Please welcome Vice President, Investor Relations, Jeff Kvaal.
Good afternoon. Welcome to the 2023 HPE Securities Analyst Meeting. Thanks for joining us in our return to New York. We haven't been here since before COVID, so four or five years now. It's nice to see so many familiar faces in the audience. I hope for you New Yorkers, this counts as an in-the-office day, I trust. And thank you to the many folks that are tuning in virtually as well. The agenda, we have Antonio first. He'll give us a look at how we're pivoting to our higher growth, higher margin recurring revenues, also positioning ourselves for sustainable and profitable growth. Through his talk, Antonio will introduce the heads of three of our new segments.
So we've got Phil Mottram from Intelligent Edge, Fidelma Russo in Hybrid Cloud, and also Justin Hotard from HPC and AI. They'll give us deeper dives into each of their own businesses. We'll then take a break until 4:00. Jeremy will come on after. During the break, please give the execs a little bit of space. We'll have plenty of time for Q&A at the end. Jeremy's presentation wraps up, we'll have approximately an hour or so for Q&A. Okay. So with that, I guess, let me mention that once the webcast does conclude, that's when we'll be issuing our press release and putting it all on the presentation materials on our website. That is at hpe/investor/sam2023. Also, we'll put a replay up on the IR website.
We'll keep that up there for about a year. And lastly, let me get to the disclosures. The event may include forward-looking statements involving risks, estimates, and assumptions. HPE assumes no obligation to update those statements. And with that, it's now my pleasure to welcome Antonio Neri, HPE's President and CEO. Antonio.
All right. Thank you. Well, good afternoon, and thank you for joining us today. It is exciting to be back at the New York Stock Exchange. I had the honor to, this morning, to do again, the opening bell. It's always fun to be here. And welcome to those of you joining us virtually via the webcast. Together with my leadership team today, I'm excited to, for the opportunity to demonstrate to you how much additional value our strategy can create for our shareholders in the near and long term. I'm very proud of the progress we have made in the last five years. The world around us has changed immensely, and so have we.
Today, as I prepare to share with you the opportunity we see for HPE in the next three years, also, reflect on the last three years and what they have demanded of HPE and enterprises around the world. They require vision, agility, and resilience from every enterprise around the globe. We all were confronted with unimaginable challenges. In that time period, HPE has transformed to become the edge-to-cloud company. We have anticipated what customers will need as they recover from, the pandemic, which we call the new world post-pandemic. We have offered unparalleled innovation, and we have executed extremely well. Since the beginning of fiscal year 2021, HPE has successfully deployed our strategic pivot to higher margin and higher growth areas of the market, which has created strong value for our shareholders.
We have delivered 13.4% revenue gross growth since then, compared to the last trailing twelve months, with particularly exceptional growth in recurring revenue. Since we began our strategic as-a-service pivot in 2018, and started tracking annualized revenue run rate, or ARR, we have tripled it. Last quarter, ARR rose to a record of 48%, exceeding a long-term target growth rate of between 35% and 45%, which we shared last year at SAM. HPE has grown our annual, annual non-GAAP diluted net earnings per share by 43.5% between fiscal year 2020 and the trailing twelve months through the most recent quarter. In fiscal year 2020, it was $1.54, and it is now at $2.21. Our non-GAAP gross margin growth has also been impressive.
We have added 310 basis points to our non-GAAP gross margin since fiscal year 2021, ending the most recent trailing twelve months period at 34.8%. Our non-GAAP operating margins improved by 280 basis points in the same period. We have made these profitability gains as a result of diversifying our business to grow the software and services-rich parts of our portfolio, while maintaining discipline in pricing, particularly in our compute business segment. We have introduced profitable subscription-based cloud services with HPE GreenLake, which enhances the stickiness of our solutions with our customers. Greater profitability has also enabled us to significantly increase free cash flow generation in the last three years. From the start of fiscal year 2021 through the most recent trailing twelve months, we have more than tripled free cash flow generation.
We are on track to deliver within the range we guided during fiscal year 2023 Q3 earnings. Over approximately the last three years, we have strategically focused our portfolio on areas of high growth and high margin. We generated a more than 121% total shareholder return, as compared to a 37% return for the S&P 500 over the same period. Importantly, this return represent a CAGR of 31%, as compared to the 11% CAGR for the S&P 500. As we produce higher profit and greater cash flow, we deliver on our commitments to provide direct capital return to our investors. Through our disciplined capital allocation framework, we returned about $11 billion to shareholders through dividends and share purchases from the start of fiscal year 2018 to our most recent quarter. Obviously, that included the spin-off proceeds.
But this capital return benefit represent 122% of our free cash flow over that period. These strong, consistent results would have been admirable in any period, but I'm particularly proud of them coming off such a turbulent three years. Frankly, I believe the market should give us more credit for the accomplishments, because they are a testament to the relevance of our strategy, the differentiation of our portfolio, and the strength of our execution. We deliver when it matter the most, and as we continue to execute against this strategy in fiscal year 2024, and over the next three years, we will continue to accelerate value for our shareholders. Today, I will discuss HPE's compelling market opportunity for sustainable, profitable growth. At the conclusion of our time together, I hope you walk away with a deeper understanding of a few things.
First, how HPE's bet on intelligent edge is paying off, with a sizable proportion of our revenue growth and segment operating profit coming from this business segment, and how we plan to sustain our momentum through TAM expansion and market share gains. Second, how our Hybrid Cloud business is positioned to become a sustainable center for growth through the scale of HPE GreenLake, with investments to make us the number one leading Hybrid Cloud provider. Third, how AI has expanded our overall market opportunity to drive what we expect will be profitable growth in the next fiscal year, adding to our overall compute in HPC and AI businesses. And finally, how our continued mix shift to those areas of our portfolio, combined with operational discipline, will drive profitable revenue growth for our shareholders.
HPE's strategy is aligned to the significant market trends we see today around Edge, Hybrid Cloud, and AI, all which will create profitable market expansion opportunities that will help us fuel our growth. Customers continue to validate our strategy, turning to us to power critical business transformations. Even in this macroeconomic environment, we continue to see them prioritizing data-first digital transformation initiatives. Those initiatives increasingly include AI, which is invigorating today's IT spending. We anticipate the overall total addressable market across our portfolio will rise by nearly $100 billion from the end of fiscal year 2022, to more than $340 billion by fiscal year 2026, when we eliminate some of the crossovers, because obviously you have some elements of the portfolio can be added in, in different segments.
We intend to capture this growing TAM across the mega trends of edge, Hybrid Cloud, and AI through a mix of HPE's business segments. Our approach will include innovation in key markets, where we already have a very strong position, as well as customer acquisition and expansion in high-margin adjacent markets. Let's start at the edge. At the edge, we are expanding our innovation in security, private 5G, and data center networking. In Hybrid Cloud, we continue to expand in private cloud, the segment of the market we anticipate growing at approximately 10% CAGR. We can also capture a larger portion of the Hybrid Cloud market with our entry into the growing AI-powered IT operations management segment through our recent acquisition of OpsRamp.
In AI, we are well positioned to capture the sizable growth with our full stack AI native architecture. That includes AI infrastructure growing at a 23% CAGR, and AI platform software growing at a 32% CAGR. Ultimately, we anticipate the AI market, where we play, growing by roughly 2.4 times to almost $150 billion by 2026. Clearly, there is a great opportunity for HPE in those growth segments. So let's detail each, each one today, so you can emerge with a better appreciation for our differentiated innovation and the strength in each category, and our plans to further capitalize on the opportunity ahead of us. Our intelligent edge business is a prime example of what HPE can do to maximize future growth with the right investment at the right time.
We have invested nearly $6 billion in our Aruba, HPE Aruba Networking business, organically and inorganically, since I became CEO. It is on track to be a more than $5 billion annual business in fiscal year 2023 and beyond, poised to sustain, to generate sustained revenue growth and the highest profitability of any of our business segments. It is a critical part of our business, both from a strategic and financial standpoint. Our broad portfolio delivers what we call security-first networking, from edge to cloud to exascale, which is very important component. We're gonna talk later about that. In fiscal year 2024, we will build on our successes in our HPE Aruba Networking, the core businesses of campus and branch, with a goal of continuing to take share in the enterprise, mid-market, and SMB customer segments.
We are expanding our offers in growing markets like security, private 5G, and data center networking, which we have entered through organic innovation and acquisitions this year. We intend to increase our already high gross margin in this business. By fiscal year 2026, we expect our total addressable market at the edge and networking to grow to about $94 billion. Customers will continue to expect a unified networking experience that is secure, automated, and high performing, as well as flexible subscription and consumption-based solutions. Our value proposition is absolutely suited to address these needs. This adds up to the significant opportunity for HPE and for our shareholders. Phil Mottram will now provide more details about our leadership position and priorities in this very important segment. So I would like to welcome Phil. Phil, please.
Thank you, Antonio. So, my name is Phil Mottram, and I've been running the Intelligent Edge business now for a little over two years, and it's been an incredible journey for me and my team. As you look around this room today, and you admire the great decoration, and you're scanning the ceiling occasionally, I also want you to pay attention to the Aruba access points that have been installed in the alcoves and the corners. So NYSE make great technology choices, and we're proud to be the wireless provider here at the New York Stock Exchange. So nice one, NYSE, if that's a phrase we can use. If you've been following HPE for the last few years, you've undoubtedly heard Antonio refer to us as the edge-to-cloud company.
The first part is all about the connectivity, and that's where our intelligent edge business comes into play. In the last 2 years, we've seen significant growth in this segment, and by the end of 2023, we expect the annual revenue of our intelligent edge business to increase by nearly $2 billion from FY 2021 levels through consistent growth over that timeframe. The third quarter of FY 2023 marked the 11th consecutive quarter of year-over-year growth, and this segment represented 20% of the company's revenues. In addition to the top line revenue growth, we've also seen an increase in profits since 2021. Through intentional efforts towards operational efficiency, as well as portfolio management and a focus on margin-rich products, we've increased our profitability. In Q3, we marked a record level of operating profit margin, and we accounted for 49% of the company's total segment operating profits.
So we've done this by capitalizing on the high demand in the market for our products and solutions, strategically adding more product areas to expand our TAM, and focusing on our delivery mechanisms. We're giving the customers the option for consumption or subscription purchasing through our network as a service offerings and through HPE GreenLake. With Aruba Central on HPE GreenLake, we're one of the only companies giving customers the ability to manage all of their network solutions from one single platform. As we go forward, we will continue to follow the network in order to identify selling and upselling opportunities, from the access point in the ceiling, to the campus switch, to the SD-WAN network. Selling these incremental products to our existing customers will continue to drive revenue in core business segments.
We'll also be following the network even further into the data center, integrating it with security and adding private 5G and delivering it all through the same unified cloud native experience. Additionally, we are weaving sustainability into everything that we do. This is something we feel is vitally important, and we've heard loud and clear that it's a priority for our customers as well. To give you an example of this, our asset lifecycle management program within HPE Financial Services reclaims and refurbishes used assets from our customers. Over the last three years, we've given more than 8.2 million devices a second life, ultimately reducing the amount of material that's directed towards landfill. Bringing all of this together, we're confident that we have the right strategy in place to drive sustainable, profitable growth and deliver value for our shareholders.
As we look ahead, we believe that the trends in the marketplace around hybrid work, security, and data center requirements will create new opportunities for us. While there are some headwinds related to the overall networking market growth projections, we expect to continue to sustain our Intelligent Edge business segment performance through TAM expansion and continued market share gain in key segments. We continue to innovate and enhance our offering and receive validation from leading industry analysts in this regard. In the edge networking space, we're targeting a TAM of about $94 billion by 2026, and that's 1.5 times the TAM that we pursued in 2022. Within that, I see four distinct areas of growth for our business. The first area is security.
We believe that the networking market is converging with the security market under a framework called SASE, or Secure Access Service Edge, and we're not alone in that belief. Gartner believe that 60% of enterprise customers will buy SASE from a single vendor within three years. SASE has five key elements, and we already had two of them with networking and firewalls. The acquisition of Axis Security earlier this year allowed us to add the missing three elements of SASE, namely ZTNA, SWG, and CASB. Within HPE Networking, HPE Aruba Networks, I'm sorry, we've always been great at protecting things on the network. Adding Axis Security now gives us the ability to be able to protect people. So we can protect people and things across multiple network devices, all from one location on HPE GreenLake.
We have a compelling and highly differentiated offer, and we're poised to gain share in this high-growth market that will represent a $3 billion TAM in 2026. The second area of growth for our business is in private 5G. So the rollout of 5G has advanced mobile networks around the world, and in many countries, governments have now assigned 5G spectrum to be used by enterprises privately, which is what is referred to as private 5G. We believe that private 5G and Wi-Fi will exist side by side, and enabling new customer use cases where outdoor coverage and latency is important, such as in ports, defense applications, and other scenarios like mining operations and sporting events. Indeed, we recently provided private 5G and Wi-Fi 6E connectivity for the Ryder Cup in Italy.
So we provided always-on connectivity to 250,000 spectators, while also delivering a secure private network for the operations staff across the hundreds of acres that made up the venue. So hats off to us for that. So private 5G represents a significant opportunity to HPE to enable a unified customer experience with security and policy management. Additionally, this represents a significant opportunity to HPE to capture market share in this space at healthy gross margins. The third area of growth for our business is in data center networking, which is an area that will grow to a $19 billion TAM by 2026. Today, our HPE Aruba Networking CX switch platform is already sold into many data centers around the world.
We've spent 12 months enhancing the capability of our switching products and have a 2-year roadmap that is highly differentiated through our HPE Aruba Networking CX operating system, which will enable enterprise customers to deploy cloud-native networking in their data centers or colo facilities. We expect to continue to capture market share from existing customers, especially those who want to extend the management ease of Aruba Central into their data center. The fourth area of growth for our business is in NaaS or Network as a Service. The NaaS market is expected to reach, to represent a $3 billion opportunity by 2026, which is good news. But the better news is that we already have a clear lead in this market.
With our customers, we can take them on a journey to NaaS, as opposed to some of the more recent startups in this space, which require customers to rip and replace all of their network technology in order to deploy NaaS. Our intelligence services capability, or the managed services piece of NaaS, offers a 98% reduction of network events through automation and utilization of AI, ML-driven analytics. NaaS also represents an important way in which we're delivering on our commitment to sustainability, because NaaS creates a circular economy by extending the life of assets for customers. For these reasons and more, we continue to see demand from customers for NaaS solutions. Today, we have a growing number of customers across different verticals, including retail, hospitality, higher education, and manufacturing. And some of the customers already consuming our NaaS services, The Home Depot, CarMax, and KPMG.
And now, in addition to large and global customers, we're now scaling into the commercial and mid-market segments through and with our channel partners. So those are the four areas where we expect to see significant growth in our business. Aggressively pursuing these growth opportunities while continuing to capture revenue in our core segments, is how we will drive sustained business momentum. And our ability to do that can all be put down to one key point: customers buy HPE Aruba Networking, because we are customer-centric. Aruba Central, which is our HPE GreenLake cloud control plane, is the industry's first AI-powered, cloud-native architecture designed to connect, protect, and automate the edge. This gives customers a simple but powerful management tool to drive efficiencies in their business. And the three core layers of our portfolio each deliver specific benefits for customers.
First, our unified infrastructure helps eliminate operational silos and streamline operations. We do this by bringing together wired, wireless, and SD-WAN technologies across campus, branch, remote worker, and data center locations into a single cloud-native management and orchestration tool, and that's Aruba Central. Second, our edge to cloud security simplifies network security through a combination of built-in zero trust features focused on users, devices, and IoT, and a SASE integration framework that combines WAN edge functions and integration with third-party cloud-delivered security services. And third, our AI and automation software enables our customers to resolve issues quickly, ideally, before they impact the business, while also helping IT departments operate more efficiently. We have a robust portfolio of products and solutions that deliver what our customers want and need to harness the power of the edge. And while we've seen fantastic growth in recent years, we are not standing still.
We have the right strategy in place, and we'll continue to accelerate shareholder value by sustaining growth in core segments, expanding our overall TAM, and gaining market share in key high-growth, margin-rich areas. Back to you, Antonio.
HPE is a pioneering Hybrid Cloud, which represents our largest market opportunity. Recognizing the strategic importance of this market and our differentiated value proposition led by HPE GreenLake, last month, we announced we will create a new Hybrid Cloud business unit at the start of fiscal year 2024. Establishing this new segment reflects the maturity of our offering and set us apart from our peers. Our Hybrid Cloud business unit will bring together into a single Hybrid Cloud segment, all storage and compute as a service, all the offers related to that, inclusive of our HPE GreenLake Private Cloud and software solutions. Our updated operating model will enable greater focus and efficiency, faster execution, and a superior cloud-native experience for our customers and our partners. Our simplified operating model also incorporates a sales transformation to enhance the execution between our business units and our go-to-market function.
Through these changes, we will engage customers more directly on the offerings they want the most. HPE GreenLake is at the heart of our Hybrid Cloud strategy, and its growth is truly remarkable. At the end of Q3, we support 27,000 unique customer logos and 3.4 million connected devices. Today, more than 1,100 partners transact HPE GreenLake Edge and Hybrid Cloud offerings, showcasing HPE GreenLake's incredible cloud strength and scale. HPE GreenLake also plays a key role in expanding our gross margin with software and services, comprising an increasing portion of our ARR mix, now nearly at 70% at the end of Q3.
Fidelma Russo, our Chief Technology Officer, has been leading our HPE GreenLake cloud platform development and will take on the responsibility of accelerating our Hybrid Cloud services growth opportunity when she becomes the General Manager of the new Hybrid Cloud business segment on November first. I would like to invite her to talk about how we will enhance our leadership position in this growing market. Fidelma?
Good afternoon, everyone. I'm honored to lead our future Hybrid Cloud business segment, which I believe will be a large and enduring growth engine for HPE and will create long-term shareholder value. Enterprises across the world are moving from a public cloud first approach to a hybrid approach.... They've come to the realization that as their data grows and becomes more distributed in nature, it is neither scalable nor economical to rely solely on the public cloud. The emergence of generative AI has made this reality even more clear, and this creates a tremendous opportunity for us. Our new Hybrid Cloud business segment accelerates our participation in several large and attractive markets across storage, private cloud, and infrastructure software.
Combined, we are looking at a $160 billion+ market that is highly profitable, and as enterprises continue to embrace hybrid, this market will see robust growth for the future. To capitalize on the market opportunity, our strategy is to take share in storage, scale private cloud with the momentum we have with HPE GreenLake, and expand into infrastructure software. Ultimately, this strategy starts and ends with our customer. We offer the most differentiated customer value proposition among our competitors, including cloud providers. We have a growing portfolio of market-leading offerings, and we are seeing strong customer adoption. As Antonio said, we are a pioneer and leader in Hybrid Cloud with HPE GreenLake.
The reason we've been successful, and the reason we've been able to differentiate versus the public cloud, is because when we show up to our customers, our mission is to solve three of their biggest challenges: transform their business through the power of data, modernize their IT infrastructure with a true cloud experience, and dramatically simplify the operations of their multigenerational and multicloud IT estates. Let me give you some more color on how we address each of these customer challenges, and how different aspects of our portfolio come together to drive an integrated value proposition through the power of our GreenLake platform. Everything starts with data, which is the most valuable asset companies have, and it is at the center of any digital transformation.
Its gravity and distributed nature are what drive enterprises to embrace the Hybrid Cloud, and this is why storage is foundational to our Hybrid Cloud strategy. With our HPE Alletra line of storage products, we offer customers a value proposition none of our competitors can match. We have engineered a truly modern, scalable platform with unified support for block and file that is capable of meeting the most demanding requirements for AI. It has been engineered for a true Hybrid Cloud experience, and it's made possible with an architecture that's built from the ground up to be cloud native and software-defined. We continue to deliver industry-leading uptime and resilience, enabled by our market-leading AIOps capabilities. And since its launch, HPE Alletra has seen the fastest customer adoption of any storage platform in HPE history.
In addition to our product leadership, we are also increasing investments in our go-to-market, both in our direct sales force and in the channel. And with these investments, we are poised to take share from our competitors. But not only are customers trusting us with our data, they are also trusting HPE to build, and in many cases, operate their Hybrid Cloud infrastructure on top of their data through HPE GreenLake. Our entire strategy around HPE GreenLake is built on three foundational beliefs which differentiate us from the public cloud. First, customers should not have to choose between the agility of the public cloud and the performance and control of their private infrastructure. Second, customers want choice and freedom from lock-in. And third, they operate multigenerational, multi-vendor IT estates that need to coexist with the public cloud.
Over the last five years, we have built a multi-billion-dollar HPE GreenLake franchise, and it is the envy of our competitors, and we are continuing to double down on that success. This year, we added a host of new cloud-native offerings and capabilities, including hybrid multicloud orchestration for virtual machines, containers, and bare metal, a full suite of private cloud offerings that enable customers to self-manage or choose a fully managed experience, and a portfolio of world-class AI infrastructure delivered as a service. The portfolio clearly resonates with customers, and it is allowing us to extend our market leadership. Now, we understand that the adoption of hybrid introduces several operational challenges for our customers...
Through a curated strategy of organic investments and acquisitions, we have built a compelling set of SaaS offerings aimed at helping customers simplify the data management and protection of their hybrid multi-cloud environment, while reducing the risk of public cloud lock-in. We have focused on three critical customer needs: AI-powered monitoring and observability for day two operations and beyond, through our acquisition of OpsRamp. Unified data access through our HPE Ezmeral Data and Analytics Suite, and that helps customers move and transform their data for use in AI and other applications. Data lifecycle management and protection through our suite of offerings, including Zerto Disaster Recovery. All of our software is built for hybrid, to give our customers the flexibility they need to simplify their Hybrid Cloud operations at any location.
Our offers support multi-vendor, multi-cloud environments and enable native integrations for both HPE GreenLake and major public clouds. Our strategy with our HPE GreenLake SaaS offerings is to drive aggressive penetration across our customer base. All of our Hybrid Cloud offerings, storage, private cloud, and SaaS, as well as HPE Aruba Networking and HPE Compute offerings, are natively delivered through our HPE GreenLake cloud platform, and we will deliver our AI offers natively on the platform over the next couple of years. The platform enables our customers to have a consistent cloud-based management experience across all our offerings. It provides a set of essential services, including consumption analytics and a sustainability dashboard, that helps customers understand and reduce their carbon footprint.
As more customers adopt our cloud platform, they are reaping the operational benefits of a truly unified, cloud-based management model, which reduces upfront CapEx and ongoing OpEx running costs. The more HPE offerings our customers consume through HPE GreenLake, the bigger the operational benefit they gain, and the higher the value we have generated for our shareholders. To bring our differentiation to life, I'd like to share a couple of examples of how our customers are leveraging HPE's Hybrid Cloud portfolio to solve their business challenges. For those of you who have a great time watching football, you will be glad to know the Dallas Cowboys were looking for a solution to allow their players and coaches to watch and analyze video footage across different locations, and our high-performance Alletra storage solution with multi-site replication perfectly suited their needs.
Danfoss loved the managed for you aspect of our private cloud solution, and it allows them to focus on delivering business innovation globally while reducing their carbon footprint. So there's no question that HPE GreenLake is winning in the market, and Hybrid Cloud is the driving force behind it. We're seeing tremendous adoption with our customers and partners, with more than 27,000 unique customers using the HPE GreenLake platform. We are creating significant shareholder value because we are expanding our infrastructure offerings into new and higher growth markets. Our recurring revenues are growing at more than a 35% CAGR, which is faster than the public cloud, and we are delivering greater profitability through our software and services-rich portfolio. In closing, HPE has a winning Hybrid Cloud strategy.
We have a highly differentiated value proposition that is grounded in a deep understanding of our customers' evolving needs with respect to data, infrastructure modernization, and Hybrid Cloud operations. The differentiation comes from being a pioneer in as-a-service with HPE GreenLake. Our unrivaled HPE edge-to-cloud portfolio, and it's further enhanced by years of curated M&A, organic investment in our SaaS portfolio, and our HPE GreenLake platform. The value proposition is clearly resonating with our customers, and our continued success will fuel sustained growth and profit expansion for our shareholders. Now back to Antonio. Thank you.
Well, thank you, Fidelma. Excited to have Fidelma now lead this, great opportunity we have in Hybrid Cloud. Just as we have a very differentiated Hybrid Cloud value proposition, we are also uniquely positioned, with a compelling high-performance computing and AI offering, as this market dramatically expands. The market's extreme acceleration in the last year and anticipated growth over the next several years reflects the enterprise realization that they must embrace AI, or they will lose their competitiveness and get left behind.... As organizations lean into AI, they are discovering a few things. First, the data intensity of the AI workloads require a hybrid-by-design solution rather than a cloud-only approach. They need a solution across the entire AI life cycle, from training to tuning to inferencing.
Given the energy and data center services required for large-scale AI workloads, sustainability must be built into the technology from the start. HPE addresses these needs with an end-to-end portfolio designed for the full spectrum of use cases, spanning large-scale AI model development, training, and inferencing, as well as unique liquid cooling data center services expertise. Customers are attracted to HPE's market-leading supercomputing capabilities, differentiated network and interconnect IP, AI-specific software and services expertise. These capabilities position us favorably in a rapidly growing total addressable market, comprising HPC and supercomputing, AI infrastructure, and AI platform software. Our pursuit of an outsized share of this market opportunity will deliver real value to shareholders, especially when with our path to AI-driven profitability charted in the next fiscal year. Our investments in our unique silicon and software are expected to drive a larger, profitable revenue base.
I would like to welcome Justin Hotard, who leads our HPC and AI segment, to discuss how we'll plan to capitalize on the booming AI opportunity. Justin?
Thank you, Antonio. AI is driving the next wave of investment in innovation, resulting in value creation in the IT market and the global economy. It will have a transformative impact, similar to what we saw with web, mobile, and cloud. We are seeing demand shift dramatically as our customers realize the potential of AI to deliver business transformation. Today, I want to make sure I cover three points with you. First, the AI market is growing rapidly, and HPE is well positioned to profitably capture that market expansion. Second, we have a right to play and win in AI, and we're already winning in the market today. We already are seeing this in our sales pipeline, in our orders, and in our revenue. Third, we deliver solutions for the AI life cycle across training, tuning, and inferencing that will drive higher growth and margin expansion.
AI adoption is fueling significant market growth. We're targeting three areas of the AI market: supercomputing, the AI infrastructure, and AI software platform. As you can see, our total addressable market for these segments will grow at nearly 24% CAGR to $146 billion by 2026. By focusing on these areas, we are empowering customers to transform their businesses to dramatically enhance productivity, accelerate innovation, and create new revenue models. Let me cover each in depth. First, supercomputing. HPE is a market leader in delivering the world's leading supercomputing and high-performance computing solutions. We hold the number one position of performance share for the world's 500 fastest supercomputers and deliver the majority of the world's top 10 most efficient supercomputers. Our leadership positions us to continue to capture share in the market's forecasted double-digit growth.
We will continue to invest in our supercomputing technology and leverage our global service delivery footprint. The second is the AI infrastructure market. By 2026, this market is expected to be 7 times larger than our core supercomputing market and reach $86 billion. A significant amount of this growth will be driven by AI model training, which is one of the most computationally intense workloads of our time. Customers require powerful infrastructure capable of the scale and performance that supercomputing technologies deliver. For this reason, key features of supercomputing technologies, including management software, networking, and liquid cooling, will become requirements for AI infrastructure. These are technologies where HPE has unique intellectual property that positions us to deliver differentiated solutions to customers. Third, we have an opportunity above the infrastructure layer in the AI software platform market.
This market is expected to grow to $49 billion by 2026, and we plan to capture share through continued investment and innovation. We will extend our current software suite and invest in new offerings to specifically target market opportunities across the entire AI life cycle. As this portion of our portfolio grows, we anticipate healthy margins that will be accretive to our business. We believe our strategy to focus on these three segments across the AI life cycle of training, tuning, and inferencing will lead to increased market share in two ways. First, through our core supercomputing business and trusted global brand, we'll create new opportunities and expand our share of wallet within our existing customer base. Second, by further building on these core offerings with purpose-built AI solutions and HPE's global presence, we expect to attract new customers to win even more market share.
In fact, it's the combination of the market opportunities across supercomputing, AI infrastructure, and the AI software platform that differentiates HPE from the competition. Now, let me transition and cover how the growing AI market will be driven by compute and data-intensive workloads, and will need an architecture that HPE is uniquely positioned to provide. This AI-native architecture is different from the cloud architecture. Traditional cloud architectures are optimized to run multiple workloads on a single server. They were not designed with AI in mind. There are four key attributes of an AI-native architecture. First, the architecture should be designed with computing and infrastructure that is at the scale of supercomputing to take full advantage of computing capacity. This is essential to train, retrain, and tune AI models using large quantities of data with efficiency, speed, and accuracy to accelerate time to value.
Second, ensuring a truly hybrid design provides the flexibility to train, tune, and deploy AI models in any environment. Customers need the flexibility to integrate data that exists in the public cloud, the private cloud, and the edge. For example, enterprises need to deploy inferencing where they can deliver real-time insights. This will be necessary for large language models as enterprise use cases scale, just as it already is for computer vision applications across autonomous driving and medical imaging. Third, an open hardware and software ecosystem delivers strong advantages. Having an open AI ecosystem expands training and inferencing market solutions to accelerate adoption for customers and support ongoing innovation in the broader ecosystem. And fourth, these architectures must be sustainable by design. By 2028, it's estimated that AI workloads will grow at a 35% CAGR and require about 20 GW of power within data centers.
Customers will need an architecture that satisfies this demand at a new level of energy efficiency to minimize the impact of their carbon footprint. HPE is uniquely positioned to deliver this new AI-native architecture through our technology differentiation and expertise. For many years, we've made strategic investments in AI and led AI-focused research in Hewlett Packard Labs. As a part of this, we have a powerful blend of strong technology, intellectual property, expertise, and talent to deliver an AI-native architecture. This positions us to be a long-term market leader and to capture market share in the nearly $150 billion AI TAM I previously highlighted. Our multi-year investment plan targets three distinct areas to enhance our differentiated position in the market that will set us apart from the competition. These include software, high-performance networking, and supercomputing infrastructure.
We expect these investments to significantly expand our HPC and AI business segment operating margins by 2026. First, our machine learning development platform is proven to train quickly, efficiently, and at scale to create reliable and accurate models that can make valuable predictions and reduce business risk. With our current software and the investments we're making across the AI life cycle, we will deliver a robust platform of open source software. This software stack also integrates with the HPE Ezmeral Data Fabric that Fidelma covered in her overview. By integrating the two platforms, we're making it easy for customers to manage their entire data life cycle. Second, we deliver Ethernet-based high-performance networking with HPE Slingshot, which is purpose-built and proven for significant AI scaling. This technology currently powers Frontier, the world's fastest supercomputer, with nearly 40,000 GPUs at the United States Department of Energy's Oak Ridge National Laboratory.
HPE Slingshot will also power the upcoming Aurora exascale supercomputer, which features more than 60,000 GPUs at Argonne National Laboratory. HPE Slingshot connects all GPUs to operate as one single large supercomputer. This makes it possible to train 1-trillion-parameter AI models in one single instance. HPE Slingshot is built upon the Ethernet standard, and it already supports a variety of GPUs and accelerators, including those from NVIDIA, AMD, and Intel. This creates market options for customers' training needs as the market continues to scale. Finally, to support the reliability and resiliency of AI systems required for large-scale training, we offer powerful and integrated supercomputing solutions to address these needs. These include our expertise in system integration and our supercomputing software. As a part of our supercomputing infrastructure, we have extensive experience in R&D that focuses on sophisticated liquid cooling solutions.
Liquid cooling is essential to deliver sustainable data center infrastructure for AI. Our solutions can drive up to a 20% performance improvement per kilowatt over air-cooled solutions and consume 15% less power. Further, we believe next-generation accelerators will require liquid cooling in every system to meet power and thermal demands. Our intellectual property and liquid cooling infrastructure, and our advanced manufacturing capabilities in this area position us well to capture this demand. However, technology alone is not sufficient. Customers training AI models don't have the time to develop the expertise to operate their AI infrastructure and platform. This is an opportunity that positions us to offer unique innovations like a virtual private cloud for AI, with HPE GreenLake for Large Language Models that we announced in June.
This is a turnkey cloud service that integrates our AI infrastructure and AI software platform for model training, where we see significant global demand. The service is also designed to lower carbon footprint. As I covered, our intellectual property in AI is proven, and it positions us well to accelerate innovation in the market. This is why customers from various verticals are turning to us. I'd like to point out a couple of examples that illustrate how our AI-focused solutions are making a transformative impact in the market today. We have a strong partnership with the U.S. Department of Energy and its national laboratories to co-design and co-develop powerful supercomputers, as we have with Exascale, to accelerate national initiatives in AI-driven science and innovation. Argonne National Laboratory, for example, is creating a series of generative AI models at the exascale level that will be trained on HPE supercomputers.
In the pharmaceutical industry, Recursion Pharmaceuticals, a leading TechBio company, uses advancements in AI to accelerate and industrialize the discovery of new drugs. Recursion uses the HPE Machine Learning Development Environment to manage its large-scale AI training jobs on its AI supercomputer. This software significantly speeds up model training across more than 25 PB of biological and chemical data and improves team collaboration. Customers like Taiga Cloud and Crusoe Energy partner with us to leverage our expertise in global supercomputing leadership to deliver full stack solutions that integrate our industry-leading infrastructure, open source software, and on-site services. Our supercomputers and AI software are also behind the work of foundation model builders that create pre-built models used by broader enterprises. For example, Aleph Alpha has built a powerful large language model in five languages using HPE's supercomputers and machine learning development platform.
This large language model is already used today as an advanced AI digital assistant across banks, automotive manufacturers, legal firms to accelerate business outcomes. They turn to HPE to scale their training capabilities in a virtual cloud versus relying on deploying and managing their own supercomputer and software. As our launch customer for HPE GreenLake for Large Language Models, Aleph Alpha will extend their customer reach globally and accelerate their growth. As this business scales, the combination of top-line growth and higher margin offerings within our portfolio will help us achieve double-digit operating margins. AI is a fast-growing market that will be fundamental to transforming businesses and accelerating economic growth. We believe we are well positioned to profitably capture the market opportunity and win in AI to drive higher growth and margin expansion.
With our trusted expertise, differentiated IP, and long-term sustained market leadership, we can capture significant value from the AI market. This is an exciting opportunity for our company, and we look forward to fueling the value creation from AI for our customers and in turn, for our shareholders. And now let me turn it back over to Antonio.
All right. Well, thank you, Justin. Each of these areas have, we just presented, provide our customers and our shareholders with tremendous opportunity, and the engine that powers our ability to capture that opportunity is compute, which produces cash flow to invest in our business and deliver direct capital return to our shareholders. As you all know, there is a cyclical nature to the compute business. Over the last several years, digital transformation drove increased investment to modernize infrastructure. Now, customers are focused on digesting those investments.... We will be very intentional about how we execute and compute during this cycle to maintain our scale and industry-leading profitability. We are focusing on capturing every unit while maintaining balance in our operating margin performance.
We are capturing opportunities in a steady compute market from cloud repatriation, from edge, and IoT workloads, demand from the telco and 5G sector, and service providers, and from the emerging needs from AI inferencing solutions. Justin talked about that. We expect continued demand next year for servers with new GPUs and other compute accelerator types. Combined with our shift to HPE ProLiant Gen11 servers, which delivers significantly greater performance compared to the previous generation, we anticipate a tailwind in compute average unit price in fiscal year 2024. We'll measure our progress in part by tracking the servers we sell with accelerated processor units or AIs. That is because customers can use a variety of compute accelerators beyond just GPUs to support AI workloads. We have a complete compute portfolio to address the entire AI life cycle across training, tuning, and inferencing.
We also continue to see strong interest in our HPE Services, which is additive to both revenue and margin. Our world-class HPE Services team help us to deliver great customer experience, and customers can gain maximum benefit from HPE GreenLake, from our software, designing and building their infrastructure, and running their hybrid IT estates with one exceptional end-to-end customer experience. Just as compute serve as an engine to power accelerated growth, HPE Financial Services continue to be a critical competitive advantage for HPE. This business creates smarter IT life cycle solutions for our customers and partners through offerings that combine insights, financial expertise, and deep-rooted focus on sustainable IT. It is also strategically important for us as we increase our other service business to HPE GreenLake.
Customers can transition to HPE GreenLake using our asset life cycle management services to assess, to access efficient technology and cloud consumption models, creating value for our business. Going forward, we see even higher demand from our customers as they put more emphasis on finding ways to accelerate their sustainability goals through our services and the circular economy solutions. For our investors, HPE Financial Services offerings and its best-in-class return on equity provide a great source of profit that expand our earnings. In addition to driving impressive organic innovation across our portfolio, we continue to be opportunistic in making the right acquisitions. So far this year, we have acquired five businesses to accelerate our strategy and enhance our capabilities. When we assess opportunities to make organic and inorganic investments, we focus on the potential to drive higher level of recurring revenue and profitability.
We maintain particular interest in investment that help us innovate and grow profitably at the edge in Hybrid Cloud and AI. We will continue to be opportunistic in making beneficial and accretive acquisitions while following our disciplined, returns-based framework and ensuring integration success. Jeremy will address more specific on the long-term sustainable value creation and free cash flow generation objectives we pursue on behalf of our shareholders. When I reflect on HPE's strong business momentum, I'm particularly proud that a great deal of the strength come from solutions that help customers advance their important environmental, social, and governance objectives. In fact, in fiscal year 2022 alone, we generated about $1.3 billion in net revenue that we can directly attribute to sustainability engagements with customers. In addition, our portfolio enhances opportunity for our customers to achieve their environmental goals.
I am very proud that HPE is one of the only 2 global IT companies to have a net zero target of 2040 or sooner, approved by the Science Based Targets initiative. Last year, we reduced emissions directly within our control by more than one-fifth from the 2020 baseline and reduced our overall carbon intensity by 2% year-over-year. We also believe a diverse and engaged workforce fuels innovation and performance. Since 2017, our employee engagement score has risen 20 percentage points to 83%. While we have work to do, and so the entire IT industry, I am proud of our progress in diversifying our workforce. Our US workforce is 32% ethnically diverse, and more than 26% of executive positions worldwide are held by women at HPE. We also have a very diverse board of directors.
Half of our independent directors are female, and we are innovating with a strong ethical compass. We win the right way. In the last year, we have put in place even more comprehensive governance around AI to enable us to seize the opportunity this technology possesses in the right way. We have the right strategy aligned with the key market mega trends. We have the right team, with strong focus on delivering business outcomes for our customers. And while the world is navigating uncertainty, we are confident HPE will continue to accelerate value for our shareholders. To speak about shareholder value creation, Jeremy Cox will take the stage after a short break. As you know, Jeremy is serving as our interim Chief Financial Officer, while we continue our internal and external search for a permanent CFO.
We will be back on stage as soon as the market close. Thank you for your time and attention.
Every Ryder Cup gives us an opportunity to reset the technology. What we're creating here with HPE is a smart city, a data-driven environment. But it's equally about how we're taking that data and turning it into critical intelligence. What we're really trying to achieve here is to modernize that fan experience from edge to cloud. We're working very closely with the HPE GreenLake platform, and that's driving real efficiency.
One of the biggest benefits of using the HPE GreenLake platform is it's giving the Ryder Cup team here the ability to manage their service, storage, and networking from a single pane of glass.
But it's also gonna have key impact for our fans and for our partners as well, making sure that that data is fully accessible on the course where it needs to be.
The pressure is always on for us to deliver. There's no second chances. We have to be ready when the first ball is hit. We'll be ready.
Good morning, Lewis.
Oh, coffee! Car is ready in five.
Got it. I need my headset.
I need my music.
Get just what you need, right when you need it. Hewlett Packard Enterprise turns data into intelligence, opening new opportunities, edge to cloud.
We are in the business of speed, and to build the fastest F1 team, we need the best technology from factory to trackside to driver.
Yeah, sometimes we only have a few days between races. Now, we can put the exact car I'm driving in a virtual wind tunnel, apply weather predictions, even make a part from scratch. It's all about harnessing that data to build the best possible version of the car. On race day, we need hot data trackside. With HPE, we have the computing speed to make real-time decisions, to fine-tune the cars, determine the optimal time to pit, even communicate race strategies to the driver.
When you act on your data at the speed of your business, opportunities open up.
Our natural world is truly remarkable, from its breathtaking beauty to its darker side. Massive, ever-evolving, and destructive forces of nature. But there also exists one true force for nature, one made by humankind to help protect us from the world around us. What can it teach us about our ecosystem or of our biology? More than we ever dared to dream possible.
... Kajaani is a smallish city in North Finland, with a past that's very much around forestry and paper production. It has reestablished itself as a hub for IT and now also supercomputing. My name is Pekka Manninen. I'm the director of LUMI, the most powerful supercomputer in Europe... and by far the greenest in the world. It uses carbon neutral hydroelectric power, which captures waste heat and uses it for heating the surrounding city. We decided to partner with AMD and HPE because the technology they are offering is exactly what we need.
AMD and HPE have a long history together. We build the silicon, and they're building world-class systems. When we started the LUMI project, what we really set out to do was to give a stable, reliable source of performance to our researchers and scientists to go and do the incredible things they are doing.
With HPE Cray, in partnership with AMD and their EPYC processors and their Instinct accelerators, new problems that hadn't been solved before are being solved at a higher fidelity and greater accuracy than they ever could before. Problems that are really the grand challenges of this time.
Because of extremes in climate change, we will see a lot more crisis situations in the future. Destination Earth is about creating a digital replica of the Earth system to play through scenarios of how to best adapt and influence climate change at unprecedented scales and accuracy. Billions of observations streaming in from satellites in real time that tell us about the state of the atmosphere. To bring all this together, we need high-performance computing. This is where these very powerful machines come in and allow us to open the door for entirely new science to make our society and our Earth more sustainable in the future.
These supercomputers we're building are allowing that more detailed and better simulation on climate and many other things.
Things that touch a lot of people's lives.
In the future, we will not have enough pathologists. People are getting older and getting more diseases. AI systems can revolutionize the whole field of pathology. The aim of ComPatAI is to train AI to detect cancer based on quantitative data rather than just visual observation.
With LUMI, we have cut down the time needed for training the models from weeks to hours. The amount of data is huge.
We show millions and millions of samples to this neural network, training it to understand histology. The more data you feed, the better the model becomes.
This technology would mean faster, more accurate diagnosis, and of course, the benefit for the patients is very big. During the upcoming years, we will see more of these breakthroughs.
LUMI is a great collaboration where a lot of different entities came together to advance humanity.
This is where the important decisions are being made that our children will see.
We're here to deliver the technology to enable these great outcomes.
We're impacting the lives of billions of people. The possibilities are endless.
This week in AI.
Now that we're getting Hybrid Cloud right, do we know who's getting AI right?
Well, models are huge, and the data keeps growing. ... Still raining on the parade, huh, Fancy?
The models say, sunshine, Gary.
You need the right volume of data and the software to train it.
Magenta is on trend.
Magenta is so five minutes ago.
You need massive compute power because real-time insights are crucial.
Traffic update. Looks like every car is on the same shortcut.
This week in AI.
AI, éxito o fracaso?
What happened here?
Another one bites the dust.
AI.
One side is long, and one side is short.
Step back!
Maybe it's a drone. It looks like it's an AI failure.
AI.
AI.
AI. AI.
But with HPE GreenLake, we have access to supercomputing to power AI at the scale we need.
Helping to give us better insights.
Helping leave our competition in the dust.
Let's go!
Please welcome Senior Vice President and Interim Chief Financial Officer, Jeremy Cox.
All right. Well, good afternoon, and welcome back. I hope you guys enjoyed your break. You know, as an 18-year veteran of HPE, it's a privilege to be serving as the interim CFO, and I'm very happy to be with all of you guys today. Now that our leaders have presented our vision, strategy, and differentiation, let me translate that into our financial thesis and our commitments. We'll demonstrate how pivoting our mix to higher growth, higher margin, and recurring revenue is accelerating value creation for our shareholders. My main goal today is to help you understand how our strategy will deliver additional value through sustainable, profitable growth and increased capital returns. We're pleased with our progress throughout FY 2023. We've managed through some ongoing macroeconomic challenges, yet our business is performing well.
We're reiterating our Q4 guidance for revenue of $7.2 billion-$7.5 billion, and our FY 2023 guidance for revenue growth of 4%-6% in constant currency. We expect currency to be approximately a 300 basis point headwind for the year.... We're also reiterating our Q4 and FY 2023 non-GAAP EPS guidance. We continue to expect Q4 non-GAAP diluted net EPS of $0.48-$0.52, and FY 2023 non-GAAP diluted net EPS of $2.11-$2.15. We also reiterate our free cash flow guidance of $1.9 billion-$2.1 billion.
Given our business performance and the scale of opportunity in front of us, we have chosen to make some targeted investments in Q4 to accelerate our pivot in Intelligent Edge, Hybrid Cloud, and HPC and AI, which we've offset with higher OI&E than we originally expected. We've also incurred some additional GAAP expense within our cost optimization plan, including certain real estate charges, which we don't expect to be meaningful in future periods. Consequently, our full year GAAP and non-GAAP operating profit will be slightly lower than our guidance, with non-GAAP operating profit growth to be approximately 4% versus our prior guidance of 6%-7%. FY 2023 GAAP EPS is now expected to be $1.42-$1.46.
We are accelerating these investments and yet remaining within our long-term financial framework, where we continue to expect non-GAAP operating profit to grow faster than revenue over our outlook period. Our strategy at HPE is definitely accelerating value creation for our shareholders. We are improving our growth and margin profile by shifting our mix towards higher growth and higher margin segments. We're adding as-a-service software and service revenue across all of our segments. Our capital allocation strategy balances investments to drive further long-term revenue, and growth in our free cash flow, and consistently returns capital to our shareholders. Here's how you can expect to see this stronger mix come through in our financials over the next three years.
We expect our growth businesses, our Intelligent Edge, Hybrid Cloud, and HPC and AI segments, will contribute increasingly more revenue to HPE, and the combination should exceed 50% of our total segment revenue by FY 2026. Already for FY 2023, they should be more than 45% of our total segment revenue. That's an increase from approximately 40% in FY 2022, and our operating profit trajectory is even more dramatic. We expect the mix of total segment operating profit from these segments to reach over 60% in FY 2026, and that continues our recent trend. We expect to see our growth businesses, approximately 45% of total segment operating profit in FY 2023, driven largely by improvements in the Intelligent Edge, compared to under 30% in FY 2022. HPE GreenLake is a key element of our portfolio mix shift with healthy customer demand.
Continued ARR growth means our as-a-service products will represent more than a significant percentage of overall revenue and at richer growth margins. We are rolling forward our long-term ARR target CAGR by one year through FY 2026, and thus reiterating our commitment to 35%-45% CAGR growth. This puts ARR on track to more than double and reach approximately 10% of our revenue by FY 2026. In our new segment structure, we'll continue to disclose ARR in the same way. ARR growth is set to drive further gross margin expansion for the company. Gross margins for our as-a-service offerings are already meaningfully higher than the same products and services sold through a transactional CapEx model.
As we build momentum in standard offerings and SaaS, our mix of high-margin software and services within our ARR has risen to 61% in FY 2022, to 68% in the most recent quarter, and should reach the mid- to upper-70% range in FY 2026. So let me step back and discuss how that rising mix of software and service revenue appears in a representative HPE GreenLake deal. This compute and storage HPE customer opted to purchase HPE GreenLake in lieu of a transactional CapEx deal. The deal had a normal profile of revenue growth, margins, and term length. Customers in a transactional model typically overprovision hardware capacity and purchase that lower amount of revenue from software and services attached. We have normalized the software and services revenue in this deal example at $100.
This HPE GreenLake customer was happy with the flexibility to purchase less capacity to start, though with more than twice as much software and services revenue. These additional services delivered an improved experience and operational savings for the customer. Over time, rising usage led the customer to increase its spending with HPE, including three times as much software and services revenue at a similar hardware level. The GreenLake model meant we received more high-margin revenue over the deal life cycle. And it created a happy customer, who's likely to stick with us as our annual customer churn is only 3%.... Along with higher software and services contribution, the rising usage also pulls through additional hardware. And we get another bite at the apple when the customer decides to replace their hardware. Our HPFS lifecycle management business enables us to repurpose the fleet of depreciated assets into other customers.
This lowers their costs and thus increases our margins. So the bottom line, HPE GreenLake drives a better value for our shareholders through higher revenue, margins, and customer retention for HPE. Our end markets and product sets are evolving. Our new segment structure, effective November first, will improve our portfolio alignment and accelerate our go-to-market motion. We'll report our Q1 2024 results under the new structure, and prior to Q1 earnings, you can be assured we plan to file restated historical segment financials for the new reporting structure. While there are several product lines and shifts between existing segments for various optimization purposes, most of the changes are not overly material to the financials of the existing segments. The more material change is establishing the Hybrid Cloud segment, which Fidelma spoke to earlier.
It combines the existing storage segment, and the as-a-service compute business, and the software business previously reported in our corporate investment and other segment. So now let me take you into a deep dive of our current segments. You've heard we expect mega trends in Edge, Hybrid Cloud, and AI to drive our TAM to rise higher than $340 billion by 2026. In the coming slides, I'll show the financial goals of each of our segments. We'll also provide a table of TAMs by our existing segments in the appendix for your reference. The Edge and networking TAM is growing at a 10% CAGR to $94 billion by 2026. The portion addressable by our Intelligent Edge segment is growing in the mid-single digits, and our business is performing well above this.
We delivered 53% year-over-year constant currency revenue growth in Q3 2023, and we expanded our operating margin by more than 1,300 basis points to 29.7%. Last year at SAM, we talked about the mid-20% range by FY 2025. Well, well, we're ahead of schedule on operating margin. However, we're keeping this mid-20 operating margin target for this segment for the outlook period. Improving supply is now allowing us to make progress against our order book and deliver on deals we won in prior periods. We're confident that the combination of shared gains and the expanded TAM that Phil mentioned, will allow us to continue to outgrow our end markets. We expect revenue growth to be in the low double digits from FY 2022 to FY 2026, a figure that's front-end loaded, given our growth this year.
I also wanna highlight our as-a-service business in the Intelligent Edge, such as Aruba Central and Security, because it's now both sizable and growing rapidly. As growth here continues, we will see more of a trade-off between current period revenue and future period revenue. Such strong growth in FY 2023 sets a high bar for comparison in FY 2024. Even so, we expect FY 2024 revenue to be slightly up relative to FY 2023. We're forecasting a continued tailwind to revenue in the first part of FY 2024 from our significant order book. Our HPE GreenLake and storage portfolios are well positioned to capture the TAM growth we expect to see in the Hybrid Cloud, where we expect to see the TAM to grow at a 7% CAGR to $164 billion in 2026.
We estimate the TAM for storage alone will grow at a 3% CAGR to reach $66 billion in 2026. We believe we'll grow at market rates from FY 2022 to FY 2026. We haven't included significant AI contribution in our growth outlook, but we do believe AI will pull through storage demand. Our revenue driver and storage will be ongoing strength of our HPE Alletra product line, which has grown in triple digits pace over the last 5 quarters to now numbers that are no longer small. Also, as-a-service is now the fastest growing portion of the business, and like Intelligent Edge, trades some current period revenue for future period revenue.
Our SaaS business growth is one reason we expect operating margins to improve to the mid-teens range in the outlook period for this segment, which will be evident in the operating margin structure of the new Hybrid Cloud segment over the same period. We have a significant opportunity in HPC with AI, as we're seeing great momentum through the overall HPC and AI segment. We expect the AI TAM to grow at a 24% CAGR to $146 billion by 2026. We estimate the HPC and supercomputing TAM, which this segment addresses directly, will grow at a 22% CAGR to $78 billion in 2026. So we talked about AI activity after Q1. We talked about wins after Q2 and order book after Q3, when HPC and AI order book rose to more than $3 billion.
We recognized a modest amount of AI revenue in Q3, and we expect a similar amount in Q4. However, given the specific interest in AI demand, we will begin to report a new metric, total HPE orders booked that include accelerated processing units, or AIs. AI includes GPU-based orders within both the HPC and AI, and compute segments. However, HPC and AI represents the large majority. Year-to-date, total HPE AI orders are over $3 billion. This metric captures all AI orders across a range of suppliers. We expect the strength of our order book and the pipeline opportunities we see in front of us to drive near and long-term growth. We forecast solid double-digit revenue growth in the HPC and AI segment between FY 2022 and FY 2026. For modeling purposes, you can assume for now this CAGR will be front-end loaded.
We said at last year's SAM that we expect this business to carry a double-digit operating margin over time. While we've been investing and will continue to invest in AI, as we can see in our demand growth, our long-term expectation for this margin structure has not changed. We continue to expect approximately 10% operating margin by FY 2026. Operating margin improvement is a focus for this business, as Justin noted. When we expect to benefit from an increasing investment in software and continued learnings from the supply chain challenges to improve supercomputing execution, gross margins, and capital intensity. While it's early days, our commercial AI wins on average have come at margins above typical, above levels typically commanded for relevant product lines. And you know, this business benefits from scale, which AI demand is absolutely supercharging.
Our compute business is steady, value contributor to HPE, and it continues to be a critical component of our overall company scale. We estimate the compute TAM at $63 billion in 2026, excluding the Tier 1 market, where we have chosen to limit our presence. We expect the market to grow at 4% through 2026, but to be mostly flat when excluding China. We're preparing for growth in the inference market and are expecting to capture that upside through this segment. Earlier this year, we introduced several AI-optimized and GPU-intensive HPE ProLiant Gen11 servers. However, we're not yet embedding significantly above normal GPU growth for inferencing in our revenue outlook.
If, or more likely when, such growth does materialize, our product portfolio between HPC and AI and Compute is well positioned to capture the entire life cycle of opportunities from tuning or training, to tuning, to inferencing. We remain focused on capturing profitable unit share with stable revenues while managing our long-term and market-leading operating margin target of 11%-13%. This includes an assumption that component costs will rise in FY 2024. Let me also reiterate that our Compute segment comprises a portion of our total... of HPE's total server revenue. We also recognize server revenue in our HPC and AI segment. We expect HPE's combined server revenue to deliver mid-single-digit growth in FY 2024 to approximately $16 billion. HPFS remains a steady and strategic driver of healthy returns on equity across economic cycles, and it facilitates our as-a-service pivot in multiple ways.
This segment is particularly critical as a support for our HPE GreenLake business because it creates investment capacity for our customers. It supports our as-a-service pivot with its best-in-class lifecycle asset management business. The fleet management concept is becoming way more powerful as customers' mindsets are shifting from paying for specific equipment to paying for capacity. We expect the business to sustain a mid-teens return on equity with a mid-single-digit revenue growth CAGR and an upper single-digit operating margin through FY 2026. This revenue growth is slightly above our long-term trend, given the rising interest rate environment. Okay, let's translate that into a multi-year financial outlook. Our three-year revenue CAGR is expected to be 2%-4% in constant currency.
This outlook incorporates the current macroeconomic picture and FX risk, and considers a revenue base, revenue base year that benefited from an order book consumption, which means we expect demand growth to be above revenue growth in FY 2024. We aren't incorporating significant further growth in AI demand beyond what we have current line of sight to. We expect non-GAAP diluted net EPS CAGR of 3%-5%. Our structural non-GAAP effective tax rate is 15% from 14% in FY 2023, largely due to removing H3C earnings from our pre-tax non-GAAP earnings.... Excluding H3C and the tax rate change, we forecast our non-GAAP diluted net EPS CAGR to be 7%-9%. And our philosophy remains that we will grow our operating profits faster than revenue.
We expect our mix shift, including as-a-service revenue and the rising profitability on that revenue, to be strong contributors to our operating profit growth over this time. Our mix shift to higher growth revenue has lifted our visibility into free cash flows. As a result, we're increasing the amount of free cash flow we intend to return to shareholders to approximately 65%-75% between FY 2024 and FY 2026, above our recent historical target return of between 50%-60%. This includes a combination of targeted dividend increases and larger share repurchases. But we'll be thoughtful around the target on both directions, given we'll continue to follow our returns-based framework. And for cash flow, we expect our FY 2024 free cash flow to be between $1.9 billion-$2.1 billion. That guidance is flat year-over-year with our current FY 2023 guidance.
However, it's important to note, FY 2024 free cash flow will rise 10% year-over-year when adjusting for approximately $200 million in cash flow we receive from H3C and FY 2023. This year-over-year improvements on an adjusted basis will be driven by lower cost, lower cash transformation costs. We also expect improving working capital, although the surge in capital-intensive AI slightly constrains us for now. Importantly, we expect free cash flow growth to significantly exceed net income growth over the next three years. Our conversion of Non-GAAP net earnings to free cash flow is steadily improving, as you can see on our slide. We are on track for approximately 70% in FY 2023, a sustained improvement from FY 2021 or FY 2022, FY 2021 and FY 2020. We expect to reach approximately 90% by FY 2026. Let's now turn to our FY 2024 outlook.
We expect revenue growth of 2%-4% in constant currency, and FX to be a modest 50-100 basis point headwind. We are expecting AI demand to be a key driver to above-trend HPC and AI segment growth, and Intelligent Edge revenue to be slightly up year-over-year. Non-GAAP gross margins is forecasted to remain in the 35% range. Continued contributions from the Intelligent Edge, storage, and HPE GreenLake are to balance the pressure on compute gross margins after an above-trend FY 2023. We are prioritizing investments in higher growth opportunities we've talked about today, balanced with cost discipline and lower growth businesses. For modeling purposes, you can expect OI&E to be approximately -$300 million. We expect our structural non-GAAP tax rate to be approximately 15%.
So GAAP diluted net EPS is expected to be between $1.83 and $2.03. And non-GAAP diluted net EPS is to be between $1.82 and $2.02. Finally, we plan to generate FY 2024 free cash flow of between $1.9 billion and $2.1 billion. This does place us below our initial plan of more than $6.5 billion between FY 2022 and FY 2024. The biggest changes to our assumptions are, one, we no longer are assuming cash flow from H3C, and two, robust AI growth suggests supply chain and working capital won't normalize as planned. We expect free cash flow to follow our typical seasonal pattern of negative in the first half of the year and significantly positive in the second half of the year.
We intend to increase our dividend by 8% in FY 2024, given our rising confidence in free cash flow. We expect our FY 2024 corporate performance to be weighted to the back half of the year. While we are expecting the first half to benefit from normalizing order book in the Intelligent Edge business, our HPC and AI revenue growth will likely be governed by GPU availability well into FY 2024. Our non-GAAP diluted net EPS is expected to grow by 4% from a normalized level. The bridge between the midpoints of our FY 2023 and FY 2024 non-GAAP diluted net EPS includes a few non-operational headwinds I'll walk you through. The H3C put exercise creates a $0.17 non-GAAP EPS headwind. We are not planning to accrue dividends from H3C in FY 2024, nor will we report H3C earnings in our non-GAAP income in FY 2024.
However, we're required to continue to include, in GAAP income, our proportionate amount of H3C earnings generated in FY 2024, while we still own the interest. We continue to expect to receive H3C cash proceeds in the first half of the calendar year 2024. Other OI&E is likely to be a $0.10 headwind. We benefited in FY 2023 from FX hedging and other gains, and interest rates are higher. As I mentioned, our structural non-GAAP tax rate is forecasted to be 15% in FY 2024, and this equates to a $0.02 headwind in EPS, and again, is largely attributable to removing H3C earnings from our pre-tax non-GAAP income. We also have some operational headwinds in the coming year, including normalizing margin structure in Compute segment, following order book consumption during FY 2023, and planned investments in our growth businesses.
The combination should reach approximately $0.34 at the midpoint of our guidance range. However, we intend to more than offset these headwinds with operational improvements and revenue growth, specifically in our higher growth, higher margin businesses and ongoing efficiency gains. Operating margin target gains should be particularly evident in our HPC and AI and storage segments. As you know, we follow a disciplined returns-based framework to ensure we maximize shareholder value. Our rigorous investment evaluation process balances investments for growth with capital returns to shareholders. Our top priorities are investing to capture high return opportunities while remaining committed to dividends, opportunistic and material share repurchases, and retaining our investment-grade credit rating. Our acquisitions, going back to Aruba Networking, and more recently to OpsRamp, Athonet, and Axis Security, have followed our disciplined ROI-based framework.
We are confident in our ability to grow organic revenue, EPS, and free cash flow, with the premise of our M&A to be to accelerate the growth with acquisitions that are value accretive. We're expecting to receive a significant amount of cash when our H3C deal closes, and we'll update you with our plans for that cash at that time. However, we do not intend to hold excess cash over the long term, though some reserve is prudent in uncertain economic times or during a technology inflection, such as the AI. So let me close with a recap of our financial messages. We see healthy growth in the IT industry, and we have products across our portfolio that leave us very well positioned to capture that growth. As a result, our expectation is for sustainable 2%-4% revenue growth in constant currency.
We're pivoting our portfolio to higher growth, higher margin revenue, led by our efforts in the Edge, HPE GreenLake, HPE Alletra Storage, and AI. We intend to expand operating profit faster than revenue over the outlook period, which translates to a 3%-5% non-GAAP net diluted EPS growth, or 7%-9% normalized for our H3C divestiture and tax rate change. Finally, we expect free cash flow growth in excess of net income growth. This gives us line of sight to increase our target percentage of free cash flow return to our shareholders to 65%-75%. It also allows us to target an 8% increase in our dividend in FY 2024.
I've thoroughly enjoyed working more closely with all of our business unit leaders in the last couple of months as we finalize our FY 2024 plan. I have great confidence that our thoughtful strategy will provide very attractive investment returns. We look forward to your questions and feedback in a bit, but first, let me turn it back over to Antonio to close us out.
All right. Thank you, Jeremy. Thank you, Jeremy. We believe we have presented a clear case for the compelling value HPE will deliver to our shareholders. With a total addressable market opportunity growing at about 1.4 times, I am confident in our ability to capture that opportunity in a way that delivers for investors for several key reasons. First, while some microeconomic challenges remain, I'm optimistic about demand. HPE is more relevant than ever because we have innovated in the categories that are most important to our customers now and in the future. Second, we have made very sound investment choices over the last 5 years to pivot our portfolio to a diverse set of businesses with greater profitability potential, and they are paying off. As we unlock greater growth from these markets, our investors are poised to share in higher returns.
Finally, we have an experienced and passionate leadership team who execute with vision, a sense of urgency, and a commitment to culture. Our company recognizes the need, now more than ever, to continue investing in our workforce to attract and retain the necessary talent to execute on the growth strategy we communicated today. As we left the same stage last year, I shared with you that HPE does not wait for the next big thing to happen. We accelerate what comes next for our customers, our company, and our shareholders. Our foresight and spirit of innovation has always served us well and positioned us to realize the full benefit of what is in the horizon for us and for our industry. We expect HPE to capitalize on the opportunity we discussed here today at the edge, Hybrid Cloud, and the exciting opportunity that AI presents.
We believe strongly in the long-term profitability potential of HPE and what we offer as an investment opportunity. I would like now to take your questions. We will give the crew a little bit of time here to reset the stage, and then we will invite the presenters back on stage so we can take your questions. And also, the audience on the webcast can submit questions through the bottom of the screen. I guess, we're gonna sit here. The middle here, perfect. Sit here. You sit here, and then Jeremy, Justin there.
Yeah.
Okay, great.
All right. All right, thank you all for your presentations. I wanted to let you all know that in addition to the executives that we have on the stage today, we have a few other members of the executive committee that may help us out. Gerri Gold is over here. She runs our HPE FS business. Neil MacDonald, in Compute, has an important wedding anniversary today. So representing Compute is Krista Satterthwaite, who is our Compute SVP and General Manager. We have some mics floating around, so if you would like to... Yep. Okay, great. You guys know the drill.
As expected.
Super. Let's go with, let's go with Wamsi first, right here in the front row.
Yeah. Okay, and then-
Thank you for the presentation. Wamsi Mohan, Bank of America. I guess I'd want to kick off around the assumption of Compute, because you kind of broke the Compute TAM into non-AI and, you know, the AI opportunity for inferencing. You're seeing a lot of your competitors talk about AI servers or, within sort of the standard category for training as well. So kind of wondering why you're not categorizing any of, those kind of servers within, for training, specifically within Compute. And do you see the entire AI opportunity necessarily in the HPC AI TAM only?
So first of all, we don't see the AI opportunity just on HPC and AI. That's why Jeremy made a comment in his remarks. You have to think about the server category, okay? The server category in our company has two distinct segments: the Compute segment, which is the traditional enterprise, and what I call the cloud-centric kind of infrastructure, whether it's tier one, two, or three. We see that also in other verticals like telecommunications, and obviously more at the edge as well. And then you have what I call the HPC and supercomputing side, which includes the traditional HPC business, the supercomputing, now what Justin talked about, AI infrastructure. Much of the AI infrastructure for training and tuning is captured in his side, but there are aspects of training that may be captured in Compute as well.
But most of the inference, in the end, is in the traditional compute, because you may deploy a small cluster, or you may deploy one server with 8 GPUs. And that's why Jeremy was very good in saying, when you think about the entire server category as a, as a whole, we expect that business to grow mid-single digits, approximately $60 billion, I think you quoted it. And, but the important message you need to take away is that for AI, we cover the entire life cycle, and in particular on the training and tuning, because we have large set of capabilities across supercomputing as one aspect, AI infrastructure, which is gonna grow 7 times, you know, supercomputing, and then the AI software platform, which is a huge differentiation, to be able to deploy this system at scale.
Fact of the matter is that we're gonna give you an order for what we call AIs, which includes GPUs and other accelerators, which includes both Compute and HPC and AI.
Okay.
Yeah.
I'm gonna go with Aaron in the front here. Thank you.
Yeah. Thanks for taking the question. So first of all, I just want a clarification. The $3 billion number that you talked about-
Mm-hmm.
For the AI, I think it was new orders. There was numbers in the past you've thrown around, pipeline at $3, orders at $1.6. I think it started at $800 million. I just want to clarify, what's the comp of that number previously? And then on the AI narrative, I, I'm gonna get a little bit technical here, but the Slingshot is a core competency of the company, so I'm curious of where we stand. As we think about these large AI models, you know, infrastructure deployments, there's a lot of inertia around InfiniBand versus Ethernet, and there's this migration from 400-800 gig. I'm curious of where you stand on Slingshot from a technology perspective-
Yeah.
to keep up
Yeah.
with that, that different architecture.
Yeah. So maybe, Jeremy-
Sure.
You want to talk about the orders, and then you can talk about the Slingshot.
Sure. So we have talked historically about wins, AI wins, $1.6 billion. We've also talked about how that contributed to our order book in HPC and AI, that was around $3 billion at the end of Q3. I think the better compare of what we're saying now is these are, this is representative of what we have orders booked in the year. So that obviously will feed into the order book, but that is, there may be other factors that feed into that order book. So these are GPU-based orders that are booked within the year, and that's over $3 billion, and we see path for that to continue to grow. I would say also on top of that, that the $1.6 billion was more capturing what we were seeing in the near-term demand around the H100 NVIDIA piece.
I would say that a meaningful majority of the over $3 billion order number is made up of that category.
So think about it this way: it is $3 billion orders, which are in our books, and we will take time to build, ship all that time. And that's why we were very clear in our comment there, that we want to give you the order... Right? Because sometimes between the win and the order, there is a time delay. You win it, but then you have to book it, and takes time and all that. So $3 billion is the order for the balance of the year so far.
Year to date.
Year-to-date.
So, okay, Slingshot.
Slingshot. So I, I think first of all, we, we get caught up a little bit in mixing interconnects, network interconnects that are used for east-west traffic and within a cluster, versus 400 and 800 gig at a data center switching side. So if you think about interconnects in a cluster, there's two players in the market today: InfiniBand from NVIDIA and Slingshot. Those are the two things providing high performance interconnect for east-west traffic and clusters. If you think a little bit about what Wamsi was asking earlier on a box, there's not a lot of people buying one box of GPUs today, right? They're buying them in 512 GPU building blocks, and so that whole thing has to be connected by a network, right?
And that network is InfiniBand, the default for the NVIDIA reference architecture, the H100 reference architecture. However, we have demand outside of that in our supercomputing business for Slingshot, for interconnecting both H100s, but also Grace Hopper, as well as MI300. And so we obviously are running that on, you know, on the Frontier supercomputer and Argonne supercomputer, which is powered by Intel GPUs. So we've got broad demand for that interconnect. That's different than the data center network that sits on top. And if you think about why that's different, it's that at the data center level, I've got to handle a whole bunch of different kinds of traffic. And so the speed matters because I got to manage the inefficiency.
The interconnect is all about tuning and optimization for latency and loss, 'cause I've got to run one parallel workload. I'm trying to run one large training run. In the supercomputing world, I'm running one large simulation for weather prediction or something like that. And that's why they're very different. To give you a technical metric, we did a little bit of analysis on Slingshot, maybe a year or two ago. We found that Slingshot at 200 gig was 92% effective when running RoCE, RDMA over Converged Ethernet, for those of you that enjoy the technical side of this, versus 50% for Ethernet. So that means that, you know, packets data efficiency was far higher because we're running one very specific workload. We're trying to address GPUs or CPUs across distributed memory.
That's why east-west traffic and the interconnect for the clusters is different than the top network. Does that... I know I went deep on the tech, so I want to make sure I -
The other thing I will say, part of the investments we are making, right? We talk about investing, you know, in the supercomputing IP. Obviously, that supercomputing IP, a lot is the what he talked about, which is the networking piece, which is Slingshot, and the software to manage that latency. That will follow, you know, 400 and 800, right?
Yeah.
But that's because of the internet. But the architecture is so different, right? Because you're connecting every node to every node, and if you have a Frontier, is 9,408 nodes, times 9,408 nodes. Because you have to look as a one entity, and that's where our advantage is, and that's why we can use it elsewhere.
Okay, why don't we go to Toni, right next to you?
Yeah.
He'll come right down the line. Can I just remind you to introduce yourself on the mic, please.
Toni Sacconaghi from Bernstein. So I just wanna follow up on that last question.
Yeah.
So you're saying you have $3 billion in orders year to date for accelerated compute. Sounds like the vast majority of that is in HPC and AI.
Correct.
You've said your HPC and AI backlog is $3 billion. It's been $2 billion or more every quarter for the last 12 quarters. So is the $3 billion in orders, like, how, how do we think about that? And I, I don't think you addressed the specific question, what was that exact comparable metric last year through this date? Because I'm still struggling to jive the huge order number with the backlog that you're attributing to A-
Yeah.
AI.
Yeah.
Then secondly, just related to that, can you comment on the wait times for delivery on your AI servers? How concentrated your customer base, and what kinds of customers have generated the orders? Thank you.
Yeah. Maybe we start with the first one and then back to you, because the second is interesting and ties to the customer-
Yeah
stories we share.
Right. When we thought about this metric, Toni, we wanted to make sure we were being thoughtful, not about just the near-term demand, like in the H100, which again, was more of the focus of our prior discussion of the $1.6 billion wins. We wanted to capture the full life cycle, and you can see that again, within the lower inferencing area, within the compute area, all the way up through the supercomputing space, and then obviously within the high-performance space, where you're seeing the H100 activity happen more. And so, you know, the prior period compared to that, I would say you can think about the compute portion of that has not significantly increased on a year-over-year basis, less than 10% of that total of that total order base.
Supercomputing at this point, a little bit more than that, but not, again, not a significantly high percentage. So the vast majority right now is falling in that HPC, that HPC area that's attaching with the H100s. And this is the metric that we're going to continue to provide to you guys. I appreciate that some of the order book dynamics, that had a multi-year component to it. As you know, some of these supercomputing deals-
Yeah
take multi years to resolve. So we thought this order metric would be a better way for you guys to digest, the activity that's currently happening.
... in the prior periods, Toni, there, there's also the fact that you may have like a Frontier or Aurora or some of those that-
Yeah.
just wipes out quite significant backlog in one go, and then you build again. But this is what he's referring to, that 2023, the mix of the bookings has changed quite dramatically.
Yeah.
Okay, so you wanna talk about,
Yeah, the customer.
Yeah, the customers. What type of customers demands we're seeing?
Yeah. So I touched on some of the segments. You know, and I would say the way to think about it is, if you think about some of those cloud service providers, they tend to be buying in these 512 GPU chunks that I touched on earlier. So you know, in terms of concentration, I would think of them buying and buying, you know, large clusters in that, in that stage for, for training. Now, they may buy multiples, but that's a, that's a good building block to think about. Then, the foundation model builders that are, that are buying directly, they... I would say again, they're, that's probably the largest size they're building.
So I think, you know, they may be buying something that's, you know, half or a full pod of GPUs, I think 256 or 512. And then we have customers that are buying, you know, subsets of that. And so I think about those more like traditional enterprise customers, existing HPC and AI customers that are adding capabilities. What we haven't seen yet is material demand from national AI initiatives and national AI centers, and I think we'll start to see some of that come in. That may look like more of a mix of our traditional supercomputing systems. So they may buy Cray EX systems, they may buy Cray XDs, or they may buy a combination.
An example of that would be, you know, if you look at the win we announced with KAUST, they bought a small Grace Hopper cluster, which is basically for AI training. The reason that's so important and so differentiated is that there's a lot of customers that are looking at GenAI, not just for, you know, broad LLM applications, and I touched on this with Argonne. There are customers looking at training scientific models, doing expert models with, you know, deep technical knowledge. Those customers realize that they may need something that's a little different than just the standard building block today, and that's why they're coming to us. And I think you'll, you know, hopefully you'll see us announce some more, you know, some more news on that shortly. But there's, you know, we see these—each of these markets is very different today.
You have to think about different models, right? A lot of the momentum we have heard in the market is about this large language model. That's one way to think about it, but the other ones are traditional AI models like computer vision, ModSim , molecular docking, climate. Climate is pretty significant as well, and diffusion, text, picture to text.
Yes.
We have all this, and one of the things we have, which is a key differentiator, we have been supporting those models for a long time. Now they become more,
Simon.
Pervasive. Right.
Thank you. Simon Leopold with Raymond James. Two, if I might. First one, I wanna see if we could get a better understanding of the impact on the financial statements of rising and high interest rates, how that factors into the business. And the second one is, probably for Phil, what are your market assumptions when you talk about slight growth in the Edge business? What are you assuming is going on in the world around HPE, and what's your basically assumption on your share position? Thanks.
So you wanna talk about the interest to start?
Yeah. Yeah. I would think about that, Simon, in two parts. As you likely know, the large majority of our company debt is attributable to our FS business, and there, that business is effectively able to capture the high interest rates and pass that on through to customers as it's attracting on the margin there. And so that business is actually driving higher revenue performance in the HPE FS business that I mentioned earlier, as a result of that direction. On the core business, we mentioned a $0.10 headwind to EPS. That is definitely a component of that headwind, are the higher interest rates as we're rolling over bonds, and we're having to incur higher interest rates on those bonds as we roll.
And then on the market share assumptions that we are making, so we're assuming that the market is negative next year, so the market drops next year, and we offset that through continued market share gains, which we have been gaining market share the last couple of years, and then also the expansion of the TAM into the new areas that I talked about in my presentation, namely security, data center networks, private 5G, and NaaS. So that's the assumption that we are making.
Asiya, right, right up front. Thanks, Kate.
Thanks. Asiya from Citi Research. Just if I can, on the storage, which I understand now is in the Hybrid Cloud segment, I think, you know, the underlying assumptions would be as you shift more to Alletra and to your own software IP, the margins on that would be higher than the mid-teens that you've been reporting. I think in the revised model, that's kind of what you're reiterating, that the hybrid segment would still be in the mid-teens. So maybe you can walk us through, are there some underlying investments still that are going on in storage? And then you talked about market share gains as well, specifically as it relates to storage. Maybe you can kind of lay out, you know, where you see a strong opportunity to gain share there and, you know, who would be the traditional players?
Is it the legacy players that you're going after? Is there some of the more newer software offerings there?
Can we start?
Go ahead. Yep.
Okay. So, in terms of storage investment, we've invested a lot over the last number of years in the new architecture. We continue to invest on the, on the protocol build-out on the architecture. So, we're really poised in the block space, to take share, and, you know, our investment, in the go-to-market side of it is now complementing the, engineering investment we've put in. So there is continued investment in, new protocols and also the investment in the go-to-market side. In terms of taking share, you know, we've got about a 10% market share, within storage. We've got a lot of runway to go, and, we're very confident about, the, cloud-enabled, and the platform-based, attractiveness of the architecture with our customers.
We see us not just against traditional players, but also when customers are thinking about, "Should I put my data in the public cloud, or should I put my data on-prem?" And so, especially as AI is, you know, becoming more top of mind on storage buyers as well, we are seeing more and more conversations about how do I expand my storage footprint on-prem.
Part of that investment, as Phil might talk about protocol, is the file piece of this, because it ties directly to the AI piece.
Can we go over to Meta, over here on the side in the front?
Thanks, Meta Marshall, Morgan Stanley. You know, numerous of you guys talked about kind of customers exploring different paths on AI. I guess I just wanted to get a sense of, you know, where do you think customers are on that journey of even trying to kind of discover what their approach is gonna be? And, you know, when do you see the timeline of some of those decisions? And then maybe a second question, you guys talked about a double-digit operating margin on HPC. Just what are kind of the critical components to getting from where you are today to that double digit? Thanks.
I'm gonna take those two.
Yep, okay. Go ahead.
Meta, I'll answer the double-digit one. It's pretty... First, operating margin is really simple. It's scale in the business, which gives us leverage, and then it's just the mix shift to more of our IP-rich offerings. And that includes the penetration of our software stack. Obviously, continued penetration of Slingshot as a part of that, the growth and demand we see. And then, the complete solution we talked about, where we're integrating, you know, our services and our supercomputing IP and that software stack in the HPE GreenLake for LLMs. So that's the driver of margins. On your first question, the journey, look, I think where we are right now is there are a number of customers spending, you know, making massive investments around foundation models.
A lot of it is around, LLMs for, you know, for broad commercial applications, and that has not, that has not broadly deployed into massive investments in the enterprise yet. It's largely, I think it's largely the early days. We think the enterprise build-out is gonna be massive. We also think that, the initial models, which are largely general-purpose language models across many languages around the world, are gonna be, are gonna be the, the beginning of a much broader and more robust build-out of models and training. And that will be things like technical use cases. So you think about scientific models, you think about, obviously, financial and trading is one that, that, you know, we'll, we'll probably continue to build, just given the, the dynamics and the economics. But that will go on and on and on.
I think that's what we think... That's why we think we're at the beginning of something very big. It's also why we're being very conservative in terms of how we're thinking about the business and our forecasting, because we're really looking at the demand that we see today. But we, we think this is gonna be the beginning of a long investment. And, you, enterprises, most enterprises I talk to, and Krista can comment as well, are very productivity focused today. As I touched on, I think productivity is the start. You know, over time, I think we're gonna see people look at new revenue models, and we'll see that part of the business emerge as well.
Yep. That's why I want to make the comment that a lot of the new customers are about developing these foundational models. Most of the enterprises are gonna leverage these foundation models and tune those models with their own data. But they are right now more focused on productivity, like we are in our own company, right? How we leverage that, whether it's in the coding base with our R&D team, whether it's in our operations, in our services team, with bots and, and, and the like. But I think the biggest focus for them is how they maintain control of the intellectual property and the data in a way, in a way that allows them to deploy these models with confidence and the accuracy, because that's the other important piece-
Mm-hmm.
We talked about.
Yeah.
Has to be accurate, right? So...
Okay. Why don't we go with David over here? Two more over here live streaming, and then we'll come back up to the front, you folks.
Great. Thank you. David Vogt from UBS. Two, if I may, one for Justin, one for Phil. As inferencing becomes a bigger part of the AI sort of story going forward, how is Slingshot sort of competitively positioned against the legacy, sort of Ethernet vendors as we move in that direction? And then for Phil, you talked about four different categories, sort of driving growth in Edge and Aruba. You talked about networking, convergence with security, also data center switching. Those are two pretty crowded markets. So can you maybe touch on what you bring to the table besides sort of the operating system, which you touched on briefly, in terms of how you're positioning yourself vis-a-vis the competitors in the marketplace already kind of pursuing a similar strategy? Thanks.
Yep. Yeah, so I think if you look at it, again, as you think about Slingshot, think of it as, you know, as east-west traffic within a cluster. So as inferencing workloads scale and there's more demand, there probably are parts of inferencing that will have, you know, large capacity, right? I mean, think about content distribution networks as a parallel in the internet. If you think about, you know, the other part of inferencing, which is why we're focused on the entire life cycle, getting down to the edge, you know, these are, you know, this could be down into your mobile device or into, obviously, a car, as we see with autonomous driving. So there's gonna be a lot of places that inferencing gets used.
I mean, we do think there's gonna be a demand for, you know, a cloud, you know, a cloud service or a large cluster of inferencing solutions, as part of the broader market build-out. Phil, I'll let you answer the other one.
Yeah, so I think the question was, crowded markets in both campus switching and security, and then how do we believe that we'll be successful in those parts of the market? So, on the campus switching element, I mean, our strategy there is to follow the network. So obviously, the access point connects into the campus network, and we find customers that use the platform, Aruba Central. You know, once they're comfortable with that, it's a logical extension to go into campus switching. And we find it... you know, there's a big player, obviously, in the campus switching space that we come up against quite regularly, and customers like our customer centricity and the fact that we have one platform linking all of the products together, whereas some of our competitors have different platforms for different products.
So that's on the campus switching side of it. On the security side of it, I think that's a market that's gonna go through a number of changes, isn't it? Because on the security side, you've got the three elements, being ZTNA, SWG, and CASB. And most enterprise customers right now don't buy those three from one player. They're often buying them from three different players. And then on the network side, you know, you've got network firewall and the SD-WAN side of it. So I agree that there's a lot of players in the market, but there aren't many players with the full range of five capabilities. So I actually think we've got a really, really good opportunity there, albeit the market is gonna have to go through a number of steps to consolidate.
Sidney, please. Yeah.
Great. Thanks. Sidney Ho with Deutsche Bank. Just wanna follow up with that question. You talk about intelligent edge growing about low double digits over a three-year period. Which of these opportunities that you highlighted are more near term, call it one to two years, versus longer term? Do you have all the product and services already in your portfolio? And in the near term, are you expecting the operating margin of that business to go back to that mid-20% already in fiscal 2024? Thanks.
I mean, just on the, on the capability side of it, in the products, I mean, on the data center side of it, we have been investing in the products in that space for the last 12 months, and we've got a pretty good defined roadmap for the next couple of years. And we do $ low hundreds of millions already in the data center space. And it's kind of difficult, actually, for us to— customers buy switches. We don't always know whether they're going into a data center or a campus location. But for the most part, we know that some of the products are already going into data centers. We've got a good starting point there, and the sales teams are pretty accustomed to selling in data centers.
So data center is a very real opportunity for us today, and we've got good growth there, and we've been adding more resources, so I feel pretty good about that. private 5G is an earlier space. I mean, you know, the company that we acquired is pretty low in revenue terms, but there is a lot of interest, because, you know, what private 5G is allowing companies to do is, companies with very outdoor locations, like, I don't know, oil refineries and those sorts of places, they haven't really been able to benefit from network technology before because they've struggled with covering large outdoor spaces, and private 5G enables that. So we've seen, since we made the acquisition, a lot of interest from many, many different vertical sectors, albeit the starting point from a revenue perspective is relatively low.
And the same is true on security. You know, the company that we bought, we're very, very pleased with the technology and the people. And when customers go through proof of concepts of the technology, every proof of concept that we have had with the customer since we bought the company, they've always proceeded to contract. So we're very happy with the technology, but again, it's relatively low revenues at this point, so we've got a way to scale. And then with regards to the margins and OP, I mean, on the security side, they would be better than the margins that we currently see. On the private 5G, would they be there or thereabouts, maybe slightly below. Data center would be typical to the margins that we, again, we currently see.
And then on the OpEx side of it, you know, this market, we should benefit, you know, on the top line, potentially, things get more aggressive next year because the market contracts, and therefore, potentially have to do a bit more discounting discuss. I mean, we feel like we've got a pretty strong proposition. But then when you go to the OpEx line, you know, during the supply chain crisis, we had to fly everything around the world by air, which is more expensive. We used to have to pay expedite fees to suppliers to get equipment moving. And as the supply chain has now opened, we now have the opportunity to move equipment around by ship and not pay expedite fees.
You know, there's puts and takes in the overall P&L, but I think the, the statements that Jeremy-
Yeah
... made earlier are very doable.
I would just add to that. I would say, yeah, it's a dynamic of similar to compute, we had high price backlog that was, as it was being worked down with lower logistic costs, we saw, you know, above trend results. And so that's gonna start to normalize a bit, and we're continuing to make investment in this business. And so while I expect us to be in FY 2024, maybe on the higher end of that mid-20s range, over the three-year outlook, we would expect to be in that range.
Why don't we take one from the webcast? It's been staring at us here.
Yeah, there's one here, actually.
Yeah.
You wanna read it?
I'll paraphrase it, really.
Yeah.
Actually, I can read it. It's well, well, well written. So it's $200 million in H3C cash flow you mentioned for fiscal 2023. How does that compare to prior years, and is that a full year number?
Sure. So maybe a bit of the confusion here is that we have to differentiate between the earnings that are incorporated for EPS purposes and then the cash that we ultimately include in our free cash flow, which are really from dividends, which are paid in arrears. And so that may be the disconnect for the person that asked this. And so the adjustment we've made to the $200 million on a year-over-year basis is reflecting the cash received in the dividends. The adjustment we made to EPS was reflective to our amount of earnings that benefited from the addition of those H3C earnings into our non-GAAP EPS.
... But it has been fully?
Yeah, over time it would match. We do have some dynamics with the, with the disposition, where, you know, under, under rules in China, there are certain dividends, or dividends aren't allowed to be paid after a certain period of time, and so, that will, that will be an impact in 2024 as well.
Is that number essentially flattish? Has it been growing? How can they look back a year or two?
I think that's the relative adjustment for 2023.
Okay. Why don't we go with Samik, up here in the front? Thanks, Kate.
Hi, Samik from JPMorgan. Had one for Justin and one for Jeremy. Justin, when you look at the different parts of the AI stack that you're offering, software, networking, and then the compute solutions, like, how are you thinking about the go-to-market in terms of, do you want to sort of remain the full solution provider, or is a better sort of adoption and quicker adoption to be chased if you disaggregate that pieces and sort of use your differentiation in each? And for Jeremy, you talked about H3C proceeds, you'll update the investors on that.
When you think about excess cash, can you just define that for us, how you think about what sort of cash on the balance sheet that you need, and what are your priorities once you get that cash, in excess of what you want to, what are the priorities?
I think the way to think about this simply is, in supercomputing, we're a full provider, and that's necessary because it's almost impossible to deliver a supercomputer without delivering the whole system. It's really, really challenging. Probably there's some people who tried it, but it's hard. In the AI infrastructure space, you know, we're selling. We obviously aspire to be a full provider in that space, but we're also selling parts of our stack where it makes sense. So in some cases we're obviously selling H100 clusters. I mentioned, you know, Recursion Pharmaceuticals. They're a software customer only today. They bought our machine learning development environment, and you've got a customer like Aleph Alpha that initially bought infrastructure from us, then bought software, and is now our launch customer for HPE GreenLake for Large Language Models.
So on the cash flow side from H3C, look, I think with, you know, we're constantly balancing between, you know, our desire to return more capital to our shareholders, but also making sure we have the right amount of capital to make the strategic investments to grow our business. So we assess that balance through our ROI-based framework that we've talked about. I think as the H3C proceeds come in, we have now directionally suggested to you guys that, hey, we have a better confidence in our long-term outlook on free cash flow. That's driving our expectation of increasing the amount of share repurchase and dividends as a total percentage of cash flow back to our shareholders to the 65%-75% range.
I think those things will be factored in when we get to the cash position, when H3C is delivered in kind of the mid-calendar 2024 period.
So again, I think it's important when you step back and look at what we guided for capital return in 2024 and the direction for the next three years, is all the operational side, right? So we have better line of sight. The pivot is happening. Obviously, we have seen the gross margin increase in our portfolio, and that gave us the confidence to increase from, you know, the 50%-60% to now 65%-75%, and in that, increase the dividend by 8% at the time when the dividend comes to be paid. And ultimately, when we receive the proceeds sometime in 2024, we're gonna use the same capital framework.
But, you know, the confidence is not the $3 billion-$3.5 billion after taxes, it will be less, obviously, but is about our ability to execute the plan that we just gave. And then when the cash comes in, we will assess the best return, but we don't expect to hold excess cash in our balance sheet.
Why don't we go with Mike, and then we'll go with Louis ?
Yeah.
Thank you. Mike Ng from Goldman Sachs. I just have two. The first one is just on Intelligent Edge. I was just wondering if you could talk a little bit about the curve over this midterm outlook. Obviously, very front-end weighted. Is that just a function of some of the backlog reduction that we'll see over the next two years? I'm just trying to reconcile that with, some of the, you know, growth initiatives that you have that we mentioned before, and you know, why that doesn't, why doesn't that help to, you know, accelerate that growth in, in fiscal 2025 and fiscal 2026? And then just a quick follow-up on H3C.
Yeah. Yeah, so I mean, yeah, you're exactly right. I mean, obviously this year we have benefited by a significant backlog retirement, and we have some of that next year, albeit at a lower rate. But we do benefit from this portfolio expansion. And as I say, next year, we're lapping, you know, very significant growth, obviously, for this year. So probably next year is gonna be more subdued, as Jeremy highlighted. But then, you know, as you go out year two, year three, after that, we would expect to build way more momentum in the new areas that I highlighted today, so data centers, security, and private 5G. And I think some of those could be quite meaningful in the sort of medium term.
Yeah, but I think it's important to not miss a point. In the last two years, we added $2 billion to this business. That doesn't go away, okay? Let's be clear. We added $2 billion. We are building a new platform, if you will, and we're gonna grow from that new baseline, the numbers that Jeremy explained. In that, there is obviously continued market share in the campus and branch, which includes wireless and switching and wide area network. Security will be a driver of that, obviously, with SD-WAN, but then there is the new areas that takes longer to mature because you need to either fully integrate in the platform, customer use cases will come to life, go-to-market is ramping. So that's why we gave you the guidance that we gave.
But, but let's not walk back on the fact that we just added $2 billion, and that stays as the new baseline.
Great. That's very clear. Then my next question is just around some of the comments around compute ASPs benefiting from accelerated servers in fiscal 2024.
Yeah.
You know, I guess how meaningful of a benefit is that? Are you seeing, you know, demand for accelerated servers right now? And then, if you could provide some additional comments around the GPU availability point, that'd be very helpful. Thank you.
Yeah, I mean, so the AUP, so average unit price, when you have some sort of accelerator on particular GPU, is significantly higher than the traditional CPU. The question is the margin side of the house, right? So that's why our constant approach with the operations team, Mark Lines and Krista here, representing compute on pricing discipline, is very, very critical. So as the shift continue to happen, and as I said, you know, we see evidence of that, but it's still very small relative to the size of the CPU base in traditional compute, call it single server. You know, the structural side of that will grow. And also, remember, with Gen11, we also have another structural change, which obviously we can add more memory, more storage, and more everything associated with that.
And also, there is a component of subscription as a part of GreenLake to life cycle that server. That's why we said on an aggregate, the average unit price in 2024, plus the rise of cost on the traditional components, which is just a functional demand and supply, after the capacity kind of shortages, is gonna happen. So the question is how that transition happen is what we built in our guidance, is what we have line of sight, but in the context of the full life cycle.
I would maybe just add one point to that. On a year-over-year basis, though, we'll still see some pressure in AUP. While we're seeing some of these tailwinds that Antonio described, you know, we're coming off a significantly inflated AUP position in FY 2023. So part of the compute story will be some of that compression in AUP, even though we'll see tailwinds start carrying that out of FY 2024.
In the second half.
Yeah.
Just on lead time, I mean, I think what I would say is, I think Toni asked us, and I may not have answered it earlier, is we continue to see demand well in excess of supply, right? And we don't see that changing right now. So, so that, yeah, that continues to be the view we have going into 2024. And that's why Jeremy touched on it being governed by availability.
Yeah. Our mix of demand is, can be quite significant, right? So if you have a supercomputing with 40,000 GPUs, that's pretty significant versus a customer has 512.
Yeah.
Right? Yes, Louis .
Thank you. Louis Miscioscia, Daiwa Capital Markets. So, asking a little bit of a different question. You know, ARM recently went public, and, you know, you've got companies like NVIDIA, Apple, AWS using their chips.
Mm-hmm.
Just wondering, is there any opportunity for HP to include any ARM products that could lower your cost? Or alternatively, do you see a possibility, any risk to any area?
To include ARM?
ARM products, yeah.
We already do.
ARM products.
It's Well, we already have a great partnership with a company called Ampere. We already integrated that in the Gen11 platforms. We already have interest from customers. They have unique workloads. They actually were not typical AI workloads. With other type of workloads, maybe the, the core count and the power and all of that was perfectly suited for that workload, but it's still early. And remember that part of the, the, the challenge, I will say, but I think as an opportunity, is the entire ecosystem of the application compilers and software have to come all together. I think we are at a different mature level than we were maybe 3-4 years ago.
Yeah.
But we already have in the supercomputer.
Yeah.
I mean, you can talk about that.
Yeah, I would add two things. I mean, we were. When NVIDIA announced Grace Hopper, they announced the win that we had with them with CSCS in Switzerland or Grace Hopper, and so there's a solution with Cray EX on the supercomputing side that is ARM-based. The other thing is, in May, we announced a system at University of Bristol in the U.K. It is a Grace Superchip system, which is ARM-based. So, you know, as Antonio said, it's very much about the tool chain. But we're, you know, across whether it's the compute portfolio or in HPC and AI, we see demand. We're supporting the demand strategically and working closely with partners like Ampere and NVIDIA.
But we have that capability already in our products to support.
I think we have one sort of back center. Yes, please.
Yeah. Hi, Sid Nag from Gartner Research. I had a quick question in terms of your longer-term strategy. It seems like you're at a crossroads as a corporation. So on one hand, you're competing with the incumbents, and you're competing with the hyperscalers, right? So in other words, do you look at the hyperscaler space as a co-opt and compete strategy and continue to compete with the incumbents? And where do you see the growth coming?
I mean, we know, we talk about the world being hybrid, and in that experience, you have to include the public aspect of that experience. And hence why GreenLake, as a platform, as our cloud platform, which is truly designed with hybrid principle, we actually support the three largest cloud, 'cause customers have data workloads in those, in those public cloud. And our goal is to give them the best cloud native experience, whether on the public cloud or on-prem.... And at Discover, we made several announcements with Amazon Web Services. We had a partnership with Azure for different things, and so forth. So we have to continue to include that in our thinking, and in other aspects, we are gonna compete, right?
When it comes down to deliver specific outcomes by delivering a on-prem or a colo or edge infrastructure that have the same principles, but ultimately give to them an edge to cloud hybrid experience that matters. I think the growth, I mean, private cloud, as I said, right, is growing at 10%. That's a great opportunity for us, honestly. That's why we enter the market with a series of offerings that I think are better suited for what the customers are looking for, with choice, because we support all the runtimes available in the market. So it's just a balance, right? But ultimately, it's what the customer is looking for and how we address those needs in the best possible way.
If ultimately selling services through a hybrid solution, through data protection services that may they back up in a public cloud, that's fine, you know? It's, it can be either. But ultimately, we are driving to a cloud-native experience that delivers the best economics with the best experience that's truly hybrid.
Did I see a couple over here? No, I guess not.
I have one here.
Yeah. Why don't we go to the web first, and the question is, you noted $0.18 in investments in 2024. Are these products, these products or go-to-market investment, which businesses?
Yeah, it's a combination of both. So obviously, in Phil's business, right, we continue to make investments in, in the product side. Phil talked about the continued progression of our data center products, right? Whether Aruba or CX, which gives the same experience if you're in the data center or the campus and branch, but obviously have these different features and functionality, but it's all managed through the same platform. Also, it has integration that is doing with 5G and Axis Security, but also is investing, as well in the go-to-market, continue to expand that coverage and specialization. All our investment in go-to-market are about specialization. You know, much deeper specialization security in Hybrid Cloud, obviously growing the coverage. Phil might talk about the storage in HPC and AI. Solution architecting is important. Pre-sales is important.
Then on the R&D side, obviously there is a sizable investment that we're driving in AI, is silicon. When you are in the silicon, you ask the question, right, it takes multi-year investment in silicon. It is not like a quarter and you're done, right? So it takes a continuous investment. And the software, because the software we are developing is the machine learning development environment, is what we call the PaaS layer to make sure the data automation, the data pipeline for AI takes place. 'Cause one of the biggest pain points customer have is that how they bring together the data to be trained on that model, right? And so, so those are the things we are investing, as well.
So there is a combination, but it's ultimately is edge, Hybrid Cloud, and AI, and obviously, Fidelma has other investment in GreenLake as well. But also we've made some investment in compute, silicon, Root of Trust , and other software elements that are very important, which we can monetize through our subscription-based models.
All right, we've got a follow-up from Wamsi, and we have time for one or two more. We'll get you, Matt, for sure.
Thanks for taking the follow-up. I was wondering if you could share any color on the $3 billion AI orders in terms of just average and median order size, so we can understand if that's, you know, how much of that is largely HPC driven. I mean, are these kind of order sizes of $100 million each, or are we talking about, you know, $250,000? What's kind of, kind of the distribution look like?
I don't know. I mean, it's obviously large size deals.
More sizable than-
More sizable.
Yeah. I think I covered the... I think Toni asked the question about the customer profile. I think you could sort of think about the mix in that construct, the types of customers, and obviously the, you know, the large, you know, the more SuperPOD -sized, 512 GPU-sized clusters people are buying, you know, the more disproportionate they shape that mix.
Yeah, but think about high up number.
Yeah.
Yeah.
We've got, yeah.
Oh, here, that side. We have a man on that side.
Thank you. Matt Sheerin from Stifel. A question on your guidance for fiscal 2024. You talked about a $0.42 tailwind in terms of operational improvements, that it's gonna offset some of those headwinds-
Yeah
... you talked about. Could you go into that? That's a pretty big number. Are we talking about restructuring? Will there be charges, headcount reductions, things like that?
Yeah. Well, we have... It's almost like the same number we had last year, if I remember correctly.
Uh.
It was $0.45, I think last year, or $0.40+ in operational improvement. But it's actually the combination of the mix shift, right, which is driving higher gross margin dollars as the mix of the business continue to shift. Also, there is ongoing improvements in our operations effectiveness, which obviously we drive as a part of our ongoing improvements. We do that all now within our domain. We build those investments in our typical, you know, plans, not the typical way that was done many years ago. And so also we continue to be way more efficient from the back-end operations with our COO. But, you know, we have done a great job in driving efficiency in our real estate, in our supply chain with digitization and all the things you expect.
But I think the vast majority of this is the continued mix shift in our portfolio; it drives higher gross profit. And also the go-to-market. The go-to-market's pretty interesting evolution, right? Because since I became CEO, I have taken deliberate action with our Chief Sales Officer to simplify the structure, eliminate roles, drive simplicity in our sales compensation, more focus, productivity at the sales rep level. Those are all bundled together in that $0.42.
Thank you.
Okay, sure. We've got Aaron up front.
Yeah, Aaron Rakers again, Wells Fargo. Thanks for the follow-up. I just wanna—I keep going back to this AI stuff, so I apologize. I think in the past, in the commentary, you've talked about a lot of that being under the GreenLake umbrella. And so as we all try and think about this path and the pace of monetization of this large order book-
Mm-hmm.
How do we think about that in the context of GreenLake, right? Wouldn't it come out and actually rev rec over a period of time under the contractual GreenLake model?
Yeah.
Do you take the capital intensity related to that as well? Do you take the upfront CapEx?
So, you know, through the GreenLake model, you know, we have the benefit of our HPFS business that ultimately incurs that leverage cost at the end of the day. And that provides the capacity for us to grow that portion of the business. From a revenue recognition perspective, it depends, at the end of the day, on the term length in the contract with the customer. You know, the shorter term periods, you would expect to see while we recognize revenue up front in the segment, we would be eliminating that and recognizing it over time for the company. Whereas you have longer-term arrangements with the customer, those would require upfront recognition, both for the segment and for the company.
So, but overall, and to your broader question of how much of that GreenLake business was within that $3 billion-plus base, there's a fair amount. At this point, I would say the majority is CapEx, transactional CapEx, but there's a good chunk of that GreenLake business embedded in that number.
Yeah. We are extremely aware, Aaron, of the capital intensity, so we drive in different commercial terms in the way they pay us, so that we can ensure that there is ongoing, sometimes prepayments of that.
Yeah.
Okay? But, you know, ultimately, as I think about the type of customers, right? So I think you're gonna have a mix of different type of transactions in Greenlake. You're gonna have those who say, "Antonio, Justin, I need a virtual private cloud with this size for me for the next three years, because I'm gonna continue to build a model, training, retraining," and it's not a multitenant solution. It is in a public instance that leverages our data center services and power, efficiency and expertise that we have to run the system of scale, but they are dedicated to them, right? So that's, I think, is gonna be-
Yeah
... quite the vast majority of it. And then there will be those enterprises that may want to tune their data for a period of time, tune their model with a period of time. That may be for a month or two, and then they said, "Okay, I don't need that anymore." But it's still, you know, from the data intellectual property perspective, they want to make sure that's contained, right? They don't want to put data in a multitenant type of environment. That's very obvious.
I mean, maybe just two things. I mean, given the demand that we've been talking about, most customers, when they're coming to us, even to buy a cluster on HPE GreenLake, it's more about getting the operating expertise and getting it up and running faster. They want control of the cluster, right? They're not interested in using it for a period of time. They're actually willing to commit because of the limited supply. The second thing, and Antonio touched on this, is the multitenancy, and this is different than the public cloud. And I touched on this a little bit in my comments, but public cloud, you're running multiple applications on one server, right? In most cases, or you maybe scale it out to a few servers. It's kinda like what we do on our iPhones or laptops.
In a multitenant environment, in HPC and in AI, it's exactly the same. You're allocating a large portion, if not the entire cluster, to one job, and the application's running in parallel over the whole thing. And that actually is where we have a lot of experience because we built system management software that does this for supercomputing customers, and it's exactly the... It's very parallel to what we do in that space. It's almost exactly the same. And so, as Antonio touched on, as the market matures and we see demand for that tuning workload, we think that's gonna be another opportunity. We haven't, you know, we have the core technology for it, but we haven't seen that demand emerge yet.
Yep.
Okay. We have time for one more? All right, we've got one right up front. Samik, please.
Hi, Samik Chatterjee. Justin, going back to AI again, liquid cooling, a lot of discussion nowadays on that. How do you see that as a differentiated to where the competitors are on that front? Also seems like that ecosystem to provide liquid cooling is not really completely developed yet. When you think about partnerships or even M&A, are you looking to enable - what are you doing to enable that ecosystem so that you can sort of build on that differentiation?
Yeah. There's multiple elements to the ecosystem. This is a place, you know, Cray has a history of working in liquid cooling, you know, for decades, so we've actually got lots of experience in implementing it. When you think about the liquid cooling ecosystem, you need to think about it across a set of dimensions. The first is the... Let's start with the data center. Data center has to be liquid cooling ready, and so that's a place where there's a bunch of capital build-out that's happening, but it's an important part of the investment, right? Because you actually need to plumb the data center for liquid cooling. Second, for us, in terms of IP, is how you design the systems and the architecture and how you manufacture them. We have a factory in Chippewa Falls.
We've set up a factory in the Czech Republic to manufacture this. To my knowledge, they're the two largest scale factories in the world for building liquid cooling systems. Then you have to design the systems for liquid cooling, which is quite different than just simply air cooling, because you're dealing with heat dissipation across every port component. So you think about fans, and you're like, "Okay, well, I can just replace the fans." You can't just replace the fans, 'cause you've now got to make sure your liquid cooling, the CPU, the accelerator, the memory, the drives. If you're gonna put a full liquid cooling system in and have it be fully, you know, room neutral, you actually have to liquid cool the fabric and the switches, right, as well.
So that's what we do in a supercomputer, and that's the kind of capability in the IP we're bringing more broadly. The third part of it is then you need people that actually know how to deploy these things and manage the services and implement them. And that may sound simple, but it turns out, if you pull out the wrong plug when you're servicing one of these things, you can have a pretty significant water leak, right? And the other part of servicing is you actually have to maintain the water. The water running through these things, if you think about it, is getting heated and cooled. It's getting subjected to different thermodynamics. Sometimes it's getting subjected to light, and so we have to manage that, so we have to chemically treat the water.
We have expertise in all these areas, from manufacturing to system design to the service and deployment. And then, of course, there's the ecosystem behind that, of components and parts. But I think when you look at it, the three things that we do today, we're well ahead of the market, and we understand how to make this repeatable and scalable. And I see it even in how fast we can deploy a supercomputer today versus that's liquid cooled, versus what we were able to do just a couple of years ago when I started.
The bottom line, we are unique in that space. This is what make us attractive, because customers who need to deploy at the scale, they need the services. Our HPE's services business is a significant point of differentiation. You can go buy commodity and then have someone else build it for you. We deploy the entire solution for the customer from beginning to the end, and that IP is very unique for us, and that's what it requires in an AI-native architecture. I think that's something that we're gonna leverage as we go forward. So I think we are-
Yeah, why don't we hold it there?
Yeah. So, so we appreciate all your time and investment. So we are happy to be back at the New York Stock Exchange. You know, we want to thank the New York Stock Exchange for the hospitality. I think we, we have walked you through our strategy in very detail. We have given you, a lot of information. I know it takes time, but, you know, one of the feedback you have given to us, is the fact that you want to understand how we compete in the market. I think we have laid the case very clear in each of the business segments. What is our differentiation? So I hope you, you got greater insight on the value HPE has captured through our strategic pivot, which obviously you see in the numbers, through higher growth and higher margin areas.
You know, as the market expands in these areas where we have made investment or we're gonna double down the investment, I think it's making us not just relevant, but also will create greater value for our shareholders. I believe HPE is an excellent opportunity to invest. I think our guidance is logical, considering some of the things we have to deal with, HPC and not. But we are very confident in what we're doing. And the reason why I'm confident is because customers are coming to us. That's the bottom line. When you spend the time I spend, as I do, more than 50% with customers, it gives me the confidence that we can go and execute what we show you and accelerate our value for our shareholders. Thank you.
I know there is a little bit of mingling here. Thank you for the audience-
Mingling?
on the webcast for staying with us.
Not too much.
All right. Thank you.