theCUBE + NYSE Wired: CMO Leaders | Dinakar Munagala, Blaize

Clips
More from theCUBE + NYSE Wired: CMO Leaders

Dinakar Munagala

CEO

Blaize

play_circle_outline Blaze Revolutionizes AI Infrastructure with Edge-to-Cloud Connectivity for Efficient Data Processing

play_circle_outline Breaking Cost Barriers with Flexible Programmable Processors for AI Innovations like DeepSeek

play_circle_outline Intelligent Edge: Focus on Edge intelligence over traditional IoT for AI applications.

Info
Transcript

Dinakar Munagala, Blaize

Dinakar Munagala

CEO Blaize

Dinakar Munagala, CEO of Blaze, discusses the evolution of Edge Artificial Intelligence (AI) computing with theCUBE analysts John Furrier and Dave Vellante at the CMO Leaders event. Blaze focuses on empowering smart cities, autonomous vehicles, and public safety through innovative processor architecture and partnerships with Independent Software Vendors (ISVs). Munagala discusses the transition from AI being largely cloud-based to leveraging data from a significantly larger physical world. The analysts explore this evolution, querying Blaze's role in enabling... Read more

explore Keep Exploring

What is the secret sauce behind Blaze's processor for enabling AI at the Edge? add

What is the significance of having a flexible, programmable processor at the Edge for AI deployment, according to the discussion about DeepSeek and Blaze? add

What are some reasons why Edge AI is becoming increasingly popular compared to traditional IoT methods that send data back to the cloud? add

bolt Powered by CUBE AI

Dinakar Munagala, Blaize

search

>> [Music] Welcome back everyone to theCUBE. I'm John Furrier your host. Here at our Palo Alto Studios. It's the CMO Leaders events, part of theCUBE and the NYC Wired community, bringing all the top minds together in a network effect to bring all the data to you. Dinakar. Here's the CEO of Blaze. Thanks for coming on, theCUBE. Appreciate it.

>> Thank you for having me here.

>> So I want to get into quickly the company. I want you to explain what you guys do. You guys just went public and then also with CMO series here, it's a lot going on. What do you guys do?

>> Thank you. So Blaze is an Edge AI computing company. So the core of our animation is a novel processor architecture and software. And we partnered with the ISV community to enable complete solutions for the physical world. So it could be smart cities, defense, automotive, public safety, all of these, outside of the data center. How do we actually serve them with AI? That's where the company's focused.

>> So AI infrastructure, clearly. Outside the data, it's Edge as they call the Edge.

>> They call it the Edge. Nowadays, they call it the physical AI, and essentially that's where we are.

>> So in devices and cameras, stores, anything that's connected?

>> Exactly, yes.

>> So it's got connections. Okay. So it talks to a data center, it talks to the Cloud. That's distributed computing devices, basically.

>> Okay.

>> It runs software. Okay. So now that's going to power the applications.

>> Right.

>> Okay. So you're at the network layer, physical layer all the way up. How do you see the AI? Because you're enabling more AI.

>> Yes.

>> Inference at the Edge. Maybe work with some training, but you have more data coming in too. So you have two challenges. You have the challenge inference at the Edge, and by the way, it's a device, so it's getting more data. So it's getting data in which is new data. How do you train it? How does that work? There's some technology involved in managing all this. How do you guys view that and how does that propel the AI wave?

>> So if you take a step back, it's a massive, massive opportunity and inflection point. The way we like to describe it is AI 1.0 moving to 2.0. So 1.0 was all about AI being done in the Cloud, where they had access to data, compute and talent. But really the opportunity outside in the physical world, if you measure it by the amount of data, it's about a million times larger. We're talking about everything from education, healthcare, safe cities, defense, manufacturing. The amount of data is a million times more, and hence the opportunity is that much more. But there are challenges, right? The challenges are very different for the Edge. It's often efficiency, things like cost and things like a knowledge gap. So a solution for the Edge has to be addressing all of these in order to win. And that's what we are focused on.

>> What's the secret sauce?

>> The secret sauce comes down to-

>> Give it away....

>> It comes down to the actual processor itself. Blaze's processor is innovative, in that it focuses on data flow graphs and how we process them efficiently compared to GPU style computing. That's where we get the efficiency. And then we layer in the, remember I also mentioned the knowledge gap. So we've actually built considerable amount of software on top to help our customers adopt Blaze into solutions. It's a set of rich libraries as well as a code free platform called Studio. So that a factory worker can actually take our product and apply it. So what we build is very similar to a Macintosh of AI, right? Efficient processing power and easy to use software, so that a doctor can take it into healthcare, right? Without being an actual data scientist. So that's!

>> And run applications on it.

>> And run applications.

>> So are they writing apps for it? Are they going to use existing ecosystem apps on it? How do you guys see that developing? So you guys have a platform, especially like a Mac. What A Mac is exactly? We got to run something on this.

>> Right. That sounds right.

>> What you guys running on... What do you see running on it?

>> So we have partner applications on the platform and these could be for example, around video security for a smart city and a smart city planner trying to understand traffic patterns. So we have partner software that runs on the platform to make such applications come to life.

>> Scope the opportunity and the complexity that you guys help manage. Because smart cities, I mean people see cameras on light poles, traffic, whatever. But when you start getting into now more computer vision, you've got much more telemetry coming in, multimodal across multiple different data sources, some more real time than others. Scope the problem space in terms of what you guys are solving. How big is it? Because a smart city things are connected. You've got databases that are centralized. You might have Edge database, you mentioned knowledge, graphs. You get all kinds of inference capabilities. How big and complex is this?

>> So if you measure it by the amount of data, it's massive. Think of us as, there's about a billion cameras installed and shipping every year. And the push now for these cities, et cetera, is to make value out of these cameras. So the three kinds of values, right? One is, "Hey, how do I actually generate revenue?" Right? It could be automated traffic violations based on number plate recognition and traffic violations. It could be how do you potentially dim the lighting not in turn. So you want the safety preserved, but if nobody's there, dim the lights in a certain area. Save electricity, right? Cost efficiency. The third category is around crime prevention and first responder services. So these are all possible by video analytics, camera analytics, applying AI. And the idea is to do them near the camera, real time and then process, and of course, send an alert back to a common command and control center. So those are the class of use cases. And it's massive because of the sheer number of cameras that are out there and shipping. And most of the cities and municipalities are paying attention to this now.

>> One of the big things that I love about this market we're in right now is, you're starting to see a flywheel of innovation that's showing performance and cost reduction gain. Similar to the old processor days of, remember the PC came out, Dave Vellante and I were arguing on our last pod, not really arguing, debating. But riffing like, is the DeepSeek style innovation more like a Google innovation? Like search, PageRank or, I took the side of the PC that was my angle, but you're seeing innovation around the constraints. Edge has always been about power and connectivity. Okay? So you got to connect to something. How do you guys manage the constraints? Because there's a lot of innovation going around by the constraints and the footprint. Because you never know how you're going to be deployed in Edge because physical.

>> Right.

>> Can you share your thoughts and vision on managing through those constraints? Because a lot more engineering is going to get done. More software will come out. It'll be faster, smaller, cheaper, but better.

>> Right. So DeepSeek is, first of all, I think the same way as you do that this is a huge thing, right? It's about how IBM mainframe went to PC kind of revolution. Basically, they said, "Look, hey, truly democratize AI, bring down the cost barriers and you can deploy it to where the data is near edge as smaller computers.

>> Off the shelf.

>> Off the shelf and it can just work." Right? That's the beauty. And in that, right? In the DeepSeek kind of... And there'll be many of these by the way, in the coming weeks, months here, right?

>> Like the PC then someone will leapfrog the next one.

>> Absolutely. Exactly. So having fundamentally at the Edge, a processor that is flexible to adapt to the likes of DeepSeek is very critical. Because a lot of the AI efforts at the processor layer, they've truly not built programmable processors. They've taken a shotgun approach of fixed function. What is really needed is a programmable approach. And Blaze from day one, consciously we saw this coming, we knew that innovation will happen and this will happen. So we built programmable processors. We built making sure that we can run things like mixture of experts, right? Which is the philosophy of DeepSeek. It's all about, as an example, when you have a Math problem, "Hey, what is two plus three?" You don't wake up the whole world, right? You just wake up the educator model and say, "Hey, what is two plus three?" And so being nimble and trying to get that, and so our processors solved that. So that's the level of innovation that we see going forward. And how the Edge can be connect... Connectivity was the other piece, right? All of the wonderful work that's happening in the Cloud should continue to happen. But how do you actually get smaller nimble models? They're calling it small language models. That's the big thing that we see coming to the Edge. And the Edge being able to run that efficiently and seamlessly connect to the Cloud. I think that's the thing that we see.

>> And the other thing too, again, we had a great conversation on theCUBE yesterday and we've been having it for over a year. There's general consensus now, it's not about the Cloud versus On-Prem, it's Cloud, On-Prem, Edge as a distributed computing architecture.

>> Absolutely.

>> So okay, that's known science. It's the networking. But now the software is getting smaller and smarter. The chips and the hardware is getting faster and smarter and smaller, less power. So it could do things. So that's clear. Okay, now you've got to run software on these things. So now we believe, and I'd love to get your thoughts on this, our research team, and Dave and I are riffing that we think that this DeepSeek moment, I say DeepSeek moment, style moment, because I think it's going to be replicated by other people, is the chat GPT for the enterprise. Meaning chat GPT woke up the whole world. "Oh my God, user experience like iPhone." Whatever. This is more infrastructure. Using the innovation, what they did they were just clever. They didn't really, there's no breakthrough, other than their engineering through the constraints. That's going to happen more and more. So now if this happens, all these POCs that are stuck in purgatory, in the enterprise and some of these Edge industrial and mission-critical applications. There's going to be a tsunami of apps.

>> Absolutely.

>> Do you agree?

>> Hundred percent agree, right? So because the cost barriers are broken down, the knowledge barrier is the next thing that has to be broken down, right? Because a lot of people who are in the physical world, they're not necessarily AI experts. They're good at what they do. They run factories, they help get domain experts.

>> Another mixture of another domain expert, the more domain expertise.

>> Right.

>> Kind of the wired network.

>> Right, exactly. So how do you actually get the technology to serve them, right? Is a key component. So investing in a layer of software, just like the iPhone, right? The iPhone movement happened because they had an app for everything and the interface was so simple. So somebody investing towards that, I think is the right!

>> Well, the app... The iPhone had enough apps to make it so usable.

>> Exactly.

>> But then the app store made it unstoppable.

>> Unstoppable.

>> Even today.

>> Right. So we see the world the same way. Having a marketplace, having ability to take an application that you've built for a certain industrial, let's say factory worker, floor safety, et cetera. You should be able to deploy a Blaze box, click on it, and it just works. And we are actually working towards such things. We've just announced a smart city in a box. So the idea is the box comes shipped. You just plug it in and you have a menu of applications that you can just click and it works.

>> So we're at the CMO leader summit. We haven't gotten to any CMO content yet. But I will bring this up to connect over. We were just at the NYSC for a media week during NRF which is retail. I call it retail. It's the retail show. Now, we don't cover retail, we cover the Edge. Retail is the Edge. So there's cameras in there, you got walkout purchases by Amazon, so you got computer vision, you got analytics. Marketers have to at some point be in position to be successful. And certainly retail where consumers are, whether they're online or not, they're out moving around the physical world. You're starting to see the connection between your physical footprints literally, and digital footprints. So now that's got to come together. So where are we in your mind of reality of full convergence, singularity of that coming together? Because that's what marketers will get to soon with what you're doing.

>> So absolutely, right? Retail, there's a lot of ad-based buying. And for example, right now you're in a store and you actually have to figure out buying A versus B, and imagine somebody pushes a coupon to your phone, saying that, "Hey buy Kellogg's and stuff this." Right? Those things are going to happen. But also that camera-based analytics like, "Oh, there's a consumer who's stuck on something." Alerting a shopper to go help them. All of those things are going happen thanks to AI.

>> Yeah. And I think we've been waiting for the intelligent edge. I think it's been on our coverage on SiliconANGLE for 15 years. The industrial Edge, certainly critical infrastructure, you see factories, that's obviously happened. Now you're getting into the consumer lifestyle aspect of it. Where are we in your mind on the progress bar of mainstream adoptions for connecting in closed-loop marketing? If you had to take a guess.

>> Give me an example of that. When you!

>> I'm a consumer, I might have a wearable, I've got my phone, now I got my watch, I might have a Fitbit or an iWatch, or I'm connecting, and I'm in a store, I'm walking around. They're tracking everything now.

>> They're tracking everything. Yeah. And healthcare is, I think the big beneficiary, right? In that whole space. And there's massive, massive amounts of data, and I'm assuming that's going to be put to good use, right? And giving you an alert before something catastrophic happens, right?

>> Yeah.

>> So those are the positive things. But absolutely, I think there are companies, I know a friends company is working on data and the healthcare, based on the wearables and so on.

>> Are you seeing a correlation? And obviously you got sell edge devices, more billions out there, better for you. Is there certain verticals like healthcare, like retail, where there's more devices? Which ones would you stack rank in terms of there are more IoT, I mean, what do you call these? It's not IoT, right? It's everything. So do you guys even use that term IoT or is it like everything's an IoT device to you?

>> We don't necessarily use IoT because this is all about really intelligence at the edge where you're able to apply AI and take actions. IoT was, traditional IoT was move everything back to the Cloud still, right? Get the data. So I think Edge AI is slightly more!

>> So a little bit more, a little nuanced to IoT definitions tied to the Cloud. Here you're just like... Like a smart city, if you do a smart city with Blaze, it's thousands and thousands of devices.

>> Also smart cities and defense for example, right? For a reason, they don't want to send the data back to the Cloud. They want to do everything On Prem, air gapped, use cases and so on. So those are all perfect use cases for the Edge. But to answer your question, right? Massive, massive amount of sensors, massive amounts of devices, all of the data. The other big thing that we see is inter-usability, right? So a use case that you've learned, let's say on a factory floor, recognizing cracks on a magnetic tile. That use case can be applied to agriculture, where you're trying to detect weeds on a... So there's a reuse. So potentially there's this whole reuse across use cases as one gets deeper into the Edge. So that's something that we see.

>> Do you have visibility into a lot of the techniques around training inference and reinforced learning? Do you guys see a lot of that going on?

>> Yeah, yeah, of course.

>> How do you see the new data coming in? Because this comes up a lot in my conversations. Like, okay, we're going to train it and then deploy inference at the Edge, but the training's centralized, but then you've got new data coming in that's not trained, or maybe it is. How do you guys think about that? What's the processor do? Can you program that on the spot? Can I just co-locate data there? Is it highly available? Is it high availability? I mean, there's all kinds of nuances.

>> Right. So we are part of a medical engagement. Let me give it to you maybe the context with an example where the hospitalist, they wanted to apply AI to x-rays, right? And as they learn, as new patients are seen, so they already, AI applies, tells you, "Hey, this is an anomaly, look at this," et cetera. But then as new patients come in week over week, month over month, they want the data to be retrained. So they want what is known as an all-in-one solution, right? Everything training plus inference with a common software layer. And they don't need the training to be real-time, but over a period of a few weeks or something have a new model. So that's an example I think of practical value. And data can't be sent back!

>> So the cadence of real-time changes based upon what the application is?

>> Correct. Exactly.

>> So, x-rays, you don't need to be up-to-the-minute?

>> Exactly.

>> It could be a day a week, just whatever the trickles in.

>> Correct.

>> So you don't have to worry about data egress and all the costs around moving data.

>> Exactly.

>> You can write software for that.

>> You can write software for that. Yeah.

>> All right, so what's the business like for you guys right now? Give us a taste of performance, momentum, areas that you're succeeding in. What are some of the hot areas? Can you share without breaking any public company rules?

>> Sure. So we're seeing massive interest from the likes of defense, and government spending, perimeter security, right? Those use cases from smart cities, all the use cases that I described, things like airports, right? Their use cases like "Hey, how do you track baggage handling?" Because these are fines that the airlines otherwise pay. So getting AI to solve those problems is an operational expense reduction and many more. We get hit by a lot, but as a young company we're trying to stay focused on a few and then replicate them across. So that's where we see.

>> You can die by eating too much business.

>> Yes, that is true.

>> And it's an old expression. How big is the company?

>> We're about 225 people right now.

>> That's awesome. Congratulations. One of the things I want to ask you before we wrap up is, AI infrastructure is the hottest area. Obviously you know how I'm pretty much on that side of the camp. And I'm really excited by the developer action coming. I think I like this DeepSeek style, and I intentionally use that word because I think it's just style. There's going to be more of that innovation. That's going to open up a renaissance and a Cambrian explosion of development. It's already waiting, it's going to burst out in the scene. But the AI is super critical. Things seem to be stopped outside of these areas that are in huge demand, airport, military tech and whatnot. So once that AI infrastructure comes together, there's three kind of areas I want to get your thoughts on. There's the one, approach which is I want to drive cost reduction. Cost reduction, airport, whatever, they don't care about revenue. Maybe they do, but maybe it's just a different area. Cost reduction. And then there's the other one, it's like I have money, I don't really, I'd love to cut costs, but I want to optimize for productivity and revenue. And then there's the third use case I'm seeing, which is very tight production window. Not a lot of stuff comes through production, high bar resilience. But I got tons of sandbox app and stuff. I got Sprawl everywhere. So I got to manage all this stuff. Prototyping could be like digital twins. Those are the three archetype motions. Do you see the same thing? And how does each one of those, well first of all, do you agree and if so, how do they go to the next level?

>> So we are part of all three such!

>> Do you agree with all three?

>> I agree with all three, yes. I agree with all three. And I would say that we are even part, as part of our customer engagements, we are part of all these three categories that you mentioned. And they're all necessary, right? That's how you validate AI for a business, especially at the Edge. Their main goal business is something else. They're manufacturing or something, right? Why do you need AI? Right? Is the first question. So these things validate the business need and then they engage. So absolutely. I do agree with all three.

>> All right, so final question. We're here at the CMO leaders. What advice would you give a CMO that they're leaning towards being more strategic, drive more sales, more customer retention, there's workflows, they have data. As someone who's working in the AI infrastructure, and enabling them, what would you give for advice for CMOS out there that are watching, how to approach their job for the next five, 10 years?

>> So the advice I give to myself and to our team is look at the value that AI is creating, right? Focus on the value, that will ultimately drive your sales or drive your customer adoption, right? So we focus on, "Hey, are we truly creating value for the end customer?" I think that's how I look at it.

>> Dinakar great advice. Compete on value, and you get customers. It just be raining business.

>> Right.

>> Thank you so much. Great conversation. Blaze, really at the front edge of the innovation wave. Smart cities, you start to see computers everywhere and all devices, but the intelligence is going to be everywhere. It's distributed computing, of course, theCUBE is here, bringing all the action. I'm John Furrier with theCUBE. Thanks for watching. [Music]