In this interview from the Nvidia GTC AI Conference and Expo in San Jose, Vrashank Jain of Dell Technologies joins Steve Kearns of Elastic to talk with theCUBE + NYSE Wired's Gemma Allen about how their partnership is turning decades of untapped unstructured enterprise data into fuel for AI agents and production AI factories. Both companies earned a mention in Jensen Huang's keynote, and Jain explains that with over 4,000 AI factory customers, Dell has watched enterprise challenges shift from power and cooling to the data problem. Kearns highlights how Elastic's hybrid search — combining keyword and vector retrieval — ensures agents pull the right context from massive unstructured data stores, preventing compounding errors as agentic workflows loop through multiple reasoning steps.
The conversation also explores the security imperative driving a dramatic shift to on-premise AI. Kearns details how Elastic enforces document-level security at the data store itself, ensuring that agents can only access what their credentials permit regardless of the application layer calling them. Jain describes a fundamental change in how enterprises think about data: the destination is no longer a dashboard but an agent, pushing organizations to build modular "data products" scoped to specific contexts and governed at the product layer. Real-world use cases range from manufacturing technicians querying decades of equipment documentation to healthcare providers summarizing clinical notes. The pair also touches on sovereign AI and smart city initiatives, where municipal planners in emerging cities are beginning to leverage data platforms for real-time decision-making. From Dell's position as the first OEM to ship the GB300 to Elastic's GPU-accelerated vector database, the discussion provides a practical roadmap for how tightly integrated infrastructure and retrieval technology can close the gap between AI promise and enterprise production outcomes.
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
NVIDIA GTC 2026. If you don’t think you received an email check your
spam folder.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Register for NVIDIA GTC 2026
Please fill out the information below. You will receive an email with a verification link confirming your registration. Click the link to automatically sign into the site.
You’re almost there!
We just sent you a verification email. Please click the verification button in the email. Once your email address is verified, you will have full access to all event content for NVIDIA GTC 2026.
I want my badge and interests to be visible to all attendees.
Checking this box will display your presense on the attendees list, view your profile and allow other attendees to contact you via 1-1 chat. Read the Privacy Policy. At any time, you can choose to disable this preference.
Select your Interests!
add
Upload your photo
Uploading..
OR
Connect via Twitter
Connect via Linkedin
EDIT PASSWORD
Share
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
NVIDIA GTC 2026. If you don’t think you received an email check your
spam folder.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Sign in to gain access to NVIDIA GTC 2026
Please sign in with LinkedIn to continue to NVIDIA GTC 2026. Signing in with LinkedIn ensures a professional environment.
In this interview from the Nvidia GTC AI Conference and Expo in San Jose, Vrashank Jain of Dell Technologies joins Steve Kearns of Elastic to talk with theCUBE + NYSE Wired's Gemma Allen about how their partnership is turning decades of untapped unstructured enterprise data into fuel for AI agents and production AI factories. Both companies earned a mention in Jensen Huang's keynote, and Jain explains that with over 4,000 AI factory customers, Dell has watched enterprise challenges shift from power and cooling to the data problem. Kearns highlights how Elasti...Read more
exploreKeep Exploring
How do vector databases and vector search help get the right parts of large amounts of unstructured data into an LLM's context window, and why does that matter for LLMs or agents that perform iterative retrieval and tool-calling?add
How has the emergence of vector/semantic search changed how people search documents, and why is a hybrid approach (keyword/metadata plus vector search) valuable?add
How did the partnership between Dell and Elastic come about, and why did Dell choose Elastic to address the data challenges involved in onboarding users to AI applications?add
What developments have made it easier for models to retrieve and reason over the right information from unstructured organizational data, and why did earlier approaches sometimes fail to deliver?add
>> Welcome back to theCUBE. We are here on the ground in San Jose. It's NVIDIA GTC 2026, and there is so much action happening. Joining me now are two folks who made the keynote yesterday for both of your prospective companies. We have Vrashank Jain from Dell Technologies and Steve Kearns from Elastic. Welcome guys.
Steve Kearns
>> Thanks for having us.
Vrashank Jain
>> Thanks for having us.
Gemma Allen
>> Well, I think there's a lot of folks here in this huge ballroom that would love to have their companies mentioned in Jensen's keynote, right?
Vrashank Jain
>> Yes.
Gemma Allen
>> The guy is practically Jesus, let's be honest. So talk to me a little bit about this problem that you have come together to solve. It has seemed to me as though data and getting value from data has been a long-term problem with a lot of promise and not a whole lot of realized outcomes. What's changed? I'll start with you.
Vrashank Jain
>> No, 100%. You're right, it does take an absolute team of effort and many weekends to get it into that level of the keynote. So we do appreciate NVIDIA doing that for us. I think Jensen sort of made the case yesterday, which is, he started off by talking about the data. He started talking about CUDA, and he immediately pivoted to talking about cuVS and cuDF. And that's basically because I think even NVIDIA and everybody else has started to realize, the way to get AI factories into production at enterprises has to be starting with solving the data problem. NVIDIA has a vested interest in doing that, obviously, but it requires an entire ecosystem of companies and tools coming together to actually make it happen. I think unstructured data, to your point, it's not a new problem. We've been talking about this for 20 years at least. Enterprises have been storing videos and documents, and images for years, hoping that there's a killer use case that's going to come around at some point. It's here, and now we're trying to go solve that problem. It was about time to finally come up to trying to solve it, and we're here.
Gemma Allen
>> And for you, Steve, to your point, it's a problem that's existed for a long time. It's also been considered a behavioral problem. Elastic has kind of really tried to fundamentally shift that to almost force behaviors a certain direction with vector image search. Talk to me a little bit about what you've been working on, Steve.
Steve Kearns
>> Yeah. So I think when we think about the work that we've done as a vector database, the work that we've done with Dell in partnership with NVIDIA, we really think a lot about how do we take that massive amount of unstructured data that businesses have. And LLMs make it useful, but how do you actually put it to work? How do you actually get the right part of all of that unstructured data to the LLM to solve an actual task that somebody's trying to accomplish? And what vector search does is it gives you a better set of tools to get the right, relevant context into that LLM's context window so that as you're asking a question, the model, the LLM, the agent, can go retrieve the most relevant subset of that information, bring that back, and then answer your question in an accurate way. This was always important. This is important when we were showing people 10 blue links. But when you start to say not just here's an agent with one shot question or a one shot answer, when you start to let that agent loose to call these tools in a loop, run retrieval, take the answer that it got from the first question, ask a second question. At any point in that process, if you're giving it the wrong context, the wrong information, it's going to make a wrong choice, and that starts to become a bigger and bigger impact on the business.
Gemma Allen
>> So talk me through an example for a second.
Vrashank Jain
>> Sure.
Gemma Allen
>> I am working in FP&A team. I am searching, as a human, contract. I want to know what contract do we have with Dell. In this world, I am asking this machine as an agent, who do we do business with in this space? What sort of ways is it changing how we think about the context behind the data?
Vrashank Jain
>> Yeah. I think it changes the way that you would think about the problem. A few years ago, we would have set up a search database in which we put all the documents, and the only way that people could search was searching for Dell, searching for contract, searching for older than 30 days. And that's kind of it. The only keywords were anything that the metadata was indexed on. Now the search changes to tell me anything that has Dell, Dell Technologies, Michael Dell, it has a contract about PowerEdge. All of that is contextually semantically similar to Dell. So it democratizes your ability to go search for any document, and that's where vector search really, really comes in. So the combination for metadata search or keyword search, and vector search is really the key here, which is I think... Elastic calls it the hybrid search technology. I think that's one of the big reasons why we picked Elasticsearch as our key partner in this effort, because we were looking for a partner who could bring both of those technologies to bear.
Gemma Allen
>> So talk about how the partnership came to be. You at Dell have over 4,000 AI factories customers, right?
Vrashank Jain
>> And counting.
Gemma Allen
>> And counting. I am sure, as those models are reiterating, learning, Dell is also gathering a lot of knowledge about what's working and what's not working across enterprise. How did you guys decide, okay, we have a problem here, and Elastic can perhaps solve it?
Vrashank Jain
>> Absolutely. I think the fact that we have 4,000 and counting customers gives us a unique view into the enterprises, and we've realized that the problems that were there two years ago, power, cooling, rack space, have gone away. It's really come down to how do we bring more users onboarded to AI applications, and that always comes down to a data problem. We were looking internally, we were looking externally to look at the ecosystem and say, which technology partner can help deliver that best outcome. We had a lot of internal experience with Elastic already. And like I said, we were looking for the best of breed, a best keyword and vector search company, and we landed with Elastic.
Gemma Allen
>> And for you, Steve, I mean, you have been solving this problem in different eras of tech, right?
Steve Kearns
>> We have.
Gemma Allen
>> Data, like we said, is not a new problem, but AI, it seems as though has created this magic solution, almost inserted data fairy that can fix everything for us. What happens underneath the hood? How is this suddenly becoming such a viable change?
Steve Kearns
>> Well, I think a couple of things happened simultaneously. I think the quality of retrieval, the quality of being able to retrieve the right information for a given question, it got a little bit easier as embedding models and ranking models, and vector database technology became better. And then you also saw at the same time, these reasoning models, they can start to run for longer. They can take a question that the user has and then start to call tools around the organization to figure out, hey, what data do I actually have in the system. What data might be able to help me answer this question? Now, how do I go find the piece of information that I need? And it was one thing when this was fully structured data. In some ways, it's like running SQL queries is still hard, but fairly straightforward. But when you start to do this now on unstructured data, if you don't have the tools, if the engine that you're using, the tools that are available to the LLM, if they're not smart enough, if they don't have enough capability, you're not going to be able to get good results. And that's where the promise sometimes hasn't matched to the, I don't know, the delivery. And I think what's changing is the set of technologies have now reached that point with Elastic as a vector database, accelerated with NVIDIA's GPUs, we can now solve a much wider range of these problems.
Gemma Allen
>> So talk to me about some enterprise use cases. I'm interested to understand some real tactical examples of how this has been super effective.
Vrashank Jain
>> Sure. We can talk through a few. Think of yourself as a technician in a manufacturing environment, walking around the shop floor, and you're looking to solve a problem with the particular equipment. And the answer is buried in decades-old documentation and standard operating procedures that you don't have the patience or the knowledge to go search for. It's a perfect use case for this. Being able to have a chatbot at your hands that you can use to search through all the documentation to find that one nugget of information that tells you this is what's wrong with that equipment, here's how you go solve it. That's tremendous. You can extend that to healthcare, where doctors are dealing with millions of notes for every single clinical conversation they're having. Being able to search through and summarizing all of that information is tremendously valuable for doctors.
Steve Kearns
>> And one of the interesting things, I had to give a talk earlier today with Rama, who's the head of AI and ML at NVIDIA's IT department. And one of the things that she had pointed out that's critically important as another layer of that story is actually the data security part of that. And so for these enterprise systems, part of it is like, can I get the data? But then the second part is, should I be able to get that data? And so when you think about what does it take to build these enterprise platforms that are suitable for businesses that are in regulated industries or really any industry, you have to be able to respect every aspect of how that organization operates right down from document level data security all the way up through auditing and traceability of all of the activities that happen within that agent.
Gemma Allen
>> It's so true. The exposure element of this. If you think about OpenClaw as an example, if you were to allow OpenClaw into your enterprise tomorrow, my God, this exposure would be absolutely huge, right?
Vrashank Jain
>> With great power comes great responsibility.
Gemma Allen
>> For sure. So when you think about that, about the security challenge and realizing value, but also in a governed way, how does that play into the conversations you guys see, especially in terms of companies building their own version of Copilot, leveraging the LLMs? Where does the rubber meet the road there? Where do we see the cutoff?
Vrashank Jain
>> Yeah, I think that's one of the big reasons we're seeing a dramatic shift to on-premise. If you look at the most closely guarded secrets and the most valuable information for any major enterprise, it's within their data center, it's within their control. They don't want to let that go, but that also doesn't mean that they don't get access to the amazing piece of technology that's available. They just need to find a way to point it at that closely guarded data and keep it secure within their environment. That's really why Dell is really focused on this, because frankly, a large majority of those data centers are running our equipment, that data is sitting on our storage. It's our imperative to go and help those customers get value, and that's really why the partnership.
Steve Kearns
>> And in these environments, you have to think about security at every layer of the stack. In the olden days, when we used to build big applications and things like that, you would implement at the application level, this kind of user-level security and things like that to filter the results that they can see. But when you start to unleash an agent on all of these different data sources, who's implementing that security? And so, one of the nice things about the partnership and the way that Elastic operates, we actually implement the security at the data store level. So when you show up with your credentials, you can only see the documents in that deployment that you yourself should be able to see. When you grant your OpenClaw privileges to go see things, it can only see what it is allowed to see. And so that idea of enforcing the security at every layer is really, really important. We actually sort of think about it the other way. You shouldn't enforce it at the lowest possible level, and then it doesn't matter what the higher levels are. OpenClaw's not going to get access to any data it shouldn't because it's enforced by the credentials of the OpenClaw to access the system in the first place.
Gemma Allen
>> And let's talk for a second about the orchestration layer of having multiple clean data sets that are highly valuable across enterprise. In some respects, some of the arguments right now about this SaaS apocalypse, the term we hear a lot at The New York Stock Exchange, is the fact that these systems are modular, they're siloed. You have a world where an LLM can do one thing for you in one and can do one in another, but actually, if you could realize data value from elements of those systems in a modular way and knit that data together, it could drive a huge amount of value perhaps, because it's contextual and it's relevant to you and your industry. What are you seeing there? How do you think enterprises are thinking about this?
Vrashank Jain
>> I think there is a change coming, I'll describe it in this way. In the traditional way, most of our data sets ended up in a dashboard. Most of our data sets ended up in a data warehouse or a data mart, and it was destined for a dashboard. In the new way of doing things, our destination is an agent. It is not a human consuming this data anymore; it's actually an agent. And so what we're now starting to see are customers pushing us into allowing them to build data products. Data products that are particularly suited for agents with a given amount of context, and that makes the whole data exposure modular, because each data product is based on a particular vertical, a silo, a context, and that's really what you're exposing. And if you apply your security rules on that data product layer, you now have a much easier way of exposing this data in a controlled manner than saying, "We'll give you access to the whole warehouse, we'll give you access to the whole index." That's not really where people want to go next.
Steve Kearns
>> Yeah. And I think that one of the things that's changed, it's back to that: Are we building an application, or are we making data available to agents? Because when you had to build an application, you had to go and build it. You had a team, you had people, you had to plan it, it had a life cycle. Where now it's saying, "How do I safely, securely make the right data available to these agents?" And now the agents can now decide how to use that data
Vrashank Jain
>> Because that application maybe hasn't even been built yet. It's about to get built because an LLM is about to think about it, which means we can't design this for any one given application anymore. We have to design it for agents who can do pretty much anything we give them.
Gemma Allen
>> So let's talk for a second about the growing customer journey here and the buyer persona for this. You guys do a lot of work with enterprise. We hear a lot about sovereign AI, too. It's a very, very hot space right now. There are governments, in my opinion, of all use cases that could really value from having a better data management system, and leverage their data in a way that affects citizen outcome. What are you guys seeing in that space? Is it a space you're combinedly going after? Do you think there's good opportunity there?
Vrashank Jain
>> I actually do think so because I think it's... Maybe about 10 years ago, the concept of smart city really came out, but nobody really knew what smart cities really were.
Gemma Allen
>> I mean, I still struggle, but, yeah.
Vrashank Jain
>> We still struggle. Because people thought is it big brother? Is it video surveillance? Now I think the smart city concept is becoming a little bit better because city managers, municipal planners, traffic planners now want to be able to search through and access citizen data, anonymized, of course, but they want to be able to go search through because they want to make better real-time decisions on cities. That's what it means for a city to be smart.
Gemma Allen
>> For sure.
Vrashank Jain
>> It means it to be agile, which means it needs access to good data. So a data platform is now starting to become very relevant for smart cities all over the world. We have some experience with some emerging cities in Africa that are starting to do this already. We think this is going to grow quite a bit.
Gemma Allen
>> So very exciting week for you both. I mean, it was a great GTC for Dell and Elastic. You guys also had the Blackwell announcement, first OEM to ship GB300, big, big deal, right? Dell is definitely in the line. You're in the futuristic space in Jensen's mind, which everyone wants to be. So talk to me a little bit about where you go from here, how you stay on top of a curve where competitors are coming at you so fast.
Vrashank Jain
>> Oh, yeah. That's the one thing we haven't talked about, which is, at this point, everybody has a data platform. I can understand how it gets a little harder to distinguish between one vendor versus the other. But no, you're right. I think for us, really, the opportunity is continued success with AI factory drives demand for data platforms, drives demand for best to breed capabilities, drives demand for partners like Elastic.
Steve Kearns
>> That's right. And I think it all comes back to the outcome. If your customers are successful with the combined products, we're going to have a lot more customers, and I'm a big believer that the state of the technology today, we're going to see that happen very quickly.
Gemma Allen
>> Well, folks, great to chat to you. Happy St. Patrick's Day as well, here from San Jose.
Vrashank Jain
>> Happy St. Patrick's.
Steve Kearns
>> Likewise.
Gemma Allen
>> I hope you're going to have a Guinness later. I know I will.
Steve Kearns
>> Absolutely.
Vrashank Jain
>> For sure.
Gemma Allen
>> Thanks so much.
Vrashank Jain
>> Thank you.
Gemma Allen
>> I'm Gemma Allen here on the ground NVIDIA GTC 2026, live on theCUBE with Dell and Elastic. Stay tuned.
Gemma Allen