Name: Jeff Denworth, VAST Data
Uploaded: 2024-10-01T16:00:00.000Z
Duration: 31 min 59 s

Clips
News
More from VAST Presents: Enter the Cosmos

Jeff Denworth

Co-Founder

VAST Data

Three insights you might have missed from the Vast Data ‘Enter the Cosmos’ event

Enterprise AI infrastructure faces unprecedented demands today. As AI-powered applications scale, the need for seamless data orchestration across hybrid environments is becoming critical.For Vast Data Inc., the goal has been to operationalize the entire AI pipeline. The company has emerged as a central player as companies seek infrastructure that can rapidly optimize data flows and quickly scale resources to support next-generation enterprise workloads.The company’s new developments in enterprise AI infrastructure were a central focus of the Vast Data “Enter the Cosmos” event, which ran on October 1 with an exclusive broadcast with theCUBE, SiliconANGLE Media’s livestream studio

Vast Data announces InsightEngine, Cosmos community for faster AI adoption

Vast Data Inc. sees a world in which all data is accessible, and the company took the next step toward making that a reality by announcing the Vast InsightEngine in partnership with Nvidia Corp.Vast Data is a platform that merges the concepts of data storage and database to create a transactional data warehouse on top of an unstructured data store. With Vast InsightEngine, the company now enables agentic AI to make autonomous decisions based on real-time data.Vast Data’s Jeff Denworth (right) talks about founding the Cosmos community.“For 40 years you’ve had structured and unstructured data,” said Jeff Denworth (pictured), co-founder of Vast

play_circle_outline Revolutionizing Data Storage: VAST Data's Partnership with NVIDIA and the Evolution of AI

play_circle_outline Real-time data processing and transactionality in AI infrastructure

play_circle_outline Conversations with customers on building next generation AI capabilities

play_circle_outline Infrastructure constraints and the need for cloud and next-gen cloud solutions

play_circle_outline Partnership with NVIDIA to enable real-time search on datasets with recency

play_circle_outline Streamlining AI Deployment: Universal Operating System for Infrastructure Simplifies Complexities for Customers

play_circle_outline Relevance of VAST Data for new foundation in AI and data strategies

33:44

Info

Jeff Denworth, VAST Data

Jeff Denworth

Co-Founder VAST Data

>> We are here for an exclusive interview with Jeff Denworth, who's the co-founder of VAST Data. You just saw their stream. They kicked off a lot of great news. Exciting frontier-like execution around the future of AI. Jeff, great to see you. Congrats on the news and everything you guys got going on. It's a rocket ship, the VASTronauts, as you guys call your employees and staff. Congratulations.

Jeff Denworth

>> Yeah, yeah, it's been pretty crazy times for us.

>> So Cosmos has just had ... you guys just did your presentation, Jensen Wong was on there from NVIDIA.

Jeff Denworth

>> Yep.

>> He is so authentic, but he really highlights the relationship with NVIDIA and VAST, and that was a big part of the news. You guys have that product now integrated with NVIDIA. He talks about this next great adventure of AI, this frontier that you're both on. But one of the things that stuck out at me, and I want to get your reaction right out of the gate here is it's an eight and a half year relationship with NVIDIA. And a lot of people are jumping on the NVIDIA bandwagon now, and so know NVIDIA's stock's high, and they're the proxy for this next bridge to the future. And this is clear certainly from a financial performance standpoint, stock price. But you go back eight and a half years ago, it was not obvious to everybody what was going on around metadata and this new computing paradigm, which even Mark Zuckerberg was quoted this week in saying that this has yet been capped in terms of capability. Where compute used to be finite and have constraints. Now we're on a whole nother adventure. Your tagline here. What's your reaction to that and how do you explain the theory back then? And obviously it's playing out beautifully for you guys.

Jeff Denworth

>> Well, it's pretty wild to think that you can just spend money on intelligence now. That wasn't something that we could do in the past. And so organizations like Meta are jumping onto that to make their services a lot smarter. If you go back eight and a half years, the VAST idea was not possible without NVIDIA. We started in the early days by working with some of the advanced networking concepts that their team was developing. And then following that, we've got so many different integration points and different areas that we collaborate. And when we started the company, Renen and we were talking about companies like thinking machines, if you go back. And we had this idea that as deep learning took hold in the market, that it would be possible to build machines that can start to think for themselves. But when you start a business in 2016 or so, betting the farm on deep learning was a little bit premature. So we just started building out enterprise infrastructure using technologies like what comes from NVIDIA. And then over time, obviously everything now is exploding because people are utilizing the power of deep learning.

>> On a personal level, you must feel like as a team, you guys made that bet. Thinking machines for the folks who don't know was the beginning of supercomputing systems, big iron as it used to be called, not the mainframes, but as you got into more of those specialized systems, HPC, high-performance computing. Jensen's vision of democratizing supercomputing is playing out. That bet has come in.

Jeff Denworth

>> Well, yes and no, I would say. So supercomputers are being democratized for training. And earlier we announced that we're working with the X AI team on some of the stuff that they're doing to train Grok. And these are just fantastically large systems that we get to work with. But every time I meet with a CIO or a chief data officer for let's say a Fortune 1000 class company, they ask me the same question all the time after we show them the platform. They say, "Your database, does it support vectors and can I do vector search on it?" And really what that individual is telling me is that they are still very early in their journey towards bringing in AI agents and pumping all of their enterprise data through them to make their business run faster. So I agree with Zuckerberg, which is we're still in the early, early days of large scale AI adoption. I think NVIDIA, I don't know if you're tracking that, I didn't catch the last earnings report. But maybe 50% of their businesses coming from inference, which means we're still making the model smarter. And then once we do that, then you could imagine 1000 times larger market opportunity for inference.

>> Yeah, I love that metaphor you use about training. You're kind of in school. I don't think you used the school analogy, but I've heard him say that before, we say it on theCUBE. You train, you go to school, then you get out of school, then you infer and reason from that knowledge, you go back a little bit for classes, that's reinforced learning, fine-tuning. That's a human thing. Jensen actually mentioned that LLMs and foundation models will start to think for themselves. And what I love about in a quote he said, "AI is the first invention that invents new versions of itself, and that's where agents come in." So you bring in the fact that now that trend is obvious and you've got a developer market that's frenzied right now on open-source software. You look at Hugging Face, they've got zillions of models now. So the developers are hungry. The data now is kind of being identified as, "Hey, metadata is still great." The secret sauce is metadata, which you guys have been part of. This is the big part of the news of InsightEngine, the community you're putting together. Those are the two big elements of your news. And the fact that you've got the NVIDIA partnership ties it together because you can see where the platform comes in. Unpack all that for us because I think this is a killer point on the trend here, because as agents do come in and they will come in, agents, sub-agents, we've been quantifying that on theCUBEResearch. This is a key part of your news today.

Jeff Denworth

>> So if you think about it, there's two problems that hold enterprises back from using large-scale AI across the business. The first is a model can't be trained with all of an organization's business. I was with a Fortune 10 customer the other day. They said, "We probably have about 100 petabytes of unique data." GPT-3, just to give you a comparison, was trained on like 45 terabytes. And so there's just more data sitting within an organization than what can be cost-effectively trained within a model. The second thing is when you train a model on any amount of data, the thing that it can't do is preserve the security policies on that data that was trained on. So if you and I get the same model and then we have different permissions at the source data level, it doesn't come through. And so the way that we saw this is that there was an opportunity to make what's called RAG or AI retrieval, a lot more simple to deploy across the enterprise. And it starts with unstructured data, which has always been our specialization. And you used the term metadata, and I really like that because I think there are different ways to approach retrieval, but at the end of the day what's happening is that file systems are just evolving. And if you think about it, you now have another form of metadata in the form of vectors that makes your data conversational. When you go down to your source data, you can get answers out of that data, not just lists of files or things like that. You've got a laptop here that presents all the files to you, but it doesn't say what's in those files. It doesn't allow you just run an LLM to ask questions to that data. And that's what's happening. So for 40 years you've had structured and unstructured data. The unstructured data is 20 times larger than the structured data market that sits in enterprise data warehouses and databases. Well, let's suppose that over the next couple years the unstructured data industry will disappear. Why is that? Because now you can query on all of your data, and I think it's just a massive opportunity for organizations to unlock their data.

>> Data should just be vanilla, just data's out there, and in what form, it's abstracted away.

Jeff Denworth

>> Data and metadata in the form of indexes and additional tokens and things like that.

>> Yeah, we've been doing theCUBE for 15 years as you know, and Dave Vellante and I always talk about metadata. Because that was the key to storage, file systems and having data about data. And when you guys talk about the vector embeds in this AI retrieval, which is basically a better word than RAG, which is retrieval augmentation generation, which is an industry term, which by the way is the killer app right now.

Jeff Denworth

>> Yes.

>> And all new trends have a killer app right out of the gate, which is essentially search in this case, but there's more coming. So vector embeds is metadata in a new format. Neural network, whatever you want to call it, it's not the old way, but it's still new data. It's still something that will scale. And I think this is where I think the Jensen tie into LLMs reinventing themselves or technology that will reinvent itself, because the agents will get smarter. So this is where the inference crosses over the causal AI right around the corner. Other trends that are out there are just evolving. So it's going to take time, but you got to get it in the right format. So what you guys have presented, I think what I see is the metadata piece of the new format.

Jeff Denworth

>> And that's where AI also comes in because you have embedding models which are basically LLMs and large video models and other types of models that understand unstructured data that can now read through your source data and create descriptions of that data in the form of a vector. And so that was a big part of it. And the InsightEngine basically puts us together with NVIDIA where our software now supports deployment on GPU based systems with what are called NIMS, which are NVIDIA inference microservices. You have NIMS for video, you have NIMS for text, you have NIMS for proteins and other bioinformatic data. It's going to be crazy. And if you look at the exponential rate of improvement for all these new AI inference models, that all then just gets directed towards your enterprise data and allows you to get a lot smarter about what's inside of it. But the other thing that we announced, and this is a quiet one, but I think it's as radical of a change in the market as the earlier part is we now support for the VAST database a Kafka broker. And so if you think about it forever, you've had data warehouses where you put your analytics data and then you have your event buses where you put all of your real-time data streams and then you try to get one to the other. And so what if you could just stream directly into your data warehouse to the tune of millions of inserts per second? Now you can correlate everything within your environment, and if you can expose that all to LLMs and to retrievers, then it gets really interesting for basically just opening up your business using AI.

>> I think the big innovation is that it's disruptive. You're disrupting features that were companies, now they become features. We're starting to see this new reconstitution of value in the tech stack as well as process. Two things happening at the same time. Process improvement's hard enough, nevermind adopting a new tech stack that might make a company you work with a feature of something else. What's your reaction to that? Because I've never seen it in my career this happening at the same time, a front-end process improvement and back-end implementation so radical and so transformative.

Jeff Denworth

>> We think about trade-offs all day long. And if you look at that tech stack that you're talking about, so many different products represent some trade-off around scale or performance or transactions or analytics and things like this. And we built this new architecture that just allows us to not even worry or think about that stuff. And so as a result, we take the opportunity to make things more real-time and to make them more efficient so that customers don't have to deal with that complexity. But complexity is still a big thing. That's where Cosmos comes in. So today we're also announcing this community of AI data and infrastructure practitioners that can come together and share best practices and we bring together a lot of vendors, a lot of users, even venture capitalists are coming into this community to start to build this AI ecosystem that will help move us all forward a lot quicker. And there's only so much that VAST can simplify, but we all really need to work together to make it better for everybody.

>> It's interesting when you have horizontal disruption that's innovative, community's critical. How can people get involved in this community? Have you guys thought about the onboarding yet? Is it open? How do people get involved?

Jeff Denworth

>> Yeah, so we hired a gentleman by the name of Jonas Rosland to actually be the community facilitator, and he's great. He's worked with a lot of very popular communities in the past. And we're building out all these different platforms where people come in. So for example, we have Discourse and Discord server and we have all sorts of other different venues that organizations can jump into and immediately start just sharing knowledge. Then on top of that, we have both virtual and physical events that we're going to be hosting throughout the course of the next couple years to just bring people together to force them to share insights.

>> And that aligns obviously with the open source community. Again, it's collaborative sport. And I want to get back to AI retrieval, which I love the term by the way, because RAG is the industry term. It's the killer app. But some people will say, "Hey, RAG's not really AI. All it is vector embeds. That's been around for a while." I might answer that is, "No, it's the first use case where there's value being created." And I think the agent thing that ties next, and I want to get your thoughts on where retrieval and agents play together, because essentially faster knowledge sharing and decision-making has been some of the benefits of the productivity we're seeing in AI. So AI retrieval sets the table. So how do you connect the dots between AI retrieval, which a lot of people are getting the vector embed format, the metadata for AI, if you will. RAG's an easy low-hanging use case.

Jeff Denworth

>> It is....

>> Fruit use case. Agents are right around the corner, obviously Salesforce betting the ranch on Agentforce, things like that. So you're starting to see people making the not pivot, but extension to their value. Where does AI retrieval and agents connect?

Jeff Denworth

>> So there are two parts where AI comes into the retrieval process. First, you pointed out vector databases are not new. They've been around for a very long time. And similarity search, which is how you search through adjacent vectors is also ... For example, VAST has been working with similarity algorithms for eight years now in terms of just doing that for data reduction. But in order to create a vector, you need something that understands data and that's where AI first comes in. You have LLMs and protein models and video models that all start to understand data and create those vector signatures that then have to go into a database. And then you have the second one, which is an agent that basically you can use as a translator. LLMs can take what a user may ask and then turn that into a vector search operation or a text to SQL operation. So LLMs obviously take a lot to train, so you can't say that there's no AI in that part of the process, but it's a pretty pedestrian use case anymore because we've all been using LLMs so much that it's just part of our everyday life. We forget how much AI is embedded into that. So yeah, they're used for search, but there's two forms of AI that are really, really hard to build that are deeply embedded into that.

>> And when you look at the data infrastructure, one of the things we've loved about when you guys launched with us in theCUBE a year ago, and again, the results have been great on the VAST side is that it's a new data infrastructure for the next generation AI. That's what we saw. Now you've got the data engine, you've got InsightEngine, you've got the deal with NVIDIA kind of tied in, integrated in. LLMs will talk to each other. We predicted that on our CUBEResearch in our power law. How do you guys see this differentiating in the marketplace? Because you mentioned you had scale on, I think it was trillion transactions you mentioned on your keynote there. What's the differentiation versus the competition? How would you answer that question?

Jeff Denworth

>> I think it's the definition of real-time. So first we think about retrieval as, "Okay, a human has asked a question to some complex data source to get an answer back." That makes sense. But as you pointed out, agents that are chaining with agents means that businesses are going to start working in GPU time. And at that point, I don't think that you can abide a lot of stale data. And so if you think about the way that retrieval engines have been built today, or data warehouses, you always have some other source data that has to be periodically indexed back into where you're asking your questions to. That doesn't work for us. The way that we see it is we don't want AI agents to ever be exposed to any stale data or any stale permissions data so that CISOs are happy, you're always secure, but businesses are powered by real-time data all the time. And that could only be done by building a transactional vector database, a transactional data warehouse and an unstructured data store that is coupled with a runtime and a triggering system such that every piece of data that flows into the system is instantaneously indexed and then updated into this vector database that ultimately LLMs can then go ask questions to. And that's never been done before. And the reason it hasn't been done is because architectures up until VAST, were never designed for scale and transactionality.

>> Yeah, I saw Renen talking about the community piece and how there's a lot of collaboration around these use cases. We'll get to the enterprise side of it, how they adopt it and what the best practices are, but I want to get your thoughts on that whole model being stale approach. You mentioned that again in the keynote. Recency and relevance seem to be the key variables right now in this conversation. And this also ties to the governance piece, which I think is an offensive opportunity for companies to use that, not be defensive with governance, but use as an offensive way to move the needle on value. Compliance, I call it defense. It's like you got to play defense with compliance, do the reporting. But you got this idea that I got to have to make sure the data is recent and relevant. What's your reaction to that? How would you explain that to a customer who's got all of this 3D chess going on in their reset as they think about AI?

Jeff Denworth

>> So a user stores a file called passwords, and then nobody notices and the AI engine goes and does a training or fine-tuning on that data, and then indexes that data. Let's say if it can't be fine-tuned, it indexes that data into a vector database. So you have the answers already in your vector database. Somebody realizes that happens and they say, "Okay, go delete that file ASAP." Now because of retrieval, you also have to go and re-index your vector database. And if that's not directly tied into your source data, that becomes a problem where you do have a defensive problem, I think. And so we basically said-

>> You mean on the governance side?

Jeff Denworth

>> Yeah, for sure. It's a huge risk window that gets created for organizations. And so that's the problematic use case, which is, "Okay, I've got a lot of data that I didn't mean to train or index into a vector database." And so that has to be resolved, which we have. But the second thing that needs to be resolved is, okay, so you've got business decisions that are being made through these AI agents that you've created, and if they don't have recency of data, then they're always seeing some sort of historical look of an organization. And that doesn't work if you want to basically power a business with real-time for recommenders or for security systems or for anything that needs real-time data. It's never been a problem that's been solved until VAST.

>> Yeah, one of the things that's come up, we've talked about vector embeds and we're even doing our CUBE AI stuff. We're playing with our small little model. But when I talk to big customers, they think about scale and then they think about, "Okay, architecture for a foundational rebooting IT." It's not IT in the old way, it's the new IT. And they think, "Okay, what do I got to do?" So there's a little bit of guardrail mentality. It's very DevOps-like, and I want to get your thoughts on how you're experiencing the customers because when they go look at AI, they don't want to think about vector embeds because that's not what they're trying to solve. They're trying to solve a large-scale distributed computing infrastructure problem of, "How do I set the table for the agents that are going to be coming onboard and/or any new application?" We're seeing scalable apps doing stuff with climate that we've never seen before. For example, Climate Week last week in New York, the big conversation was I'm using cloud technology and large-scale data sets. That was an un-gettable application just five years ago. Now you have SaaS apps being agentic-enabled, that's going to be innovative. So the top of the stack looks like it's coming together. I just want to make sure my stuff works. And that's most of the mindset of the top companies we talk to.

Jeff Denworth

>> So we've been doing a lot of customer discovery as we start to build in vector capabilities into the platform. And I remember we had a call the other day with one of the more popular SaaS applications that's used AI to understand data. And they said, "We're going to tell you all the goods and all the bads about every vector product that we've worked with." And at the end of the call they said, "We don't want to ever deal with this again." And so I think that's true. And if you think about this as advanced forms of metadata that are redefining unstructured data systems, then it is basically something that customers shouldn't have to deal with.

>> Let's zoom out for a second. Let's talk about your role at the company as co-founder. I know for a fact we see each other all the time at the key events.

Jeff Denworth

>> We do.

>> I know you're always booked in meetings. So what is the common day in the life? Create the high order bit of what the conversation is like with consistent conversations you're having with people trying to build this next generation adventure with AI and actually getting it right. What are some of the conversations you're having? What are the key issues and what's the key innovation that you guys are presenting?

Jeff Denworth

>> So I think it's fair to say we have two types of customers. We have the foundation model builders that think about just fantastically large scale. Coreweave was part of the Cosmos event, and we've been fortunate to partner with them for over a year now, and these guys are building the big of the big. And so they don't really worry as much about enterprise deployment of AI across their pre-existing datasets, they're building for tomorrow. But then we have essentially the Fortune 5,000 that we work with alongside those organizations that are taking the models that come from them and starting to deploy them at some level of modest scale. And so here, I think it's still very much experimentation. I don't know what your perspective is, but it's just really, really early. And what's happening is, as you go down this rabbit hole, people are realizing that the amount of tools that are required to go and put in this new plumbing is way more than what people would've expected. And so we're every day working to minimize the number of tools that they have to deploy by just putting more and more capability into the platform. And the constant term that keeps coming back is AI readiness. And I don't think everybody totally understands what that means, but they just want to be ready to react to a line of business or something like that that comes in that has some crazy idea that can move their business forward so much faster.

>> It's interesting, we were talking about this on the research team about this point about, "What do we see?" It's interesting because there's different perspectives. The most common perspective we see, and I want to get your reaction to this, is that ... There was a phrase, I forget who said it on theCUBE. "Our innovation strategy is driven by our constraints." And if you look at the constraints right now in the market, the infrastructure capabilities from compute, GPUs are getting stronger and faster, so you're going to see infrastructure increase in performance. The data layer is an area where people are really trying to just get their house in order. So there's a lot of that going on. And then at the top of the stack, like I said, there's a huge developer frenzy, and the agentic wave is here. So I think people see clearly, "Okay, agents will be part of it." Because they can just do the math in their head and saying, "Wow, I can just automate this. I'm more productive." So they can connect the dots on the value proposition of what the app will look like. But when they turn internally, they go, "Wait a minute, we have all these silos." So it becomes a whole nother infrastructure resetting of the foundation of what's capable. So we're in this tipping point. Do you agree with that? Is that consistent?

Jeff Denworth

>> Computationally, you have capital constraints, you have power constraints, you have space constraints in certain cases. And so there is very much a conversation that's happening right now about cloud and next generation cloud. Jensen said to me once, he said ... And I won't name the customer. He's like, "Think about these guys. They're never going to have a data center with 10 megawatts of power that they can just use for new GPUs. So they have to go to cloud." And what's happening is the cloud market is responding to this really quickly, and VAST is powering the next generation cloud with this multi-tenant capability. But the question then is, "Well, all my data's over here and now you're telling me to put my machines over here. What do I do?" And that's why we built something called the VAST DataSpace, which is this new ability to federate all the different places that you compute. So you can keep your data here but compute on it here. And the term oftentimes comes up with data gravity, and I think we've solved for data gravity because customers can't solve for power gravity. And that seems to be the big sticking point right now that most organizations deal with outside of the developer problem, which we don't really work into.

>> Well, the big news is the InsightEngine, and obviously the community piece that you guys are announcing, again, those things go hand in hand. It's a one-two punch. NVIDIA, obviously Jensen Wong with Renen, it was a great part of the presentation. He's great because he's not scripted. He's just off the cuff.

Jeff Denworth

>> He's amazing.

>> He'll play whatever music is on, but it's always about AI and Frontier. And talk about the relationship with NVIDIA, because I think this is important and speaks to some of the things you're talking about, which is obviously they're looking at not just serving those big customers, they're also looking at the enterprise, they see NIMS, their piece. What part of the platform, take us through what the relationship with NVIDIA is ... Obviously eight and a half years. But on this news, what's the innovation? What's the key element of the NVIDIA platform that's accelerating your growth?

Jeff Denworth

>> Sure. NVIDIA likes to use the word real-time as much as we do, because they see that requirement from customers. And so what we've done is about ... I don't know, maybe around GTC time, so I'm going back to March or so. We started having conversations about how our platform can be extended. And if you think about this not as a storage system, not as a database system, but rather just like a data flow computer, then what they realized is that there would be an opportunity to actually not just do fast search through a dataset, but rather to do fast search on a dataset that has ultimate recency, to use your point. So the first thing they did is they exposed us to their NIMS agenda, and now you can deploy those in VAST systems using us as basically the runtime or operating system for AI. The machines that deploy these NIMS obviously need to be powered by high performance NVIDIA GPUs. So we've added support for that within our system. And then what you have is enterprise infrastructure with all the enterprise features, both from a file management and from a database management perspective that you can deploy at massive scale for that consolidation point that you talked about earlier, now with all this NVIDIA instrumentation and software that just runs directly in the system and is deployed by the system. So this is big for them because from a partnership perspective, they think, "Okay, the next big hill we need to climb is enterprise." And if you look across the enterprise, you just have legacy systems that aren't designed for this level of processing and reprocessing. So yesterday I was talking on Slack to the team and I used the term, "Retain to retrain." And if you think about that data retention, as these embedding models get better and better, it's going to be the key to making organization smarter because now you just have this massive corpus that you can just go train and retrain on all the time. And I think NVIDIA is excited for that. They now have an infrastructure partner that can match to the scale and ambition that they have.

>> I love that, "Retain to retrain." I think that points to the wave we're on. And my final comment I want to get your thoughts on is that if you look at the innovation, it's not just an IT issue, not just get the technology and get the data right, there's all this process, because the workflows are going to be affected by the applications that either have an agentic or more large scale to them, obviously powered by the data. And it's like watching the construction of a new bridge to cross over to the next wave.

Jeff Denworth

>> While the cars are on it.

>> Now, so you got the old bridge, and now the bridge to the future, the AI adventure that you guys are talking about. You got new players like NVIDIA ... Well, they're an existing player, but you got the innovators are new set of players, and some existing ones. So the customers are seeing this bridge to the future. What's the roadmap look like? If I'm the customer, I'm thinking, "Okay, I love VAST, I love NVIDIA. I love what they're doing. It hangs together with what we're trying to do." What's that bridge look like? Take me through the roadmap, what's next? Obviously NIMS with NVIDIA, what's that going to enable? What's your vision for what's next? What's on the roadmap? Can you share your thoughts of as this bridge gets constructed to cross into the future, what's that look like?

Jeff Denworth

>> Well, I think actually Cosmos gives you some breadcrumbs to that because as we point towards this future where you can envision this geo-distributed AI operating system that is basically the runtime for all this, then we recognize that customers will have extremely diverse infrastructure and you're going to have just a flood of new embedding models and different types of AI models that come into agentic organizations. And so our objective is to essentially be universal. We want to be on every platform and we want to be in every single data center, on-prem or in the cloud that a customer may want to compute in. And we want to minimize the amount of complexity that customers have to deal with so that they can essentially get this thinking machine's experience that we're trying to build towards. And so that is a ton of work that we have to do on all different levels in order to build out to what customers require for all their data, all their infrastructure, and all the places they want to compute. That's a lot of work.

>> Love that, thinking machines. It's like, I got to think, I got to learn, I got to get reinforced. And all that is right in line with what's going on with AI. It's beautiful.

Jeff Denworth

>> Yeah, if you think about it, you have the data that you have in your organization, those are your memories. You have the new data coming in, those are your experiences. And when you correlate these two, this is when you have realizations. And that's the thing that we're excited about.

>> Well, Jeff, congratulations on an awesome event, and again, continued success. The VAST Data team's continuing to grow. I know you guys are working on some really big projects and as well as plowing the fields for the enterprise. Final point, I'll give you the last word. For the folks out there watching that are AI builders or people trying to set the table for the future, why VAST? Why are you guys relevant?

Jeff Denworth

>> Why VAST? Well, if you think about this idea that machines are now working through data at the pace and the scale that NVIDIA is, I think it's very clear that legacy architecture concepts aren't necessarily applicative. And so we've been on a mission since day one to start from a new foundation that allows people to have their data strategy rise up to their application strategy. And I think our success to date is evidence of the fact that we're doing some things right. So the guidance is don't do it alone. Talk to us about what you want to do, not the constraints that have been imposed by infrastructure of yesterday, and it'll be a fun conversation.

>> A lot of builders on the back end and front end. Jeff, thanks for coming on.

Jeff Denworth

>> Thanks, John.

>> This is exclusive coverage here with Jeff Denworth, the co-founder of VAST Data, part of Cosmos, their big program. We're announcing big news here. Up next, theCUBE analysts are going to break down all the action, give their analysis of what's going on with VAST Data. We'll be right back.

Jeff Denworth, VAST Data

Jeff Denworth

>> Yeah, yeah, it's been pretty crazy times for us.

>> So Cosmos has just had ... you guys just did your presentation, Jensen Wong was on there from NVIDIA.

Jeff Denworth

>> Yep.

Jeff Denworth

>> Data should just be vanilla, just data's out there, and in what form, it's abstracted away.

Jeff Denworth

>> Data and metadata in the form of indexes and additional tokens and things like that.

Jeff Denworth

>> Yes.

Jeff Denworth

>> It is....

Jeff Denworth

>> You mean on the governance side?

Jeff Denworth

>> Let's zoom out for a second. Let's talk about your role at the company as co-founder. I know for a fact we see each other all the time at the key events.

Jeff Denworth

>> We do.

Jeff Denworth

>> He's amazing.

Jeff Denworth

>> While the cars are on it.

Jeff Denworth

>> Love that, thinking machines. It's like, I got to think, I got to learn, I got to get reinforced. And all that is right in line with what's going on with AI. It's beautiful.

Jeff Denworth

>> A lot of builders on the back end and front end. Jeff, thanks for coming on.

Jeff Denworth

>> Thanks, John.