In this segment from theCUBE + NYSE Wired’s AI Factories event, Eran Kirzner, founder and chief executive officer of Lightbits Labs, joins host John Furrier to discuss the critical role of software-defined storage in the next wave of AI infrastructure. As the industry pivots from massive training clusters to real-time inference, the demand for agility and low latency becomes paramount. Kirzner details how Lightbits Labs leverages NVMe over TCP to transform commodity hardware into high-performance, scalable storage systems, effectively replacing rigid appliances with flexible, cloud-native architectures. The conversation highlights the necessity of "feeding the beast" – ensuring expensive GPUs remain utilized through autonomous provisioning that reduces setup times from hours to mere minutes.
The discussion delves further into maximizing data center efficiency, with Kirzner explaining how software-defined approaches allow for dynamic workload orchestration between training and inference tasks. He outlines how Lightbits helps enterprises and neo-clouds – such as Crusoe Cloud – reduce their storage footprint by up to 50% while maintaining high reliability and security standards. From the concept of the "AI Garage" to the complexities of multi-tenancy and hybrid cloud sovereignty, the interview explores how data-centric strategies are enabling organizations to optimize resource allocation, eliminate idle GPU cycles and build the resilient infrastructure required for the future of AI factories.
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
theCUBE + NYSE Wired: AI Factories - Data Centers of the Future. If you don’t think you received an email check your
spam folder.
Sign in to AI Factories - Data Centers of the Future.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Register for AI Factories - Data Centers of the Future
Please fill out the information below. You will receive an email with a verification link confirming your registration. Click the link to automatically sign into the site.
You’re almost there!
We just sent you a verification email. Please click the verification button in the email. Once your email address is verified, you will have full access to all event content for AI Factories - Data Centers of the Future.
I want my badge and interests to be visible to all attendees.
Checking this box will display your presense on the attendees list, view your profile and allow other attendees to contact you via 1-1 chat. Read the Privacy Policy. At any time, you can choose to disable this preference.
Select your Interests!
add
Upload your photo
Uploading..
OR
Connect via Twitter
Connect via Linkedin
EDIT PASSWORD
Share
Forgot Password
Almost there!
We just sent you a verification email. Please verify your account to gain access to
theCUBE + NYSE Wired: AI Factories - Data Centers of the Future. If you don’t think you received an email check your
spam folder.
Sign in to AI Factories - Data Centers of the Future.
In order to sign in, enter the email address you used to registered for the event. Once completed, you will receive an email with a verification link. Open the link to automatically sign into the site.
Sign in to gain access to theCUBE + NYSE Wired: AI Factories - Data Centers of the Future
Please sign in with LinkedIn to continue to theCUBE + NYSE Wired: AI Factories - Data Centers of the Future. Signing in with LinkedIn ensures a professional environment.
Are you sure you want to remove access rights for this user?
Details
Manage Access
email address
Community Invitation
Eran Kirzner, Lightbits Labs
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.
In this segment from theCUBE + NYSE Wired’s AI Factories event, Eran Kirzner, founder and chief executive officer of Lightbits Labs, joins host John Furrier to discuss the critical role of software-defined storage in the next wave of AI infrastructure. As the industry pivots from massive training clusters to real-time inference, the demand for agility and low latency becomes paramount. Kirzner details how Lightbits Labs leverages NVMe over TCP to transform commodity hardware into high-performance, scalable storage systems, effectively replacing rigid applianc...Read more
exploreKeep Exploring
What is the main focus and topic of discussion in the AI Factory series being featured on theCUBE?add
What is the significance of the NVMe over TCP protocol in modern data centers and its impact on AI data processing?add
What makes storage a strategic component in a data center architecture, especially in relation to GPU utilization and the use of commodity hardware?add
What is the relationship and differences between training and inference in the context of GPU requirements and resource management?add
What are the advantages of using elastic storage and provisioning in a model that requires frequent changes?add
What advancements have been made in software-defined storage, particularly by Lightbits, in terms of ease of deployment and performance?add
>> Welcome back, everyone, to theCUBE. I'm John Furrier, host of theCUBE here at our NYSE studios, of course. We have our Palo Alto Studio connecting Silicon Valley and Wall Street, part of our NYSE Wired program and community. This is our AI Factory series where we feature the leaders making it happen, building out the infrastructure to bring in the era of AI as we enabled more tokens, more tokens per second, tokens per watt. As intelligence becomes the output of the factories, data is critical and all that and how it's organized and how it's built will enable in this next wave of agents. Eran is here, Founder and CEO of Lightbits Labs. Great to see you, thanks for coming on theCUBE. We're going to talk storage, obviously the hottest topic, memory, storage. Thanks for coming on.
Eran Kirzner
>> Yeah, thank you for having me. It's exciting, the energy here, everything is kind of pumped up. Yeah? Thank you.
John Furrier
>> The option floor behind us is very active, trading floor. Obviously on the equity side, it's still programmable, it's all machines now, but a little bit different. But I mean, the market's pretty hot too. I mean, the enthusiasm right now for AI infrastructure. We just came back from the GSA Semiconductor Awards program. The reset's happening, but the game is still the same. The data centers, the large scale data centers, the neoclouds, you got the hyperscalers all still talking about storage, networking, and compute, and database. As large scale systems become these large AI factories or super servers or supercomputers, the role of storage is key, you guys are a critical piece of that. Talk about what your firm does, because you guys take a classic approach to software defined in an era where it's all software defined and all the hotter stories around NVIDIA, for instance, is about CUDA, CUDA-X. You're starting to see the role of software and hardware continue to be emphasized.
Eran Kirzner
>> Yeah, yeah. So what it's like, we're doing, basically we founded the company, we invented a protocol called NVMe over TCP, and that's basically allow you to scale data center. And between data center, basically storage can be anywhere within the data center. So, we build a protocol and on top of that, we build an enterprise grade storage. Now in the context of AI, as you mentioned, it's super important to get the data on time, and latency is important, especially when you move from training to inference. It's not just about how much data you can get, how fast you can get the data, how interactive you can get the data.
John Furrier
>> Talk about the new requirements, because it's funny because the game is still the same. You still got computers, you just have servers, but now that's just all clustered together. You got the interconnect, super hot right now on optics. You're seeing that these interconnects are bringing machines together, processors. The compute is clear, we see with the accelerated compute side. Now, the storage becomes more strategic.
Eran Kirzner
>> Right.
John Furrier
>> Talk about the nature of the why storage is strategic, where it sits in the architecture, GPU utilization. These aren't inexpensive devices.
Eran Kirzner
>> Yeah.
John Furrier
>> They need to have that be utilized.
Eran Kirzner
>> So, let's talk about two elements. First of all, when you're building a neocloud, when you're building a data center, you're using commodity hardware, right? So, you cannot use appliance for that. So, your storage need to be act exactly as your other kind of element, your compute, like your networking. Everything is software defined. Network software defined and GPU software defined, and also you want storage to be software defined. Why you want it? Because you don't have SSD today, for example. It's very tactic, but you can get access to 1536 SSD, you need 38 terabytes SD. So by having a software defined, a true software defined, you can run it with whatever exists today in the market. So you are not bounded by any kind of DDR limitation or SSD limitation. The other point of software defines that you can really make it scalable and agile as you need. Right? So you can have it running on 10 rack, 20 rack, 1,000 rack, right? And in order to satisfy your kind of requirement. And you talked about the shift in requirement. Shift in requirement is basically what we see today is that people start to move from only training 70% of their revenue of the neocloud today was about training, and now it's becoming more and more inference. When you have inference, the requirement are different. It's not just about the volume of data you can feed into the GPU. It's about the effective cycle you can get from the GPU. And the effective cycle is about the latency and it's very interactive.
John Furrier
>> Talk about the dynamic between training and inference, because you start to see pattern emerging where it was like, okay, big clusters for training, inference has always been the killer app, it's been called. Now you have as things data comes in, proprietary data is the crown jewels. Training and inference is going hand in hand because once you're trained, then you infer. And then once you have the inference, you still got to do some reinforced training and learning. So you have this cycle of constantly iterating through and then you got specialty models emerging and vertical models are emerging, financial services, you name the vertical. There's domain specific things going on. So, you have a blending of that.
Eran Kirzner
>> Yeah. So we have three type of kind of leading segment customer. We have e-commerce customers, we have financial customer, top financial customers, and data center. Right? And as you mentioned, when you build a data center, you need to be agile enough, you need to become flexible enough to switch between training and infrastructure. When you look at the hyperscaler, I agree that we have one chunk for training, one chunk for inference, but when you are mid-sized data center, you need to scramble between the two. You need to have the ability to provision and de-provision, to checkpoint and recover. And this is back going to us, to the storage. How fast your storage can checkpoint everything to the storage, remember exactly where we are in the training, then get the burst of inference for some rush hour, and then bring the training again on the same machine, same GPU. Now, if you take it one step further even, then you will have customer in silicon. You will have customer chip for different tasks, for different model of inference, right?
John Furrier
>> Yeah.
Eran Kirzner
>> So basically, the architecture has change rapidly.
John Furrier
>> So, define agility because what you're basically bringing up is a use case where you could have general purpose training clusters. Hey, throw all the training jobs at this system, throw all the inference here. But then you're talking about more agility where you have to orchestrate things with software.
Eran Kirzner
>> Exactly, exactly.
John Furrier
>> Balancing workloads between what they need at any given time, the context. It's like IO, right?
Eran Kirzner
>> Yeah.
John Furrier
>> I mean, in a way, these tokens have to deal with these situations. Talk about why agility is important, define agility.
Eran Kirzner
>> Yeah. So I think you kind of mentioned it, you hit it, right? So if you look fast forward, like 18 months from now, 12 months from now, this agility have to take place because you have different model, multi-tenant, you need a quality of service, you need different security. All of that need to go all the way from the endpoint into the storage. Right? The agility is important because basically the model will be changed, you need different permutation simultaneously to support. Right? So if you go with kind of a very fixed hardware size for the storage, that's what you have, right? You bought like one petabyte, 10 petabyte of storage, one fixed, one set of characteristics. If you go with the software refine, you can basically do whatever you want with it. It's like the-
John Furrier
>> It's elastic.
Eran Kirzner
>> It's very elastic, it's like you have your own cloud and you have full flexibility on the cloud. And basically, we are running on any standard machine. So when you decommission us, you can run compute on those machine. When you need us, you commission us again, you deploy it again. Now, when you look also on a real kind of GPU or AI factory, provisioning is also important. So, it's not just pumping the data in and out fast, it's how fast you can provision. Think about you have like 8,000 or 20,000 GPU now that you need to provision. It's bears that need to come from the storage, basically. You need to get all the images, all the model into, all the CPU and GPU, simultaneously. I can tell you what I hear from my customers. Before they started to use Lightbits, they basically had to provision hours. So, think about it. You run your credit card, you start to get access to the GPU, but you have zero cycle, zero effectiveness. Right? So, if I tell you that now your provisioning can take instead of eight hours, it will take eight minutes.
John Furrier
>> I mean, I'm smiling because the term that we used to say in the industry was, "Let's stand up, cluster." That's basically provisioning.
Eran Kirzner
>> Exactly.
John Furrier
>> And it's a lot of work. I mean, it's manual labor, there's configurations involved. What you're getting at is software define is, it's all done autonomous.
Eran Kirzner
>> Yes.
John Furrier
>> So, this is where the agility comes in. I think this is the key point.
Eran Kirzner
>> Exactly.
John Furrier
>> All right. Talk about what this enables because I used to use the word disruptive enablement means something's disrupted, but people don't like that term because it's more accelerated enablement. I buy that, I can see that point, but you're disrupting the old architecture with the new architecture. You got open source software is booming. Commodity hardware is booming still. Yeah, there's some custom chips out there for use cases, but the need to manage that custom, I mean that off the shelf hardware, is where the software comes in. Talk about that piece because I think this is where people have to deal with either preexisting hardware or they're going to make purchases. So okay, I don't want to stand up, I don't want to have to provision. I just want my workloads to run. And what is that going to enable on, say agents, for instance? I can see agents being dependent upon having low latency software orchestration. Take us through your vision on how the enablement kicks in.
Eran Kirzner
>> Okay. So, we talked about the shift from training to inference, right? It's all over, right? And inference, basically we're talking about AI factory. I would say let's call it AI garage, okay, because every enterprise will have AI factory, right? So, it will not be in the scale of huge xAI and Meta, but everyone will have a factory. Okay? So, constraint-
John Furrier
>> Factories, plural. They'll have multiple factories.
Eran Kirzner
>> Multiple.
John Furrier
>> Edge factories.
Eran Kirzner
>> Exactly, exactly.
John Furrier
>> Micro factories.
Eran Kirzner
>> Exactly. So you can understand the requirement change and also the constraint of, can you accommodate this in a small space? Can you get in a less power consumption? Can you get it more effective? Okay? So today, basically what people are doing, taking their blueprint of training and running inference on top of that, right? But when you try to squeeze everything to a low footprint, lower power consumption and make it more effective, as you mentioned, it's going to be very interactive. Okay?And now inference also by nature, it's different. The way they consume from the storage point of view, it's different, right? In training, all the data is already well known, right? You know, you have all the Bible, you have all the kind of Wikipedia, you have all the movie in the world, and you know then we just need to pump in into the GPU. In inference, it's very interactive like the discussion we have right now. I don't know what you're going to ask me, we haven't planned it. You don't know what I'm going to tell you, right?
John Furrier
>> It's generative.
Eran Kirzner
>> Right, it's generative, and you need to process it in real time and you need to get a response. Now think about it, as you mentioned, you have a lot of agent, each one of them waiting for something. So, latency is the king. Okay? So, switch from throughput to latency and to effectiveness.
John Furrier
>> Yeah. We have an expression on theCUBE pod, Dave Vellante and I, we call it feed the beast. The feed the beast is the GPUs. They need to be fed the data, that's what you're getting at. So, I have to ask you about software defined storage because that's a term that goes back a decade. Software defined networking was really a key part of the hyper-converged. Now we're un-converging them, or if you will, or connecting them. We got this hyper-converge at the edge coming quickly, that's going to happen. But now on these central AI factories, you have all this gear waiting to do stuff and now agents are going to be software. So, feeding the beast is-
Eran Kirzner
>> It's the king....
John Furrier
>> really the key, and that's where software comes in, right?
Eran Kirzner
>> Yeah. So you're right, software defined storage was here for ages, right? Right, at least 20 years already. Everybody familiar with Seth, for example, from Reddit, IBM. What Lightbits did, is two things from in the context of software defined. A, make it really easy to deploy. Everybody can deploy it, you don't need an army of engineer. It simply works out of the box immediately. That one, make it easy, we don't need tons of professionals. And the other point is, make it really fast. Because previously if you want something fast, you have to buy an appliance. You want something reliable, you want to go and appliance. We actually shifted to the point that software defined storage is now reliable, fast, and easy to use. Okay? And scale, because when you're looking at the deployment model today, it really need to scale. So that characteristic is plugin and with the type of customer we have, it really need to be kind of-
John Furrier
>> The benefits really is the provisioning speed, the agility on orchestration of workloads, making sure you get the right resource. What's the action item for your customers and prospects? Because people watching are trying to squint through all the hype. It's the same system architecture, but little bit of a different twist. Obviously you got tokens, the intelligence there. What is the to-do action item for folks who are thinking about extending their factories, building these factories? What's the consideration? What should they think about?
Eran Kirzner
>> I say, if you wake up tomorrow morning and say, "Okay, I can make my AI factory more efficient," you can look at Lightbits, okay? Because we can basically run on your existing server, we can augment what we have today. You don't need to replace it, you just augment it. We can augment what you have today. You can just connect it like you connect to any other local SSD, because what we built is basically standard. Okay? So it's coming in-box with all kind of the... If it's Linux, if it's VMware, if it's OpenStack, if it's Reddit, if it's Kubernetes. Whatever orchestration system you're using today, we already plugged in. Okay? We have zero footprint on the client side, everything we put on the client side, it's actually coming, it's coming from the community.
John Furrier
>> So if a customer is invested in cloud native, you guys are an ideal solution because they can use software defined storage for their existing stuff and then bring in some of the AI factory stuff.
Eran Kirzner
>> Exactly, exactly.
John Furrier
>> They have some flexibility there.
Eran Kirzner
>> Exactly. Yeah.
John Furrier
>> All right, talk about efficiency because I think this is something that's come out a lot. People in this hype market will just buy a bunch of GPU. I got my GPUs, but you don't really need GPUs for everything.
Eran Kirzner
>> Yeah.
John Furrier
>> You can use compute. We're seeing a lot of use cases where this orchestration or intelligence and software can match resource allocation, whether it's HPM, storage, networking, and move things around with the workloads. Talk about that efficiency piece.
Eran Kirzner
>> Yeah. So on a single footprint from the storage point of view, we are going to be at least five times the performance of what you have today. So, seriously faster, and same thing in latency. And from efficient point of view, it's also consolidation. We can squeeze more kind of density. So we can run on TLC, we can run on QLC. That means that from efficiency point of view, we have customers that actually reduce their footprint by 30 to 50%. And this is also data reduction. Okay? So data reduction, more kind of high density kind of type of SSD, and going to the next level of kind of .
John Furrier
>> All right Talk about the company, talk about some of the momentum you have, obviously the use cases. Talk about some of the stats and run some of the .
Eran Kirzner
>> So basically we have three segment of customers, e-commerce, including the largest e-commerce worldwide. We have financial, top financial, including bank, a hedge fund, and credit card companies.
John Furrier
>> A lot of transactional activity.
Eran Kirzner
>> Transactional, a lot of transact... Everything is transactional we excel. Analytic, fraud detection, all of this stuff. And the last one is cloud. Cloud provider include neocloud provider. And I can mention some of the public customer we have, like Crusoe Cloud is our customer and Moonlight and Nebul in Europe. So, there's a lot of neocloud providers.
>> Yeah, yeah, yeah. So, our customer love the solution. It's easy to use and it's running on the commodity hardware. So they don't need to buy some fancy stuff. If they're using supermicro to continue with supermicro, using Dell, HP, Lenovo, all of that we're running. Right?
John Furrier
>> They want to get their hands as much gear as possible and then squeeze as much value. I mean, we're in a squeeze the resource, maximize efficiency. I mean, people don't want to pay big bucks for GPUs when they can't get the best out of them. All right. So what's the roadmap look like? What are you working on now? What are some of the cool things you're working on?
Eran Kirzner
>> Yeah. So basically, the few things happen in parallel in the company. First of all, we are increasing the number of customers, right? Especially cloud and cloud service providers, edge cloud and so on. That's one. From the product point of view, continue to enhance the product, a lot about kind of security, multi-tenancy, scale out. So multi-cluster, , cluster simultaneously across multi-data center. So this is one aspect that the product is moving forward. The other aspect is inference focus. We are basically coming with a solution, which will be, I would say killer for inference. Okay? That basically will keep your GPU zero idle cycle. Okay? Because if you look at inference today, it's like 50% of the time, the GPU is either crunching the same thing again and again or waiting for data. We are going to solve it.
John Furrier
>> Yeah. I mean, so you're bringing up two use cases. I want to double click on that. So the GPUs are sitting there turning back on basically. We can all relate to our ChatGPT thread that's sitting there waiting.
Eran Kirzner
>> Large context, yeah.
John Furrier
>> It's like, okay, reboot the thread. That's got to kick off a GPU cycle. And a lot of people are using OpenAI to do basic math, they really don't need a GPU for that. 1.1 is two, I get that. And so that's one. And then the other one is the wasted cycles waiting. So waiting and then turning on, which means crank up.
Eran Kirzner
>> Yeah. So, let me double click on that. So two things, as you mentioned. One is repetitive work. Okay? Think about it. You mentioned the context or you gave the example of ChatGPT. So, thousands of millions of people logging in, right? The context is there, but you need to select the right GPU. So the data actually need to move from one GPU to another. We are taking care of that. Okay? That's one thing. And the second is recalculating, right? Because your stuff was there and now you need to re-memorize that, re-initiate it, right? So, if somebody will give you the information already been calculated, bang, less your effective cycle versus just cycle.
John Furrier
>> It's funny, they use the term memory, but it's not memory-memory. It's like remembering what was in the context window at that time.
Eran Kirzner
>> Exactly, yeah....
John Furrier
>> which is why everyone kind of starts a new chat because they just get a fresh brain, basically. This memory piece is super important. That also speaks to, if you'll squint through that concept, the important role of storage, because you can do a snapshot. There's a lot of offload and acceleration concepts going on in storage where we're seeing storage becoming key. I was just at AWS re:Invent talking to some of the S3 team and they're like, "Things are changing very fast. Why move the data around? Let it sit there, just move an agent to it." So you're starting to see storage become also an architecture enabler for managing data.
Eran Kirzner
>> Yeah, yeah, yeah. You will have, taking it even one step further, you will have some kind of computation and some sophistication, all the storage itself. Okay? So analyzing pattern, allowing the things to be feed in into the GPU before we actually ask for them.
John Furrier
>> All right. So question for you is, for the folks watching, what should they not do? What is the third rail, if you will? What is the thing that they should not do with storage? Just buy a bunch of SSDs or, I mean, what is the best practice not to do? Because a lot of people are buying gear, like I said, and people are used to that, they can just procure, "I got some storage, we're good," or obviously software will play a role. What should they focus on? What should they avoid?
Eran Kirzner
>> Yeah. So, it's a great question. I would say a few things. One, if you care about security, just don't use the vanilla kind of SSDs that you have today. Somebody need to get everything secure. That's one. Second, reliability. Again, same thing. It's you have so expensive kind of workload running now, you don't want it to fail. You want everything to continue to run, so reliability is not less important than the kind of performance. So you want something that actually give you the five, nine, 10 mind, always available, always up. Okay? So, go back to what not to do. Go away from the appliance. It's over.
John Furrier
>> The appliance era is over.
Eran Kirzner
>> It's over. You want software defined. You want to go, everything is software defined anyway, why you are going back to the appliance? Okay?
John Furrier
>> It's like going back to fourth grade. You don't want to go back in time. This is the thing, people are used to the rack and stack mentality.
Eran Kirzner
>> Right, when you have traditionally-
John Furrier
>> Provisioning, standing up systems.
Eran Kirzner
>> Yeah. When you have traditional workload and it's smaller footprint, yeah, you can still use the appliance. If you go data center scale and what kind of neo kind of type of application, you need something else.
John Furrier
>> I mean, the common thread we're seeing in this AI factory series is that whether it's centralized or edge based, which is coming fast, being data centric is really a key thought process. Being data centric, understanding what resources are available and avoiding the provisioning problems. Would you say that summarizes that?
Eran Kirzner
>> Yeah. I think people now start to realize why it's called data center. Right? Yeah, it's all centered about the data, nobody is throwing away anything. So, data is just going to explode and explode and explode, and it will be harder and harder to manage all of that.
John Furrier
>> Talk about hybrid cloud, because hybrid cloud now is becoming essentially the architecture for data center cloud and edge. And the role of sovereignty, because sovereignty is a cloud term now around governments, but we're seeing, you mentioned multi-tenancy. We see multi-tenancy becoming a huge deal, and multimodal becoming a huge deal. Sovereign brings that governance kind of perspective. I mean, the country stuff, I can see that because it's got boundaries. But ultimately, if you have a hybrid computing environment, I mean, sovereign has to be a built-in feature because apps could be sovereign. They have the unique needs, they have unique governance.
Eran Kirzner
>> So you see that a lot of Europe, we see that, that a lot of centralized kind of cloud solving the problem, different regulation, different security requirement, all of that out there. We see our customer also using multi-cloud environment because they want to burst into the cloud and do some stuff there and then coming back and running it on their own centralized cloud. If you look at an enterprise, a lot of the financial enterprise, they don't actually shift their data into the public cloud just to run there. And also, you don't have enough capacity, GPU capacity, right? So you end up, if you are a large enterprise, running on neocloud, running on the public cloud, running on your own cloud, and you need a mechanism to use the same kind of concept. Another reason, as you mentioned, use software defined. If you're using everything kind of the same, you don't need to re-orchestrate it. You can use the same profile, on-prem, edge cloud, your public cloud or neocloud. All of that can work.
John Furrier
>> Continuity is key there.
Eran Kirzner
>> Yeah.
John Furrier
>> Eran, thanks for coming on. I appreciate sharing your insights. Put a plug in for the company. What are you guys working on? You're hiring?
Eran Kirzner
>> Yeah, we are always hiring, especially on the AI project we're adding more people. And we are moving forward with our customer, helping them to build the next generation stuff. It's super excited and .
John Furrier
>> Well, it's great to have you on the Cube here. A little quiet on Friday here inside the NYSE, the Cube Studios. Appreciate your time.
Eran Kirzner
>> Thank you.
John Furrier
>> AI factories continue to be the hot conversation as it brings in an enablement model and storage continues to play that role. You got to feed the beast, feed those GPUs. Also, computing is clear, it's not just GPUs, it's XPUs. And again, off the shelf standard hardware is becoming, again, the key focus here. In theCUBE we're doing our part to bring you the data, thanks for watching.
John Furrier