theCUBE + NYSE Wired: AI Factories - Data Centers of the Future | John Roese, Dell Technologies

John Roese, Dell Technologies

In this interview from theCUBE + NYSE Wired: AI Factories – Data Centers of the Future event, Glean co-founder and CEO Arvind Jain joins theCUBE’s John Furrier to unpack what’s really working in enterprise AI today and what comes next. Jain explains why knowledge access remains the first successful AI use case at scale and how Glean’s enterprise search brings AI into everyday work. He details the past year’s lessons with AI agents – from the need for guardrails, security, evaluation and monitoring to democratizing agent building so business owners (not just data scientists) can create production-grade agents. The conversation dives into Glean’s vision of the enterprise brain powered by an enterprise graph, highlighting the importance of deep context, human workflows and behavior to reduce “noise” and drive outcomes. Jain outlines core building blocks – hundreds of enterprise integrations and a growing actions library – that let agents securely read company knowledge and take actions across systems (e.g., CRM updates, HR tasks, calendar checks). He discusses how organizations are standing up AI Centers of Excellence, prioritizing “top 10–20” agents across functions like engineering, support and sales, and why a horizontal AI data platform that unifies structured and unstructured data – accessed conversationally and stitched together via standards like MCP – sets the foundation for AI factory-scale operations. Looking ahead, Jain says Glean’s upgraded assistant is evolving from reactive tool to proactive companion that anticipates tasks and accelerates productivity.

Share this session

Clips
More from theCUBE + NYSE Wired: AI Factories - Data Centers of the Future

John Roese

Global Chief Technology Officer & Chief AI Officer

Dell Technologies

play_circle_outline Discussion focuses on AI factories and Dell’s work in AI infrastructure transformation.

play_circle_outline Enterprises are ramping up their AI infrastructure for greater ROI and efficiency.

play_circle_outline Corporate sovereignty and governance are becoming more integrated with enterprise AI strategies.

play_circle_outline AI systems’ operational resiliency and disaster recovery strategies need to be modernized.

play_circle_outline Companies must focus on governance and ROI while identifying valuable AI applications.

Info
Transcript

John Roese, Dell Technologies

John Roese

Global Chief Technology Officer & Chief AI Officer Dell Technologies

In this segment from the theCUBE + NYSE Wired: AI Factories - Data Centers of the Future event, Dell Technologies CTO John Roese joins host John Furrier to discuss the rapid mainstreaming of AI factories and the shift from experimentation to production at scale. Roese analyzes how global enterprises are finally connecting the dots between infrastructure investment and tangible ROI, moving beyond simple compute build-outs to reimagine the data layer with knowledge graphs and vector databases. The discussion also highlights the explosion of sovereign AI strateg... Read more

explore Keep Exploring

What has been a significant development in the adoption of AI by enterprises, particularly in relation to Dell's initiatives? add

What trends are being observed in enterprise adoption of AI and how are companies finding value in their investments? add

What are the emerging trends and predictions regarding sovereignty, governance, and networking in the context of enterprise technology and its impact on factory operations? add

What are the considerations for ensuring the continued availability and resilience of AI systems in an organization? add

What key strategies should companies focus on when implementing AI technologies to ensure successful outcomes? add

bolt Powered by CUBE AI

John Roese, Dell Technologies

search

John Furrier

>> Hello. I'm John Furrier with theCUBE. We are here at our NYSC CUBE Studios. Of course, we have our Palo Alto studio out in Silicon Valley connecting Wall Street and technology. John Roese is here, the CTO of Dell Technology as part of our AI factory series. This is one of our most popular series. John, welcome back to theCUBE. Great to see you coming in remote to the NYSC CUBE Studios, part of our NYC Wired program. Thanks for coming on.

John Roese

>> Yeah, great to be here.

John Furrier

>> So you've been a big catalyst. We've had many conversations over the past year and a half around AI factories, the work Dell's doing. Certainly in the transformation inside Dell, but also outside with customers. You've been seeing the AI factory. You and Michael Dell have been on many customer calls. We've chatted about that. It's now going mainstream. It's really one of our most popular content programs. As enterprises, it's still the large scale. Hyperscale is the neo clouds. They're still ramping up the AI infrastructure today here at the New York Stock Exchange. We're expecting to hear the Time person of the year. It's going to be, it looks like, not a person but all AI architects. It speaks to this generational shift. It's a big bang to light speed acceleration, a topic you've been writing about, so what's your take looking back through the year of where we've come from and where we are?

John Roese

>> Yeah, I think we're making fantastic progress, but we're not done yet. Obviously, we're seeing the enterprise adoption continues to ramp. I was doing some dialogues yesterday with Europe and Asia, and there are chief information officers and chief AI officers in major companies everywhere in the world that are starting to tip into production, and when they do that, then they start to scale their infrastructure which puts code forward, then they start really scaling their AI factory architecture, and most importantly, they're doing this not randomly. It's not about just deploying infrastructure. They're doing it because they've found a path finally to ROI. They're transforming their cost of sales, they're transforming their supply chain, they're changing the way they develop products, and that's really the magic. And it took a little while to get everybody aligned that you connect the dots between a value proposition of impactful ROI to your business and your investment in AI, and that has largely started to fall into place over the last year in the enterprise. Parallel to that, by the way, the sovereign world has taken off. Beginning of the year, I predicted that every country in the world would develop or at least be on their journey of having a sovereign AI strategy, and most of them didn't a year ago. Today they all do. You can't talk to a government that isn't trying to figure this out and they're trying to work through it, and that correlates to infrastructure build out and technology and societal impact.

John Furrier

>> Yeah. We're seeing sovereignty also bleed into corporate enterprise because sovereign data is becoming a topic with agents, and as the edge starts to develop, you're starting to see the dots to connect for the hyper converged edge where you have Nvidia has their software development environments out there, reference architecture is around, wireless, wire ethernet and unlicensed spectrum with spectrum, collapsing in factories. That brings in a hybrid cloud/private sovereign vibe to it, so sovereign is extending into the kinds of policies and networks that need to be in place to make the factories work. You predicted that, and also you predicted the not slow down on the enterprise but the buildup of the confidence in the enterprise. This year, you're predicting a re-engineering of the fabric of the enterprise. You got sovereign, you got governance, these things are in play. You got the networks, a big part of the factory is network connections, interconnect and the factories. Now you got distributed networks at the edge. We're expecting MWC to be a big theme there. What does that look like? What does that re-engineering the fabric of the enterprise look like at this lightning speed?

John Roese

>> Yeah. There's two... Think about the enterprise infrastructure has a set of layers. There's a network layer, there's a compute layer, there's a data layer, there's a storage layer, there's an application layer, and the architect understands that. A lot of those layers are now actively being rearchitected for AI. Two of the most interesting ones, in my view, the compute layer is largely done. We now know accelerated compute is the answer. That's why we have AIPCs and AI factories, but the data layer is being rethought, and the way it's being rethought is to not try to take our legacy data and make it the platform of AI. That's just the data AI uses. What you do to create data management in an AI environment is you build a knowledge layer, and the knowledge layer is a different set of tools. It's graph databases, knowledge graphs, vector databases. It's all of the information necessary to feed chatbots and agents and other tools, and those are new. They also behave differently. They need higher performance, they're extremely mission-critical, they're incredibly valuable, so we are seeing that happen. But the other one, the other layer architecturally that's being rethought is, to some extent, the communication network. Not necessarily just ethernet and low level functions, but the way communication happens, and the reason for that isn't because of chatbots, it's because of agents. Agents have a characteristic that makes them different than a chatbot. They talk to each other, they can collaborate, and if we believe we're going to have billions of agents, we have to build communication systems for those agents to securely connect to each other, find each other, interact, share information. And so all of that is happening with some new moves in the Linux Foundation with protocols like A2A and MCP for tool use and data access. All of that is a very active space, and in 2026, we're going to see a different kind of networking. We're going to see a different topology of communication that's optimized for agents talking to each other and agents talking to people, and people talking to each other.

John Furrier

>> I think that's going to really open up the edge, which I mentioned. You pointed out, networking is the operating systems. Dynamo, KV cache, that's the big discussion all last year. We extend that out to the protocols and the frameworks that the developers are using for agents. Agents can do a lot of work like manage things, manage access, governance, routes, so I think the edge is interesting. The other thing that you predicted, and I want to get your thoughts for 2026, is you did predict the mainstreaming of GenAI. Obviously, some of the chatbots we saw, some of the low hanging use cases, low hanging fruit like RAG, knowledge basis, starting to see that data layer. I want to get your thoughts on what you see in 2026 with agents, because we're already seeing some of the traditional sacred cows of storage, for instance, where people aren't moving the data. They're keeping it where it is and then letting agents do the work, so you're starting to see new patterns emerging where agentic infrastructure is going to manage a lot of this plumbing, glue layer, whatever you want to call it, abstraction. What's your view on this? Because I think this becomes an architectural opportunity to really scale up these agents in the factories.

John Roese

>> Yeah, you're absolutely right. We fortunately have been working with fully autonomous agents, not chatbots called agents, but real agents, agents that actually do autonomous work and talk to each other and use tools for over a year and a half. We were very early in experimenting and developing these technologies, and what we learned this year as one of the early adopters is whatever you thought agents were going to do for you and how they were going to exist in your organization, you're wrong. It's going to be different. It's going to be bigger and more disruptive, and so to give you some example, our prediction for next year is that the incorporation of agents into the enterprise will not manifest as agents being tools for people exclusively. Today, we think about an agent as, "Oh, it's an AI that'll book my travel for me." That's interesting, but that's not actually what's going to happen in totality. It turns out, yes, they will provide tools, but more importantly, when you bring autonomous AIs that can actually do work independent of you and can have digital skills and actually play roles, you start redesigning your org structure, you change the relationships, you think differently, and so a couple of things that we've seen. The first is when agents show up as a first class citizen in a team where it's a team of people with agents working together, the agent may start out as a tool, but eventually, you start to use it for other things like coordinating the entire team. It might actually look to some extent like your boss. It might amplify the skills of the best person so everybody gets access to the advice of that expertise and a knowledge graph. We also found that agents, because they're autonomous, allows us to go after problems that we historically wouldn't have solved because it was too expensive to throw bodies at it. Think about all the problems that you haven't solved because just too much time and too much energy and too much cost to use human beings to clean up your CRM database, polish your data, whatever. The reality is autonomous agents make that almost free, and so we're seeing entirely new use cases that were ignored in the past but are very valuable starting to happen. And then most importantly, we're seeing agents start to interact with each other where it's not just about the agents in your enterprise working with each other. Protocols like A2A allow your supply chain agents to talk to your supplier's agents and have conversations with each other, do very complex tasks. So the moral to the story is I don't know all the things that are going to happen, but what we've found is everybody who puts autonomous agents into production realizes very quickly that they underestimated the impact and the change that happens to organizational structure, to operating model and to the technology stack when these autonomous entities become part of your workforce.

John Furrier

>> It's interesting you bring that up because one of the things we're seeing in the data is that three to six months in, you start to see that domain in adoption, the intelligence in the domain specific task or workload or wherever it is. We're hearing even Jensen Huang, I interviewed last Thursday at the GSA awards, he talked about AI factories as a system of intelligence, pumping out intelligence, not just tokens. It's going to move from tokens to intelligence. And you mentioned earlier, I want to get your thoughts on this because the building blocks that you're seeing, and correct me if I'm wrong, but the data management piece is the backbone, if you will, the substrate, and then the agentic is the operational nervous system, the runtime environment. Is that right, and how should we think about these two pieces? Because I think those are fundamentally the building blocks for scaling up and getting that productivity, getting that value creation and extraction from agentic.

John Roese

>> Yeah. Yeah, you're absolutely right. The agent itself is just a software system and it has a number of components around. It has LLMs, it has knowledge graphs, it has protocols like 8A and MCP. It can do work, but it's a software system. What makes it a specialist or an expert, what makes it good at doing something, interestingly enough, is not usually the large language model. Large language models just provide general purpose, collective skills. Whatever's on the internet, the LLM can help you understand, and what really makes it specialized are the knowledge graphs that you use, the fact that you can take your proprietary data, data that isn't on the internet, that really is unique to your company or your skill or a particular capability like being an oncologist or a hematologist in medical, and you can take that data, add it to the agent, and now the agent becomes intelligent at an expert level versus just a good enough generalist, and that data layer is not magic. It's a real thing. You have to actually build it. You have to build an infrastructure that supports knowledge graphs and maintains them and can feed them, that also can do things like agentic RAG as other techniques. The bottom line is we have been unambiguous. Michael has said this for years that data is the most important asset in your enterprise, and if you control the data, you will get value. Up until now, it was generally direct value to human beings. In the AI cycle, especially with agents, if you control proprietary data that is unique to your company, that is the manifestation of your expertise, you now have a tool called autonomous agents that can scale that. You have an operating system that can take that intelligence and make it pervasive and even monetize it. That is an incredibly powerful thing, but if you gave up your data and didn't have any, then agents are just a general purpose tool. And so it's very much data is the difference between expertise and value and just a general purpose thing that everybody can do when we talk about agents.

John Furrier

>> You mentioned at the top of this interview that you guys have been doing autonomous agents for a while, and I just came back from AWS's annual event and they released frontier agents they call them, which is essentially smart agents that can do more autonomous things as you mentioned. But one tell sign I want to get your thoughts on that could be the real trend this year is that you look at Dell technologies, look at what AWS just did. You guys are building agents for yourselves as a service for your customers. Amazon built a Kero code assistant, autonomous coding thing, a DevOps agent in security. Why? They do a lot of that DevOps work in the cloud, so they have data that they made their own. You guys did it for Dell. This is, I think, where customers are going to start to lock in the values saying, "Hey, we have agent opportunities that we could take from our learnings, from our IT infrastructure, our day two operations on cloud native, on hybrid, security, governance postures, data they have there." So I think this is a trend that might translate into the broader market because I think Dell and, say, AWS are use cases saying, "Hey, we could do this. Why not just do it?" Why? Because you can just create so much... Why wouldn't I want a Dell technical assistant all the time doing stuff for me, like actual work, not just like helping me figure out things, but like actually do the heavy lifting? Do you agree with that? do you see the similar thing where customers will start to identify some of these tasks that they have domain specific knowledge on?

John Roese

>> Yeah. Yeah, I think there's two parts of that. The funny thing is that that discussion, I'm glad other companies are now talking about things that we showed at Dell Technologies World at the beginning of the year. Remember, we demonstrated a whole bunch of agents running infrastructure on behalf of our customers and painted this picture of that's the future, that infrastructure for instance is not just infrastructure, it's the intelligence to run infrastructure, to troubleshoot it, to operate it, so we've done that ourself. We know a lot about infrastructure, and therefore, we're creating the embodiment of that intelligence as an autonomous entity called an agent as part of our overall portfolio and offerings. I think you're absolutely right, every company in the world for the most part that has some unique skills and some proprietary knowledge has that opportunity, and agents as they mature, which they're now fairly... They're not fully mature, but they're definitely implementable, it is not a massive technical effort. It is really the connection between your information, understanding what behaviors you want, and using tools that are relatively straightforward to use today, and that I think will become a sustainable source of differentiation for almost every industry. On the flip side, companies that ignore it and quite frankly don't do any agentic work might find themselves in a battle between automation and human effort. You know which one wins every time? Automation. And so it's super important to realize that the only thing that prevents you from using autonomous agents to scale your intelligence, your special skills to a much larger population of customers and users is quite frankly being able to control your data and connect it into these systems and then start to express them into your market as part of your offering, and that's exactly what we did. We showed it at the beginning of the year and we've been continuing on this journey. We're seeing other people do the exact same thing, so I think the trend, which I think is real, is the definition of a product is no longer hardware and software. It is hardware, software, and intelligence, and I think we've been saying that term for a while. Most people didn't understand it. Well, now it's manifest as, hey, a system, a technology, an architecture that doesn't have this level of autonomous intelligence as part of it is probably at a disadvantage and not as valuable as one that does, and that does seem like a pretty big inflection point in the industry.

John Furrier

>> It's interesting, in our little social group here in theCUBE and our community, we're talking about AI as autonomous, this autonomous nervous system, but remember the term day two operations used to be very popular. In a way, these autonomous agents are essentially automating day two operations in real time, and governance is the long pole in the tent. And you mentioned some of the things that you guys have done. I know for example, we've covered Dell's disaster and recovery, cyber resilience, an area that you guys have done very well in for instance as great business. But now, you can apply agents to things like the critical assets involved, so things like disaster recovery, cyber resilience, a lot of these... Security is another one where you guys can bring in. Can you give an example of how you guys see that being redefined, because it's not just a product anymore, it's actually an operational day two thing, always on and always being managed by autonomous agents?

John Roese

>> Yeah. One of our predictions for this year is that it's just a logical progression. Once you build an AI factory, once you start to use AI systems at scale, you become quite dependent on them. You don't want them to go away. At Dell, we have a number of systems that are very important to the operations of our company now, and they save us a lot of money, they create a lot of value. They are entirely novel AI architectures. They're not legacy systems with a little veneer on. And the result is once you get into that state, you then ask the question, "Well, how do I make sure that they don't go away, that they're always available?" which is the discussion of cyber resilience, of survivability, of disaster recovery, all of those terms. What our prediction is though is because AI systems are very different, they're not the same. You don't operate them the same. You're using agents, you don't build them the same. They're built on GPUs and accelerated compute. You don't have the same data, they have knowledge layers as opposed to traditional systems of record. You do need to pause and ask, is there a better way to make them resilient? And so what we're advocating for is take a step back, and even though many of the technologies like cyber vaults and things of that nature will be incredibly important, you can actually architect disaster recovery and survivability and cyber resilience for an AI factory in an actually very different way and actually a lot simpler than doing it for an entire application of state and brownfield. And so in addition to be having an opportunity to rethink the architecture and make it modern, we also have some new tools. You mentioned agents, so we can automate things, but the other tool we have, which is fascinating, is the CSP infrastructure, the sovereign infrastructure. When you're trying to back something up and have a second one, you can either build it yourself or there are these gigantic pools of GPUs sitting there out in the world in sovereign infrastructures and CSPs that could become very valuable as a place to activate your AI functionality in the case of a disaster, because those GPUs are easily available, they are pivotable to new use cases. And so we don't know exactly what the topology is going to look like and the perfect way to do it. What we do know is survivability and cyber resiliency are going to become a critical conversation in every AI factory discussion.

John Furrier

>> And-

John Roese

>> Number two, we have new tools.

John Furrier

>> Yeah, and also the speed game that you were talking about, lightning speed, just the things that are jumping off the page to me at the end of the year as we look back is migrations are hot. Why? You can do them faster. This backup and recovery used to be a process that can be managed beautifully with agents, so all these things that you predicted and we talked about last year have come true, so hats off to you. And I love the 2026. I think sovereign is going to expand in scope to be much more just basic hybrid, private capabilities that's going to happen on the fly. I think agentic will bring. We'll watch that very closely. Really appreciate that commentary. I have to ask you, for 2026, what's the action plan for leaders as you talk to them? I know you guys, you hit the road all the time with Michael, you talk to your top customers, you have adoption with AI factories. What is the action plan? What should leaders do now for 2026? What's the game plan?

John Roese

>> Yeah, I sound like a broken record because I've been saying it now for a year and a half. The action plan is focus, governance focus, pick your battles, focus on ROI. We now have tons of data and reports over the last year that say the wrong answer is just go pick a random AI technology tool and roll it out to all your employees and hope for the best. That does not work. You might want to do that, but that's not a strategy. On the other end, every study, every use case we've seen where a company identifies where the real value is, where the ROI is, the places they can transform their business in a narrow scope and then they concentrate their energy on getting those into production at scale, that's the winning formula. My first prediction for 2026 is this will... Last year, I said agentic would be the word of the year. I think governance is going to be the word of the year this year because without proper governance, without a set of rules, without guidelines, without the ability to prioritize, you never get unstuck from POC purgatory, and so that is going to be... It continues to be the emphasis. Now we have lots of examples of that, we have better tools, but governance, governance, governance. Have focus, get something running. If you are still meandering around with a thousand POCs, you are late. If, on the other hand, you have one project, just one That is tipped into production at scale and it's saving you tens of millions of dollars and transforming a piece of your business, you're doing great, because then you just got to replicate that over and over again. So that tipping point is the thing people must focus on, and it's hard because there's a lot of noise encouraging you not to do that, but every piece of data we have says you win with an AI factory when you use it for something that matters.

John Furrier

>> Yeah, and then pumping out those intelligence tokens. John, great to have you on. Congratulations on a great year. A lot of your predictions actually panned out, and again, continuation into 2026. Love this edge development, we're going to see that explode this year. I love the networking aspect of it, love how storage is just continuing to be a very relevant part, and of course, memory. We need more compute, so keep ranking. Thanks for coming on. Thanks for sharing on the AI Factory series. Appreciate it.

John Roese

>> Thanks, John.

John Furrier

>> I'm John Furrier with theCUBE. We are here at our NYSE CUBE Studios. Of course, theCUBE in Palo Alto and New York City, bringing all the action from the field, doing our part to bring the factory of content to you. Thanks for watching.