NVIDIA GTC 2025 | Haseeb Budhani, Rafay Systems

Clips
More from NVIDIA GTC 2025

Haseeb Budhani

Co-founder & CEO

Rafay Systems

play_circle_outline AI systems and AI factories adoption in the industry, focusing on GPUs and software.

play_circle_outline Collaboration with Nvidia to enable self-service consumption of GPUs and AI applications.

play_circle_outline Accenture collaboration for sovereign AI clouds, addressing orchestration layer gap.

play_circle_outline Urgency in utilizing infrastructure investments and delivering applications to customers

play_circle_outline Tech-Savvy C-Suite Leaders Drive Business Transformation at Nvidia's GTC Event Through Expansion into US Market

Info
Transcript

Haseeb Budhani, Rafay Systems

Haseeb Budhani

Co-founder & CEO Rafay Systems

Exploring AI Systems and Infrastructure at NVIDIA GTC 25 with Industry Expert Haseeb Budhani

Haseeb Budhani, the Chief Executive Officer and co-founder of Rafay Systems, joins theCUBE Research for an insightful discussion ahead of NVIDIA's GTC 25 conference. Budhani, a respected figure in platform engineering and artificial intelligence infrastructure, delves into the advancements and innovations driving the next wave of AI systems and architecture.

In this engaging conversation, John Furrier from theCUBE introduces Budhani, a seasoned expert in... Read more

explore Keep Exploring

What is the industry currently focused on in terms of supercomputing, chips, AI systems, and the adoption of GPUs with new software? add

What is the news with Nvidia in relation to enabling self-service consumption of GPUs and AI applications in enterprises? add

What kind of collaboration agreement did we sign with Accenture and what specifically are we working on together? add

What are the two pressures on the buyer in this situation? add

What are the expectations for Nvidia at GTC this year? add

bolt Powered by CUBE AI

Haseeb Budhani, Rafay Systems

search

>> Hello, welcome to theCUBE here in the Palo Alto Studios. I'm John Furrier, host of theCUBE. We've got a special CUBE conversation. We've got some news going into next week's GTC, which is NVIDIA's conference. It's kind of an industry show. It's kind of like ADOS reinvent for supercomputing and chips, and a lot of action's going on. But what` the real conversation will be around is AI systems, AI factories, and how the industry is adopting the GPUs and putting in the new software that run on all these things. We've got some great news here. Hasseb Budhani is here, the CEO and Co-founder of Rafay, friend of theCUBE, been on theCUBE, CUBE alumni. Back in the years, and going back to the CNC of KubeCon. Hasseb, great to see you. And such a lot going on around your company. The news you guys have with Nvidia is super special, because we've been covering AI factories, and asking the question, what runs on it? Which is, everyone's talking about that as the core next step. You guys got some news. Thanks for coming on.

Haseeb Budhani

>> John, thanks for having me. I really appreciate it.

>> So one of the things Dave Vellante and I have been talking a lot about, certainly with companies like Dell, Dell Technologies, and Nvidia, and others, is that, okay, we love AI infrastructure, we love Cloud Native. That's where we first met, KubeCon CNCF. But as platform engineering solidifies around Kubernetes, and AI factories, which is a re-architecture of the data center, and you've seen the cloud doing the same thing, you know the role of semiconductors, it's changing. We all see that. Nvidia stock price is leading the market. They replaced Intel. Everyone knows what's going on there. And it's going as fast as it can. But it's ushering in this next wave, which is the software that runs on it, and then that's going to open up the data agent wave. So you're seeing the progression. You guys have made great progress. I mean, you've been grinding in the Kubernetes area, building a great software stack. That's now matured. Okay? That's solid. And now that's a lead in to what's going to sit on, and orchestrate, and manage all the services that are going to run on these AI clusters. This is something that you guys have been working on at Rafay hard, and you got some serious momentum, and you got some new news.

>> That's right. - Give us a quick overview.

>> You got in a relationship with Nvidia

Haseeb Budhani

>> and you got a special new thing with Accenture. Let's unpack that. So first, let's start with the news with Nvidia.

Haseeb Budhani

>> Absolutely. So the news with Nvidia is that we worked with them to get a reference architecture published, that talks about how enterprises and GPU clouds can enable self-service consumption, of not just the GPUs that they have available, but also the AI applications that run on them. So as an enterprise, if you want to use, for example, NVIDIA's NIM ,or DataRobot, or any other agentic application, your developers want to be able to go press a button and start using it, like a classic CSP experience. How do you do that? How do you go from a bunch of GPUs, to an actual GPU cloud? And the software that we're building forever, and you know that the big thing with Rafay for many years now has been, how do you enable self- service consumption of compute? The same premise that applied to the CPU world applies here, but in a bigger way, because GPUs are really expensive, and they're not that many of them. So making sure that the utilization is high, making sure there's multi-tenancy in place, making sure that people can get them on demand, is a really important requirement in the industry right now. And we happen to be in the right place at the right time, with a great stack that sits between the hardware and the applications. Like back in the day, we used to call it middleware, right? But this is, the Nvidia team uses the phrase platform as a service, which is why the reference architecture is called the GPU PaaS, the reference architecture for Nvidia hardware.

>> Yeah. And Nvidia is so hardcore too, as you know, they just, it's an enablement, it's acceleration, accelerated computing. I love the middleware term, because I was just joking with someone online, this morning actually, around the term middleware, because we got, someone was calling me up, "Hey, you and Dave used the word middleware. " Now we use it tongue in cheek, because we kind of use the old school definition of hardware, and middleware, and apps. But again, it's a paradigm. It's not necessarily... But you can apply the middleware concept to hardware. Okay? But it's being sliced. There's layers of middleware now. It's a little bit more complex, because the underlying infrastructure isn't a box anymore. It's boxes, plural, and other interconnect components. So these AI factories, which we love, by the way, we think that's the way to go, you're seeing Amazon, the hyperscalers already doing it, is the system architecture. So that's where all the work's been done. Now, what's interesting is, I don't believe in luck, by the way. So I think you guys have done the right thing with Kubernetes, because the platform engineering that was being done in the enterprise, that's going to run the workloads. That's a developer layer, okay, that was being worked on. But now the new environment and hardware comes out, so you guys were smart to move over there. Talk about that dynamic, because I think this is a nuanced point in the industry, but certainly, it's not a shift, it's just an extension of what platform engineering's intersecting into. So your world, you are riding down that wave of Kubernetes, into the perfect storm of innovation. Because what came out over here was the AI infrastructure.

Haseeb Budhani

>> Indeed. - And it still needs platform engineering.

>> Cloud-native workloads are going to run on that.

>> Indeed. So this is the fascinating part that,

Haseeb Budhani

>> I'll give you examples, right? So we closed a deal with a large GPU cloud in Indonesia. And the customer there, the users there, they're platform engineers. They're really, really sharp guys, actually. They've built some incredible stuff. But they had the clarity that, for them to deliver self- service consumption, it will take 18 to 24 months, beyond where they were today. So when they use our software, they're up and running in three weeks. So we help them just shave that time to market significantly. Because the problem is, these GPUs, they have a shelf life, right? They don't live forever. So if the shelf life is, let's call it three years, and it's going to take you two years to build the software, that's a problem.

>> Yeah, I got some H100s today, I want Blackwell tomorrow.

>> Exactly. - But again, they're in a cadence of,

>> I won't say Moore's law because it's a little accelerated,

Haseeb Budhani

>> kind of Moore's law, but every new version builds on the other. So it's an acceleration of performance.

Haseeb Budhani

>> Correct. - That's their job. The number one question

>> that comes up, and I want to get your reaction to this, and talk about how this relates to your news with Nvidia, is the number one problem enterprises have right now that we're talking to is, I'm going to the AI clusters, I'm going to the AI factories, and what do I run on it? So it's hardware. At the end of the day, it's system hardware that has to run stuff. How did you guys see this? What was the opportunity recognition? Was it something that just you ran right into? Or is it something you saw? And how did you react with the company? Because I think this is, obviously, makes a lot of sense now looking back, but it wasn't obvious to the world what you were doing a year ago.

>> Indeed. When I told the board that, "Look,

Haseeb Budhani

>> there's something here, but I don't know exactly what it is. And allow me the breathing room to go figure that out, with a handful of team members," it worked out. But it may not have. Right? So two things happened, for us to say, "There's something here. " One is that we have a large customer who's a hedge fund. And their board asked them to look into AI, so they went and bought a bunch of Dell hardware, and that sat in a corner because they don't know what to do next. I mean, they're smart guys, they just, they're like-

>> Yeah.

>> And Dell sells great servers, and they dropped a bunch of servers on the dock, so to speak. Maybe not a dock, but shipped it to them.

Haseeb Budhani

>> Yes. And then we're like, okay, now where do we go from here? So they spent a lot of time looking into that. And they asked us the question, "Do you have something that could help us? " And look, fundamentally, it's Kubernetes. Sure. But we found that there was a gap. And the gap was that Kubernetes is great, but there was no platform as a service layer on top, so people couldn't consume it like the way a cloud would deliver services. So we worked with this customer and they really helped us think through the gaps. And then, separately, we found this great partner, actually through Nvidia, in Indonesia. That was our first real GPU cloud customer. And they helped us really think through the gaps as well. I think we found, we got luck. No, you don't believe in luck, but something happened where we got customer-

>> The right place at right time, that's good luck.

Haseeb Budhani

>> And a cloud customer, right?

>> So then we were able to see a pattern. And then the question was, are there more of this? So we started asking our customers, "Are you buying hardware? And what are you going to do with it?" It turns out many of them are. And then separately, the clouds out there, we found a number of them, data center companies of old, or telcos of old, who were now investing in GPUs. And they were trying to launch a regional cloud service or a sovereign AI cloud service. Many of them here, many of them here. And all we had to do is extend the platform with the past experience to give them the service they were looking for. And that led us to the other thing, which is the extension thing we'll talk about momentarily, but it's actually quite amazing how quickly this market is moving in this direction.

>> Yeah. And that's, the demand for GPUs fuels innovation.

>> People go, "Okay, great, I like it. I want to do it on-prem for my AI, my intellectual property is. " Okay, we'll come back to that. I want to come back and unpack a little bit further in some of the market dynamics as GTC is coming. So great. Congratulations on the reference architects with Nvidia. Second news is the Accenture relationship. This is a different kind, but it builds on what you're doing here, for more people on the streets deploying, because there's a deployment wave coming.

Haseeb Budhani

>> Yes, sir. - Talk about the Accenture deal

>> that you guys just announced.

Haseeb Budhani

>> Yeah, absolutely. So we signed a collaboration agreement with Accenture, to globally work on sovereign AI clouds together, where we go into these accounts together. Accenture has an incredibly well-thought through practice for these sovereign AI clouds, starting from consulting, all the way to operationalization. And they also had identified a gap in their stack, which was the orchestration layer. So they call it orchestration, and different people call it different things. Middleware has orchestration-

>> It's software.

Haseeb Budhani

>> Right, software. - Middleware.

>> Yeah. So they had a gap

Haseeb Budhani

>> that their technologists recognized.

>> They spent time with us, learning about the platform. And they moved really fast. It was quite shocking. I think in four or five weeks they'd made it, come to the conclusion, we got to work together. And now we're basically working on deals together. So that solves a lot of problems for their customers and also gives us the scale.

>> Yeah. And you got the professional services layer. So Accenture looks at this as, okay, I'm now deploying, so that's what they do. And they put it together. And sovereign cloud, just from a definition standpoint, that's what, that's in-country clouds?

Haseeb Budhani

>> Correct. So the top priority for Accenture seems to be the sovereign cloud. So in my mind, there are three classes of customers. So there's a sovereign cloud, which is sort of a large cloud in a certain country, that is thinking about the country's best interests, promoting AI in their country, maybe the local language, LLM, et cetera. Then there are these private data center companies, who are now building GPU clouds, and providing them at a lower cost than the CSPs. And then there are enterprises. So there's three classes of these customers. There's probably, I don't know, hundreds of the sovereign clouds, maybe. But there could be a thousand of these GPU clouds, and there's tens of thousands of these enterprises who are buying hardware. So the market opportunity is very, very large. And certainly, organizations like Accenture, because they have such scale, they see it before some of us see it. They saw it well before saw it. And they saw the gap and they said, "All right, this fits in."

>> It's funny you mentioned about the naming. Some people call it neo clouds, sovereign clouds-

>> Neo clouds, yeah. - ...

>> private clouds. So it's all kind of the same thing.

Haseeb Budhani

>> It's essentially a resource center where there's GPUs and systems there, empowering it, software on top. So clearly, software is now coming into the fold on these cluster large scale systems, powering where the demand is, which is application. So congratulations on Nvidia and Accenture. Got distribution with Accenture, people in the streets. You've got Nvidia, certainly a great technology enabling. Dell technologies, instead of shipping boxes sitting there, they can sit there and get them up and running faster, so they can get time to value. So I see the logic, and it's a great success formula. The question is, is that, what's it do for your business? You got a little latitude to go see what was there, you poke at it, you do some entrepreneuring, you put it together. What's the impact to the pipeline? Your business? Can you share any momentum data around what this turning into? Because it looks like it's going to accelerate, that should yield faster deployments, because we're in that phase now, it's crossing over to, I got my systems, POCs have been running, let's start moving stuff into production. What's been the pipeline impact?

>> So I'll give you one specific example, that's...

Haseeb Budhani

>> I'll give you two, actually. So we closed the deal in Turkey with a partner who I met over Thanksgiving. And we signed the deal on the 31st of January, two and a half months later. Zero to let's go live. Closed the deal in India, met them late October, closed it on December 31st. All right, these things are moving super fast. So when they see it, they go, "All right, I got got to do this now. " And this has definitely changed a lot of things for us in the company. Like, you know us very well, you know that my style is, let's run a small sales organization, let's go slowly. That's been my style forever. But now I'm like, man, I wish I had like a hundred more people.

>> Yeah, you got to go fast. - Like sales and marketing.

Haseeb Budhani

>> Well, the market flipped, it spun to you doorstep.

>> And that's what I was trying to get and quantify some

>> of the momentum, because this brings up the market force. So Paul Nishwari, who's on our team and who covers AppDev, he and I were having a conversation, and saying, "Wow, the confluence of two worlds are coming together. " AppDev, which you mentioned developers, which is, we've been working a lot on, when you've been on theCUBE, at KubeCon and CNC Cloud Native Foundation. Cloud Native World is all about, let's clean up the platform engineering side, which is essentially servicing in the CICD pipeline for developers. That's AppDev.

Haseeb Budhani

>> Yep. - Okay. AppDev will be the end game here very

>> quickly, because your progression is, the hardware systems, AI clusters, need software to run on it, a stack. And then once that's in place, then just go. You got data in there, you're orchestrating, and then the AppDev will kick in. So you're actually setting the table for the AppDev market. Do you agree? Is that-

>> Oh, a hundred percent.

Haseeb Budhani

>> In fact, I'll tell you something that has surprised us inside the company in the last few weeks. So for GDC, our friends at Accenture were preparing a demo, which they will do every day next week. And they really wanted to show these agentic applications that they have built. So the question is, okay, I have these apps, how do I get them delivered to these users who want to, can press a button and consume them? So we worked together a couple of weeks and they used our platform to do the app provisioning and delivery to the developer. So we standardized, on our platform, 27 of their agentic apps in a week. And next week you're going to see them, like, you can press a button and get them. Right? So if you can deliver these very valuable applications, so there's an application that says, "I'm going to do a credit check on somebody, I need an agent for that. " Or "I'm going to underwrite a mortgage, I need an agent for that," et cetera, et cetera. These are very valuable business applications. But now there's an opportunity to deliver them and an enterprise can come buy them.

>> It's interesting. See, I got to tell you, how things come full circle, it's interesting. I was in New York City at the NYSE where our new studio is, and I was talking to some of the big banks about AI. Nothing to do with KubeCon, Kubernetes, nothing. What models are you using? Are you going to do it on-prem? And in the conversation, Kubernetes comes up, randomly, because it's lingua franca for them for what's going to run. They weren't talking about Kubernetes. We used to talk about Kubernetes all the time, "Hey, all this cool stuff. " But they're talking about how they're going to get the models running. So their conversation and their eyes were on, okay, I have models, that's the app layer. That's the AppDev piece. So the AppDev piece has now got platform engineering and model integration, because that's what they were scoping for the system.

Haseeb Budhani

>> I love that. - It's almost like they forgot that they

>> needed an operating system to run it. So again, this is where I think you're kind of breakthrough here, because if this continues, your deals are closing fast, the question is, why? And the answer, to me, seems to be because the dev's coming and the models are coming.

Haseeb Budhani

>> Yep. Yeah, look, the two pressures on the buyer, right? One is, they have the investments already made, in some cases, right? And there's more investments coming, so they got to utilize this infrastructure fast, because, otherwise, they're paying depreciation for nothing. And the other thing is they have customers waiting. That's the best part. The customer doesn't want to say, "`Okay, give me a... " "What do I do with a GPU? I need an app, right? I need something." Whether it's the Nvidia application stacks, or Nvidia has Nvidia AI enterprise, which is a bunch of really great applications. Or Accenture may have their own applications. Or locally, regionally, like our friends in Turkey, they have a whole ecosystem of developers who are developing apps. How do you deliver them? Simple, right? So now we have a solution.

>> It's funny, when you're in business, there's two types of meetings. There's project updates, how are we doing? Check in. And then there's business hardcore urgency. And I think what you're saying is that you're seeing urgency because the business is on the table. This is outcome driven. So the pressure coming in is like, okay, how fast can we get this up? Because, they can quantify the ups and downs of not having it. And I think this is a phenomena we're seeing, and it's a good one, because, again, platform engineering has been in this, okay, setting the table, the foundation's good, how are we doing? It was moving fast, but now it's like, no status update. Where's the-

>> It seems slow now.

>> Where's the apps? - Now, in hindsight,

Haseeb Budhani

>> that seems slow, my friend.

>> Yeah, that- - It was going great, but this is crazy.

Haseeb Budhani

>> So how has that changed your culture?

>> And this is a good segue into GTC, because that's next week.

Haseeb Budhani

>> You said your style's methodical, a little slow. You said slow, but I think just more calculated. Now, is it straight and narrow for you? You're just going to gas it so it's more clarity? Were you slow because you were just trying to figure things out, going around hairpin turns? Or the market hasn't been clear? Is it that? Or is it just more of, okay, now we have a straight and narrow, which we want to gas it?

Haseeb Budhani

>> Yeah, I think we are at the let's gas it moment in this company. Which is why I said I wish I'd hired a hundred people a year ago. How could we have known? But now, because the opportunity is here, this is such a great team. My colleagues, like in engineering, so my team leads who are running their own teams, they're helping with customers. So we have guys on the team who are basically managing the customer relationship, because they are so close to the stack. So they actually know what's going on better, sometimes, than I do, for sure. For sure, I do. But even our CS team and sales team, they're still learning. So the engineers are helping. So the culture of the company has gone, I mean, it's really changed. Everybody's a sales guy now. It's kind of amazing to see, actually. I see-

>> for sales training for the engineering team.

>> It's incredible. - It's probing questions, the line.

>> Yeah, and they're enjoying it. Right?

Haseeb Budhani

>> And I'll tell you this story.

Haseeb Budhani

>> So our friends in India, their CEO shows up to these POC calls. Like, we've never seen that. The platform engineering stuff, no CEO of a bank is showing up to figure out, "Okay, how's my implementation going for this? " Right? But now in this world, because it's so expensive, showing up. And the CEO of this company, he's providing direct feedback to the engineers. So now, they're getting this kind of level of feedback from the top guy in the customer account. Right? Completely different, right?

>> Yeah. And I think the CFO, the CEO, the CXOs are getting involved because the dollars are on the table. We had a CFO event as part of our NYC wiredCUBE relationship. We've never done that before. But we did it because, not because we want to talk to CFOs, because the business transformation risk management/ business models are changing on the customer side. So they're trying to get in and understand the technology, because they have to model it out. So I mean, if you're a CFO, and you're waiting to get your AppDev team tweaked for AI, I mean, that's input to their job, because they're the architects for the business. Business model is finance, it's money.

Haseeb Budhani

>> Yeah. Everybody cares, at a C level, about this, for all the right reasons I would say. So this is making people talk about, "This is like the dot com. " I think this is very, this is substantive. There's real work happening here. There's real applications being built here. Look, I'm just very glad that we .

>> Yeah, I would agree. I would agree with you. We just came

>> back from HumanX. We just had a Crypto Trailblazers event here. There is real use cases. Yeah, a little bit bubbly and hype in some areas, but agents are coming super fast. That's not a hype cycle. It mean, it's a hype cycle, but it's reality. There's low hanging fruit enough to get business value.

>> Truly. - And I think we're going to see that.

>> Let's talk quickly about GTC. It's coming up.

Haseeb Budhani

>> You got a relationship with Nvidia, you have the inside baseball there. Can you share with us what your view is on Nvidia at GTC this year? What do you expect to see? Obviously, AI factories, this is year two. I mean, last year they had the big announcement, Michael Dell's in the front row. I'll actually be interviewing Michael Dell in his booth on Tuesday. I want to find the update from him. But I'm expecting to see more of this, more maturization, more evolution, faster, software running on hardware systems.

>> Yep. I would agree. I think software is

Haseeb Budhani

>> where everybody's minds seem to be now, because that's where you get consumption. Maybe you want developers to be happy so they can consume more, and buy more, and everybody wins. That's what I would expect. Like Mr. Huang always does a great show, puts on a great show. And I'm sure he will again. But software, I suspect, is going to be the most important thing in this event.

>> Yeah, and I'll be interested to see how the NIMS are playing out. But I think, ultimately, people still want the GPUs, and they're looking at architecting them differently. Okay, final question, what's next? You're going to come out of FTC, get pipeline, you're hiring folks, put a plug in for what's going on at Rafay. What's happening there? I'm sure you're going to have all the investors coming at you with all this new growth. What's-

Haseeb Budhani

>> Yeah, we're hiring as many solution architects as we can. I mean, this is really fascinating, right? This is a very, at some level, it's a very technical conversation, which is why the engineers on the team are enjoying it, and actually succeeding at it. So we're hiring as many solution architects as we can, like quickly. We have been very focused on Asia. I would say, Asia, India, there's been a lot of business there, Middle East, et cetera. And we need to focus on the US. So we're definitely starting to hire a lot of people in the US. I mean, this is going to be a crazy year. I'm hoping I can get a day off after a GTC, like a day off. But, my God-

>> No, you can't sleep.

Haseeb Budhani

>> My God, it's been a crazy six months.

>> And I think it's going to be another crazy nine to 12 months.

>> Well, at the end of the day, the scoreboard about revenue, value, I mean, outcome-oriented solutions. And I still think that we're going to see a lot more narrative around the collision, in a good way, between AppDev, and platform engineering, and these AI factories, because that's going to open up and unleash more speed for deployments. So Assentia's going to win. Everyone's going to win. Hasseb, great to have you on theCUBE. Congratulations on the news. Appreciate you coming in.

Haseeb Budhani

>> Oh, thank you very much, John, for having me. Good to see you.

>> Get the news here. It's CUBE conversation. Again, GTC next week. The new model starting to show the proof, and that's in the results on the outcomes. People are getting involved, the CXOs are coming in, the investments are being made, and now they're harvesting. And software's going to be power. You got to run software on the hardware. It's almost that simple. I wish it was. Of course, we're bringing all the action here on theCUBE. I'm John Furrier, host of theCUBE. Thanks for watching.