 Welcome back everyone to theCUBE's live coverage in our Barcelona studios of theCUBE. I'm John Furrier, host of theCUBE with Dave Vellante. We're extracting the signal from the noise. Dave, 14 years of theCUBE, Barcelona, big set here. Mobile World Congress is now called MWC, a lot of action, and the biggest topic that's being discussed obviously is AI, but connectivity, and one of the big topics around connectivity is what's going on at the edge, AI at the edge, cloud to edge, our next guest, Jeff Sharp, Senior Director, Edge, AI, super micro. Welcome to theCUBE. Welcome to theCUBE. It's great getting out of the aisles where all these thousands of people are. So, yeah. It's great to have you on. So, first of all, a lot of people know super micro because they make great boxes, servers, but you're in the edge group, you're in the product side, you're dealing with as the edge comes out, because this, you're in telecom show, it's about the devices at the edge of the network where there's factories, intelligent edge, or retail. There's a lot of action going on, especially around architecture. I got to put devices there, how's the AI going to work? So, every company is working on, what do I do at the edge? That's what you're doing. So, tell us, what's going on at the edge, and what's super micro's angle on what you guys are doing there? Well, the key is, it's not just about the edge, it's edge to cloud, right? And it's everything in between. But what we're seeing is, is from a customer's perspective, edge is where a lot of the value's coming from, right? That's where the customer's experience is coming in. That's where they're engaging things like avatars and wayfinding, and from airports to retail stores to security. The cloud, it has its purpose, but the edge has its purpose as well. And from a super micro perspective, the key is, is how do we put more workloads on a device? How do we use AI as a generative engine for not just LLM generative and predictive AI, but how can we use that hardware and our core partners to make it happen at the edge? You know, one of the things that we've been covering, obviously we can follow on the AI side of it, the computing power and training and inference, compute needs, GPU, et cetera, is coming down to the devices. You look at the Lama models, the performance is getting better on whether it's a PC or a device. Now certainly on the higher end, you need multiple GPUs, thousands of servers, and a cluster, there's GPU clouds for that. What does the AI look like at the edge for you guys? What are you seeing from customers and developers around what they're thinking? What are they deploying? How are they experimenting with AI at the edge? What are some things, could you share some use cases of what's experimenting? Because experiments are going to turn into production. Right, so one of the key things is, is number one, cost, right? So if you think of the edge, what you want to make sure is, is that you're not deploying something that the customer can't afford, because it's then exponentially multiplied out into the industry. The other key is, is a lot of these applications, a lot of people are seeing AI, but they truly don't understand the revolution of AI, where it's heading and what's the functions of it. They hear chat, GPD and all that. But for the edge, the key part is, is delivering something from Supermicro that's cost effective, but also it's powerful and enough to do not what they want today, but also what they want tomorrow. They don't have to replace hardware constantly. So we try to build these bigger workloads and work models. So for instance, think of a grocery store, walking into a grocery store, right? We all experience a self-checkout. You know, you go in and you look at an Apple, is this a Fuji, Macintosh, Gala? What kind of Apple is this? But grocery stores are starting to move to computer vision, right? They say, okay, they've already pre-trained the model, they know what a Gala Apple looks like versus a Macintosh, and you put that under a camera and boom, there it is. So the quality of experience is there, but the quality of experience could be worse when it says it's a Fuji Apple, right? And it's, no, hey, I need some help over here. So we're trying to do better in regards to that human machine interface and being to provide those analytics up into the cloud to continue to train, to continue to learn. It's like our kids, right? So our kids, you know, have you ever had a kid that drives a car for the first time, you know, or sweat and we're freaking out? Same thing with AI, you got to train, you got to do all these different models. And it takes time. It takes time, exactly. It needs some supervision. It needs supervision, right? And then they'd be unsupervised. Exactly. Red means stop, green means go. Machine learning, Dave. Supervised machine learning and unsupervised machine learning. Exactly. So, okay, so you guys consume a lot of, obviously, processors. We were just talking earlier about the XPUs, the NPUs, the CPUs, the... NPUs, there's a PU for everything now. That's right, it's called XPUs, accelerators, all kinds of, look at it. So, what does the edge look like? It's not just a general purpose X86 that you throw over the wall and say, hey, name it, you know, edge class server or whatever. It's different, it's specialized. Can you explain, kind of, take us inside and paint a picture of what the capability looks like? Absolutely. So, we're taking the approach. So, Supermicro has a building block model, which works really well for us. We have common elements that we build chassis around based on the environment that it's going to go into. So, what we do is, is we look at, we have so many SKUs, oh my God, we have a billion SKUs out there. And what we do in the edge is we take a subset of that SKUs and look at it from a T-shirt size. I'm a small, you could tell, right? So, we go from small all the way to extra large. And each of those are not just the physical size, but it's the capabilities of that. It's like you were saying, the XPU. How many GPUs can I put in the system? How many MPUs can I put in? What's the smart nick? What's the broadband connection that is required for this edge system? So, we work very closely with our customers to try to figure out that. And also, it's not going to go into, a lot of times in a nice sterile data center room, it's going to go in a convenience store next to the toilet paper, right? It's going to be next to the bathroom where the mops are. So, we build these systems that are hardened that has the ability to manage those. Restaurants, great example. Think of the kitchen, steam, particulates in the air. They need a fanless product versus a high-end big rack mount server. So, we work very closely with our customers with that T-shirt size mentality. And then think of, okay, here's your workloads, but what are your workloads in the future? And we build that system around it so they don't rip and replace. So, talk about some of the use cases. A couple years ago, I wrote a piece. This is a before-chat GPT. And we said, we had this slide in there. As AI matures, inference is going to dominate at the edge. So, we try to come up with all these examples. We had autonomous driving in there, traffic optimization, AI power grid, new payment systems, retail, onshore manufacturing, machine diagnosed, smart factories, all, we were just going to- All the buzzwords, right? We were just kind of spit-falling, right? Bingo. And then, but the funny thing is, we had like the crossover point is 2025. Actually, it's not bad. And then inference kind of takes over. Give research right again, Dave. Well, this was really one of those conceptual diagrams. I tend to get the research right all the time. Thank you. Dave's humble. Make it real for us. What are people actually doing? Where's the money? Where's the money? So, the different market, we take a vertical approach to everything, right? So, we have content experts, industrial experts in each of these industries. And some examples would be, we'll take enterprise edge as a great example. So, businesses, they hear the chat GP, I can ask a command and a beautiful PowerPoint pops out. I look at things such as security around that. How do I make sure that that inferencing model is secure because you're pulling from pretrained models to the edge? How do I make sure there's zero trust to those attachments? Retail stores, C stores and quick serve stores, making sure that I'm running multiple use cases, whether it's a digital concierge, a kiosk. How do I order? But instead of touching it, maybe I'm talking to an avatar or futures holograms, right? You're seeing these spinning LED wills with holograms. I'm now speaking to a fake person, but it looks real. We're also seeing things like AR VR introduction and the stores and home use. And with all that, that needs inferencing. That needs high speed edge, low latency, 10 milliseconds or less technologies. And all these different use cases, the end customer wants a common workflow. How do I pull in computer vision? How do I pull in digital concierge avatars? How do I pull in interactive kiosk that's doing conversational AI? We want to do that in containers and how do I manage those containers remotely? That's the challenge that our customers see is they're looking at super micro and our partners. It's how do I deploy this, right? I don't have an IT guy in every one of my stores. How do I manage this remotely? And how do I do it with a single pane of glass? So we're working with core software partners that can enable that. And with all of our silicon partners, NVIDIA and Intel and AMD and Ampure ARM, we're working with all of them and trying to enhance that experience of not just the consumer or the user to that machine, but also the support to that machine as well. And talking to carriers here at Mobile World Congress, the carriers want to make money on this, right? They want to have the ability of, they see that gleaming AI object out there, but how do I make money? And how do I manage that to provide the best quality of service? The retail angle is interesting. You mentioned deploying and not having IT staff. That's the norm. So AI is going to help there. The other area that AI I'd like to get your thoughts on is how do you make that retail outlook? It could be big store, a small store. Again, no one's no IT, you're not racking the stacking. The device is going to be smaller. You've got to be more agile and deal with a fixed physical footprint. Right, right. So one of the things that people are talking about, and we've been reporting it, is that personalization is going to be a big part of AI. So if I got a store, I want to personalize that for the right workloads and workflows and data that I need for my system. If I'm on an airplane, I'm going to need to have the connectivity come in to my first-class module. Or, which we were talking to Boeing last night about, they love that stuff. So you're going to start to see this edge become very intelligent, but also be its own data center. Yep, be its own data center. So it's going to get smaller, faster, cheaper. But the performance is going to increase. We heard Charlie Quass from Broadcom talking about as the silicon gets better, lower power, more scale, more action. Exactly. So what do you see for this personalization? What are some of the other early signs of it? Is the toe in the water? People are just trying to figure out how to make it work right now. Are we too far in the future? I think we're still, if you think of the curve, right? The hockey stick curve of AI. We're still at the bottom of intelligence and where we want to go. It hasn't kicked up yet. It's kicking up. And one of the reasons of it kicking up and to accelerate that is for humans to be more, not scared to death of this thing, right? Personalization. All of a sudden, people are thinking, oh my God, they know that's Jeff Sharp and all my background and all the parties I went to back in college and all that stuff. They didn't have Instagram when I was in college. Or YouTube, so I have no social footprint. I have no digital footprint. Thank God. But the core is, it's all about that experience and companies and the retail are looking to differentiate themselves. Think of all the convenience stores, sea stores that are out there, fast food restaurants. They want a technology that a consumer feels confident with. They're used to and they want to be able to differentiate from sea store A to sea store B. Things such as, I pull into, to get gas. Buckeys, great example. Buckeys, 300 pumps. Have you ever heard of Buckeys? It's a monster sea store based in Texas that's now moving to the east coast of the US. And I pull into the gas stop. It does license plate recognition. It knows that Jeff Sharp is a Buckeys preferred customer. Before I even get up to the pump, the pump has a display. Hi, Jeff, you have a big monster truck, diesel. We'll go ahead and pump. You got a message in your inbox. You want to play it now? And do you want to order something? You know, while you're pumping gas, you want to order your favorite hot dog with cheese, chili and all that. And we'll go ahead and prepare that order. Yes or no? Yeah, sure, I'll do that while I'm pumping. I go in, my hot dog's ready. And oh, by the way, we're offering a great discount on your favorite beer or soda. And we'll go ahead and do that. And it's all personalized for Jeff Sharp. Airport, same way. You get off your plane in a foreign country. How do we deal with the next gate I need to get to? And instead of looking for the screens, maybe it is a digital avatar. Maybe it is something on my phone that I'm interacting with at the edge. So I love that Buckeys example. So just trying to think about the sort of anatomy of where super micro plays. Buckeys has a data center somewhere. I'm sure you got some equipment in there. Maybe they're using the cloud. But it's all at the edge, by the way, so. Yeah, okay. But somewhere that data's going back, right? I presume. Exactly. Okay, so that's right. But in real time, I got Buckeys 300 pumps. And so you play there, you're all over that. Now, is there a spectrum to that edge? You know, you got kind of a big machine, maybe you're coordinating things and you got stuff on the pump. And where do you play across that spectrum? And how do you think about the sort of near edge and far edge? Even within Buckeys, there's like a near edge and far edge. Right, right. So that's one of the key benefits of super micro. And I'm not doing a big super micro pitch here, but really what differentiates us is that we do have the equipment that goes in the gas pumps. We do have the equipment that goes into the back room systems that are the size and shape and environmental conditions that it can work. And by the way, we're also in their cloud, right? We're managing their entire data center with all of our hardware. So true edge to cloud from the NNU's device, compute device to maybe an IoT gateway to a storage. Even we do routers, we do switches all the way up to the data center. So we can help support that whole flow. And because super micro, we get a lot of the silicon before anybody else. We actually are really proud of our first to market a lot of things, then we can work with Buckeys and say, oh, we got this new technology. Let's do a proof of concept with your, maybe the lowest risk store you have or a brand new store and let's test it out together. So we try to partner as much as we can with our customers. We don't want to just be a supplier. We really want to partner with them, understand their wants and needs and their desires and work really closely with them. And your skews, a billion skews. Oh my God. So, but those skews like you might have a vision system to read the license plate and that might have an arm processor in it. I'm just going to making this up, but then you've got a, you know, it maybe is a bigger server that's doing all the transactions. And I mean, it's a wide spectrum of capabilities. Like you said, you've got every piece of silicon under the sun that you've got visibility on and it's all getting spread across the edge. It is. How do you size this market? You just put a big dollar T on it? That would be the simple way. So what we've done is we've invested in a tremendous amount of manpower, top talent, architects that it's not just, realize also it's not just about the great piece of hardware that's sitting there. Think about, let's use a convenience store as another example. They want to put 50 cameras all around their store. They want to put cameras outside. Every gas pump's going to have a camera. Well, Supermicro, how many systems do I need to do that? What's the right GPU, CPU, NPU combination? Do I go high-end? Do I go low-end? And it's not just the hardware capabilities but the logical capabilities. How do we invest in future-proofing that system but supporting it in today's mode? So we have a whole organization that supplies those services to say, okay, 50 cameras, 1080p, 4K P, 30 frames rate per second, blah, blah, blah, blah, blah. We now have the ability to funnel that into and say, you need this system, this is the perfect combination, and this is the cost. Oh, you want to go down? Well, let's think about the traffic flow of this. Maybe we need to reduce the smart neck. So we work very, very closely with our customers on not just the cost profile, but the needs profile as well. You mentioned GPU, everybody's crazy about it. In video, H100 can't wait to get H200, I love it. But there's a lot of other GPUs. I got a GPU in here, I think I got a GPU in here. All kinds of alternatives that are out there. Again, full spectrum. There are. And that's why I'm interested in the size. I mean, I think it's really as somebody who used to and sometimes still sizes markets, it's hard to get your hands around this one. I think several years ago, we took a stab at it, kind of top down, bottom up, and it's almost incalculable how large it is. I mean, it's literally trillions. And it's growing like crazy. It's not just your big, you know, NVIDIA brands, Intel, Flex cards. Now you have others out there too, and it's growing. And each of those are carving out their niche. You have ones that are very specific in computer vision. And they're trying to beat the big boys, and they're trying to be very agile. The key to all those GPUs and MPUs is really the SDKs, the way that I train the models, the way that I infure those models. There needs to be a very common subset. So we coach these startups in these areas that say, hey, you want to do this? Just realize customers don't want to go with a very purpose-built SDK software development kit or an inferencing model. They want to use a standard aspect of that. Jeff, thanks so much for coming on theCUBE. Last minute we have left. Put a plug in for your group. Super computing, what are you guys working on? What are you looking to do? What's your goals? What are some of these you're excited about? You put the plug in. Well, thanks so much. So our organization is heavily focused on the telecom industry, not just from a network element, but also the enterprise edge and their edge. How do we make telcos money? We are very, very focused on that edge to cloud mentality with our team more on the edge, but how do we work with the cloud? Working with our silicon providers, with different NIC cards, GPUs, different form factors. We're trying to differentiate ourselves not with just the broad scope of products we have, but the additional value that we provide to the market. Like I said, how many cameras, how many sensors, how many, what is the infrastructure that attached? I need help doing that. Our customers don't know how to do that. So our organization does marketing, product delivery, product NPI. We do business development, market development, all in an entity for the edge. All right, Jeff, thanks so much for coming on theCUBE. Hey, thank you so much, man. All right, we're going to be, time for lunch here at theCUBE. Dave, Dave Three of Ford lays a live coverage. My name's John Furrier with Dave Vellante. Extracting the signal from the noise, we'll be right back with more live coverage after this short break.