theCUBE + NYSE Wired: Physical AI & Robotics Leaders | Prag Mishra, Armada

Clips
More from theCUBE + NYSE Wired: Physical AI & Robotics Leaders

Prag Mishra

Chief AI Officer

Armada

play_circle_outline AI on the edge as convergence of physical and digital worlds

play_circle_outline Distributed Cloud Hyperscaler Models for Deploying AI in Oilfields, Mines, Vessels and Remote Industrial Sites

play_circle_outline Real-Time Edge Processing of High-Throughput Camera, LiDAR, LWD and MWD Data for Low-Latency Safety and Control

play_circle_outline Specialized and frontier models require local deployment to avoid latency and bandwidth limits

play_circle_outline AI factories/micro-factories: modular, repeatable, power-efficient nodes for edge deployment

play_circle_outline Sovereign/private AI needs: data locality, security, and operational continuity during outages

Info
Transcript

Prag Mishra, Armada

Prag Mishra

Chief AI Officer Armada

In this interview from theCUBE's New York Stock Exchange studio, John Furrier sits down with Prag Mishra, chief AI officer of Armada, to explore how edge computing is evolving into the next frontier of physical AI. Mishra outlines Armada’s vision of the edge as a fully orchestrated operating environment, combining compute, storage, networking and security to deliver hyperscale-grade AI in remote and mission-critical locations. From oil rigs and mining sites to maritime operations, he explains how distributed infrastructure is enabling real-time intelligence w... Read more

explore Keep Exploring

What are the current trends in AI and what will today's conversation focus on? add

What does your company/solution do, and what kinds of environments or use cases does it target? add

What are the main components, challenges, and design considerations for deploying AI (including GPU-accelerated compute) at the edge in high-throughput industrial environments such as oil rigs and mines? add

What are the implications and challenges of using frontier, hybrid, or specialized AI models that are not local — specifically regarding latency, whether they are trained locally, data transfer and storage tradeoffs, and the infrastructure needed to deploy them at the edge (for example in high-data scenarios like seismic processing)? add

How will AI be deployed at the edge (e.g., via smaller "AI factories"), and what will be the main constraints and considerations (such as power, energy, and price/performance)? add

How can co-locating compute using an AI grid enable distributed intelligence while ensuring data sovereignty, security, redundancy, and continued operation during power or grid outages? add

bolt Powered by CUBE AI

Prag Mishra, Armada

search

John Furrier

>> Welcome back here. I'm John Furrier with theCUBE here at our NYSE studio. We're here with our AI Factory AI Leader series. Obviously robotics, physical AI is hot, agents are hot, but that's just the preamble to physical AI, which is the full convergence of physical and digital world. We're seeing that also in crypto infrastructure as well. Kind of lagging a little bit behind AI, but AI is front and center. You're seeing all the action. Of course, we have a series coming up soon called AI On The Edge, which we're going to release after Mobile World Congress or MWC. But this conversation is really going to be about AI on the edge as the edge starts to become a really critical point for the kinds of serves that's going to change society, our lives, and the value. We got Prag Mishra here, Chief AI officer, Armand, great to have you on. And we were chatting earlier.

Prag Mishra

>> Thank you, John.

John Furrier

>> Thanks for coming in.

Prag Mishra

>> Thank you for having us.

John Furrier

>> So Dan Wright, the CEO, co-founder, was on, came in remote, but we had a great chat and it was very clear to me that the vision is long. It's long term.

Prag Mishra

>> Yes.

John Furrier

>> And what does that mean in this market? But still, it's not the short term. And you guys really have a unique perspective on AI on the edge because AI is going to infuse in everything and certainly the edge infrastructure and ultimately the AI native experiences.

Prag Mishra

>> That's right. That's right. And think of us as the distributed cloud for the edge. And we are the hyperscaler for deploying all kinds of AI and high performance, high throughput data processing applications at remote locations. And when I talk about remote locations, think about the oil fields of Saudi Arabia or the mines in Mojave or the outback or the outskirts of Perth. And think about vessels operating on open ocean and also just about anywhere where you have industrial operations. And there are a lot of places in this world where very critical mission, critical operations take place. And we would like to bring the same kind of compute connectivity that is resilient, secure, and sovereign for all of these different use cases that we want to enable for a lot of our customers.

John Furrier

>> Yeah. I mean, the edge is a term that's been around the industry. I mean, I go back 20 years even when you had the consumerization of IT.

Prag Mishra

>> That's right.

John Furrier

>> The 2007 timeframe, the beginning of the cloud wave. I mean, Edge was an office, it was a desktop.

Prag Mishra

>> That's right.

John Furrier

>> Now the edge... Then it went to industrial edge factories. Now there's a whole nother definition of edge because you have so much going on there. You have the proliferation of mobile, wearables, everything's happening.

Prag Mishra

>> That's right.

John Furrier

>> The wireless technologies are getting much more robust and high speed. Ethernet with wifi, you got Wi-Fi seven coming. So you got tons of kind of connectivity laying out there.

Prag Mishra

>> That's right.

John Furrier

>> But it's changed. And you're talking about a use case that's like bringing an AI factory to an oil right or a boat, like a vessel. So that's not like a box. It's not like a router we see at our homes. It's like we're talking about a full-blown data center, supercomputer system.

Prag Mishra

>> That's right. See, John, very nicely put. See, Edge is not a physical device. It's not just a set of devices that you deploy. It is a whole operating environment, as you said, right? It is networking plus storage, plus security, plus compute, plus connectivity. All of that put together and orchestrated well at the edge. So it's not just a single thing, a device. It is the whole ecosystem of everything that is required for deploying applications and running them reliably, safely, securely. And of course, in many of these places-

John Furrier

>> I was talking to Dan-...

Prag Mishra

>> disconnected.

John Furrier

>> Your CEO, and he and I were hinting at this. Maybe I might have said it, but it's certainly a systems thinking game now.

Prag Mishra

>> Sure. Yes.

John Furrier

>> But the computer industry was computer systems. When the mainframe went to the mini and the mini went to client server, and then ultimately the PC and server Wintel, those were systems. The servers were systems. And now the large scale systems become AI factories. And an AI factory today, we see these big data centers, billions of dollars of CapEx, and the functionality has changed so much. It's really, really good right now and getting better, but those are monster data centers.

Prag Mishra

>> That's right.

John Furrier

>> How do you take that paradigm-

Prag Mishra

>> Very nice....

John Furrier

>> and move it to the edge. We got the middle. We'll come back to metro factories later. But right now I've got centralized monster factories on prem or in a data center in Texas or wherever, but now you bring it to the edge. How do you make the AI work in a distributed computing, heterogeneous infrastructure environment with software? Because that's a kind of standard concept.

Prag Mishra

>> That's right.

John Furrier

>> How does that work in your mind?

Prag Mishra

>> Yeah, that's a great point. So when we talk about AI at the edge, there are several components to it. First of all, it's the modularity, right? We have to make all these processes and some of these processes at oil rigs or mines, they're extremely complex and they have very strict latency. They're very niche in terms of the knowledge and the expert systems that are deployed there and the control loops that are deployed at the edge. So we have to understand all of that in order to deploy this hyperscale grade compute to solve a lot of the problems that we are probably solving in the cloud. But when you deploy this at the edge, you now have to take a look at, okay, what is the data throughput, right? When you are getting, you have cameras, you have LIDAR, you have LWD, MWD logging and measurement while drilling, all these very complex CaDA systems that's streaming data at very high throughput, what does it take to take the data? And by the way, a lot of this data is now sitting at the edge.

John Furrier

>> And by the way, it's new data. It's not trained.

Prag Mishra

>> It's fresh data. Exactly.

John Furrier

>> It's not even trained yet.

Prag Mishra

>> Yes. And the half life of data is very, depending on what kind of application you're looking, it could be from a few seconds after which it's stale. If you have a security or a safety concern, there's no... I mean, you have to act on the data right away. So when you bring that kind of infrastructure and GPU and accelerated compute and an extremely resilient environment, the physicals, the temperatures, the humidity pressure and so on, all the environmentals around it, you're now enabling all those applications that were running in very controlled environments of the hyperscaler cloud. Now you're running them at the edge as reliably. And the best part is that you are able to process much more data than the bandwidth that could handle of streaming all the data to the hyperscaler cloud. So that turnaround time is cut down by a magnitude, right?

John Furrier

>> And by the way, if you add in another layer of complexity or thinking on that, you'd say, okay, what if I'm using models that are either frontier or hybrids or specialized models? That could be either your models or someone else's or theirs.

Prag Mishra

>> That's right.

John Furrier

>> If they're not local, there's latency to go get data.

Prag Mishra

>> Exactly.

John Furrier

>> Okay. Then, okay, is it trained or not? So there's like a whole distributed computing, kind of like an old storage paradigm of high availability versus highly available concepts where it's like, okay, I can get the data. It's available, but it's not like real available.

Prag Mishra

>> That's right.

John Furrier

>> I got to rephrase that.

Prag Mishra

>> Your data.

John Furrier

>> That little nuance kills the latency.

Prag Mishra

>> That's right. Suppose you're talking about seismic processing. You're collecting a lot of data from seismic sensors and you have this huge terabytes of data that you're collecting in a few hours. Now think about it. You want to process them or you have to go back onto land and do landfall and basically transfer them to a cloud compute system there. Or do you want to just cut down the latency to making quick decisions at the edge and then say, look, I'm going to process this data quickly and reliably and run the same kind of applications that I would run elsewhere. And now you're making decisions, right? When you are talking about physical AI, John, it's not just about devices or robots or large models. Models exist, the silicon exists, but the infrastructure to deploy them at the edge doesn't quite... And we are filling that gap.

John Furrier

>> You guys are filling that gap. And that's why I like your company and I like the fact that you're seeing beyond what's today around the corner. Yeah, you knocked down the low hanging fruit. Dan was talking about the go to market. Good. I'm obviously the company's doing successful, obviously doing well.

Prag Mishra

>> Absolutely.

John Furrier

>> Yep. Good golf clap for the team, but seeing around the corner-

Prag Mishra

>> Dan and John-...

John Furrier

>> has questions like, okay-...

Prag Mishra

>> are good at that....

John Furrier

>> networks are networks, networks have latency. So I can design the best network on the planet, but if I can't infuse it, if I can't AI enable it's useless.

Prag Mishra

>> That's right.

John Furrier

>> I see terrestrial networks that are beautiful, but have no AI. Nokia did a deal with NVIDIA. Why? Because they have a network infrastructure, but they don't have AI, but NVIDIA is bringing AI to Nokia.

Prag Mishra

>> That's right.

John Furrier

>> And that's kind of what's happening between the two worlds of distributed computing headers, infrastructure players, but software.

Prag Mishra

>> That's right.

John Furrier

>> Now you have an AI component. Explain this dynamic, because I think this is the big thing that's going to happen at MWC and this year at all the shows is, are you bringing AI natively into the design and the functionality? Explain this. I got infrastructure, but now I got to bring AI to the table.

Prag Mishra

>> So that's a great point, John. See, we have to think about this as a AI native infrastructure, right? When you are thinking about deploying very sophisticated frontier models and they have been fine-tuned for certain custom proprietary data sets, right? Now, when you want to deploy them at the edge, you of course have to enable it by using the right kind of GPUs and the accelerated compute. You do that piece, but then also there's a power angle to it, right? So it's a very interesting mix between power, bandwidth, and compute. And playing those three things, you have to balance them right at the edge in a way that you can run these models with the right kind of connectivity, the right kind of compute, the right kind of the power that's-

John Furrier

>> I was talk about to robotics guys earlier today. We're here for obviously defense tech.

Prag Mishra

>> Sure.

John Furrier

>> A16, Andreessen Horowitz, A16z was here around America, democracy and revitalization manufacturing. But in my interview with another entrepreneur, a founder of a robotics company, we were talking about the models. And I always use that line of The Matrix, that movie, The Matrix, everyone loves. If you're a little older, you know it. If it's a cult movie, if you're young, probably should watch it. He gets jumps into the helicopter.

Prag Mishra

>> That's right.

John Furrier

>> And he's upload how to fly a helicopter. That is the kind of use case we're going to see at the edge. And robots are already kind of doing this. So this brings up the question of how DeepSeek and some of these viral things that are happening with agents where in real time you can distill information. And I want to ask a specific question because I think the trend that we're seeing, and I want you to give me your reaction is, if I'm at the edge and I require precision, low latency data modeling available for AI, I want to go back and get the stuff that's just perfect. That's in essence, distilling and filtering and getting the right model to the edge. If it's a networking, you can come down. So here I walk into a location, I'm moving around the world as a human being, a physical world, and camera sees my face and goes, "There's John. He's with Prag. They know each other. They did a CUBE interview together."

Prag Mishra

>> That's right.

John Furrier

>> Go to... Get John's model and bring it down on that context or if it's a retail outlet, bring down his agent.

Prag Mishra

>> That's right.

John Furrier

>> So these are the things we're going to start to see. Do you agree with that and what's your reaction?

Prag Mishra

>> Oh, absolutely. No, I'll give you a good example of the thing. Suppose you're talking about, let's say a drilling operation in the North Sea, right? It's a offshore drill rig that's operating and they're drilling and then you have all kinds of data that is coming in and they have experts who are sitting on shore making decisions like, okay-

John Furrier

>> That's good data. Which we do make decisions.

Prag Mishra

>> Exactly. What happens if the connectivity is lost? And now you're operating at the edge, either you shut down the operation, the drilling operation, or you have enough compute and that kind of intelligence, as you said, an expert intelligence and the model that you are talking about, this expert large model that can give advice and recommendations to the drillers on site. And that's an export system that has been fine-tuned and understands about all the history of the operations that has happened at that site, that rig over time, and it can recommend the right set of instructions and advice to the human operators, right? It's an advisory system that's sitting there. Think of it as a copilot, a drilling rig, copilot that's sitting there. Now, if you are enabling that, even when you have loss of connectivity or bad connectivity or a cyber attack, you can still operate this drill rig in real time and continue your operations and you'll cut down your NPT, which is the non-productive time, you can increase your ROP as well as the well placement continues as normal operations would, even without these connectivity and .

John Furrier

>> That's a great example. And I want to follow up by saying that I'm seeing other examples of that kind of discussion, but I also want to add in simulation to this.

Prag Mishra

>> That's right.

John Furrier

>> Because NVIDIA and I had a briefing with their head of healthcare-

Prag Mishra

>> Oh, absolutely....

John Furrier

>> and the AI operating room, AIOR, they call it, they have run simulations in advance. So as the doctor's doing the surgery, you got computer vision everywhere, there's some surgery going on, not yet a robot, but it could be a robot, whatever. Based upon the condition of the patient at any given time, based upon the simulations, there's agents ready to deploy action and give doctor real time feedback or bring in other doctors from around the world via camera. So you're starting to see this kind of network effect and simulation, not just agents deploying because they have to act on some data.

Prag Mishra

>> That's right.

John Furrier

>> Take me through your AI thinking around this, because I think this is going to be a template.

Prag Mishra

>> This is a great point. So if you employ an AI platform right at the edge that does simulate, it has access to all the large models, it has access to a lot of simulation results and it can continuously keep on learning and as at the same time come up with all scenarios like, okay, here's a simulation of what if this happens and this happens at every single time, given what the state is thus far, it starts predicting what's going to happen in the next few hours or few seconds, right? And when that happens, it comes up with the right set of, okay, I need help for this. I need an expert. I need a doctor who very well understands how this surgery can be best done because that person has done similar surgeries in the past. So depending on what the simulation results are, and depending on what the state of that patient or the surgery operation that has taken at that point, it can start recommending who to bring in, how to orchestrate the whole thing. But you need that kind of compute as well as reliable connectivity to make that happen at the edge.

John Furrier

>> You got NVIDIA. You have a relationship with NVIDIA, they got Alpamayo, vera Rubin, AI infrastructure, inference, AI factories, agentic and open models, networking, physical AI, a lot of success in vertical industries and ecosystems. So this is an area I want to get your thoughts as an AI leader.

Prag Mishra

>> That's right.

John Furrier

>> As you guys and your team look to do the edge, which I'm big bullish on, this is going to be a huge growth area. And when it hyper converges talk about that at MWC. What is your relationship with NVIDIA and how do you see the market evolving? Because I think NVIDIA has made... We were just talking about bets, I think it's a sure thing.

Prag Mishra

>> That's right.

John Furrier

>> They kind of made the bet years ago, but it's kind of playing out physical AI. What's around the corner for you on this one? How do you see AI playing out and what's your relationship with NVIDIA?

Prag Mishra

>> Sure. John, so with NVIDIA, we have a deep partnership and we are working with several of the teams. Of course, we use NVIDIA's Exterior Compute, GB200s, H200s, H100s, and even the past generations. So we bring all the NVIDIA as well as, of course, other kinds of compute as well. We have a very close tie. We also employ a lot of the software and also the platform that NVIDIA has built. But in order to get them to work at the remote edge, we need to build that kind of infrastructure and make that kind of infrastructure in terms of the environmental monitoring and the racks and the cooling and the power and so on available. So with NVIDIA, we have a close partnership. Now, your second part of the question is what's around the corner, right? As we get deep into all these really very large industrial operations like oil and gas, mining, logistics, manufacturing, we are seeing a lot of the problems that they have been facing for decades, right? And they have closed loop control systems, they have IT, OT divide that they are trying to solve. So now when we bring all these specialized compute from NVIDIA and others, we now start thinking, okay, what does it take to integrate OT systems with IT systems? And because the OT systems are extremely closed and rightfully so.

John Furrier

>> And brittle.

Prag Mishra

>> And brittle.

John Furrier

>> Yeah.

Prag Mishra

>> Exactly. And so now it becomes a very interesting problem to solve is how do you balance these embedded systems?

John Furrier

>> It's called a DGX box from NVIDIA, put a DGX box in there and get Thor, develop on it, and you're good, right?

Prag Mishra

>> Well, yes.

John Furrier

>> NVIDIA, you owe me one for that. I'll tell , he owes me one for that. No, but this is a question because this comes up a lot. So the transformation, I believe, is going to be very rapid accelerated.

Prag Mishra

>> That's right. Absolutely.

John Furrier

>> And that's why I asked about Thor because their development environment is hot right now on the edge. Everyone I talk to who's doing real edge coding, either on the device side or writing software for radios or whatever, you're seeing a lot of Thor, and NVDIA is rushing to get some reference implementations together. So it's kind of emerging.

Prag Mishra

>> Very true. Yes. Yes.

John Furrier

>> I mean, it's clearing up fast, but like it's emerging. So I'm trying to settle this in my mind for GTC and MWC coming up, what that's going to look like from an AI perspective because that's going to open up the floodgates of innovation on the AI side. Your job turns into, okay, how fast can I get capabilities up and running?

Prag Mishra

>> That's right. And for a lot of our customers, that's exactly the same question, right? When they are talking about this kind of compute, they're always talking about, okay, how do I create very stable networks at the edge? How do I stream data? How do I process what data at what nodes? So that is being worked out, right? And that is where you think about all these models that are distant, pruned, quantized, and you get these very specialized models that are specialized for certain applications, certain tasks, and they are run on some specialized compute hardware that's available. But we are trying to make it so modular, just like a AI factory, we are trying to make it modular repeatable so that we can deploy them at scale and still be resilient and still be extremely secure.

John Furrier

>> And then at runtime, I use the word run time, I'll put that in quotes. So as the chief AI officer, you got to start thinking, and I'd love to get your thoughts on this, is what does a runtime like environment look like? Again, it's distributed computing. So you're swapping out models at it all the time, you're making decisions about what's coming in, because remember new data's coming in.

Prag Mishra

>> That's right.

John Furrier

>> It's like, what do I do with it?

Prag Mishra

>> And by putting all this computer at the edge, a new thing is happening is that people are putting more and more sensors at the edge because they said, "Oh, you know what, my network wasn't not able to handle 10 ATP frame and 30 frames a second or .

John Furrier

>> And we've seen that move before. The more horsepower you bring to the device or product, the better more software gets better. Yeah.

Prag Mishra

>> Yep. It opens up more applications, it opens up more sensors, it opens up more modalities, right? So when you have new modalities of sensing that is being deployed, hyperspectral imaging or seismic, acoustic, LiDAR, all of that specialized sensors, several different modalities are being brought in just because they have now the ability to process and act upon the data, that is now opening up so much more in terms of what kind of applications, what kind of computer. So it's almost like a feedback loop, right?

John Furrier

>> It's a good one. I think it's going to be one of the biggest ways because AI on the edge, which we're going to showcase a lot of this year, is really a hybrid of physical and networking basically.

Prag Mishra

>> That's right. Exactly.

John Furrier

>> So you bring AI factories, which I won't say fungible, but like the model of an AI factory today is going to be smaller, power efficient, all the things that are needed at the edge, depending on how big the edge is.

Prag Mishra

>> And that's a great point. The word I would say is power, right? How much energy is available? We are always going to be throttled by energy, right? It's a very interesting thing, right?

John Furrier

>> It's called a mini factory. I mean, it's just power, both power energy and power performance, price performance. It's dependent. I could drop a DGX box and make that a factory and call it a factory, whatever you want, a factory node.

Prag Mishra

>> That's right.

John Furrier

>> I mean, it's like a micro factory.

Prag Mishra

>> That's right.

John Furrier

>> But it's doing something, but it's talking to the factory.

Prag Mishra

>> That's right.

John Furrier

>> So this is where I think the distributed computing heterogeneous networking aspect of our computer science over the decades is going to shine because cloud is just one big fat edge node.

Prag Mishra

>> That's right. Yes.

John Furrier

>> It's a big node.

Prag Mishra

>> It's like slightly farther away, right? And the cloud is always going to be there, right?

John Furrier

>> Yes.

Prag Mishra

>> We're going to do all the specialized training and all kinds of validation in the cloud.

John Furrier

>> Yeah. Matt Garman's all over. He was just speaking at an event yesterday we were at with Cisco and he's saying that provisioning is going to be a generative process.

Prag Mishra

>> That's right.

John Furrier

>> So remember the old provisioning infrastructure.

Prag Mishra

>> That's right.

John Furrier

>> He's like, they spin up services on Amazon. So I would see more spinning up things.

Prag Mishra

>> That's right.

John Furrier

>> Generatively.

Prag Mishra

>> That's right. And once that happens, right, you're doing something of course it's a hybrid edge cloud play, but I think that's unlocking of what can be processed at the edge in real time, data gets converted into information or even at times knowledge.

John Furrier

>> I mean, my working theory right now, based upon the data that I've had and some of the things we're working on research side is hyper converged edge is going to happen with the wireless protocols will be completely invisible.

Prag Mishra

>> That's right.

John Furrier

>> Factories will be connected to them. They'll just be transceiver antennas, radio networks, and then you're going to see hybrid, private cloud, multi-tenants-

Prag Mishra

>> Very . Yes.

John Furrier

>> Because if that happens, then you have multi-tenancy by default.

Prag Mishra

>> That's right.

John Furrier

>> That's going to also scale into the network.

Prag Mishra

>> That's very good. Yep.

John Furrier

>> Central to edge and then fills in on metro clouds, building clouds, you're going to have all kinds of conversation if the networks are ethernet, cable or wifi or spectrum, and it's converged, that's invisible.

Prag Mishra

>> That's right.

John Furrier

>> And then everything's up for grabs.

Prag Mishra

>> And think of Armada as the control plane for the hyper converged edge, all the physical AI that needs to run at the edge, right? So Armada is filling-

John Furrier

>> Can you imagine an AI factory in a building in Manhattan that has power?

Prag Mishra

>> That's right.

John Furrier

>> And a lot of wasted power, by the way.

Prag Mishra

>> That's another point. Yes.

John Furrier

>> And you say, okay, grid them all up. Now you have energy sharing.

Prag Mishra

>> That's right.

John Furrier

>> Ethernet backhaul, wireless networks. So this is really going to be a very interesting time if the telcos and the NVIDIAs can pull together. And I think it's a huge opportunity for NVIDIA specifically to bring AI to the dumb pipes.

Prag Mishra

>> Totally.

John Furrier

>> And you guys benefit.

Prag Mishra

>> And John, the one point you mentioned that is extremely important, there are places where we have excess energy, right? Think about the flare gas in Bakken or Permian Basin, right?

John Furrier

>> Yeah.

Prag Mishra

>> You have plenty of energy that's not being used efficiently and wisely. And now, what if you were to bring the compute there and be able to convert all that excess energy, as you just mentioned, all that excess energy to compute, and then because you don't have the infrastructure to take all that energy and distribute it into the grid.

John Furrier

>> I'm smiling because you reminded me of that buzzword. I think in the late '90s or early 2000s, maybe in the mid '90s, grid computing.

Prag Mishra

>> Yes.

John Furrier

>> Was a different definition, but the idea was everything's on a grid. I mean, I really believe that if Jensen and the team and-

Prag Mishra

>> AI grid. Yes.

John Furrier

>> You guys continue to do what you're doing, you can instrument everything and create a grid of resource.

Prag Mishra

>> That's right.

John Furrier

>> Compute, resource, energy, controlled by the agents in the factory.

Prag Mishra

>> That's right.

John Furrier

>> That to me is a very viable scenario.

Prag Mishra

>> Very much. And you basically, you push compute to wherever there's excess energy and convert that energy into compute. And so we could be able to run a lot of our models and training them and validating them and doing all kinds of sophisticated fine-tuning on top of that.

John Furrier

>> And to illustrate that point further, I'll say that the big buzz this last week was the Claude Bot now, multiple, whatever they're calling it, started by a couple of friends of mine, actually, one of my good friends, it would create a social network for you.

Prag Mishra

>> Oh, totally.

John Furrier

>> Okay. And it was very gimmicky, but it shows the power of how agents can sell form and reason. Why couldn't it do that for energy? Go get me some energy.

Prag Mishra

>> Actually, that is a very interesting point. Once we are able to kind of look at that level where we can orchestrate different AI agents-

John Furrier

>> It's the sign of the future....

Prag Mishra

>> and start... So that's the thing, right? So if we have a AI grid, now just the riff on-

John Furrier

>> Yeah, riff. Yeah, let's riff it.

Prag Mishra

>> The AI grid that you are talking about, right? So now it's a complete different grid of compute and it overlaps on the grid for power. There are all these nodes of grid power where power is generated, like think about hydroelectric or solar or wind farms or even coal and nuclear power plants. You have pockets of power that is generating basically all these sources of energy. And now think about co-locating the compute using the AI grid wherever you have-

John Furrier

>> You're building an intelligence layer into the system.

Prag Mishra

>> Distributed intelligence, John, is distributed.

John Furrier

>> Distributed intelligence.

Prag Mishra

>> So you not only get resilience and have redundancy, but you also make things secure because now you're able to basically constrain these models, keep basically the whole aspect of sovereign AI, the sovereign-

John Furrier

>> Sovereign AI and default is private cloud too. What is sovereignty? Private.

Prag Mishra

>> That's right.

John Furrier

>> So everyone's saying, oh, private cloud, but sovereignty will be a feature.

Prag Mishra

>> That's right. And there are many places where-

John Furrier

>> Not a platform.

Prag Mishra

>> Yep. There are many places where you cannot take the data out of not just the boundaries of the nation, but boundaries of the confines of the factory, right? They're so proprietary, they're so sensitive. So now think about keeping all those, and as you said, these areas could have their own power sources and even if the grid wet be down for some reason, right, maybe it's a cyber attack or maybe it's a grid imbalance that has created an outage, these pockets of compute, along with co-located power, will still be able to operate in a very sovereign and private manner.

John Furrier

>> Well, we'll put that on our list of startups to do after we're retired or tomorrow. Prag, great to have you on. You're living on the edge. Your company.

Prag Mishra

>> Thank you.

John Furrier

>> Great job. Congratulations, Dan for me.

Prag Mishra

>> I will. .

John Furrier

>> Our AI On The Edge series will be kicking off this quarter after MWC. Maybe we might wait until after GTC, but we'll be exploring what's going on at the edge, living on the edge, AI on the edge. It's going to be a very dynamic, robust, but lucrative opportunity to extend functionality of factories, AI factories, micro factories, all kinds of new use cases. If you're interested in that topic, hit me up on DM or LinkedIn. Again, the AI Factory series, our AI Leader series, robotics is all happening here in theCUBE. I'm John Furrier, your host of theCUBE. Thanks for watching.

Prag Mishra

>> Thank you.