Building AI Infrastructure for the Token Economy — Dell and NVIDIA at Dell Technologies World 2026

As enterprises scale AI from pilots into production, the economics of AI are increasingly being shaped by the movement, structure, accessibility, and efficiency of data. The variable that determines whether AI delivers measurable ROI is no longer the model, it’s the data layer. Specifically, how data is structured, stored, accessed, and delivered to GPU infrastructure at the speed and efficiency required by the token economy plays a crucial role. Enterprises that experience performance limits and cost overruns in production AI typically identify bottlenecks in their storage architecture and data orchestration, rather than in the models they use.

Patrick Moorhead and Daniel Newman sit down with Ihab Tarazi, ISG CTO and Senior Vice President at Dell Technologies, and Jason Hardy, VP of Storage Technology at NVIDIA, at Dell Technologies World 2026 in Las Vegas, to examine how the token economy is reshaping enterprise AI infrastructure strategy from the ground up. They cover the Dell AI Data Platform and Exascale Storage announcements from DTW26, what co-engineering between Dell and NVIDIA actually means at the architecture level, and what enterprise leaders need to prioritize over the next 12 months to build production AI environments that scale without compounding cost.

Ihab and Jason also address why the deep engineering collaboration between Dell and NVIDIA produces different outcomes than standard integration partnerships, and what the combined platform means for enterprise customers trying to close the gap between AI ambition and operational reality.

Key Takeaways:

• The token economy has made the data layer the primary design decision in enterprise AI architecture. Token throughput, data accessibility, and storage efficiency determine whether GPU infrastructure delivers its theoretical performance or sits underutilized.
• The Dell AI Data Platform and Exascale Storage address the performance and scalability demands of production AI workloads. The platform is designed around the specific requirements of agentic AI environments where data must be continuously accessible and high-throughput.
• Co-engineering between Dell and NVIDIA means validation at the component level. Workload performance, storage configuration, and GPU efficiency are tested and optimized together before the customer sees the stack.
• Stranded GPU performance is one of the most expensive and underreported problems in enterprise AI. Storage bandwidth and data orchestration architecture directly determine whether organizations are extracting the performance they are paying for.
• Enterprise AI performance is determined at the data layer. The Dell and NVIDIA partnership is built around that reality. The enterprises building AI-ready data infrastructure foundations now will hold a compounding advantage as agentic AI workloads expand and token economics drive cost scrutiny to the infrastructure level.‍

Watch the full video at sixfivemedia.com, and be sure to subscribe to our YouTube channel so you never miss an episode.

Disclaimer: Six Five Media is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded, and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors, and we ask that you do not treat us as such.

IHAB TARAZI:
You cannot deploy today's GPUs if you don't deploy storage with it immediately. If you're doing it for training you have to put all the data in a storage cluster and be able to feed it. If you're doing inferencing you need fast access to the data. You're not going to get the utilization out of the GPUs that you invested in without storage.

PATRICK MOORHEAD: 

Welcome back to day two of Dell Tech World 2026. I am Patrick Moorhead and this is Daniel Newman. We are the six five in Las Vegas. Daniel but a great show so far. I think my biggest takeaway is that agentic AI particularly for on premises enterprise is getting more real.

DANIEL NEWMAN: 

It is getting more real, and enterprises are really thinking about how to drive and deliver ROI from those investments and infrastructure, and basically how to make tokens turn to value. That's right.

PATRICK MOORHEAD: 

And as both our research suggests, data strategy is still top five for impediments to getting stuff done in agentic AI. And I can't imagine two of the best people to talk through that than Ihab with Dell and Jason from NVIDIA.

JASON HARDY: 

It's good to be here.

DANIEL NEWMAN: 

Yeah it's so great to have you both here. But I'd love to get started with you. You know we're hearing a lot over the last couple of days. You know I. You know let's just say we're moving past experimentation. Jensen and Michael had that great exchange up on the stage talking about the utility that now comes with a I. But with such utility you know the architecture is changing. Organizations need to think differently about how they're going to scale. And of course having their data ready and every transformation that's gone on data preparation has been one of the big bottlenecks. So with all this going on you know why is the data layer so important and what should enterprise be doing to basically deliver the best token economics with their implementations.

IHAB TARAZI: 

Yeah I think you summarized it well. What's happening this event with the availability of new ecosystem partners that you saw yesterday. As people try to implement and get the value out of these systems they need to be able to orchestrate the data and they need to be able to make connection to where the files are. That's where everybody is to get the value out of the models and be able to get to the token economics.

PATRICK MOORHEAD: 

Jason. Sorry for botching your name. No it's all good. Apologies. It's a tough one.

IHAB TARAZI:

It is.

PATRICK MOORHEAD:

 It's really tough.

JASON HARDY: 

It's OK.

PATRICK MOORHEAD: 

I don't botch the question. No, seriously data and storage is so so important. And I'm curious from an NVIDIA point of view, why is it so important for you? And quite frankly, why are you sitting up here on a stage with with Dell tech?

JASON HARDY: 

,Yeah, I mean, you're absolutely right. Data has been one of the biggest fundamental challenges in the success of adoption from an enterprise perspective. The compute from a GPU perspective it's there it's ready to go. But if you cannot feed it fast enough if you cannot feed it with the right data, whether it's cleansed or structured or labeled or the whole pipeline for that, then. Inferencing in the enterprise will struggle. And not having that pipeline established and then getting the proximity to it and getting the security right, every little piece of the data challenge impacts in such a high way on the inferencing end of it. And that's ultimately what we want to do is generate the tokens to create the intelligence to be able to drive innovation.

DANIEL NEWMAN: 

And in terms of that deep collaboration with Dell, what role does that play?

JASON HARDY: 

Yeah. So Dell is amazingly positioned to be able to deliver into the enterprise ecosystem everything from the compute side. We saw on stage yesterday the whole lineup, the very sexy suite of capabilities there. That's that's awesome. That gets it all the way to the end user. But then it's the data center challenge. It's the proximity to data itself and then the software ecosystem on top of it. using things like KUVS for acceleration of search or using things like Rapids for better integration into the structured data ecosystem, or all of the other libraries and capabilities that we've created. Dell has taken on and integrated into a truly end-to-end offering that allows for the data to be addressed as much as the infrastructure, as much as the compute. And it's driving that ultimate value to the enterprise end user, which is accurate information, creating that intelligence.

DANIEL NEWMAN: 

Yeah, let's talk a little bit about the announcements here. You know, you guys have the Dell AI Data Platform, Exascale Storage. Talk about what, you know, overall is being announced between Dell and NVIDIA here at Dell Technologies World.

IHAB TARAZI: 

,Yeah, we made two big announcements yesterday. The first one of them was the AI Data Platform. And the AI Data Platform enables you not only to connect to all the data sources you have you need, but you can put security governance lineage. When you think about agents, not only they need to get to all the data, but they need to know what's the most recent file is. When somebody updates a month, a software program, or a file, it needs to know this is the latest. So all the tools for data management orchestration are part of the AI data platform. Included in it our most recent acquisition with Data Loop, which automatically orchestrates all of them in one framework. What's also new is that you can now apply CPU, GPU acceleration to the data. So with these CUDA libraries, Jason talked about. This is a big deal, as we showed with Start Burst. It could be 10x, it could be 20x the speed, you know, of what you do today. The second thing was Exascale. It is our large-scale storage designed specifically for AI and CSPs. So now you can get large scale data center hardware, liquid-cooled or air-cooled, massive scale, latest generation, and now you can put on it object scale, power scale, you can put on it block with PowerFlex, but you could also put a lightning for CME, context data memory, and the AI data platform. So you can take all the software on common data center hardware.

-

PATRICK MOORHEAD: 

Can you talk a little bit about, by the way, I'm sure everybody out in the audience knows exactly every product you're talking about, maybe not, but why is all this important for scaling AI?

IHAB TARAZI: 

Well, first, you cannot deploy today's GPUs if you don't deploy storage with it immediately. If you're doing it for training, you have to put all the data in a storage cluster and be able to feed it. If you're doing inferencing, you need fast access to the data. You're not going to get the utilization out of the GPUs that you invested in without storage. And then most people need a combination of object and file. The thing with Exascale, you never know. Do you need mostly object? Do you need some file? It's application-dependent, so what we announced yesterday allows you to mix and match as you go. You can start using most of your deployment as object, but then over time, if you have a lot of inferencing, you can shift, and that's why this is critical to scale it.

PATRICK MOORHEAD: 

Yeah, it's the most significant thing that I've seen is, you know, we always used to call it storage data, but storage wasn't data, okay? They were two separate things. Now they've just, they've meshed together. Correct. So, Jason, Jensen loves to talk about co-engineering. Yes. Co-engineering with partners, you know, time to market, kind of really associating, you know, getting to the underlying true needs of the customer. Why. What does it. What does that co-engineering look like between Nvidia and Dell? And you know why it does matter.

JASON HARDY: 

Yeah. No, I mean it's great. It's like it's Dell has an amazing route to market with a vast enterprise customer ecosystem. And we want to get the technology into those hands as quickly as possible. From a co-engineering perspective, it's as much as what we do across our portfolio, as much as it's Dell right there with us, side by side, co-developing and pushing forward and innovating a new space. He mentioned the file and object side. It's them taking, at day zero, our software capabilities to accelerate on file and integrating it right into their next-generation products. So it's not just saying, hey, we have some features, take these. It's like, hey, this is our customer feedback. This is what we can do from a capabilities perspective on our side, on the NVIDIA side, and then working together hand-in-hand to deliver something that's relevant to the market. And then doing that at scale repeatedly, and then innovating on top of it as customer demand changes or as the market changes, and always coming up with great new ideas to work together on.

PATRICK MOORHEAD: 

So I heard a lot of benefits. I heard time to market, performance, reliability. What did I miss?

JASON HARDY: 

We know each other's names, which is really good.

DANIEL NEWMAN: 

Now that makes sense. And Ihab, quickly touch, I know NVIDIA likes to talk about extreme co-design, but I mean, this is an example, right, where, you know, deep engineering collaboration is utilized. Like, why is that deep partnership or extreme co-design, why is that so important for basically accelerating these build-outs?

IHAB TARAZI: 

Yeah, I would say that is the secret of success between the two companies on from the Dell side. We know what are the blockers for the ecosystem to evolve and for customers. One example we needed confidential compute. for most of the frontier models to be able to deploy on-prem. So this is an example. We went to NVIDIA and said, this is what the market need. This is how fast we need it. These are the specifics. Or in the case of data, we're saying, we do need accelerators for all these ecosystem. We need it for these databases. We need it for this. And this is direct customer input. So without it, some of the announcements you saw yesterday could have been delayed another year or two.

JASON HARDY: 

Starburst is a great example of that. It's like it was just go in and do that at light speed.

DANIEL NEWMAN: 

So let's let's fast forward to the future. We know there's so much going on as it relates to you know, I like the term tokenomics. But you know we're seeing this heavily subsidized sort of token experience in the market. But enterprises that are going to start deploying at scale the bill is starting to come in. The ones that are like hey, experiment everyone and then the five or 10 million dollar bill comes in because everyone did what you said. So companies are trying to figure out how to navigate this whole movement. I mean what over the next 12 months you have do you think is going to take place that enterprise are going to figure out how to write size, implement, move forward, and build for you know the rules and rails of a business wanting to truly deploy and succeed with AI.

IHAB TARAZI: 

Yeah that's an excellent question. Every discussion we're having with customers about token economics right now because they they love the tools. They have the ROI. They know what they need to do now. It's about navigating the economics. you The first thing we're doing with the A.I. data platform and storage is giving them control over the cost of compute and storage by being able to put the data where it needs to be and be able to control that cost of infrastructure. This is significant. Half of the expectations that agents are going to generate enormous data and they need significant compute. So that step is mostly what we announced yesterday. You know all of the different models. are deployable on-prem on a data platform. So the ecosystem is moving. The next step after that is how many models themselves can come in. So you can have different economics other than by token. You saw we did announce that with SpaceX yesterday. So some movement of that and some new open source models. Those are the two biggest activities that we see happening.

DANIEL NEWMAN: 

And Jason, I mean, NVIDIA has been really outspoken about open source. I mean of course, you have great partnerships and recent deepening partnership I've heard with them. It seemed with Anthropic and others, but like open source seems to really have a big part of the economic story, and also smaller models that are more target-specific. Like, what do you see sort of happening over the next year as enterprises go from you know sort of proof of concept to you know exponential scale across across their deployments.

JASON HARDY: 

Yeah, I mean, definitely open source is a big piece of it, and we'll continue to contribute into that even with what we do on our names and blueprints and everything like that today. Being able to continue to evolve that forward. Working with great partners like Dell to continue to drive that into innovation and lower that cost per token so that the customers can realize that benefit while also having the best capabilities possible on-prem, that hybrid approach where necessary. But ultimately, it's like what Michael and Jensen said yesterday, it's about addressing the demand today and then being able to focus on how we scale that into the future.

DANIEL NEWMAN: 

And the partnership between NVIDIA and Dell, what comes next? Yeah, be sure to pre-announce everything. I'd like to know that, and I'd also like to know Jason's gross margins today?

PATRICK MOORHEAD: 

Tomorrow?

DANIEL NEWMAN: 

Yeah, tomorrow. Just kidding.

IHAB TARAZI: 

I would say making agentic work with security, with guardrails, that's a much harder thing then. So you heard a little bit about open shell NEMO class. That area is massive, and that was just the beginning of it. And also, I would say the work we're doing on disaggregated inferencing and KV cache, those two are the biggest areas of innovation. You'll see more. Absolutely.

PATRICK MOORHEAD: 

Yeah. My brain went ding ding ding on the open shell. That one slide. I don't know. I'm digging this Newman claw claw.

DANIEL NEWMAN: 

That's good. Well, Jason, I want to thank you both so much for sitting here with us today. We've got a great rest of Dell Technologies World here. Keynotes to come. I'm sure more announcements. You guys have a little announcement tomorrow that the whole world is going to stop. I can't believe you wouldn't give us the margin.

PATRICK MOORHEAD: 

I know. Can you believe it?

DANIEL NEWMAN: 

So selfish. Great to hear from both of you. Congratulations on the announcements, the partnership, and I wish you a great rest of your event. Yeah. Excellent.

IHAB TARAZI: 

Thank you very much. Appreciate it. Thanks for having us on your show here. Yes.

DANIEL NEWMAN: 

And thank you, everybody, for being part of this Six Five On The Road. We are here in Las Vegas at Dell Technologies World 2026. Stay with us. More coverage to come.