Great. Well, thank you, everybody. Welcome to the Hewlett Packard Enterprise keynote fireside chat at the Goldman Sachs Communacopia and Technology Conference. I have the privilege of introducing Antonio Neri, who's the Chief Executive Officer at Hewlett Packard Enterprise. Prior to becoming CEO in 2018, Antonio spent 23 years in various leadership roles in the combined Hewlett-Packard Company, including as president, executive vice president, and general manager of HP's Enterprise Group. Before we start, I want to read a disclaimer on behalf of HPE. Antonio's remarks may contain forward-looking statements, so please refer to the company's SEC filings, including its most recent Form 10-Q, for a discussion of risk factors that relate to its business.
My name is Mike Ng, and I cover hardware and communications technology here at Goldman. We have about 35 minutes for today's presentation. So first, thank you so much for being here.
Yeah. Thank you.
Antonio, it's really a privilege to have you on stage. So you've had a front seat to some of the biggest transformations, both of the company, but also the technology industry more broadly. I wanna ask about the technology shift that we're undergoing today, which is really about accelerated compute. First, you know, how would you characterize this technological shift relative to prior ones, such as the initial build-out of public cloud? You know, what can we learn from the past, and in what ways is AI different?
Well, thank you, Michael, and good morning, everyone. Thanks for having me today here. You know, my generation is my where I come from. I think I have lived all the major inflection points you know, in the history of IT and computing itself. You know, obviously, from the original mainframe to PC client servers, to mobility, to cloud, and today, AI. And I will say that this feels a little bit different. There are a lot of similarities, but they are different. In many ways, in the past, it was about connecting the world. It was about driving digital transformation or digitized enterprise. This is about transforming the really the way we work and obviously our own personal lives.
I think it's, in my view, it's gonna be the most disruptive technology, at least in my lifetime, and it has the potential to really transform everything, including solving some of the biggest societal challenges we live, and HPE has been at the forefront of many innovations, including supercomputing, which obviously has used some of these unique capabilities for a long period of time, but now, with generative AI, it's being democratized and be accessible to all, including consumers, so it's an exciting time, but also we are early in that journey. Clearly, today, the companies that they are invested in building this large language model models are driving a lot of the demand we see for accelerated computing, but we already see a significant uptick in enterprise and sovereign clouds.
And I think on the sovereign space, obviously, it's not just a matter of sovereignty. It's about bringing the cultural aspect to this new technology, which needs to be in the context of what those countries represent as well. And so Hewlett Packard Enterprise is well positioned to participate in all of them with our unique innovation and the ongoing investments we continue to make in the business. So again, I lived many of those transitions, and this is as exciting or more exciting than the previous one.
That's great. Let's dive a little bit deeper into this AI opportunity. Last quarter, HP Enterprise demonstrated a lot of momentum in AI systems with orders of $1.6 billion, AI systems revenue of $1.3 billion, a backlog of $3.4 billion. How would you articulate HP Enterprise's AI strategy today, and who are the main customers who are consuming HP Enterprise's AI products? You mentioned enterprise and sovereign cloud as an area where you're gaining a lot of momentum.
Yeah. So clearly, we see a lot of momentum. We're very pleased with our results, and we have done that with a lot of discipline, including the fact that, you know, The Street's still digesting our results, I will say. The fact that on a year-over-year basis, we actually improved profitability when it comes down to the server business. In fact, we improved our operating margin seventy basis points, notwithstanding that the mix of the server was obviously heavily weighted to the AI servers. But at the same time, we saw great momentum on the traditional servers, where also we saw double-digit growth, both sequentially and on a quarter-over-quarter basis. And it's fair to say, when you look at the profitability of our server business, it's the most profitable business on the server segment today in the market. We see tremendous demand.
Our our pipeline is a multiple of what we already, you know, converted, and the backlog is very healthy, but you know, as I think of the AI market, I think about, you know, three unique segments, and it's important we differentiate those segments. One is the service provider segment, which obviously include the hyperscalers and what I call the Tier two, Tier three segments. The Tier two, Tier three are more focused on hosting. A lot of them are not in the United States. In fact, many of them are in Europe and in Asia, and they are there to serve a purpose, which is basically giving access to enterprises and other customers, and some of them are, you know, small language model builders to accelerated computing.
Obviously, that's where a lot of the demand is, and also driven by these model builders, which consume a lot of accelerated computing and a lot of power. And so getting access to that is very important on a time basis. The second segment is the sovereign cloud. And by the way, the first segment is, you can argue, if you take the model builders aside, is maybe 10, 20 customers at the most.
Right.
And they're gonna consume hundreds of thousands of GPUs, if not, at some point, you know, exceeding a million GPUs. The second segment of the market is the sovereign cloud. And think about countries or geographies where they are gonna build AI clouds of scale to give access to enterprises and the local customers to that technology. And those are maybe in the tens of customers, maybe hundred. And they're gonna consume each time, tens of thousands of GPUs, because ultimately, that's the size of the systems they're gonna deploy. And we already won several of them, including University of Bristol, the Japan AIST, and the like. And then there is the enterprise, which is still in the early stages, and that represented for us in the mid-teens, on the order side and the revenue conversion, and that's still early stages.
But we see the maturity use cases taking place. In fact, you know, we see a tremendous traction in healthcare, in manufacturing, and services in general. And to me, that's a big opportunity. And that's why, you and I were chatting a little bit earlier, that at HPE Discover, we introduced what I call an AI-in-the-box solution. Because the enterprise, they have no time to bring together infrastructure and all the software to develop these AI applications. And that's why our solution that was co-engineered with NVIDIA is basically a fully you know engineered solution from infrastructure to the application to the workflows that enterprise customers need.
And that's early, and that's why it's so exciting, because ultimately, a lot of those customers need help upfront and on the back end of this, in addition to the technology and the software.
That's great. And I was really impressed by what I saw at HPE Discover with HPE Private Cloud AI, where you're really bringing together NVIDIA's compute, networking, and software with HPE's compute, storage, and cloud capabilities. Could you just expand a little bit more on what HPE Private Cloud AI is, and how that might be differentiated relative to what else might be in the market for enterprises?
Yeah. As you saw, it's a solution specifically targeted in the enterprise segment of the market, and the premise was about simplicity and speed: speed to time to value. That solution comes in four pre-configured solutions for inferencing, RAG, small language model training or fine-tuning, and large language model development, which not a lot of enterprises are gonna do themselves. They're gonna pick a model and then give context with the data. We thought about that value proposition from how we can speed up the deployment of this technology, and help customers in the enterprise space deliver the results on their return on investment. Therefore, it was all about the experience. The experience starts in our HPE GreenLake cloud. Today, we have 37,000 customers on the cloud platform, managing networking, servers, storage, and all the SaaS associated services, including private cloud.
And basically, from there, you just basically, with three clicks in less than thirty seconds, you are actually up and running, and we give the full stack, including the RAG, the NVIDIA AI Enterprise software, the NIMs, including the agent blueprints for specific verticals. In fact, last week, we announced some incremental additions to that, and then obviously, it's day zero, day one, and day two, all built in the same experience, and the day two is as important as day one, right, when you deploy this, so we felt, we feel this is gonna be an incredible flagship product for us, and it's part of the GreenLake experience.
We deploy that infrastructure, one of the four pre-configured, either in a customer data center or in a colo, and we can offer a customer managed or a self- or a HPE managed, meaning managed services. So the differentiation is co-engineered ready in a SKU. They can order one single product number versus a reference architecture or multiple components that customer has to put together.
Great. So HP Enterprise is a leader in exascale supercomputing, and I think the company is responsible for manufacturing and selling the top three of the top five supercomputers, including number one, which is Frontier, and number two, which is Aurora. Could you talk a little bit about how HP Enterprise's leadership in supercomputing helps support the broader AI strategy? One thing that certainly comes to mind for me is the success that you've gotten in sovereign,
Yeah
AI, right? You mentioned a couple of those projects.
I think in general, sovereign AI will look a little bit more like a supercomputer in many ways. But put aside that, I think HP has decades of experience delivering AI at scale capabilities in governments, in academia, and you know, very high-end enterprises that may need that supercomputing power. And as you said, our supercomputer business has broke the exascale barrier. That took 14 years, from the petaflops to exaflop, and the Frontier system is capable to deliver a quintillion operations per second, which is a billion billion transactions. Think about how much you can do with that and how many problems you can go solve with that. And to give a sense, Frontier today runs on 60,000 GPUs and 40,000 CPUs, all in one cohesive system.
The magic of that is actually our networking fabric. We have a coherent fabric, which we call HPE Slingshot, that allows us to bring all of this. The most fundamental metric that customers look there is the ability to start and end the model without interrupting anything. So because of that, we now have to build and run this AI system of scale. But as you know, as we go forward, the amount of energy we're going to consume is significant. In fact, some of the systems consumes tens of megawatts just for one system. As we transition to direct liquid cooling, which is foundational to what comes next, you need to have the manufacturing and the services capabilities to do it.
HPE has two amazing factories that build direct liquid cooling now for a number of years, which means the capital to build those factories has already been deployed years before. Therefore, we are ready to adopt these new technologies, whether it's from NVIDIA or any other silicon provider. Obviously, we have chatted about Blackwell now for some time, but the reality is we can build and service any system that's directly cooled. Now, in addition to that, we have hundreds of patents in that space that makes us unique and different. But ultimately, you have to service the systems, and the maintenance cycle of the systems is significantly higher. That's why I believe, as we transition to this new set of technologies, not only we can deliver better performance, which is obvious, but at the same time, lower the energy consumption.
HPE has a number of capabilities and patents there while we do the maintenance and the service, which is an opportunity for us to drive gross margin expansions as we go forward.
Great. So you've got, you know, experience in liquid cooling. You have the networking fabric with Slingshot. You're serving an existing customer in, in sovereign, and you're in a good position to deliver maintenance and services. Are there any other, you know, product capabilities or, you know, services that you like to offer that you might have to develop in-house or pursue through M&A to, you know, service what's required in, in terms of AI infrastructure demand going forward?
Yeah, I think when you think about the next generation of architectures, I do believe the networking component is going to be a core tenet of that architecture. And both through the organic investments we have made, and the pending acquisition of Juniper, we will have an amazing set of intellectual property to drive the next iteration of the architecture for accelerated computing. And I think this is where, you know, HPE think about not just the server or the rack, but we think about the whole data center as a package, and that's an opportunity for us once we close the Juniper transaction. In addition, we also have amazing software, because one of the magics, I will call it, is the software to manage contention through the networks.
So as we continue to grow and chain more and more GPUs together, and, and as Slingshot fabric already can support 80,000 GPUs in one cohesive system, we also need to make sure that the AI models run very efficiently. 'Cause the worst thing you can do, Michael, is start the model and stop. That's a waste of time, energy, and dollars, and therefore, we now have to continue to run these models of scale without interruption. And that's a value-added services that customers want and need, especially at that scale. As you go to enterprise, you know, you're going to have hundreds of thousands of customers that will deploy these much more standardized solutions, and our opportunity there is the whole stack, including the services piece.
But in the end, I don't think enterprises will deploy more than hundreds of GPUs at a time, and therefore, it's all about the speed of deployment and the ability to deliver that time to value of the use case.
Right. That makes a lot of sense. And if I could just ask about your positioning among hyperscalers for AI. You know, I think you've had some notable wins there.
Mm-hmm.
But a natural question would be, hyperscalers haven't historically worked with OEMs, so, why work with HPE for AI?
It was very simple. You know, it was the ability to provide a solution that was at data center level, ready to run the models.
Yeah.
It was not about selling just servers. We offered them, as a part of GreenLake, a complete data center solution for them to run their models. That's why we got that business from one particular hyperscaler.
Okay, great. And just kind of tying it back to financials, last quarter, HPE reported relatively stable operating margins within its server segment, where the AI systems reside, despite, you know, accelerating revenue contributions from these AI systems. How would you characterize the margin profile of the AI systems relative to the broader, you know, server segment? And what are the opportunities to improve the margins of AI systems over time?
Yeah. So, at the operating margins level, because we don't disclose the gross margin, we actually improved operating margins by seventy basis points on a year-over-year basis, but the contribution of AI was significantly higher, because last year, at this time, we only converted $300 million. This time, we converted $1.3 billion. So with $1 billion more... Let's put it this way, with $1 billion more revenue in the AI space, we actually improved seventy basis points on operating margins. And when you need to look at that, it's not just the product aspect of it, it's the services aspect of it.
Yeah.
That's why when you look at our earnings, we start disclosing in our composition between product and services, how much contribution comes from each of them. And I do believe as we grow enterprise, we grow sovereign, and we shift to direct liquid cooling, there will be more services opportunity as we go along the way. And so but we are very committed, Michael, to maintain our profitability within the ranges we guided a while back in our long-term ranges, which is 11% to 13%. And so when you look at the total server and AI segment, operating margins was 10.8%.
Yep.
But there is obviously a backlog. You mentioned that $3.4 billion. We have an amazing pipeline, and we believe we are operating with discipline, and that's the key here. I think you need to find the right balance to drive profitable growth in each of the segments and be able to deliver those results, in addition to generate the right cash through the working capital, because the amount of inventory you have to manage through these transitions.
Yeah, I appreciated those new disclosures, breaking down AI system orders and revenue by products and services—
Right
Because I think the observation was, you know, there's probably 10% to 15% of orders coming from services, but it hasn't really shown up in revenue.
It's all deferred, right?
Yep.
So that, that's the reality. Now, normally, when you sign a contract like that, you can go for three years, and all that revenue gets deferred over the three years.
Right. You know, one area that you called out and was a very, you know, was an area of strength in the quarter, was in traditional compute.
Yep.
Could you talk a little bit about what you're seeing there? Are we at a point where there should be an inflection in traditional compute demand?
I do believe the traditional server or the traditional compute is a recovery, and I think there is a little bit of pent-up demand in the market to modernize that infrastructure. But at the same time, we saw on our side that our transition to Gen 11, which is our latest generation of products, although we're already working on Gen 12, is accelerating. In fact, we said that 60% of our server business has already transitioned to Gen 11. And when you look at the Gen 11 versus the previous generation, obviously, it's more dense structurally, has a different set of cost and AUP, and then obviously, we start seeing already a slight uptick in commodity costs. So from that perspective, we feel good about it.
In fact, the traditional server grew double digits, both sequentially and on a year-over-year basis. Now, remember, that server also goes into our private cloud.
Yeah.
And so you get the benefit of selling the server as a standalone or the benefit of selling it as a part of the private cloud stack. And so in any case, everything needs a compute at some point in time. And also, I believe, which is very important to understand, is that it makes no economic sense to move a legacy workload to a server that has accelerated computing. It's not necessary. It's a waste of resource and dollars, and that's why we have not seen cannibalization from the AI business to the traditional, legacy business. I don't think that would be the case unless you rearchitect the application when you build a lot of AI.
But the reality is that that AI application, in the end, will be more used as an inferencing solution than maybe a traditional workload doing transaction type of, you know, work.
Great. That's really helpful and on the traditional server side, are you seeing any notable areas of demand when you think about individual customer verticals, whether that's government, enterprise, service provider or—
For sure, government.
Corporate.
I mean, we're a large provider to the government, and we do many things there. I think healthcare, as well. That's another area. But I will say it's more by workload, I will say. One of the things we saw, quite a bit of traction is with, you know, large IoT workloads like SAP and others, that, you know, when you think about the economics of that and think about how much data you need to move for ultimately running on the same infrastructure, pretty much, you know, unless you're running out of space and power, it is actually, depending on the size of your instance, it's cheaper to run it on-prem than off-prem.
Okay, great. Shifting gears to storage, could you talk a little bit about the transitioning that's happening in the storage portfolio, you know, migrating more to HPE Alletra, some of the mix shifts that are happening between first-party storage solutions versus third-party solutions? Maybe you can just set the groundwork for—
Yeah. Well, it's an intentional strategy. Historically, the company, if you go back 20 years, have had a mix of offerings between owned IP and non-owned IP, and, over the last, number of years, have been very intentional to drive to our own IP products, and obviously, make the R&D investments to drive that shift. This is what we refer as the HPE Alletra and Alletra MP, and MP stand for multi-protocol. But from a market trend perspective, obviously, AI would demand different type of protocols and would demand also more storage capacity. So my view is that, you know, as we go forward with Alletra MP, we have an opportunity to expand our footprint beyond the traditional market we have been participated, which is the traditional block market. And so we have multiple, I will say, inflection point.
Number one is our own install base to Alletra MP in the block space. We have a fantastic solution that can compete with anyone on price, performance, in fact, in many ways, deliver better performance with a guarantee, 100% availability and, and type of, you know, you know, as a service models that people are looking for. But that same infrastructure was architected in a scale-out software-defined way that allows us to go from block, to file, to object without changing the infrastructure. And that's a big issue for customers because you don't want to buy this infrastructure for block and that infrastructure for file and or object, and manage two different set of experiences, two different, control planes. By the way, all of this we built inside GreenLake.
So when you deploy that Alletra MP hardware, if you will, is a scale out, so you keep adding capacity. The only thing it just will change is the operating system that runs on it. And that, I call it the bits, are downloaded from GreenLake on that hardware. So you may have this capacity for block, and you need more capacity for file and object. It's the same hardware, but it's just the operating system that sits on it is now for file and object. But the control plane and the life cycle management is exactly the same inside HPE GreenLake. So this is an exciting transition for us, and we actually wrap all of that with secondary storage, with our own solution, with backup and recovery, and ransomware protection services with Zerto. And we have unique partnership.
There are very few now and very curated, where we believe are complementary to what we do, but not to offer everything to everyone, and also, we'll say, while we do all of that, it's fair to recognize that also we're changing the business model, so the hardware piece of this, again, consistent architecture, is a CapEx model, but when you download the software, whether it's for block, file, and object, that's a SaaS component. Therefore, on a, let's say, on a $1,000 average price, a portion of that $1,000 is deferred over the length of the license, which is a SaaS kind of approach, so that's important. Remember that this is also included in the private cloud for AI, so inside you have a ProLiant server, and you have HPE Alletra MP, fully certified with NVIDIA.
In both cases, you may have a switch, you know, which today is HPE Aruba for the Alletra MP offering.
Let's talk about Intelligent Edge. You know, a segment of the business that saw a tremendous amount of demand immediately following the pandemic.
Yeah
With a lot of network upgrades, but it also led to a period of inventory digestion by customers.
Yeah.
Where are we in that inventory digestion period? I believe we're mostly done with it at this point, but could you just share what you're seeing from a customer demand perspective in wireless LAN and Aruba?
I think we have done a remarkable job on the acquisition of Aruba, which I did in 2015. Just to give a sense, when I acquired Aruba, it was mainly a wireless company, a Wi-Fi company, and the size of the business in 2015 was $750 million. I reverse integrated our campus switching because Aruba is a campus and branch-focused company, and combined, they were $1.6 billion. At the end of 2023, we were $5.3 billion. So we, you know, we, we did a great job in, you know, more than two and a half times, if you will, almost three times the revenue of that business. And but you said it, right?
Obviously, COVID drove demand for, you know, modernizing the campuses as people came back to offices and the like. But now we believe we have passed the trough. In last quarter, we saw sequential orders and revenue growth, which is an important signal that the market is in recovery, and we expect that to continue for the next several quarters. And so that's another positive momentum we think we're going to have as we enter 2025, in addition to the fact that we expect to close the Juniper transaction by the end of calendar 2024 or early calendar 2025 . And that will give us, for the first time in the history of the company, and I will say even the history of HP, if you go back, a full intellectual property stack from the silicon...
Because our campus switching is 90% our own silicon, 100% of it. So, from the silicon, to the infrastructure, to the operating system, to the software, the security, and the services, to cover the edge-to-cloud spectrum through this modernization that we need because of AI, in an AI-driven approach for the $180 billion TAM, which basically we can cover the entire TAM. Remember, the combination of these two companies will be an $11 billion business, so more than double the current networking segment, as reported. It's gonna be representing 35% or so of the company's revenue, but more than 50% of the company's profit.
So clearly, we are shifting the look of our portfolio to higher growth, higher margin areas on a control point that's essential as we go through this inflection point from the architectural perspective. So we feel good. You know, the momentum is there, recovery in the market, in line with the peers, by the way. You know, the Juniper transaction, you know, closing around that timeline, I just said, and, you know, the combination of the two assets giving us a different financial profile, but most importantly, a different IP that give us the relevance in this new market.
Right, and Intelligent Edge is one of the higher margin segments within the organization—
Yeah.
you know, mid-twenties.
Yep.
Long-term EBIT margins. Could you discuss some of the factors that are gonna impact margins over the near to midterm within Intelligent Edge? Why is mid-twenties the right long-term margin profile?
A couple of things. You know, as I said, you know, at the hardware level, we own much of our IP, except the radio access piece of it, which obviously we work with Broadcom or Qualcomm. But on the other stuff, on the campus switching, is our own IP. But a lot of the momentum we see is the experience. Aruba was conceived with a cloud-native approach, and a mobile-first approach, which allows customers not only to provide connectivity, but to build experiences around that connectivity. And we see tremendous traction in each of the verticals, whether you take hospitality or whether you take healthcare, or you take, you know, manufacturing into the IoT space. And we have been very diligent in adding to adjacent market along the way. So again, Aruba was a Wi-Fi reverse integrated and campus switching.
Then we added SD-WAN through the acquisition of Silver Peak. SASE is a necessity in at the edge through the SSE, the Security Service Edge. So now we have the true consolidation of the network layer. We expanded into private 5G, which is a huge demand for manufacturing and other verticals, and into the IoT. So we believe we have a full, comprehensive approach there. We added on top of that, the cloud layer, and that's one of the reasons why Aruba is growing so much inside HPE GreenLake, and our ARR is $1.7 billion and growing close to 40%. And then obviously, the AIOps, which is this AI-driven.
But Juniper brings the complementarity with Mist in one segment of the market, in the campus and branch, and then into the cloud and the service provider space, and in the enterprise data center switching space.
Yeah. On capital allocation, HP put out a target to return sixty-five—
Yep
To 75% of free cash flow to shareholders over the next few years. Could you talk about how you're balancing the return of cash to shareholders with reinvesting in the business and M&A, and what's your appetite for more M&A, particularly given the backdrop of some of the pending deals?
Yeah. So first of all, we always, you know, return capital to shareholders. So I think about it. I've been six and a half years as a CEO. In the first six years or so, we returned almost $11 billion of capital to shareholders. That's, you know, when you think about our free cash flow, approximately around $2 billion, right? That's in line with cash generation, although we had obviously the spin-off at the time before I became CEO. That said, obviously, the first thing is to pay dividends, and dividend is a very critical component of our capital allocation. This past quarter, we paid $0.13. At the beginning of the year, we raised the dividend by 8%, so that's consistent. In terms of share buybacks, it is an allocation between share buybacks and investment, where it makes sense.
We do that on a stringent return on invested capital, and also understand we need to make investments in the business. And we felt that this time, doubling the networking business was the right time because of what we see in AI and because of the ability to change dramatically our financial profile. And by the way, that transaction is a no-brainer for shareholder, because in the end, we are committed at least $450 million of synergies, which pays for the deal itself. So, I believe that that's a no-brainer. But obviously, as we go forward, our goal is to deliver that 65% to 75%, and the first thing is that in year one post-close, the transaction is already accretive on a non-GAAP basis. Obviously, we're gonna drive the synergies as aggressive as we can.
But already, year one post-close, this is gonna be accretive for shareholders on a non-GAAP basis. And then obviously, pay down the debt, which our commitment is to bring them down to two times EBITDA, and we have plans to do that. And so that's the journey we're gonna be. And honestly, ultimately, this company has done what we said, and I'm very confident in our team to go execute that strategy.
Antonio, it's been such a privilege to have you on stage. Thank you for joining us here and at the conference.
Thank you, Michael.
Thank you.
All right.