Nutanix .NEXT 2026 | Dan Ciruli & Anindo Sengupta, Nutanix

Clips
More from Nutanix .NEXT 2026

Dan Ciruli

Vice President and General Manager, Cloud Native

Nutanix

Anindo Sengupta

Vice President of Product Management

Nutanix

play_circle_outline Accelerating AI Workloads with Nutanix NKP and AI Pass: Integrated Kubernetes, Storage, Kubeflow, Vector DBs, KServe

play_circle_outline SP Central: multi-tenant GPU infrastructure for service providers and consumption models.

play_circle_outline Nutanix Enterprise AI on AHV: Flow Networking and AI Gateway for Governed, Cost‑Controlled Multi‑Tenant Inference

play_circle_outline GPU ecosystem partnerships ensure access to latest accelerators and validated compatibility.

Info
Transcript

Dan Ciruli & Anindo Sengupta, Nutanix

Dan Ciruli

Vice President and General Manager, Cloud Native Nutanix

Anindo Sengupta

Vice President of Product Management Nutanix

In this interview from Nutanix .NEXT 2026, Dan Ciruli, vice president and general manager of cloud native at Nutanix, joins Anindo Sengupta, vice president of product management at Nutanix, to talk with theCUBE's John Furrier and co-host Alison Kosik about the rapid convergence of cloud-native and agentic AI infrastructure and what it takes to build a unified enterprise platform for the next phase of AI. Ciruli and Sengupta describe Nutanix's strategic position as the software layer between chip infrastructure and AI models — the "middle of the cake" in a fiv... Read more

explore Keep Exploring

How is Nutanix evolving its infrastructure and products to support AI/ML applications and developers' expectations? add

What is the approach to agentic AI infrastructure, and what was announced at .NEXT (for example Neocloud and SP Central) to support multi-tenant GPU compute, consumption layers, and the new inference engineering stack? add

What did the company announce to help partners build multi-tenant GPU clouds and to manage AI inference, model/tool governance, and infrastructure? add

What is the company's strategy for supporting AI workloads—particularly inferencing on Kubernetes/cloud-native platforms, gateway products, and GPU partnerships—to deliver high performance, low latency, and lower cost per token? add

bolt Powered by CUBE AI

Dan Ciruli & Anindo Sengupta, Nutanix

search

Alison Kosik

>> Welcome back to theCUBE. We are coming to you from Chicago and we are streaming live from Nutanix .NEXT. I'm Alison Kosik. We have been hosting this show all day. I mean, I have with John Furrier, he's the co-founder and co-CEO of theCUBE here, and we are about to enter another discussion here about the next phase of enterprise IT.

John Furrier

>> Yeah. I mean, I think cloud native, Alison, we talked all week about this and today in depth. What we're seeing with Nutanix is the intersection of cloud native and AI native worlds rapidly coming together with the rise of agents. And the infrastructure required to do that is front and center. And a lot of smart people, still a small community. It's growing very rapidly. And the Linux Foundation recently announced the Agentic AI Foundation. That's going to be like the next CNCF version of AI, but it's going to be part of the Linux Foundation. So it's already exploding in growth, 170 members. So that's a tell sign that these two worlds of complimentary engineering and business-minded people are going to scale AI to new heights. And in this next segment, we're going to unpack that.

Alison Kosik

>> Well, that's great analysis. So let's go ahead and unpack that with Dan Ciruli. He's the Vice President and General Manager of Cloud Native with Nutanix. Welcome to theCUBE.

Dan Ciruli

>> Thanks for having me.

Alison Kosik

>> And Anindo Sengupta, Vice President of Product Management at Nutanix. Again, welcome to theCUBE. Great to have you both.

Anindo Sengupta

>> Thank you.

Alison Kosik

>> So what are your thoughts about what John just waxed poetic on?

Anindo Sengupta

>> I think this space is so exciting with agentic AI. What we're seeing is a culmination of what as we've been doing. We make managing IT infrastructure simple. And with agentic AI, the AI infrastructure is actually a lot of complexity. And what we've been doing with building hybrid multi-cloud, what we've been doing with cloud native, it just adds up to a point where we feel the agentic AI tsunami that we're seeing, Nutanix is well positioned to really help our customers and partners in really managing the agentic AI storm and really simplifying the complex infrastructure that I head to manage that.

John Furrier

>> What are some of the things people should know about with Nutanix relative to cloud native and making agentic? What's products? And GTC does some action. We saw some stuff there. Lay it out. Lay out all the elements that are in play.

Anindo Sengupta

>> Danny, you want to talk about cloud native?

Dan Ciruli

>> Sure. I mean, I think it starts with ... First of all, it starts with compute and storage, right? Nutanix has a storage layer, distributed storage. The architecture that Google used to underpin all of its ML efforts is what you get with Nutanix. But on top of that, and what we've added in the last couple years is a Kubernetes layer. All these new applications that are being written that are going to use AI are definitely going to run in Kubernetes. That has to be part of the infrastructure. And so we've had that for a couple years now with NKP. And what we announced back at GTC was a layer on top of that, an AI path, because what developers will expect out of infrastructure keeps getting higher and higher level. At first, the developers installed Kubernetes. Now, they want it to be part of the infrastructure. And the same will be true of those projects that are going to go into that foundation that you were talking about. And things that exist today, things like Kubeflow, things like vector databases, things like KServe. The developers won't want to install and configure those, they'll think that's part of the infrastructure, so we added that layer.

John Furrier

>> And you guys also announced some neocloud capabilities.

Dan Ciruli

>> Absolutely. You want to talk about that?

John Furrier

>> Yeah, that's pretty relevant.

Anindo Sengupta

>> Yeah. So I would want to take it at two layers. So we actually have the agentic AI solution, which works on layers. And the neocloud, one is the perfect addition we announced at .NEXT. When we think of agentic AI infrastructure, we're looking at a lot of our customer's partners setting up the new accelerated computing infrastructure. And there, starting with the consumption layer, everybody wants to get managed as a neocloud, and a cloud needs tenancy. So our core focus is to kind of help our partners build multi-tenant cloud. So we announced SP Central, which is focused on taking the GPU compute infrastructure and creating multi-tenant infrastructure that they can then sell the service catalogs to end customers. There's one more thing we are doing in the layer on the top from a consumption perspective. Dan talked about the AI pass layer, but we've been investing for over two years in where we think every agent will need intelligence. And that means we need the new inference engineering stack to be robust and hardened and still be available on demand. So we have a product called Nutanix Enterprise AI, and we launched at .NEXT, the capability of doing gateway management. The AI gateway is really around cost and governance. So as agents sprawl, models and tools need to be controlled and governed. So what we've announced is the capability to really drive governance around models and tools using agentic AI. And last but not the least, we really believe that as we have more inferencing, the infrastructure has to be governed and managed. So we absolutely have a very strong Kubernetes layer on which all these applications run, but the core compute using our power of AHV and Flow allows us to be governed, managed, and really controlled more effectively. So that's what we announced.

John Furrier

>> The SP Central capability, I think, is a home run. I think the enterprise is going to explode because the hard stuff is hard. You're not going to just buy AI factories. What do you run on it? They don't load Linux anymore on machines. There are machines. There's many machines. AI factories need to have an AI stack. They got to have the containers, the Kubernetes.

Dan Ciruli

>> And I think that the reason the SP thing is going to be so instrumental is that many of these enterprises, it's going to be a long time till they can get the silicon they want. Everything from, of course, the GPUs now, even the RAM is affected. Prices are high, wait times are long, but they'll be able to turn to an SP and say, "I need some access right now." And, yeah, maybe they might move some stuff on friendly-

John Furrier

>> I mean that might be a feature, not a bug, because the utility, again, back to the cost structure of an enterprise, they have a lot of unlock to do with the data. So do you care how it comes as long as it's secure? Again, back to the concept. What are we trying to do?

Dan Ciruli

>> Right.

John Furrier

>> Do you see that playing out and what are some of the things you're watching? Because we're watching that. We think that the service provider is going to be multi-scaled, many hyperscalers. So okay, you got some GPUs, I'll get some from you and you and I can abstract it on-prem with SP.

Dan Ciruli

>> I mean, it's funny because when you walked into the .NEXT keynote, the large font letters as you walked in said, "Run anything anywhere." And that's our mission, right? And we do want to enable that. I think that customers, enterprises are going to find many reasons to run in service providers. We do see times where they have data gravity on-prem that's going to eventually require them to run things on-prem. And so our goal is to be able to say, "Hey, whatever the tools you need, do you need a hypervisor? Do you need storage? Do you need Kubernetes? Do you need that AI pass layer?" Yes, you can get it from on-prem, you can get it in the cloud. And if you need a service provider, then we're giving them that layer too.

John Furrier

>> Okay. I want to throw something out. You guys have a couple more questions, but I want to get this out now because you just brought it up. I want it reminded me. VM's great. They got their job. They're going to do a bunch of cool things. It ain't my grandfather's VM. Maybe it's kind of out there. You got containers, little renaissance going on with containers, but you've got the AI stack that's still emerging. Companies are trying to figure out what's my AI stack. So I guess my question is, what does that mean? What does an AI stack actually mean? With Kubernetes, with the cloud native, hardened infrastructure, which is getting better and better, it's pretty mature, very solid. What the hell does an AI stack actually mean?

Anindo Sengupta

>> I think that's a great question. And what we are seeing in the industry, and I'll start with Jensen's made popular the five layer cake, right? And I think it's a great way to frame the overall space. We start with the power and cooling for the data centers. Then you have the chip infrastructure. Then you need software to control the ... Then you have the actual hardware. Then you have the software to control hardware. Then you build agentic applications and there's models, right? So models and applications. And this is the broad AI tech stack, but it's too complex. Where we see we play is between the models and the chips, there's a lot going on. So when we think of Nutanix, that's the core heart of agentic AI.

John Furrier

>> Sounds like middleware to me.

Anindo Sengupta

>> And there-

Dan Ciruli

>> It's the middle of the cake for sure.

John Furrier

>> Middle of the cake. Yeah.

Anindo Sengupta

>> And the whole story between models and the chip infrastructure and how agents will use that needs an orchestrator, a software layer that can help drive velocity of innovation. So when we talk about our agentic AI stack, we start with the agent builder layer that supports agentic applications being built and those will run on the Kubernetes platform. And then the Kubernetes platform finally is orchestrating the use of this actual computing infrastructure. There, our thesis is we can absolutely run Kubernetes on bare metal. That's how, when the training world was where we had large models being trained, batch processing at scale, that model absolutely works well with Kubernetes on bare metal. But now, when you think of every employee having tens to thousands of agents, that's really inferencing at scale that needs the dynamic nature of cloud management that we've been doing for years in the traditional cloud.

Alison Kosik

>> So it needs a referee.

Anindo Sengupta

>> Absolutely. It needs somebody to operate, govern, control with security and with agility. And then that's the control plane that we are offering with Kubernetes and our compute layer with AHV and Flow.

Alison Kosik

>> Great question.

John Furrier

>> I love the referee comment, because it's like sports analogy. I love sports analogies. But the field has changed and the game has changed. And so I want to ask you guys, because you're both deep in cloud native and you guys are on both sides of the table now, AI native and cloud native. The number one question I get when I talk to my cloud native tribe and it's like they love cloud native and the AI tribe, they both have kind of like coming to the same place. The number one question I get is, what are the key linkages between the services? So assuming those worlds come together, how should people think about the connection points between where they talk to as you mentioned Kubernetes on one side, but also it's sitting on? So you guys have bare metal, you got NKP, you have other things. Lay out the linkages. What are the key connection points?

Dan Ciruli

>> Well, one of the connection points is going to be, you mentioned Kubernetes. We see customers running Kubernetes in a variety of ways, depending on their use cases Anindo was saying. We see other vendors out there saying, "Well, the right way to do Kubernetes is in a VM, in a hypervisor." Other people say, "No, no. The right way to do Kubernetes is on bare metal." And you use Kubernetes and maybe you might use Kubert to schedule a VM in there too, but Kubernetes is your primary scheduler. We think there's use cases for both. As Anindo said, often it is fine to run. In fact, it's often better to run in a hypervisor for Kubernetes. So that is going to become the layer. Here's the thing that I think that you think about applications, AI applications and regular applications. I think that in two years or three years, we don't make that distinction anymore. We just call these applications and we will expect there to be infrastructure that is smart enough to know, "Ah, I can see this workload requires a GPU. I'm going to put it someplace. And by the way, I'm going to locate this probably in a VM that it's Kubernetes scheduling in a VM and that VM is locked to a GPU so it gets optimal performance from the RAM, the GPU and the CPU, right? Oh, and here's another workload and this just regards the CPU, I'll do that over there." So we won't think of them as AI applications and applications. There will be one stack that runs all of our applications.

John Furrier

>> And these ones will be invisible.

Dan Ciruli

>> It'll be smart enough to do that.

John Furrier

>> Let's play that through. Let's play that through, because, okay, that means I need to know the resources. I got to know, I got to have software and sometimes it'll be non-deterministic. Okay. So map that out for me. What will happen? Because you have to assume, is it agents doing it? Is it which scheduler runs the VM to make sure that Vera Rubin is being used for the best reasoning and Blackwell gets the weather query? So you're starting to see the Pareto positioning.

Anindo Sengupta

>> That's a great question. And this is something where you're forcing us to talk about something that's coming up in our roadmap, but I'll talk about it.

John Furrier

>> All right, great. Spill the beans.

Anindo Sengupta

>> So in our CTO keynote, Mano mentioned that, but we see this decision across layers. One of the things that's really happening is with tokenomics becoming important, we have to kind of focus on what's the cost for token inference. The AI gateway will be really important taking the first part of the decision. So a semantic layer that says it's a simple question, we can use a small language model and it'll give a very accurate answer versus it's a deep research problem, I still want to use my biggest foundation model available. And so we'll start doing-

John Furrier

>> And the cost.

Dan Ciruli

>> And that's the thing, yeah.

John Furrier

>> The cost associated.

Dan Ciruli

>> cheap model. Yeah.

John Furrier

>> We saw FinOps come out of the cloud. Is it going to be AI FinOps?

Dan Ciruli

>> No doubt about it.

Anindo Sengupta

>> And so there's so many variables, but we can say these agents will only get access to these models. And so there is a lot of optionality at just the gateway layer. But then as you start getting into scheduling, because these are applications running on Kubernetes pods, this is the classic infrastructure problem we solve where how do you schedule the Kubernetes pods? The pods can actually run on VM on a hypervisor and you have confidential compute considerations. You have security and governance considerations. And those orchestration decisions are something that are very easy to take when you have an operating model that has been doing that for years and we just want to kind of do that across all layers.

John Furrier

>> And Dan, your point is, look, it's just going to be an application because that's going to get taken care of in the software, AKA middle cake layer, whatever you want to call it. It reminds me of the shift left conversation. Remember when secure DevOps became DevSecOps?

Dan Ciruli

>> Yeah.

John Furrier

>> Shift left was a big thing. This means applications want security. It just happens now.

Dan Ciruli

>> Yeah.

John Furrier

>> No one says shift left anymore.

Dan Ciruli

>> Right.

John Furrier

>> But that was a movement. Security came on the scene, it just happened.

Dan Ciruli

>> And it's still true, right?

John Furrier

>> We see a similar path for AI integration.

Dan Ciruli

>> Yep. And the infrastructure is going to get smarter and the infrastructure will be able to handle those things. And by the way, for years, we've had intelligence on our hypervisor that is smart enough to co-locate, because we run across multiple nodes, co-locate workloads and data, right? That's been a thing that Nutanix has done for years. It's one of the things that made us so great for VDI because we could run things close to the data. Now, you factor in the GPU and we're already good at co-locating things. So it is going to happen very quickly.

John Furrier

>> Yeah. But we got the general manager and we got the product management leads here for the cloud native. So you got the keys to the kingdom and you got the revenue. Token economics, you mentioned that. Jensen for the first time usually gives a ... He didn't say computer science once on stage, which was that I had the over at three or four. Dave won that bet, although he said token 19 times. He introduced monetization in an Nvidia keynote. Highlighting the business value of tokens. And then of course he said everyone should have an OpenClaw strategy, which is more of a hyperbole. But let's stay with tokens. The economics. How are you guys thinking about the business model and the customers in your world, because you're also enabling a lot of simplicity? Do you charge for that? I mean, how do you guys think about the business?

Dan Ciruli

>> So I think that one of the things that we've been thinking about is that right now it's very, very easy to get access to a model. It's just an API call to get access to a model, but they will charge you per token. And especially in an agentic world where one thing a user does might lead to a hundred agent actions, each of which is consuming lots of tokens. It's going to get very expensive. And so one of the reasons, I think it's a great playground, but I think customers will very quickly have to start thinking about, "Do we call an API where we're going to pay per token? Do we use some infrastructure maybe at a service provider where we're paying for time on that, but then we get to generate all the tokens, you don't pay per token? Or when does it, then does it make economic sense to buy some hardware, run it on-prem and now we're just buying electricity?" And our cloud economists will work with customers on what's the right model for you.

John Furrier

>> Yeah. That's why the finance ops .

Dan Ciruli

>> Yeah. And you said like, will there be ... Absolutely there'll be AI FinOps to help you optimize that.

John Furrier

>> Yeah. You got to own your own factory so you don't have to pay for the tokens.

Dan Ciruli

>> That's right.

Anindo Sengupta

>> Absolutely. And I think the interesting thing is there are two dimensions to this whole cost problem. One, we are actually looking at the choice on models and then how much it costs per token and when you own the infrastructure. The other is infras engineering itself is a new science where for the same model, how you actually manage the model requests, how do you manage the data layer with KV cache? How can you actually change the way you run the model with quantization, actually reduces your cost per token. So that's what we are focusing on. We're building capabilities in AI infras engine to really give customers all the options to actually squeeze out the cost here.

John Furrier

>> Guys, super exciting. I'm looking forward to continuing to cover Nutanix at a whole nother level. So obviously, cloud native is great. AI native's happening and the ecosystem's exploding around it. So I got another run, another CNCF-like run coming. We'll be playing in both ponds. What's next? What's on your radar? What are you guys focused on now? You mentioned a little teaser on the roadmap with tokens. FinOps, a little teaser there too. What are you guys optimizing for?

Anindo Sengupta

>> You want to cover cloud native? And I'll talk about it.

Dan Ciruli

>> So for cloud native, we announced a big partnership with RapidFort. Customers are more secure, security conscious than ever. Supply chain for all of this will be important. We're going to start seeing new attack vectors. So security is becoming increasingly important. We are also seeing both cloud native and AI pushing to the edge increasingly. And we are seeing people who, we've seen this in vehicles, we've seen it in retail stores talking to a government customer about a human born Kubernetes cluster. They want them in backpacks. So the edge is really a break.

Alison Kosik

>> Oh, that just blew my mind.

John Furrier

>> That's a mobile Kubernetes cluster.

Dan Ciruli

>> Is that incredible? But really, so pushing containers out to the far, far edge of something.

John Furrier

>> Hyperconverged edge company.

Dan Ciruli

>> We see a lot of growth. Mark my words.

Anindo Sengupta

>> I think the AI revolution will continue to grow on Kubernetes and cloud native. What we are focused on is really being with our customers and partners and then building through this. So a deep investment in Nutanix AI, which is really around inferencing that we talked about and the gateway product, where we'll continue to bridge whatever technology front is we have to really reduce cost per token, but at high performance and low latency. And then we are working with all our GPU partner ecosystem, because this space is moving so fast. And what we want to make sure is we are there with the latest GPU and the ecosystem capabilities to be able to solve for the simplicity of managing infrastructure. So a lot happening on that side.

John Furrier

>> A lot of great stuff. I wrote a report at MWC, I dropped a thesis called the hyperconverged edge where the AI factories will be small. And you get a DGX box. They're this big.

Anindo Sengupta

>> Right.

John Furrier

>> Yes, intelligence that can change the wireless interface, mesh up the radios. You can have disaggregated Ethernet.

Dan Ciruli

>> And it will be hyperconverged. You'll need to use local storage. You'll need to get the compute out of it, so it fits that classical definition. You have a-

John Furrier

>> You're hyperconverged, you converge, you hyperconverged, you unconverged, you're back again. This is evolution.

Dan Ciruli

>> Yep. Yep.

John Furrier

>> Good stuff guys. Thanks.

Dan Ciruli

>> All right.

Alison Kosik

>> Anindo and Dan, thanks. I know you guys have to get to a keynote. Thanks so much for stopping by theCUBE.

Dan Ciruli

>> Thanks for having us.

Anindo Sengupta

>> Yep. Thank you for having us.

Alison Kosik

>> And thanks for watching theCUBE, the leader in live technology coverage. Thanks for watching.