem, but it also runs in the hyper-scalers. It can run in other service providers. So, you get a ubiquitous platform on which you can run your AI processing. That's number one. We just released the solution roughly two months ago, and we have hundreds of customers who are now working with us. And we were right here with a big customer, who is now deploying this and wants to work with us. So, lot of traction. It's one of the hottest new products that we have introduced .

Dave Vellante

>> Definitely because the data's. There's lots happening in the cloud of course, but there's a lot of data on-prem.

Krish Prasad

>> Yes.

Manuvir Das

>> Yeah, and I think the other thing I would say, Dave, in terms of the adoption, we have to understand that yes, if you're doing a massive training job to train a big model, you need thousands and thousands of GPUs, and you can ask do, "Where do I get them?" But enterprise AI, a lot of it is about just fine-tuning, a little bit of model improvement and inference where you're running a model. And a lot of those models, they fit on a single GPU, they run really well on an OEM server with one GPU in the box. And obviously there's a lot of those out there, so there's a lot of democratization.

Dave Vellante

>> Well, what about that? Because inference is a big part of your strategy. Jensen on the earnings call, you guys announced already. Obviously we've been doing CUBE, so we haven't seen the results, but last earnings call, he said, I think, 40% of the enterprise workloads was inference, so you're doing a lot of inference obviously.

Manuvir Das

>> Yes. We're doing a lot of inference with GPUs, yes.

Dave Vellante

>> But there's a narrative out there that says, "Oh, Nvidia can't do inference." Obviously that's not the case.

Manuvir Das

>> That's not the case at all.

Dave Vellante

>> Help us square that circle. Why is that narrative out there and what's the real story?

Manuvir Das

>> Well, I think when AI use cases began long before GenAI, the models were actually very small, and so the natural thing was we'll do the training on the GPUs and we'll do the inference on CPUs. But especially as GenAI has evolved, these models are much bigger. And they're doing a lot more processing, and the latency also matters, so that shift has happened. And you have to run these models on a GPU now. The question is only: how many GPUs do you need to run the model? Some of these models, if you can look at Llama 3.1 with their 405B, you need 16 GPUs to run the thing. Other models you can fit on one GPU. So, I think yes, the world has completely changed. I think GPU is the place to do the inference, but all our partners, whether it's Dell or HPE or all of them, Lenovo, they built great servers now that are very accessible to any enterprise company to do the inference on.

>> One of the things that we put out a year and a half ago, because we have been playing it with all of the vector embeds with our data, was we saw the results, we're like, "Wow, this is..." And it's not real AI, it's just RAG. RAG is a gradient of innovation. That comes from the benefits from other people. But when we started looking at the models, we saw that the small models could be very valuable. Like ours is a small model, so now everyone talks about that now. But at that time we saw a power law that looked like a straight curve down, and then like a flat tail, but now you're seeing the specialty models in the enterprise fill in. So, the power law still got the big models like the Llamas and the OpenAIs, but now as you go down into the power law, you start to see mid-range models on the torso, the tail's got domain expertise, and then it's digital, so it's neural, so they talk to each other. So, this is something that we're seeing, you guys got the NIMS, the model, the stores. How do you explain that to people that are used to the old world, or the old world, the world a couple years ago? Because they got to start writing software, build mechanisms, and set the table for agentic applications.

Manuvir Das

>> Exactly.

>> How do you explain that?

Manuvir Das

>> Okay, so I'll put it this way. The way we think about AI at Nvidia is that it's an ongoing process. We call it the data flywheel. So, what happens is traditional applications have been static things. A software engineering team builds an app, and then it's out there and you use it. The way these AI apps actually work is I pick a model, I train it, I feed it some of my company's data. Now I build an app with it. Once I start using that AI app, it's a RAG. It's not going to be perfect, right? Sometimes it's right and sometimes it's wrong. And as it interacts with the humans, judgments are being made about whether it was right or wrong. And that is very useful information, so we call that reinforcement learning. And so as the app runs, you take that data from the app running and you feed it back to further tune the model. That makes the model better, so now tomorrow your app has become better because the model inside is better. That means more people are going to be successful using it. You're going to generate more data that's going to feed back to the fine-tuning. So, it's this continuous data flywheel that you're actually implementing. So, how do you do this? So, the first thing we did at Nvidia was we built all this software to actually tune the models, produce the models, then we did these NIMs. What are NIMs? NIMs are basically saying, "You know what? As a developer, it's going to be very difficult for you to keep up with the state of the art because models are changing all the time." We put these models inside this thing called a NIM, and all you do is deploy the NIM, and automatically every day the model that's inside the NIM is just going to become better for you. And you don't even care.

>> It's an inference microservice, basically.

Manuvir Das

>> It's inference microservice.

Dave Vellante

>> So, you talked about reinforcement learning. A lot of that is human-based, but now you're talking about machine-based. And by the way, I have to apologize to the audience. Today's Tuesday, Nvidia reports tomorrow. That's why you-

>> It's only up a point, that's why.

Dave Vellante

>> That's why you looked at me funny. I apologize for that. I thought it was Wednesday. But anyway, the reinforcement learning was very human intensive. Now you're talking about more machine intensive.

Manuvir Das

>> There's so much data being generated. And one app that's running is generating data that is helpful to tune another model for another app. It's all about that data and where it's coming from. So, we had NIMS, as you said, a great way for developers to get all the improvements, stay state of the art without having to worry about it. And then the final piece of the puzzle that we actually just announced today from Nvidia and VMware is one of our launch partner for this is what we call NIM agent blueprints. Because it's like: how do I in an enterprise actually start building my use case, whether it's a customer service use case or it's a document retrieval use case? So, given all the work we've done with various companies, our customers, we've built this catalog of apps, if you will, that are ready-made use cases. That's what the blueprints are, and we've made them available. We just announced that today. And as a customer, for example, when you adopt VCF, it's ready to go. It's ready to run those blueprints. And people can improve them, and adapt them. And John, you mentioned agentic. That's really the next step. Because as an enterprise, how do I get value out of GenAI? That's the question people ask, "How do I get value?" We see two levels of value. One is there's these generic use cases. "Okay, I've got PDF documents, can I ask questions? I've got some customer service scenario. Can I look up my records and say something?" But the next wave is really agentic, which is do things. Look this up, make a decision like this, go to another system and tell it to do that. That's how we work when we do work. That's the agentic workflows. That's what we are focused on right now. And we've produced blueprints to help people.

>> The DUC has a new category, because this would fall under Jensen's new category rule, in my mind. Rule... It's what he sees. He's said publicly, "GenAI is a new category."

Manuvir Das

>> Yeah, I think you think of it this way-

>> It's a new workflow.

Manuvir Das

>> The first phase of GenAI was basically information retrieval, a better way of doing search.

Dave Vellante

>> Request and retrieve.

Manuvir Das

>> Request and retrieve. A better way of doing search, use vector databases and all of that. The next phase is: help humans do their work, which involves actual actions. And that's where the agentic stuff .

Dave Vellante

>> So, we say large language models become large action models, and it's multiple agents. A lot of people are talking about agents, co-pilot, one agent, but we envision multiple agents working together, having role-based access as a human would, understanding big picture, understanding corporate objectives in a tree, might be OKRs or grow revenue or gain market share or increase profitability. And the agents working together work toward that objective. That's a whole new application stack.

Manuvir Das

>> That's right.

Dave Vellante

>> So, what makes VMware the platform to build those intelligent data applications? Give us the bumper sticker for that.

>> Don't give bumper... Go into detail. What we want to know.

Krish Prasad

>> What has happened is that AI has now moved from the research labs to mainstream in the enterprise. And over the past four years, we both have been really working with customers to make AI real in the enterprise. And what we are seeing now is it's actually happening with this use case. And so what customers want to do is not have a siloed separate environment to run AI. They want the same environment they use for other applications with all the controls and governance that they put on the other applications to apply to AI also. And that's where VMware comes in, because we are the infrastructure on which they run all their applications. And now they can run AI on top of our infrastructure. And with the joint work that we have done, we have made sure that the performance of AI processing is at least the same or in many cases better than bare metal running on our stack.

>> Well, I can say-

Dave Vellante

>> We saw that data today, and that's when I said, "Software mainframe," because that's what the mainframe did, it ran virtualization with no overhead. It was unbelievable magic. I know you don't like that .

>> I could just say, look, I've been in Palo Alto 25 years when I moved from the east coast. I think this relationship is really the magic of Silicon Valley because you see things early, getting in early, when the engineers are talking to engineers, like, "Hey man, that's cool. Software, hardware, software, operating leverage." And then that kicks off and then other entrepreneurs come in, then that's global. Now we're at the point of hitting the street mainstream. So VCF stands up a cloud, private cloud, I get that. You guys are used to playing the clouds with Nvidia and software and all the great hardware. How does a customer deal with this? Because is it DGX on VCF? You got the cloud. How does it work? Am I going to run my apps on VCF? What does it all work-

Manuvir Das

>> I think there's three layers of the stack. There's the hardware platform. There could be different OEM systems, et cetera, as long as the GPU's in there. There's sort of like this operating system layer, which is what comes from VCF, VMware.

Krish Prasad

>> VCF.

Manuvir Das

>> And then there's the apps on top. And I think it's not about a single hardware. I think VCF spans many different hardware systems from various vendors, and they all have architectures with GPUs in them now, so they're all supported. And so it's not a statement about one particular platform. And I think to your point about all this innovation, I'll put it this way: why is the iPhone an amazing thing? Because there's one platform called iOS, and every app builder with an idea is building on the same thing. And so as a customer, I have access to all of these apps. I think GenAI is going through the same thing in the enterprise where there's a multitude of use cases. John, you and I sitting here cannot envision every interesting use case, but the millions of developers out there can, and so we are just trying to provide them the consistent platform that they can all build their apps .

>> The scale and the horsepower and the supercomputing is there. Dave coined a term called Jensen's law. Spend more, save more. He says that on the keynotes.

Manuvir Das

>> The more you buy, the more you save.

>> The more you buy, the more you save. But if you think about it, what he's referring to is you got to invest in this infrastructure to get the ROI of the category. There's no other waylay way or mechanism because it's old. This doesn't cross that... There's no bridge to the other side. You got to build the bridge.

Manuvir Das

>> We have to understand there's a bandwagon, and you need to get on the right bandwagon because Moore's law is dead. So, if you are on the CPU bandwagon, then the curve is looking like that, but the needs of the application of the AI application are going like that. So, you need to be on the GPU bandwagon so you continually get the horsepower and the throughput that you need.

Dave Vellante

>> And that's a controversial statement that you just made, but the data plays it out. When you look at the performance that Nvidia has created over the last 10 years, it's 1000X increase in eight years, versus Moore's law is about 100X in 10 years. So, the data supports what you just said.

Manuvir Das

>> To be fair, Dave, the reason we say that is twofold. One, at the hardware level, of course the hardware is becoming much, much better, but there's a whole industry and ecosystem of people building software, and that journey is so early that there's many orders of magnitude of performance improvement still available by the software innovation. And so both are multiplying together.

>> And the cloud guys proved that on gen one SaaS. Guy with a credit card doesn't want to provision a data center or go to buy a box and host it in a rack, just goes to the cloud. Now you get scale. The problems being solved right now with GenAI are problems that are hard. Entrepreneurs will solve those problems, and you couldn't do them years ago.

Manuvir Das

>> Exactly.

>> But that's a whole mind shift.

Dave Vellante

>> But I do think the agentics conversation makes the AI or the application stack more complicated than what you described.

Manuvir Das

>> Yes.

Dave Vellante

>> It's a different... Maybe it's this three tier, but that application piece that you mentioned-

Manuvir Das

>> Yeah, let's talk about that.

Dave Vellante

>> Okay, so I'm going to put this forth. In addition, there's a federated data management. You've got to connect to the legacy data that's hidden inside all the applications. Remember, the applications contain all the data, all the business logic, and it's locked inside of that. That has to be unlocked with a federated data management system. And above that, there has to be a way to harmonize all that data, so that when I say revenue, it means revenue, not bookings or not ARR. There's a common language. And then there's a whole agent layer, orchestration agent, management of those agents, that can understand and interpret those top-down goals. So, there's a bottom-up from the connectors in the backend data systems that live in legacy apps, and there's a top-down objectives of the organization. That's a profound change in the application stack.

Manuvir Das

>> That is, Dave. So I'll tell you a view in it. And in fact, at GTC, Jensen had a few slides where he laid this out. The thing is, as you said, the data inside every application is understood only by that application. So, this problem is not to be solved by saying, "Let's suck out the data from every application and build one giant data warehouse." Because that data warehouse is not going to understand all the data.

Dave Vellante

>> It's a bottleneck.

Manuvir Das

>> But what's happening is every one of these platforms is building its own interface, like ServiceNow has Now Assist. There's a chatbot, if you will, there's a RAG in each of these applications.

Dave Vellante

>> Salesforce with the Data Cloud.

Manuvir Das

>> Exactly. And so now what the application developer does is call into all these things. And that's the code I'm writing now. I'm orchestrating... I'm pulling into ServiceNow here, but I don't need to understand the code of ServiceNow because ServiceNow has given me a RAG that I can talk to, and Salesforce has given me a RAG that I can talk to. And then I've got some custom databases of my own. And I ask Salesforce questions through the chatbot. I get some answers. I vectorize it, and put it into my vector database. I do the same with ServiceNow, and now I run my LLM across that consolidated data.

Dave Vellante

>> It's an operating system. It's an operating model.

>> But you guys are doing your job to provide the infrastructure and the capabilities, the system capabilities, that you're bringing-

Manuvir Das

>> That's why I talked about the blueprints that we announced today. For a variety of use cases, we have sort of pre-done that job, for .

Dave Vellante

>> The others have to invent that federated data management, the harmonization layer, the agent orchestration, which maybe you can do because it's a workload.

Manuvir Das

>> You go use case by use case, Dave. So, if I can tell you about... The way we did this, we focused on three use cases that we launched today and we have many more coming, but I want to share these with you, so you can understand. The first use case, which is like the broadest is you've got PDFs, you've got charts, images and all of that and that. And so we built this whole workflow for: how do you intelligently extract information out of these charts and images, et cetera, and talk to your PDFs? That's something that every enterprise company would want to do. So, that's the broadest scope, if you will. Then the next thing we did was for customer service, digital humans, how do you interact with an avatar that behind it has a RAG and is doing information retrieval and all of that? So, we put that one out there. And the third one was very vertical for healthcare, for drug discovery. Because what we see happening is in every company there'll be some adoption and building of use cases that are fairly generic, but then you got to understand what business you are in. And you have to build the use cases for your business to get the most value, to actually make your business be better. And so that's kind of what we see.

Dave Vellante

>> Well, and you see that there are some... I know we're out of time here probably, but you see some examples. For instance, Amazon.com, the retail business, not AWS, they have agents that they've had to build purpose-built that can do planning like for five years in advance what SKUs are going to be needed, what warehouse capacity is going to be needed. And they have multiple agents that talk to each other, very complicated, but they had to build that whole infrastructure stack on their own. What you guys are doing is making that... They're democratizing that, making that horizontal for any customer stack.

Manuvir Das

>> And we have the benefit that five years ago you needed humans to write 1,000,000 lines of code, and now most of that code is just in the .

Dave Vellante

>> An army of humans, yes.

Manuvir Das

>> And you write a little bit of code around it.

Dave Vellante

>> Low code as well.

>> Well, that was a master class in helping us figure out how our bot's going to be interfaced with other agents.

Dave Vellante

>> That's awesome. Thanks, you guys.

>> Let's wrap this up, and just with the future. DGX Cloud on VCF, VGPUs. GPUs are now in the workflow equation. Obviously you're managing workloads for enterprises. What's next, guys? What's in the horizon? Where's the relationship go next? Yeah.

Krish Prasad

>> Yeah, , then I'll add. Go ahead.

Manuvir Das

>> I think in order for GenAI to fly the hard way, the networking is evolving dramatically. Jensen talked about some things like Spectrum-X, et cetera. It's a full data center scale problem, how everything works together. And the work we are continually doing together is every time we make a click of innovation at that layer, the operating system sitting on top in VCF, vSphere, all of that, has to understand how all of that works. It has to abstract it on the one hand, but take advantage of its capabilities fully, like with DPUs, for example, to accelerate the network. And so that's the massive engineering work we are doing together, click by click.

Krish Prasad

>> And I would just add that resiliency is one of the key things that a platform like ours provides. And that means high availability of the GPUs. If a GPU goes down, moving the workload to some other one that is working.

Dave Vellante

>> So, you can checkpoint that, for example.

Krish Prasad

>> Yeah, checkpoint that and move it to some other GPU. So, all of that is the area of focus for us, so that the same kind of resiliency you get with the CPU you're getting with the GPU going forward.

>> And so you like this relationship, spend more, save more.

Krish Prasad

>> Oh, definitely.

>> No, it's an ROI game.

Manuvir Das

>> It is.

Krish Prasad

>> It's two data points. The NVAIE software stack that Nvidia has, it was first announced on VCF and VMware, and today all the new announcements that they have now are on top of VCF again. So, we have a very close partnership. We are the first out of the gate with Nvidia.

Manuvir Das

>> But the goal is simple. We have to democratize AI for enterprise companies, so that they can all do it.

Dave Vellante

>> Yeah, absolutely.

Manuvir Das

>> And they need a platform to do their work on. Otherwise it's too much scale.

>> All the hardware's getting faster, smaller, cheaper, better, next level. This is the future: platform engineering, data engineering. This engineering mindset, the systems revolution is upon us. I feel like we're in the nineties again.

Manuvir Das

>> Yeah. Absolutely.

>> I'm getting old.

Manuvir Das

>> sets of people. The data scientists have been in this party for a while, having fun doing AI, but there's now two new constituencies. For the IT teams, guess what, you have a new platform to deploy in your company, the GenAI platform with the right hardware and software. And for the developers who've written apps in enterprises, you can join the party. There's a new kind of app to build, which is these RAGs, these agentic things.

>> You got to get on the right side of history. And again, you guys are a great use case. Get in early, understand the software impact, leverage the infrastructure and the hardware, and build the next system. Guys, thanks so much. Manuvir, great to see you again. Krish, as always.

Manuvir Das

>> It was a pleasure.

>> VCF looks like it's going to be packed full of the one-two punch, Nvidia.

Krish Prasad

>> Yeah, absolutely.

>> Thanks guys. Appreciate it.

Krish Prasad

>> Thank you.

>> Okay. It's CUBE coverage here. We'll go whatever it takes to get the story and share that data here. Soon it'll be digitized. Thanks for watching.