Hewlett Packard Enterprise Company (HPE) Earnings Call Transcript & Summary

Hello, and wish you all a warm welcome to another addition of HPE WORLD WATCH, a series of interactive webinars that we've been running through the year, and we bring to you topics, which are relevant to the industry, to our portfolio and all things related to the trends around us. We have a very exciting topic in the next half an hour, dealing with enterprise AI from edge to exascale. As you know, AI today is embedded in almost all applications that enterprises use. And a key imperative is to embed AI into those applications which are not yet enabled for machine learning and AI. As we do this, we also find that data is proliferating to the edge. And a lot of applications demand that decision support is done at the edge without having the comfort of latency that traffic to the data center demands. These complexities and beyond will be discussed today. And the 2 speakers that we have are, first, Arti Garg, Head of Advanced AI solutions and technology from HPE; and our distinguished speaker from 451 Research, Ian Hughes, Senior Analyst, who looks after the IoT practice area. A little bit later than the program, we will also be joined by Vinod Vignani, AI and IoT practice leader for HPE in APAC. Vinod will come in and answer some of the questions that you may have. So please, as always, do chime in. There is a dialog box. You can type in your questions and your comments. We will pick them up along the way and bring them in towards the latter part of the program. With that, please welcome Arti and Ian for their presentation. Arti and Ian, over to you.

Thank you so much for having me here today for this discussion. And that's a great question. How do we define AI at the edge. Truthfully, that's a little bit of an ill-defined things. A lot of people are talking about AI at the edge, but a lot of people might be using different definitions of what the edge means. I think succinctly, what is relevant when you start to think about AI at the edge is really, it's any kind of application where you're deploying infrastructure outside of the controlled environment of a data center. And the further you get away from the data center, the more new requirements you place on top of the infrastructure that you need to support, AI workloads. And so from my perspective, all of those become edge applications, some of which have more extreme requirements to say, deal with environmental factors than others. And so once you start thinking about that, that can drive a wide range of implementations and adoption by customers in -- depending on where they're actually deploying their AI at the edge. So I'm going to use 2 extreme examples. So one is what I would call an edge capable -- or AI-capable edge data center. And this might be something that still looks a little bit like a controlled environment, a IT closet, say, in a building, but you don't have maybe the ability to place as tight controls on temperature and environmental conditions. That's one kind of edge. And we're seeing a lot of growth in that area as people want to drive more AI closer to where actions are being taken or decisions are being made. And then I think there's something that maybe seems more extreme on -- with respect to deploying AI in environment that you really maybe didn't think that you would see them in the trunk of a car, for example. And that's where I would think you're getting into some pretty awesome but niche applications of AI and AI infrastructure. And probably the biggest trend we're seeing for that ladder, far edge application is something like computer vision. And that's where the models are getting sophisticated enough, and both our hardware and software are getting advanced enough that you can support some pretty extreme computer vision and video analytics capabilities and relatively small form factors that aren't particularly power hungry. So I think it's -- there's a lot of exciting things happening in AI at the edge, and I'd like to turn it over to Ian to talk a little bit more about some of the trends that he's seeing.

Thank you very much, Arti. So at 451, we run several surveys, including our Voice of the Enterprise survey, particularly in -- here in this case, we're talking about in the IoT space because there's a -- an intrinsic link, you would seem, between the instrumentation of IoT and the need to process that data. And typically, you -- we're seeing use cases where that's happening more and more at the edge. And when we look at the personas of the people engaging with AI, the people that we call AI believers that think it's going to be a radical change and is a significant change to the way enterprises work, tend to also be the ones that are engaging in edge because edge processing happens as close to where the data is generated as possible to get the most value from it. It doesn't mean all edge has to be AI, and it doesn't mean all cloud has to not be AI, but there's very much a continuum between those things. And also, we see from the digital infrastructure approach, many of those similar people that are AI believe is NetCentric. There tend to be a correlation with them being more about on-premise integration with their IoT solutions. And if we then look, in this case, drilling down a little bit deeper into manufacturing in this case. Manufacturing is a core place we're seeing a lot of AI. Almost every conversation I have had the past couple of weeks seems to have been about AI manufacturing. And the primary use cases that we talk about are things in IoT like production monitoring and whether people are engaging in it. And as you can see from this slide, it's 85% of the 75-or-so companies in the manufacturing sector that we talk to are engaging in instrumenting their plant or at least going to in the next 2 years. But that's just data. That doesn't necessarily help you do anything. So therefore, we see the increase in the adoption of applications that engage with that data. And the primary applications really are at the very heart related to AI in some way. For instance, quality assurance, Arti mentioned, computer vision at the edge, being able to see whether something is wrong whilst it's on the line, is a key edge use case. But we tend to see, as you can see from our numbers here, about 45% of the people that we talk to are doing the quality assurance at near edge. So it's not right -- necessarily right on the machine. There is a percentage of that, but it's an aggregation process, but it doesn't need to go all the way up to the cloud because you need to be talking about your production line, and you need to be able to make adjustments quickly. And it has a contextual weight to it, that the best execution venue is at that near edge. Whereas when you talk about something like predictive maintenance, as you can see, 49% of the people are discussing -- suggesting that it's at the core where the predictive maintenance has to happen and where the decisions are made by AI and ML applications. And that's primarily because you're aggregating potentially lots of different pieces of plan and lots of different streams of information. And it might even be things about supply chain or the weather or any kind of extraneous information can help determine what might need to happen to one machine or another as you try and drive your overall equipment effectiveness. So that's less of an edge application. But then when you come down to the emerging things such as fully autonomous robotics and augmented connected workers, where you're actually trying to encapsulate what's going on, on the shop floor and do a better job, then those tend to be more edge-focused applications. You need the information there and then to be able to do the thing and make the decisions and make the change in a suitable and contextually important way. And so that's what we've seen from these adoption trends.

Yes. I totally agree. And I think we're seeing some similar things at HPE. And the ways that I would categorize some of the challenges you might see, say, in a manufacturing setting, kind of fall into 3 categories of what we consider maybe key challenges to edge AI. And one is around infrastructure, the other is around data and connectivity, and the last and sometimes overlooked, but potentially the most important piece is actually around kind of capability and experience. So if I think about the first one, we talk about infrastructure, a lot of things that might drive that are just what you're capable of deploying in usually a pretty harsh environment. And one of the things that happens with edge AI is we always say it's usually deployed in a brownfield. You're walking into an existing manufacturing facility or potentially, if it's a mine, say, some kind of resource extraction study. And so you have to think about what am I now trying to deploy. Some of the most sophisticated IT equipment that might exist, some of the most advanced processors and some of the largest and most high-throughput storage capability into an environment that's designed around people potentially needing to wear hardhats and potentially even having a lot of metallic dust in the air, that's something you might see in a manufacturing facility. And so you get into a question of what kind of infrastructure do I need to deploy. It usually needs to have some sort of ruggedization capacity or some sort of ruggedization, which means it can either withstand different temperatures, different, say, levels of moisture, potentially contamination from, like I mentioned, metallic dust in the air. So I think that's one of the challenges and it sometimes might drive decision-making of, is there something I put on the factory floor versus something that goes into a kind of IT closet that exists in a manufacturing center versus something that I need to be able to send back to the data center to bring to bear potentially very, very high-capability and high-performance computing infrastructure in order to run a very sophisticated AI model, as you talked about that not only incorporates, say, the computer vision data that I'm collecting in my facility, but a lot of other information ranging from anything to supply chain data, to maybe something you even like gleaned from social media about what might be happening in an area where was -- where a part came from. And so I think those are the nature of the types of infrastructure challenges we see. And from HPE's perspective, we offer a range of products across things that can go -- that look almost like a gateway to exascale supercomputers. And I think one of the big challenges is figuring out what needs to go where and what can go where based on environmental factors. I mentioned the second one -- the second key challenge that I think of with AI at the edge is data and connectivity. So one is, when you collect all that data, where do you store it? Do you have storage capability closer to where you collected it, closer to sensors and closer to where ideally, you're going to deploy the decisions that you make based on AI? Or do you want to have it in a data center where you potentially have a lot more capacity for storage? One of the questions that you have to balance that against, though, is connectivity. And so how easy is it for me to transfer data to the data center? If I don't have a lot of connectivity or sometimes in a manufacturing setting, I hear from some of our professional services folks that it's just -- they're already wired the way they're wired. And so there may be some limitations in terms of where you can move data off of a camera into a data center. And so that's the secondary piece is while it may be a challenge to store data on site, connectivity may force me to need to store more data on site so that I can transfer it at the rate that I can reliably do so. And also connectivity may want me to keep more data and decision-making capability on site because I don't have time or I can't risk the possibility that I lose connectivity to a remote data center or remote storage somewhere else. And that -- those 2 -- the question of infrastructure and the question of data and connectivity kind of naturally segue into the third one, which is capability and know-how. I think one of the key challenges to AI at the edge is most of the problems you're trying to solve when it comes to infrastructure and data and setting up your IT solution, there are known solutions to them, but you need to have the experience to think about them upfront and really work through that what-if scenario because when you about running AI workloads in the data center, you often have a staff there, and you often have -- can guarantee connection that you can make adjustments if something goes wrong with your infrastructure or your solution. In an edge application, it's a lot more challenging. You may not have dedicated IT staff. You may lose connectivity. And so having that experience to know what questions to ask upfront, what failure modes or risk modes to anticipate before you deploy a solution, I think, is actually one of the biggest challenges to AI at the edge. And sometimes the one that we don't discuss as much because we think about it from a technology standpoint and not necessarily a technical expertise standpoint. So those are some of the challenges that I think we see and try to address at HPE. Ian, I would love to hear your views on that and what you're seeing in your research.

Yes, certainly, when you look at the changes that we've seen in the IoT industry and what that means for data and how -- where data resides and what's going on, particularly related to AI. If you look at this chart here, we started off 5, 6 years ago and previous to that even, with sensors and actuators, often in brownfield development, delivering data to the cloud. But as you mentioned, the challenge of getting vast amount of data to the cloud to do anything with it, particularly in challenging environments and in manufacturing environments that maybe don't want to deliver to the cloud, it very quickly changed into a 3-tier architecture, where you had sensors and actuators, some degree of edge processing on premises and the cloud. So typically, the edge processing here in this example would be something local to the machine, but then the cloud model would be something across multiple plants because, obviously, you can't do edge processing across multiple plants because they're not there. So with this, this model is the one that everyone is kind of getting into now and understanding, but the reality of it is that we're moving to this much more complicated, fully distribute computing model now that many pieces of machinery, many applications actually have AI and machine learning built into them and are performing tasks. And sometimes people even don't know that maybe that's AI and machine learning in them. But that's -- those still have to be the correct models. They still have to be kept up to date. And so being able to make sure through complex data ops that the right model and the right data is in the right place across the continuum from machines or from sensors to edge to cloud, the processing can happen almost anywhere in that continuum based on the context of what you need to do with it. And so as you can see there that you might have augmented reality applications that are dealing very locally with understanding the environment around a worker. That's a lot of computer vision. That's a lot of patching in data from somewhere else, so you can see what you're doing in physical world. You might have machinery that's got its own camera attached to it. So it's trying to help it do its quality assurance. But you may also have completely disconnected tooling that is not connected most of the time that somehow you have to manage what's on it, what that thing is doing. And we live in a world of constant patching, don't we? And the same thing, then you might have autonomous robots that are engaging with -- they're doing their own job, but they have to engage with edge processing to maybe -- and the infrastructure to talk to the other autonomous systems to work a what to do, but then all those are still talking to the cloud as well. They're not sending all their data to the cloud. They're just sending kind of the appropriate updates or the particular challenges they're experiencing. So if you go from just moving data from A to B to the cloud, which is where we started, down to that very, very distribute computing model, that's quite a challenge to get the right management, to get the right things in place. But actually, the best way to manage that anyway is through AI and ML itself. So just as networks are becoming more complex and even with 5G, there's a lot of AI and ML in managing how the signals move around, how the data flows. The same thing applies to these as to where the data needs to be and where the model needs to be and how to deploy it. And also the increasing trend to locate no-code systems to help the experts on the ground, particularly like in the factories, to be able to engage with all this tooling and all these things, but to actually help them do the job, not have to necessarily have a complex application built, but work at how to pull these things together to get the right results for what they need on the plant on the shop floor.

Thanks, Ian. That lays out a lot of the challenges that we're seeing in HPE very well as customers have moved up this complexity chain that you discussed. From sensor to cloud to these very sophisticated and complex edge deployments, we see that customers are sort of facing a lot of different challenges because one of the things that I like to say is that edge applications become nasty IT problems very quickly. And HPE, that's what we do. That's what we've been doing for decades now is helping customers navigate some of the most challenging IT problems that they encounter. And we do that in, I think, a few different ways. One is with our products and technologies that we offer, including both, as I already mentioned, ruggedized equipment, some of which goes all the way down to type of 2 gateways, to exascale supercomputers in the data center and everything in between. And when I say everything in between, I think of those AI capable edge data centers as well, which may even be comprised of computing systems that look a little bit more like what you might see in a data center because it turns out that you have just enough environmental control there to make a difference or to support more standard computing equipment. In addition, we have networking capabilities in our Aruba product line to help customers add maybe more on-site connectivity to facilitate transfer of data from edge sensors and cameras to more capable computing equipment that they may have on site. But I think most importantly, where I see our ability to help our customers at HPE migrate from maybe they're living in that sensor to data center world right now to the super complex AI at the edge world of the future, is our ability to work with our customers through our professional services team and also through our knowledgeable solutions architects who can help put together solutions that will really work, not just in principle, but actually in production. And when I think of that, I think of that as designing the right architecture to get your data from where it's generated to where you can make sense of it. But I think it's also helping people anticipate the challenges that may not be obvious upfront. So as we close out and I turn this back to our host, I'd like to leave people with a few key takeaways to summarize some of the things we've discussed today. So one is, I'll start with, edge is probably a little bigger than any one person thinks that it might be. And maybe the best way to think about it is edge is any IT deployment that is outside a traditional data center because as soon as you start to move outside of that controlled environment in that controlled facility, you start to encounter problems that might need a little bit more of a specialized solution. And that -- and because of that, I sort of move into my next favorite saying, which is that any kind of edge deployment turns into a nasty IT problem very, very quickly. And you should just know that coming into any edge AI scenario. And one of the reasons that one of those more specialized edge AI IT problems that people are likely to encounter that can have a strong influence on what their edge AI solution looks like are considerations around data and connectivity. That's what makes an edge deployment different than a data center deployment in a way that doesn't just have to do with the specific types of servers or technologies that you're deploying. And I'd like to reassure people, we have a lot of solutions to these. Many of them are based on things that have been done in the data center. But the most important thing is to think about all of these challenges upfront, anticipate them and create mitigation to help solve for them before you really get into hot water. And I think that's where I really feel excited to work with customers or to help move us into this much more complicated and much more dynamic AI at the edge future. And with that, I would like to thank our host and look forward to future discussions.

Thank you so much, Arti and Ian, absolute pleasure having you with us today. And Vinod, wish you a warm welcome to today's show. Thank you for taking the time. We have some very interesting questions coming through, Vinod. And I want to turn to you as a subject matter expert. And maybe the first one is around some real-life examples. As you have worked in this space, are there certain customer experiences that you can relate to where HPE and its portfolio has gone towards addressing challenges and solutions in this area of AI at the edge. Vinod, over to you.

Thanks, Kishore. Yes, a pretty good question, right? I'd like to cite 2 engagements, 2 recent engagements, which we have worked on in the last year or so, right? The first is on the manufacturing side, like what Ian mentioned, right? Manufacturing on the edge is the trend which we see in the market. And I completely agree. I mean same in APAC, we sort of see the same trend. We were contacted by one of our customers to enhance their product quality, reduce the defects, right? Pretty similar use case, what Ian was talking about, extending it to predictive maintenance as well, right? But the special part about this engagement was, it was not just about -- not about their big data strategy but it was an overall edge-to-cloud strategy, along with data in it, right? So we designed the overall big [indiscernible] acquiring data from all the sensors, but also had the edge component, right, which is where we deployed the defect detection component, right? And then the predictive maintenance went on the -- went back on the cloud, and the results were quite good, right? We were able to -- just purely on one line, where we're able to reduce the number of defects by over 25%. And the customer is pretty excited about it, and we are planning to expand it to multiple lines, right? We've got about close to about 50-odd lines, right? So this was one interesting one. Another one which is quite exciting for us is overall -- in the transportation domain, right, where the transportation authorities are trying to avoid crowds at their stations, right? With the current COVID scenario, right, they want to monitor crowd levels at the station, right? And that's where we started working with a couple of customers in the APAC region on gearing the WiFi data from the commuters and the video analytics data, video data, merge that together on the edge and then respond back at the station level, right? So this is -- these 2 are, I would say, the most interesting use cases and the most interesting engagements we've had, specifically around edge.

Excellent. Very insightful. Thank you. Now Vinod, obviously, top of mind for everybody is this whole notion of distributed data. Obviously, edge is part of the story. But in general, today, there's a proliferation of data across, I should just call it, hybrid IT. Not just the data center. It's on devices, it's at the edge, it's with employees. Literally, data is created everywhere across the continuum of hybrid IT, from public all the way to private, all the way to the edge, right? We have a very vast portfolio of products, which addresses different parts of the challenge that distributed data faces. It would be great if you can come in and talk a little bit about the road map. Where are we heading as far as our portfolio is concerned? What can customers and partners expect in the future from a [Audio Gap] Services in this space. Over to you.

Thanks. And a great question, right? Yes, we've always had an edge-to-cloud strategy from -- for a few years now, right? So that's -- so on the infra part of it, right, yes, we've been pretty strong. We've got the ruggedized edge hardware, right, which runs in factory environments, in all these edge environments, I would say, right, where the conditions are not like data center, right? And then we've got a portfolio of products, which goes across to HPC, supercomputers, right? So that's always been there. And now for the last 2 years or so, I would say the focus has increased on data, right? We've got Ezmeral Container Platform, right, which basically provides the capability of running all the AI workloads, which were running on cloud to the age as well, right? The portability is what we've got with our Container Platform. And then comes to data, right, what you talked about, the key is the data, right? Ezmeral data fabric, again, part of the same portfolio, provides a storage back on the cloud as well as the edge, right? The replication between the edge to the cloud is much more optimized with the data fabric, right? So that's -- that I would say is the new piece. And the investments are going to go in that direction. We are bringing in the AI modeling tools on top of the container platforms, the data optimization, the storage optimization is increasing as we speak. And Arti talked about the capability part of it, right? That's something which is I sort of lead personally, right? I mean I lead the AI IoT practice for HPE, right? So we're investing a lot on resources as well, who will be able to do these kind of deployments, which are edge to cloud, right, whether it's AI on the cloud, whether it's AI on the edge, all that will be one of the focus, right, purely in terms of capability, right? Another thing, which you talked about hybrid IT, right, and Arti also talked about, right, the workload running on the edge is much more difficult to manage. On the cloud, it's difficult to manage. Cloud is pretty much okay, right? We've been doing traditionally, right? And that's where we are focusing on AI ops, right, to manage this entire infrastructure from a central location, right? And it's different from the traditional device-specific monitoring, but it cuts across edge to the cloud via a log file-based monitoring mechanism, where we can run AI models on top of it. Rather than humans doing troubleshooting, AI will do the troubleshooting across this hybrid cloud platform.

Excellent. Wonderful. Thanks for that, Vinod. Unfortunately, we've gone out of time. There are more questions and there are more queries. We'll send it your way, give you a chance to answer them maybe by e-mail. But as we draw to a close, big thank you. Thanks a lot, Vinod, for being with us. And also a big thanks to Arti and Ian for being with us today. For our audience, a quick reminder, we value your feedback. So please do make sure that you fill out the feedback form before you go. As also, the content from today's webinar is available in the form of a download link. You may want to download it and consume it at your leisure. As also, there are certain useful links that we have populated in the Resources section, which you may want to bookmark and visit them at your convenience. With that, we will come to a close for today's program. Thank you so much, and wish you all a wonderful day ahead. Bye-bye.