theCUBE + NYSE Wired: AI & Retail Trailblazers | Renen Hallak, Vast Data

Clips
More from theCUBE + NYSE Wired: AI & Retail Trailblazers

Renen Hallak

Founder & CEO

VAST Data

play_circle_outline Partnership with Nvidia in the AI era

play_circle_outline Partnership and integration with Nvidia in AI infrastructure and acceleration

play_circle_outline Importance of shared everything architecture for scalability and performance

play_circle_outline Importance of shared data plane for AI interactions

play_circle_outline Trends in AI optimized and GPU aware computing stacks

play_circle_outline Shift in computing landscape towards AI optimized and GPU aware stack

play_circle_outline Early days of Vast and innovation mindset

play_circle_outline Democratizing access to AI for enterprise

play_circle_outline Importance of resilience in enterprise AI systems, focus on data loss and downtime.

play_circle_outline Focus on providing APIs for AI application development

play_circle_outline Expanding product capabilities for data and applications, enabling startups and developers.

Info
Transcript

Renen Hallak, Vast Data

Renen Hallak

Founder & CEO VAST Data

The interview at the New York Stock Exchange focused on NRF and retail innovation, with Renen Hallak, CEO of Vast Data, discussing the company's growth and success. Vast Data has evolved its products into an operating system for the AI era, with a partnership with Nvidia proving crucial. The importance of a shared data plane for AI agents was stressed, as well as the market shift towards AI optimized infrastructure. Simplicity and security are key for enterprise adoption of AI, with a need for clear boundaries within departments. Building new applications is ... Read more

explore Keep Exploring

What are the key elements of the partnership and integration between the two companies in terms of data platform architecture and software infrastructure for AI implementation? add

What is the company's stance on working with non-Nvidia customers? add

What role does the data plane play in enabling large scale AI to progress from static learning to dynamic creation of new ideas and abilities? add

What is the importance of agents having access to the same data plane for interacting with each other and combining training with fine-tuning for AI? add

What do you think about the phrase, "The stack has to become AI optimized and GPU aware"? add

What does the stack need to become in order to keep up with advancements in technology? add

What was the moment when the speaker knew they had to innovate and be misunderstood for a long time in order to succeed? add

What are the key considerations for ensuring information security and access control within different departments of the same company? add

What is the company's track record for data loss and downtime since they started working? add

What is the focus of the company in terms of data capabilities and API development, particularly for startups building AI applications? add

What is the goal of providing more API's towards applications and becoming the standard in the space in order to accelerate the startup ecosystem and developer ecosystem for AI applications? add

bolt Powered by CUBE AI

Renen Hallak, Vast Data

search

>> Hello, welcome back to theCUBE. I'm John Furrier, your host. We are here at the New York Stock Exchange, as part of the NYSE, WIRED, and CUBE programs Media Week. NRF is happening. Really, the innovation in retail is happening. We're here for three days of wall-to-wall coverage. Of course, we're going to bring in the leaders and the experts from across the industry to break it down, of course, and get more insight into, obviously, the AI way that's here. Renen Hallak is here, the CEO of Vast Data and founder of Renen. Good to see you again. Thanks for coming on. We were just riffing before we went on about the role of AI, but great to see you.

Renen Hallak

>> Great to see you too. Thanks for having me.

>> So I think we've done, probably, this is our fourth interview together. For the folks out there watching, go check out the YouTube videos, a lot of great content, going into great detail on Vast, on the expertise. But let's just reset here. Give us the update on Vast. Obviously, the momentum's huge. We saw you guys at SuperComputing. You had more meetings than last year. Jeff was running crazy last year. This year, even more craziness. So just overall, great business momentum. Congratulations on your success. Gives us an update of where you guys are at now. Just quick progress and stats.

Renen Hallak

>> Yeah, so I like to say we grow 3X year over year. This last quarter we grew a lot faster than that. At the same time, very efficiently. We generated quite a bit of cash in this quarter, and so, beyond all of our expectations from what we planned at the beginning of the year. Growth is not meant to accelerate at these higher numbers, but it has because of these great AI tailwinds that we're seeing. And so I think the large-

>> So you are making money? You are-

Renen Hallak

>> A lot of money.

Renen Hallak

>> Okay. - Yes.

>> Go on. Can you share the numbers? Or no?

Renen Hallak

>> We have more money in the bank right now than the entirety of everything that we've raised, and it's growing very, very quickly. This year we will likely generate a triple-digit number of millions of dollars.

Renen Hallak

>> So that's real income? - Yes.

>> So you're putting the product out there, people are consuming it. What is the consumption experience like? Is it mostly a handful of top customers? What's the product mix? Can you just give a quick taste of the current state of the art?

Renen Hallak

>> Sure. So it started as a storage system, has now become the operating system for this new AI era. And so, we have the data store, the database, the data space, which is our global namespace and we're now coming out with the data engine, which a lot of new applications are sitting on top of. The first application is actually what we call the Insight Engine that we built with Nvidia, incorporating Nvidia names into the same platform. And so, the large language model builders, the foundational model builders, have now standardized on Vast. Nearly all of these new AI clouds have standardized on Vast. What we're seeing now is the proliferation of AI into the enterprise. And they require a very different set of features. They want it to be secure, they want it to be simple, they want it to be real time. And that is the next wave.

>> So you're pretty bullish on Nvidia?

Renen Hallak

>> Yes. Nvidia is one of our best partners.

>> Of course, that's why I'm being facetious. But okay, let's get into Nvidia, because Dave and I were talking on our CUBE Pod, we were talking about this in the hallway here, on the balcony on the stock exchange floor. During the PC generation, which was PCs and servers, Wintel, Microsoft Windows and Intel dominated everything. It's called the Wintel Phenomenon. And you had Mac, they had Motorola process, and other proprietary stuff. That was pretty much the PC platform.

>> Yes. - Servers came out of Wintel. Yeah, Macs had servers, but they weren't really servers, in my opinion. Sun and the proprietary guys had workstations. Okay, that was proprietary. But Wintel standardized client server at an Intel level, not PC, I mean not many computers.

>> Yes. - That was two companies dominating. So Dave and I were saying that Nvidia could be Wintel in one company. Do you agree?

Renen Hallak

>> I think they are definitely the Intel part of Wintel, and I think they have parts of the operating system. And I think the parts that they don't have, we try to complement them. So wherever they sell their solutions, whether it's GPUs, or networking gear, we come in with our part, and together, I think we've shown how to build it, what the blueprint needs to be.

>> It's funny, here on Wall Street, we're new here, but Dan Ives is the big guys. He's complained Nvidia was great. Of course, if I was here, I would've been there too. We predicted Nvidia easy. And arm too, by the way, Arm and Nvidia, two obvious. If you're paying attention, those are two companies that you go, no brainer. But it wasn't really clear to the world. Now we're in the conversation of, yeah, we all see Nvidia and it's now mainstream stock, replaces Intel on the S&P 500. Now it's like, can they go more? I, like Dan Ives, thinks yes it can. But that's not the question. My point is, to get there you got to have a product.

>> Yes. - So let's talk about the operating system, because, if they become Wintel, that means that they have to actually deliver the goods, and not become a proprietary NOS or proprietary stack like IBM had in the mainframe days, with SNA architecture, or digital equipment program with DECnet. You can't have proprietary anything, because we are in a heterogeneous world. Okay? Now multi-vendor, heterogeneous, that doesn't mean that everyone gets to play. You've got to be on the winning side of the architecture.

>> Yes. - Take me through your thoughts on this kind of thought exercise, because it's a really good one. Because I think what you're doing with Nvidia, and I've been very skeptical on the NIMS, just for the record, and Jensen knows this. But I think, I love the NIMS, but it's a land grab.

>> Yeah. - Okay, yes. But if it happens, that's good. So it's not so much a bad thing, it's just it's a bold, bold move to get microservices down and get that small language models, which I'm a big fan of, as you know. So I see the play. Now the question is, will that be the preferred future? Well, it kind of depends. In order to win the enterprise, you got to be heterogeneous. How does Nvidia become Wintel while maintaining some sort of de facto standardized stack? Is that ethernet? Is that storage? Because, you're in the middle of it. You're playing... You don't work for Nvidia, you have your own company.

>> That's right. - But you want to ride and connect in, not be parasitic, but be complimentary to the architecture.

Renen Hallak

>> Yes. It's always been compute and data. And so Nvidia, obviously, brings the compute aspects of it. And they need a data platform that was built based on a new architecture for this new era, and that's the partnership that we have.

>> Yeah. And how's that going? What's the key linkages? If you had to boil it up and simplify it to some level, I mean, not for the average business person, but for a technical person, what's the key keys to that relationship? Integration? Code? Can you take us through how you look at that?

Renen Hallak

>> Sure. We're working with their teams very closely. And what started as us using their hardware, as you say, they're moving up the stack, they're providing NIMS, they're providing blueprints. And so now, VAST is that middle of the sandwich. We're leveraging their hardware, both the GPUs, and DPUs, and network interface cards, and switches. But we're now running their software on top of our software infrastructure layer, such that it becomes an easy way for an enterprise to start using AI.

>> What about non-Nvidia customers? What's your position on that?

Renen Hallak

>> It's fine. We will work with anyone that wants to work with us, underneath and on top, both application vendors on top of us and hardware providers underneath. But today, as you can imagine, Nvidia is the biggest one.

>> Yeah. And they're running the table, triple on the train. Inference is a big hot button. We're seeing Inference come out. That's where NIMS will play nicely, I think. And by the way, I love their vision of... And I've been saying this on theCUBE for years. Dave knows this has been my rant. I believe models will talk to each other. That's clearly the power of law we put out. We were some of the first one to put the power of law out, but I think Jensen, and the big people on this side of the camp are right, that it's not one model. It's using the model. That's why LLM routing is so popular right now, because that's a data storage meets infusion layer.

>> Yes. - I don't know what to call it, what do you call this? But there'll be interactions in real time-

>> That's right. - ... under the covers with code, between data models. That's going to require a rethinking of what is a horizontal data plane, for the lack of a better description. How do you talk about this very nuanced piece? Because this is the holy grail for large scale AI.

Renen Hallak

>> Yes. The only way for agents to interact with each other is to have access to the same data plane. The only way to combine the virtues of training with fine- tuning, with inference using these human feedback loops, is by having one data plane underneath. And so the data plane becomes critical in taking AI from this static thing that just learns on top of data, into this very dynamic environment where we create new ideas, where we build new abilities.

>> So Dave just did a breaking analysis. So we did the riff on the Wintel. We just did a breaking analysis, Dave and David Floyer. The title of the post on SiliconANGLE right now headlining is, How Nvidia is creating a $1. 4 trillion data center market in a decade of AI. A trillion dollar shift is unfolding. This is real. And so, if that's the shift, stock price is irrelevant at this point. So if the cards come out that way, if he's got an ace up his sleeve, which I think he does, given his CES keynote... And Nvidia's always had stuff going on, baking in the oven, so we don't know what we don't know from the outside. You might have some insight on the inside, but this is the shift. There'll be winners and losers in this wave. Share your thoughts on what that means, because winners and losers seems like a broad term. It seems like a haymaker. But really, if you're on the wrong side of the architecture or wrong side of the street when this happens, you could be holding the bag.

Renen Hallak

>> I agree. I think architecture is key. And most of the companies that are competing today are basing their solutions on old architectures, that were built and designed 10 years ago, 20 years ago, 30 years ago, way before we could even think of this new AI era. On our side, I can tell you that that is the one differentiator. We have not yet seen a new company emerge with a disaggregated shared everything architecture that's similar to ours. And that is, I would say, the only reason why we are winning time and time again against all of these older companies that came before us.

>> What does that mean when you say they don't have that share everything architecture? I don't want to say fake it till you make it, because that's really not what's happening, but there could be a fatal flaw by not having that.

>> Yes. - What specifically about that makes them less attractive in the deals?

Renen Hallak

>> They can't scale to the levels that are required. They can't scale to the performance levels that are required. It's way too complicated to deploy. And so, as you get to these very large environments, whether it's 200,000 GPUs, or whether it's two exabytes, it doesn't work anymore. And we're seeing them struggle to compete because it wasn't designed for this new type of workload. And that's the shift that we're seeing in the industry right now.

>> You have a history previous to Vast, flash storage. So you know your storage up and down, says Cole. Now, obviously, technology-wise, you're expert. Extensive parallel computing is huge.

>> Yes. - It's a big part of post our around the 1.4 trillion. The computing landscape is changing. You're connecting into that. You're not a storage company anymore.

>> No. - You do storage data, that's a piece of it, but you don't look at this as storage.

Renen Hallak

>> We do not. If you look at the inside engine that I just mentioned, storage is a very small piece of that. It's an enabler. Having fast access to all of the information is a prerequisite, but that's just the first step. And so the data store is that first step that we took. What we realized a few years ago is that the architecture is the game changer, not the storage system. And so we leveraged the same architecture to build a new database, and to build a new data engine, and to extend those across geographies using our data space. And that is what is forming this very, very comprehensive stack that our customers are enjoying today.

>> I think that highlights some of the things that we're seeing. And I'll just throw it out there, because, I always ask about predictions every year. I usually do really well on the predictions. Dave and I have good predictive capabilities. We have a lot of data. But I didn't really want to do it this year, because I think all the action, I said, "My prediction is this, AI infrastructure is going to be hot, and it's going to be different than what people think. " And I think to your point, this is what is happening now. AI infrastructure is, if you just say the word, it doesn't mean anything. But if you unpack it and say, "What is that? What is it and why?" architecture is one. It's not yesterday's architecture, it's not yesterday's vendors. New brands will emerge. Your Vast is one of them. And so, AI infrastructure has to be re-tooled. That's why it's geek central right now for all super infrastructure nerds, because if you're in networking, storage, or compute-

>> It's all different. - ... take 25, 30 years, this is the time to start a company.

>> Yes. - Because you could actually redo and refactor other elements, and build a stack, and retrofit this for the AI era.

>> That's right. - What's your thoughts on? React to that, and do you believe that? And if that's true, what does the call to arms on that entrepreneurial wave going to look like? What's going to happen?

Renen Hallak

>> I do believe that. And I think I say this often, our biggest advantage is that we started late, and that we were able to see this deep learning revolution in front of our eyes. We were lucky enough to work with companies like hedge funds, and life science institutes, and autonomous driving car companies way before generative AI was a thing, and they showed us the future. And so we built for this moment for the last nine years now. Anybody who starts today can leverage all of those learnings in order to-

>> Through you.

Renen Hallak

>> Through us and through others that have done.

>> Now, if they try to replicate it, there is this economies of scale. Or are there? You're going to have economies of scale for the nine years of trajectory.

>> Yes. - Okay, I get that. But what about someone else starting? Is there white space? Is there areas you see where, hey, me and my friends, "Hey, went to school in the '80s together. Hey, let's start a company. Let's disrupt agents"?

Renen Hallak

>> Yes. I think there's a lot of white space, especially at the application layer on top of us. There's going to be, every single old application is going to be redone leveraging these new AI abilities. And it's just getting started right now. People are noticing this new thing. They're realizing that there's room for innovation. And this is the time to start new companies.

>> We put out, on the SiliconANGLE, we put out a story, again, on this trillion dollar shift. There's a slide that's in the story, and I'll read you the title of the slide, I want you to react to it. "The stack becomes AI optimized and GPU aware. " It's really talking about networking, storage, compute. Then you got real-time access to the business logic, a data layer, harmonization layer, and then you got the operating system. Essentially, it's an operating environment.

>> It is. - What do you think about that phrase, " The stack has to become AI optimized and GPU aware"?

Renen Hallak

>> I agree. I think everything is going to be accelerated computing. I think everything is going to be AI first. Because the old applications, the old way of doing things, are obsolete. These new agents will start to emerge in different aspects of our lives, and we, five years from now, 10 years from now, won't even understand how it was before, in the same way that today we don't understand how it would've been to live without the internet, or without smartphones.

>> Yeah. It's interesting. First of all, I love the accelerated buy into that. What's interesting, if you go back to the other waves that we lived through, we didn't really have to think about accelerated computing because Intel took care of that.

>> That's right. - And the PC manufacturers and server manufacturers built the boards, and the back plane, and all the different subsystems that go into that system architecture. But you had concepts of bolt-on offload. So what's your thoughts on offload market? Because, you have acceleration, but there's efficiencies that you can get with configuring chips, and networking, and other technologies around the system, to be what used to be as simple as memory management back in the old days. These other optimized systems, whether you call it offload, that are designed in. So acceleration's definitely happening. And by the way, it's not just Nvidia that's accelerating, it's the customers themselves, who are building their own systems. So it's not like off the shelf PC, they're building their own clusters.

>> They are. - Which is hard but good, that they get super computing.

Renen Hallak

>> That's the beginning. Because, if you look at the computing in the 1970s, you needed to be a PhD in computer science to operate a computer. And as you say, these software layers, these operating systems, make it simpler and simpler for us mere mortals to use these machines. The same exact thing is happening now with AI. We're still at the phase where you need a lot of expertise, but the software layer is coming in and it's going to make it widespread and easy for everybody to leverage.

>> That's awesome. I want to ask you a question about your origination and early days of Vast. Andy Jassy would always have a famous expression on theCUBE, both private and public. "If you want to be an innovator, you have to be willing to be misunderstood for a very long time."

Renen Hallak

>> Yes.

>> And be comfortable with that. Most entrepreneurs deal with that, "You're crazy. What?" And deal with the public ridicule.

Renen Hallak

>> If everybody agrees with you, you know you're doing something wrong.

>> "No one gets it," that's a famous line said by entrepreneurs, "No one gets me."

Renen Hallak

>> Yes.

>> Where was the moment you knew? Because you said you started the pre- deep learning, you came in. Obviously, you saw a lot of cool stuff. And we knew Extreme, and some of those high-end storage vendors, like Isilon, they were in the hyperscalers. You had visibility into some of the big self-built infrastructure. What was the moment where you knew? And when was that moment you said, "Okay, wow, we have to... I was thinking here. " It's the classic zigzag, I'm sure you went through it.

>> Yes. - When was the first time you zigged, before you made that next move into AI? Was it a customer experience? Was it something that clicked for you? When did Vast go, "We got to move everything into this area"?

Renen Hallak

>> So from the first day we were building towards AI. And the click was, it was actually two things. One, I saw a couple of YouTube videos. One was Danny Hillis, who in the '80s, built a company called Thinking Machines. And he said, "It was too early. "And I thought, again, nine years ago, that now may be a good time to revisit that ambition. The second one was Jeff Dean, from Google, who was talking about how they were starting to leverage old algorithms, deep learning algorithms, to understand what is happening within videos. They could identify cats within videos, using these very old ideas that didn't really work before. And Jeff Dean is the infrastructure guy at Google. They needed him in order to make these old algorithms work. I said, "Google has Jeff Dean, the rest of the world needs somebody to make that happen."

>> They needed a Jeff Dean.

Renen Hallak

>> That's right. - That was Vast.

Renen Hallak

>> That's what we're to trying to be.

>> The idea of democratizing what specialty skillset was needed. And then the Thinking Machines, I think it was out of Cambridge. I think one of my friends did a co-op there in the '80s. But that was a super computing venture.

>> It was. - That was like, they wanted to build the first supercomputer.

>> They did. - I think they did.

Renen Hallak

>> Yes. But they never got to a thinking machine.

>> Yeah, because it doesn't think, because there was no thinking back then.

Renen Hallak

>> Right. And there's no thinking yet, but I think we're now a lot closer to it than we were then.

>> It's so funny. You're going to laugh at this, because I got my CS degrees in late '80s. So we all touched AI at that time. You had Lisp, and you had some neural network stuff going on, but there was no need to do it. You couldn't do anything with it.

>> Didn't work. - It was like a theory. Yeah, I can build an ontology. I had to still manually set up a taxonomy. It was grunt work. And then, big fricking deal. What does it do? Apple's not a fruit, it's a company. Like what? I mean, there was not there, there.

Renen Hallak

>> The idea that we didn't need to understand how the brain works in order to recreate it, and that it was all about extremely fast access to very large amounts of information, that's the reason that Vast exists.

>> And I think, I bring this up because there's two phrases I hear a lot in the past year and a half, two years, "I did '80s. I did AI back in my day," which kind of could be true. And then there's the people like you, and others, that are there and saying, "I know what it means and I could do it now. " So there's the, "We do AI. I've done AI. We all... " I mean, people know the concepts of AI. But I think seeing that it failed because it wasn't attainable to begin with. And what you're saying, and what happened with Vast is, it's now doable.

>> It is. - And you said, "We could do this."

Renen Hallak

>> Yes, because of- - The resources are available. - ... a new architecture, because of what companies like Nvidia did underneath us, because of what companies like OpenAI did on top of us, this is the right moment.

>> Okay. So first of all, I love this conversation. So now the next question is, okay, first, that's definitely happened. And there's a lot of older entrepreneurs that are like my age, our age, my age, that are actually recycling that, "I know that's possible. I'm going to do it." And the stakes or the barriers to do it are low, relatively speaking, but the spoils are beautiful, massive. So that's cool. Now let's go to the AI for the enterprise. Now, when you look at Nvidia, saying, "We want to win the enterprise," that's what they're basically saying.

>> Yes. - a Dell AI factory. You guys are working with them. To be successful in the enterprise, what is the equivalent, "I couldn't have done that before, now it's doable"? Because now Jeff Dean, your Jeff Dean is now in Vast.

>> Yes, that's right. - So now, what is out there in the enterprise that would allow me to say, "I've been struggling with this for three decades," if I'm the head guy, or two decades, whatever's in charge, " What's now available, that's gettable?"

Renen Hallak

>> Yes.

>> That's attainable, that was unapproachable. Not even a, I mean, "Great idea, but we can never get there, " just physically, human impossible?

Renen Hallak

>> The enterprise does not know how to build a supercomputer. They don't know how to maintain it. They don't know how to operate it. We need to make it very, very simple for them, and we need to make it very secure. If you have different departments within the same company, not everybody has access to the same types of information. We need to make sure that those boundaries are very, very clear, and that the organization, the IT organization within the enterprise, does not need to have a PhD in AI or in computer science in order to operate this. And that's a big part of that next phase. We need to make it simple, we need to make it secure, so they can adopt it as well.

>> Scope the challenge, in terms of execution, operationally. I mean, you could take in pieces, hit a little beachhead, and then sequence to a much broader position. I get that. So what would be the scope of today's landscape, vis a vis, one, there's enthusiasm, clearly-

>> Yes. - ... but confidence?

Renen Hallak

>> Yeah, everyone's playing with it. Nobody really knows what to do with it yet. And I think it's a combination of generating that expertise and building new applications. Again, the enterprise, unlike some of these early AI companies, they don't build their own applications, they buy applications from software providers. And so we need new companies and old companies to build these new software abilities, such that the enterprise can consume it.

>> It's interesting, I had a conversation with Poolside, the guy from Poolside, great entrepreneurs there, hardcore AI guys, amazing team. And we were riffing on the idea that, I asked the question, "How do you know the pretenders from the players? " Everyone wants to know that. AI washing is going around everywhere. And the guy was awesome. He says, "Well, it's very simple, you got producers of AI and you got consumers of AI. " And okay, now you can put people in two camps. Are you producing AI? Which means that's hardcore. We're still in the PhD machine learning hardcore, guys getting millions of dollars a year. And then, like theCUBE, we're a consumer of AI. We don't actually make AI. So I said, "Well, you're missing one piece." He goes, "What? " "Data providers." Because now, we're a data provider. We have a proprietary data and domain expertise, but we could also feed that data to the AI, and the consumers as well, and be a consumer. So we added that third layer in. Do you agree that that's a good way to look at, producer of AI, consumer of AI, and data for AI?

Renen Hallak

>> I agree. I think that's exactly right. And data spans both of those sides. And that's where we live.

>> All right, let's get back to the enterprise. Let's just say I'm JP Morgan Chase, I'm Lori Beer over there, and I'm sure they're already a customer...

>> I can't say that. - You're smiling. They're a customer then. Okay. Okay, I won't go there. She has a $17 billion annual IT budget. Even, it's funny to even say IT budget, in the scope of 17. But that's a regulated industry. Some of it's consumer. They do about 10 trillion a year in transactions. I interviewed her at Reinvent. I asked her the question, "How do you define resilience in AI? " Now, you know storage, you know systems. Resilience, cyber resilience today, is recovery, backup, and recovery for ransomware and other things. There's no real definition for resilience. She's the only one that actually gave me an answer that I liked. Because no one has... There's no answer, how do you define it? Her answer was simple, "Well, it's just another application to us, so it's got to maintain our resilience standard. " And I'm like, actually, that's kind of an answer. That's a good answer, actually.

>> Yes. - So her answer was, when it meets the resilience standard. So what she was saying is that, hey, when the bar is high. Until then, nothing's happening. Now they're an extreme edge case, but that brings up the point of, what is resilience? When you start talking about large scale data and AI, you're the system that's providing it, enterprises need to check this box.

>> They do. - This is like table stakes. There's not resilience, not just in data and storage, but all up and down the stack, if you will. And into the application, especially when it's generative.

>> Yes. - How would-

Renen Hallak

>> And that's why you need an enterprise-grade system. These HPC-type systems don't work within the enterprise for that exact reason. They're too complicated. They're not resilient. They fall over. They weren't built for it. And so you need an enterprise-grade system that is compliant, that meets all of the regulatory requirements, that is very, very robust and secure. At the same time, you need a system that can meet the technical requirements of these new AI workloads. And that intersection is one that's very hard to do. That is the reason we've been succeeding the last couple of years, because we're right at that intersection.

>> Where is your bar in resilience? And how would you define that? I mean, it's a moving train, if you want to think about it, but it's going to get better. But where are you now, that closes those deals and gives the enterprise confidence?

Renen Hallak

>> We look at two things from a resilience perspective. We call them data loss and data unavailable. And those are the bad things that can happen in our space. We have not lost a single bit of data ever since we started working. We're talking dozens and dozens of exabytes in production. And most of our customers-

>> Not one?

Renen Hallak

>> Not one bit. And so that's-

Renen Hallak

>> Huge. - ... I think, a perfect record. We're very proud of that.

>> It's undefeated. You're undefeated.

Renen Hallak

>> From an uptime perspective, most of our customers are at a hundred percent. The ones that aren't, maybe lost access for a couple weeks-

>> So you're engineering to that resilience standard.

>> Definitely are. - And the relationship with Nvidia, what's their perspective as you guys engage with them? Obviously, they would probably say, "We're doing our best to be resilient. " But are they enabling that? Or is that mostly coming from you?

Renen Hallak

>> I think the way resilience gets built in large systems is not by relying on a specific component to being resilient, but having the redundancies in place, such that you can sustain specific failures within the components. And so, that is a big part of our partnership in the way that it's built.

>> Yeah, if you look at all the proof of concepts out there, I know you have a lot, your own dashboard, but as theCUBE research team, we look at, and theCUBE, we look and talk to a lot of folks with clustered systems, I call them, but stuff that's large scale and AI. If you map out 50 deployments, maybe two look the same. So what this tells me is that there's a lot of fragmentation in the deployments. Now, I don't think that's a bug, I think it's a feature of where we are, which is, beauty's in the eye of the beholder relative to what they're building.

>> Yeah. - Okay, I get that. But that can't be sustainable, because-

>> It's not. - ... at some point we have to get to, "I got to buy this."

Renen Hallak

>> Right.

Renen Hallak

>> Like something. - You need the cookie cutter to start working.

>> I need the package. - Yes.

>> It used to be a server. You rack it up, give me 20 of them. Do I buy an AI factory? What the hell does that mean? So what's your view on this? And how far do you think we are down the road of getting to a point where you can say, "Okay, a well-formed package of this will do this"?

Renen Hallak

>> I think we're there. I think we've been experimenting for a few years now. Every company has built it slightly differently. We've learned a lot. And now standardization is starting, and that repeatability is starting to show.

>> Yeah. Well, it's really great to chat with you over there. Love the little deep dive. Coming up, we just came off the heels of CES Robots. I was talking to someone here, NRF, there's robots in supermarkets today, not stacking, but just scanning inventory. Holograms will be there next. "Hey, can I help you? John, good to see you. Come back. " "Buying another six-pack of beer?" "No, I'm good. " But this is really becoming an industrial era.

>> It is. - It's Jensen was on stage at CES, talked about robots. He did throw a lot of cold water on quantum, which I think is a little bit more... Because he thinks he's the answer to quantum. Okay, whatever, but a lot's changing.

>> It is, very, very quickly. - What's your vision of where we are now? In the next year or two, what's going to happen?

Renen Hallak

>> I think the robot problems are getting solved through AI. I think the ability to generate ideas is starting to form. We're working on, with a lot of our customers and partners, on what's known as System Two, which is machines being able to reason, being able to design, to engineer, to solve hard problems. We're not there yet, but I feel like we're not that far away from that.

>> Yeah. Final question for you and I'll let you go. I appreciate you coming in a little long here, but I saw you talking to Anand, who runs the AI excellence set for ICE. He had some feature requests, kind of prodding you a little bit. I'm sure you get that a lot. My question is, over the past year, it's been a whirlwind, congratulations on the growth on the revenue side, and just the team expansion, the astronauts, and just overall business momentum. But you guys do talk to a lot of customers. What's the one thing you're going to work on this year, from a product standpoint, that you really want to nail down? Or one or two things that, needle moving Vast deliverables that you want to ship, or that you've heard from customers that you're working with? Because, once they get the flywheel going, I'm sure business transformation will accelerate, they'll have more needs. Maybe unforeseen needs, but now they're on the requirements map. It's a hard challenge. What are those one or two things you really want to get done that customers are saying, "I need this"?

Renen Hallak

>> So, from a data perspective, we're now at a point where all of the data can sit on Vast. We have a file interface, an object interface, a block interface. There's no requirement from a data perspective that we can't meet at this point. What we're really focused on is up that stack, and being able to provide more and more API's towards applications, such that these new startups that are building AI applications, write them on top of our SDK, and making it easier for them to get to market faster, and becoming the standard in that space.

>> And you see that feeding the accelerated startup ecosystem, developer ecosystem, enterprise AI with agents to be that enabler?

Renen Hallak

>> Yes. That is exactly right. We want to be the platform enabling all of that, and we're starting to be. And that's the big challenge for this year.

>> Well, that means you got to build an ecosystem.

Renen Hallak

>> That's exactly right. - You can get the marketing department already kind of working hard.

Renen Hallak

>> Yes. Well, we announced our Cosmos Community a couple months ago. And so that's a developer community and a vendor community that's meant to facilitate exactly that.

>> Great to have you back on. Again, more action. Again, this is the future of AI, is really breaking down the new architecture. It's going to be winners and losers. It's going to be on the right side of history or the wrong side of it. We will see what happens. Renen is in the house here at theCUBE. I'm John Furrier, your host. Thanks for watching.