Dell Technologies World 2026 | Varun Chhabra & Rajesh Rajamaran, Dell Technologies

Clips
More from Dell Technologies World 2026

Varun Chhabra

SVP, Product Marketing

Dell Technologies

Rajesh Rajamaran

VP CTO Dell Storage, Data platform & Data Protection

Dell Technologies

play_circle_outline Data Bottleneck to Breakthrough: Dell AI Factory with NVIDIA Orchestrates Curated, Unified Pipelines to Fuel Agentic AI

play_circle_outline Storage evolving into a context layer supporting knowledge graphs and operational agents.

play_circle_outline Data lifecycle perspective: multiple media, federated sources, and continuous augmentation.

play_circle_outline Turnkey AI Orchestration and Model Routing: Diverse Portfolios for Tiered Cost-Performance with Dell Automation

Info
Transcript

Varun Chhabra & Rajesh Rajamaran, Dell Technologies

Varun Chhabra

SVP, Product Marketing Dell Technologies

Rajesh Rajamaran

VP CTO Dell Storage, Data platform & Data Protection Dell Technologies

In this interview from Dell Technologies World 2026, Varun Chhabra, senior vice president of product marketing at Dell Technologies, and Rajesh Rajamaran, chief technology officer of Dell Storage at Dell Technologies, join theCUBE's John Furrier and Dave Vellante to discuss how data has emerged as the hidden bottleneck standing between enterprise AI experimentation and full-scale production deployment. Chhabra frames data as the most overlooked constraint in AI adoption, noting that across Dell's 5,000-customer AI Factory program with NVIDIA, the gap between ... Read more

explore Keep Exploring

What must organizations change about their data practices — and in what sequence — to move from proofs of concept to enterprise-scale, agent-driven AI deployments? add

How should an organization approach building and operationalizing agentic AI systems—what use cases should it start with, how should it handle data ingestion, storage/knowledge graphs and transformation, and what technical considerations (e.g., accelerators) matter? add

How is the platform/architecture designed to support heterogeneous, federated data sources and governance, provide storage-as-a-service (file/object/block), and incorporate AI augmentations without moving data? add

Yes. A neutral form of the question the text answers: How is your company making AI and infrastructure deployments more turnkey for customers, and what role do platforms (like the Dell Automation Platform and AI Data Platform) and an open partner ecosystem play in that strategy? add

bolt Powered by CUBE AI

Varun Chhabra & Rajesh Rajamaran, Dell Technologies

search

John Furrier

>> Welcome back to theCUBE's live coverage here at Dell Tech World 2026. I'm John Furrier, Dave Vellante. The AI infrastructure is at the center stage and the data is the key. This next conversation is going to tee that up. Varun Chhabra, SVP of ISG marketing at Dell Technology, and Rajesh Rajamaran is VP of CT of Dell Storage, AI data platforms and data protection. You got a lot on your plate. Thanks for coming on. Good to see you.

Varun Chhabra

>> Good to see you all. Yes, John.

John Furrier

>> All right. So people want to inject intelligence into their business and that's a high level kind of statement. It's very C-suite-like. The CFOs are now involved. The chief human resource officers are involved. The CIOs involved. They're all like, okay, how do we do this? We heard on stage, it's the data, Varun. This is the key to success, but it's not your grandfather's data-based strategy. No. It's the modern version is here. Take us through that.

Varun Chhabra

>> Data is the hidden bottleneck for enterprise AI adoption. We talked about 5,000 customers with the Dell AI factory with Nvidia. And what we're finding again and again is the difference between a successful POC, and a successful at scale deployment, often comes down to, are you able to access all your data? As Jeff talked about today, it's spread all over the place. Many different systems of record, often unstructured, needs to be tagged, curated, and fed into the data pipelines that go into your models and your AI applications. That is a very, very non-trivial task, and that's a huge part of what we're doing with the AI data platform that Dell has with Nvidia. A lot of the work, not just at the storage level, but data ingest, data curation, orchestrating the entire data pipeline, that's really the focus to make sure that data gets fed into the GPUs, so that intelligence and business context, John, which is really what data is, makes the AI outcomes more powerful.

John Furrier

>> The enterprise AI raise is really an infrastructure race. That's clear from the show. When you peel that back, it comes back to the data. So the next question is, "Yeah, but I've been doing data links. We're doing all this other stuff. What's the change? What's the sequence? What's the linkage or transfer modern pivot or not pivot, but like...

Rajesh Rajamaran

>> I think like you started, it was about the idea of business of intelligence, but the business of intelligence has to leave for an outcome. Let's say the outcomes are driven by agents. That's the end state in which you operate, which means you need to be able to create a plumbing pipeline to be able to feed that intelligence into the agents for operations. And it starts from, as Varun said, from first of all, discovering what is your data. And yes, it comes from structured and unstructured. And it isn't just about file or objects. It's also about databases or other structured data means CSVs, JSONs. For example, all of those things make up, materially, the intelligence. Now the key is bringing it all together into... Agentic doesn't know the difference between structure and unstructured. It needs to operate on data intelligence. So you need to be able to funnel both of this to a common outcome, whatever its operations is. I think from discovery to injecting the fact of provenance, what is this data? Who's this data? Where's this coming from? How do I make this data accessible into a mesh? How do I then transform this data into something more useful along the way? Because it's pipelines along the way. It's not just about agents. And then from there, you need to be able to connect and create an agentic pipeline. Therefore, context building, knowledge building, which then in each of these stage requires a lot of technology to be brought together, and like I said, hardness, to be able to materially bring this from an AI readiness perspective. And there's a lot of technology that we have put together and we work with a lot of partners. We believe that as data is important, data is not in one place, which means we have to bring together all forms of data, therefore our partnership, and then we build the scaffolding to put it all together for the agentic outcomes.

Dave Vellante

>> So the media coverage is all energy, GPUs, CapEx models. That's where all the action is in the media. You're making the case that the data is the bottleneck, that data and architecture have to be a first class citizen. That's right. A first class buying decision. So make the case to the customer. What are you telling the customers to why that premise holds water?

Varun Chhabra

>> Yeah. I'd say, Dave, it is a bottleneck, an important bottleneck. The other things you mentioned are obviously things organizations have to work through, but data is often the hidden bottleneck. It's the thing that people don't often think about. So, if you're not able to tag, curate, get your data strategy right, you're not feeding that into your GPUs, you're feeding half-baked information, right? Ultimately, what is one of a huge strategic advantage every organization has is the collective knowledge of the organization about their industry, about their competitive advantage. That is all encapsulated in data. So the more data you can feed into these models, the more likely you're going to be able to succeed. But as Rajesh was talking about, that's not easy to do. It's not trivial.

Rajesh Rajamaran

>> Yeah. And as you said, all the things you said are important. So think of it as an analogy for infrastructure as pipelines. Imagine if you have gas pipelines, they're all important. Without gas pipelines, you can't do much. But what's also important is the data inside the pipeline, for example, it's the fuel. So data's the fuel that powering all the infrastructure and they both have to come to meat. You need the infrastructure to be able to support and sort of harness the fuel. As I said, you need the engine and the infrastructure to be able to harness the fuel .

John Furrier

>> Okay. So let's take that pipeline example because we've heard the term, geez, like a decade, maybe more. Data gravity. Okay. Data gravity. And we hear you guys make this bold statement, move AI to the data, don't move the data. Love that. I think that's correct. But data pipelines have gravity. I built my pipeline and my agent doesn't know what that oil pipeline is or the gas pipeline. So there's syntax in these pipelines. Is that gravity, or is that an opportunity, or just kind of like keep it there and make that a deterministic thing? I mean, this has come up a lot. It's a very nuanced point, but it's come up a lot.

Rajesh Rajamaran

>> I mean, data's gravity is usually used in the context data. It's very important for us to live. I don't think it should be used as an inertia of any sign, right? It is what binds us together, so it's important. But at the same time, as you said, the ability for you to operate the data and sort of pipeline it, as you said, is key. So I think yes, data has gravity and then it is the fabric in which we-

John Furrier

>> But do the pipelines have gravity?

Rajesh Rajamaran

>> Yes, they do.

John Furrier

>> And how do you deal with the old pipelines?

Rajesh Rajamaran

>> So what you heard from Arthur talking about was, we had the idea of data engines. So one of the data engines curation we've taken is, you don't have to move data. The problem usually has always been around meeting to move data. So the engines that we've worked with closely does not require you to move the data. It actually did analyze in place or transform in place, which is a key part. Correct. Therefore, the gravity part where, oh man, I don't want to move my data to be able to create intelligence was the yardstick, was sort of the impediment before, which is removed. Now, let's say you have the proper vehicle for you to transform unstructured and structured data to turn into knowledge. That is a one-time snapshot. You still need something to orchestrate because it's a close loop and therefore the orchestration of data loop, the enrichment aspect of it is very important for us to bind the dots, therefore the pipeline part.

Dave Vellante

>> Well, in pipelines, I think a pipeline is workflow and workflow and process have gravity as well. And this is where you start to bring in the knowledge graph and you start to bring in context. Storage is now this context layer. That's what it's becoming. How are customers thinking about that? I think Varun, I interpret from your conversations, the hidden bottleneck. So customers really aren't tuned in to... I mean, they're focused right now probably on things like, well, we got to get more utilization out of our GPUs. They don't have time to think that far ahead, but so how do we get them from where they are today to where they really want to go, which is that really taking that tribal knowledge, that context of an organization and embedding it into agentic.

Varun Chhabra

>> I think Dave, it starts with, what we found is it starts with once you have an anchor on what your key use cases are going to be, whether it's coding assistance, whether it's could be support assist, et cetera, et cetera, that really crystallizes thinking in terms of, okay, what does my stack look like? And then the data conversation flows right from that. So it really starts with some of the conversations we have with customers around, "Hey, you have like 200 things you're looking at. Let's pick the most important ones." And then automatically then you start to think about, okay, what are the data sources that need to feed into this? Is it structured data, unstructured data? How do I ingest that data? Do accelerators play a role, like the work we're doing now with Nvidia with QDF, QVS, et cetera? How do you think about all of that? I think it all starts with just having a key understanding of, a clear understanding of what are the things I'm trying to solve.

Rajesh Rajamaran

>> Varun said it very well. So you always start with the use case first, therefore it sort of gives you a bounds of what data you need. Then there's the transformation of the data. Now usually we use agentic in sort of very nebulous terms, but actually in its use case, there's two parts. There is the use of agentic technology to create the knowledge in the first place. Then there's the operizational aspect of taking knowledge and make it functional using agents that present sort of a business value outside. And both are important from a data perspective, to create intelligence behind the scenes. And the product of that is the knowledge graph, as you said, and the knowledge graduate embedded into an operational agent, which then manages the business. So they're both integral from the use case.

John Furrier

>> We've been following storage, networking, compute, databases for many years, you guys know the old way. The new ways, the list is this long. There's a lot going on. Cyber resilient, you name it, everything's in there. But I want to talk about the data platform with Nvidia, and specifically the co-engineering piece. What's different? Can you share, because most people say, oh, another data platform. What's different, because NVIDIA is moving the needle on many fronts. We cover them too. What's the partnership engineering? Share some specifics.

Rajesh Rajamaran

>> I think the partnership with NVIDIA, I think we can break into many factors. There's a software ecosystem part which is from a data element. So we work very close with NVIDIA on acceleration of each of the, sort of knowledge curation. So let's say the QVS, QDF, for example, these are acceleration including Lightning CMX, et cetera. These are which we use software means to sort of accelerate the search of creation of knowledge or curation of knowledge. Then there is, how do you operationalize that knowledge into coding assistance, Nemoclo, for example, how do you then put this into production with OpenShield access, for example. There's the part two of, how do you then, now that you have your data, how do I run the machinery? And both of which are a very close partner with NVIDIA and there's software and hardware components that accelerate both these pieces.

Dave Vellante

>> I want to ask you guys about your philosophy. I had the pleasure of, I think it was last October, it seems like quite some time ago, given all this transpired, to get the storage deep dive. So we got the preview of what you guys were announcing today. And what was very clear is rather than saying we're going to build the God box, one box for all use cases, file, object, scale up, scale out. You said, "No, we're going to focus on purpose built." I want you to take us through that decision. There was probably some trade-offs in doing that, but why did you decide to go in that direction, and what's your expectation as to how adoption will occur?

Varun Chhabra

>> I'll start at a high level and then Rajesh, I think you can probably walk through some of the technology pieces, the trade-offs. This is not that... You guys have been following us for a while. This is not that it is completely in line with how we think about an open ecosystem, right? Especially in the world where we are today where data curation and these things are still very much in the early stage, I'd say there's a lot of work to be done to solve this problem in a turnkey fashion. It's very hard to predict what... You don't want to back organizations into particular corners by making technology trade-offs that are vertically integrated. So what we're trying to do is get the balance right between giving customers choice, whether it's at the storage level, you want to go largely with file, object, or even with extreme parallel file systems with Lightning. And then on the data curation side, how do you want to do structured? How do you want to do unstructured? How do you want to do querying? And SQL queries, how do you want to think about orchestration? We want to give customers that flexibility, but at the same time, do it in a way where when they opt into the ecosystem, it's easy and simple for them. I mean, that's a huge part of the design decisions here.

Rajesh Rajamaran

>> I think, as a technologist, you're into always build the unicorn that does everything for everybody, but we have to be practical here, right? So you have to also put the practical angle of, how does a data center operate? Who owns what, for example. So what we realized was if data is the center of this conversation, you have to say, does one person own data? No. Is data in one place? No. Is data in even in one format? No. Correct? And you asked about the data gravity. The issue with data gravity was, people try to create a system that say, "Migrate everything to me, I got everything covered." But that is not realizable in sense. So what we realized was even each layer was curated such a way that it operates in a more heterogeneous and federated way so you don't have to move data, you can connect with any data sources, you don't have to move governance aspect of those things, as an example, and that's each piece of the puzzle. So it's more broadened, and we also build the storage platforms to say we can support file as service, object as a service or block as a service, anything as a service. Now, we've added augmentations in each layer, which means we've done the AI augmentation and storage. We talked about S3 tables as an example or even KV cache, et cetera. And we also had curated data engines separating the two, because data and storage does have a little bit of separation for heterogeneous and federation aspects. Now can you do a federated query engine inside storage? You can, but not really, because it needs to scale differently. Can you do a vector search engine directly in storage? You can, but then does it scale all the needs and all the, sort of avenues in which you take data? No. Can you then build orchestration into that storage? No, because you have to grow across the real estate. So which we realize is each pocket was realistic looking at how does a real data center operate, and how do you then work this into scale into products that people use and how do you bind them together as opposed to forcing somebody saying, "You know what? I got it all covered."

Dave Vellante

>> Well, I was going to say, I just have to make a comment. It's interesting what's happened in the storage industry. When I started in this industry, there were 80 disk drive companies. And of course we've seen so many pure play storage companies come out, 3PAR and-

John Furrier

>> Yeah, we've seen it all. Yes.

Dave Vellante

>> ... others, et cetera. Even now, you're seeing storage companies change their name, pure storage goes ever pure. They don't want to be ... Vast doesn't want to be a storage company. You guys have absorbed obviously EMC. It's becoming a data platform game. And that is what's different. I don't know, maybe there's some magic storage company that'll pop up and rule the world. I don't think so because it's really the data platform and that's the trend that we're seeing here.

Rajesh Rajamaran

>> So look, I think the question was, Dell has the blessing of a portfolio. We have partners everywhere in a long corner. We're able to build a more practical solution getting everything together without having the heartache or while do everything. Other companies may not have the same luxury because a single product company, they have to elevate their pitches, but we don't have to have the same limitations. We'd rather solve the problems the right way as opposed to come from a limitations.

John Furrier

>> And the storage got different mediums to you in different places to store things. It's data at the end of the day.

Rajesh Rajamaran

>> That's right. Yeah.

John Furrier

>> All right. Final question. I know we're coming up on time, really appreciate we do a deep dive on this for sure as a follow-up because this is a super important point. When Jensen at GTC last event, he had the Pareto slide and Dave and I called that out as the most important slide, because he introduced the Vera Rubin piece. And now you have diversity of Paretos. So you can actually... It's like a product positioning map, entry level to high end. They all have different costs, benefits, price performance levels. So this brings in... We're hearing routing, model routing and keynote. What does this mean for the portfolio? Because you're clearly injecting intelligence into the platform for AI factories. It's co-designed, a lot of integration. What's the portfolio look like post-Dell Tech world with Vera Rubin? What does that unlock? Can you share your thoughts on this?

Varun Chhabra

>> I think it's very easy to go down the route of, "Hey, this rabbit hole of this piece part for compute, this platform for..." And those are important conversations. I don't want to minimize that. But if you just take a step back, John, this is really about everything that we're doing is really about how do we make things as turnkey as possible, right? There's no God box, like you guys said, not just in storage, but even in AI all up. But how do you make things as turnkey as possible? So I think things that we're doing with Dell automation platform, what we're doing with the AI data platform, really are about stitching things together and making things simpler for customers. And because it has an open ecosystem, right? It's not just, "Hey, here's the hardware. You guys go figure out the technology stack." It's really bringing in those technology partners in there, the vector databases, the security protocols, et cetera, and bringing kind of a cohesive package for customers. I think sometimes when we talk about these different brand new toys, we lose our focus on that stuff. So really important for people to remember, wherever they are in their journey, we've got something for them. But especially for enterprises, our focus is really on making things simple and easy to consume, and knowing that the ISVs and the model providers, I mean, we saw that this week with Michael's announcements, we're bringing all of them to the party.

John Furrier

>> You're accelerating the enablement.

Rajesh Rajamaran

>> That's right. Exactly.

John Furrier

>> Rajesh.

Rajesh Rajamaran

>> You said, just answering Also's question. So we usually talk about data as a singleton, but data is a life cycle as you know. So if AI brings new workloads, for example, what we've seen is every piece of technology is an enabler. So at least in the time I've been in storage, everything has been additive. There's not something we need left. How drivers are still there, flashes there, there's more things, more media's coming, no storage products really, all storage products from a portfolio are still useful for their grand scheme of what IT needs overall. So you asked question of Vera, I think there's a place for everything, where I bring certain advantages to certain aspects of data lifecycle. And I think as Jensen pointed out, we are actively working with them, enabling that part of the lifecycle for data too. So I don't think it's a choice. I think we're enabling whatever is the need of the hour for the customer.

John Furrier

>> Yeah, the name of the game, feeds the infrastructure. Rajesh. Varun, great to have you on. Good to see you.

Rajesh Rajamaran

>> Thank you guys.

John Furrier

>> Congratulations on the success and continue the momentum, theCUBE, bringing you more action here. Next up, Jeff Clark coming on theCUBE. We'll get the low down on what's going on in the infrastructure on the main stage after this short break.