Rackspace Technology, Inc. (RXT) Earnings Call Transcript & Summary

Good day, and thank you for standing by. Welcome to the Rackspace Investor Conference Call. [Operator Instructions] Please be advised today's conference is being recorded. I would now like to turn the conference over to Sagar Hebbar, Head of Investor Relations. Please go ahead.

Good morning. I'm Sagar Hebbar, Head of Investor Relations. Joining me today are Gajen Kandiah, our Chief Executive Officer; and Mark Marino, our Chief Financial Officer. As a reminder, certain comments we make on this call will be forward-looking. These statements involve risks and uncertainties, which could cause actual results to differ materially. A discussion of these risks and uncertainties is included in our SEC filings. Rackspace Technology assumes no obligation to update the information presented on the call, except as required by law. In particular, our discussion today will include forward-looking statements regarding our recently announced definitive agreement with AMD, including, without limitation, the ability to dedicate, maintain and make available an aggregate of 30 megawatts of AMD products contemplated by the GPU as a Service agreement, which may not be achieved in full or at all or may be achieved on a materially different time line. The anticipated benefits and performance of GPU and CPU compute deployments, the expected delivery of enterprise AI cloud, Enterprise Inference Engine, inference as a Service and bare metal AMD instinct capabilities, anticipated end customer demand, the expected commercial and financial benefits of the collaboration to each company and the parties' respective outlooks on the AI industry. While the parties have executed a definitive agreement establishing a commercial framework for the collaboration, individual deployments authorizations are subject to separate execution and certain commercial terms, including pricing and financial parameters remain subject to further agreement between the parties. AMD has no obligation to agree to any particular deployment as being within the scope of the framework. Any third-party financing required to implement planned deployments is subject to availability on terms acceptable to the company. The GPU-as-a-Service agreement is subject to certain financing, operational and legal conditions and provides AMD with the right of first refusal that may affect the company's flexibility in selling capacity to third parties. There can be no assurance that deployments will occur on the anticipated time line that financing will be obtained that AMD will agree to future deployments or that the anticipated benefits of the collaboration will be realized. Deployments are subject to the availability of and lead times for AMD products from third-party original equipment manufacturers. Our discussion will include forward-looking statements relating to the company's workforce realignment plan, including without limitation, the expected number of employees affected the anticipated timing and implementation of the reduction in ports across jurisdictions. The estimated onetime expenses associated with the workforce realignment plan and the anticipated Bruce annualized savings reinvestment plans. Actual expenses, savings and reinvestments may differ materially from these estimates as a result of changes in the scope, timing or implementation of the workforce realignment plan, variations in severance obligations across jurisdictions, the timing of employee access, regulatory or legal requirements applicable in certain jurisdictions, certain or actual litigation and other factors. There can be no assurance that the company will realize the anticipated savings from the workforce realignment plan within the expected time frame or at all. The company undertakes no obligation to update or revise these forward-looking statements, except as required by law. With that, I will hand the call over to Gajen.

Thank you, Sagar. Good morning, everyone. We are announcing 2 items this morning. First, we have signed a definitive agreement with AMD to deploy 30 megawatts of compute phased from late 2026 through 2028. Second, we are bringing the company together to go to market as one Rackspace. One company with our people and our investments pointed at the same strategy we've been building towards. Rackspace is rebuilding itself as the operator for governed enterprise AI designed around how production AI is deployed operated and scaled inside regulated enterprises. The AMD agreement further demonstrates this shift. Today's announcements are intentionally concurrent. Infrastructure without an operating model is capacity an operating model without committed infrastructure is expiration. Together, they established a scalable platform for disciplined growth. This is not a course correction. We have been deliberate about sharpening our strategy and executing with greater focus and accountability. Unifying as one Rackspace is the alignment of our structure to that strategy. The AMD definitive agreement is proof that the market is responding to the choices that we've been making. Focused efforts clear accountability and an integrated company designed to move enterprise AI into production, reliably and at scale. Enterprise AI has advanced beyond the experimental phase. Agentic workflows are now embedded in production systems across banking, health care, energy and government. These are regulated mission-critical environments where governance, data sovereignty and operational continuity by not selling points, they are the price of entry. Customers are no longer asking where they can access the compute. They're asking which operator can govern AI responsibly, securely and at scale inside their organization. The hyperscaler delivers compute, a systems integrator deliver services. Neither is accountable for government AI in production end to end. That is the gap Rackspace is built to fill. We believe Rackspace is uniquely positioned to answer that question through trusted customer relationships, deep operational expertise and a global infrastructure footprint. Increasingly, Customers also want to avoid dependence on any single model of provider. For us, this is not a future capability. We operate a model agnostic stack in production today. Customers run and switch the models they choose through a single orchestration layer, our context-aware inferencing keeps their domain knowledge and section context intact across that switch, and we own the SLA across whichever models they run. If a model becomes unavailable or no longer fits the workload, the customer is not stranded because the orchestration and the context sit about any 1 model. That is the continuity of government-operated delivers and the hyperscaler or an integrated is not. Today's agreement is the latest in a deliberate sequence of partnerships and each 1 is a building block in the same strategy. With Unifor, we deliver enterprise AI applications running in production in our private cloud, on infrastructure we operate and remain accountable for. With Palantir, we entered a strategic partnership in February and are building a Palantir-certified forward deployed engineering capability across foundry and AIP. And now with AMD, we secure the accelerated compute foundation beneath all of it. Our partners bring leading technology and Rackspace integrates it, operates it and remains accountable for it as a single accountable operator. The foundation beneath these partnerships is an enterprise-grade technology stack built for the demands of regulated production environments. VMware serves as the control plane, providing the virtualization, workload portability and network fabric that governed enterprise AI environments require. Blue Brick provides the cyber resilience layer, ensuring that data is protected, recoverable and auditable across hybrid and multi-cloud environments, which is nonnegotiable in health care, financial services and suberin cloud. And our forward deployed engineers are the human layer that binds it all together embedded in the customer environment accountable after go live and the reason our SLAs are a commitment rather than a ton. This is the stack that differentiates Rackspace. Every partner in our ecosystem sits inside a governed operating model that we own end to end, 1 operator accountable for the full stack. This model comes to life through 4 integrated capabilities, enterprise AI cloud, the enterprise influence engine, inference as a service and bare metal. Each is accelerated by the AMD agreement, which I will address directly. The delivery layer behind all 4 is forward-deployed engineering. Engineers who stay embedded in the customer environment and remain accountable after go live to ensure outcomes are achieved. Since we established the public cloud business unit a few years ago, we have made significant progress building from an infrastructure-led operation into a services-led organization with deep capabilities across cloud delivery, platform engineering and managed operations. The capabilities we have built are the foundation we are building on. Our private cloud business has equally demonstrated the value of this model. operating some of the most demanding regulated workloads in health care, financial services and sovereign environments. With discipline, governance and accountability we have built in private cloud, is the operating template for everything we are now scaling across the enterprise AI platform. What has changed is where those capabilities need to be directed? The customers we serve are moving from cloud adoption to AI in production. And that shift requires an operator who can manage the full stack end-to-end, not just the cloud layer. Our public cloud business is aligning to that imperative, concentrating investment on data and AI-led enterprise transformation, AI ops-driven managed services and forward deployed engineering talent that operates across hybrid environments from edge to core to cloud. This includes a reduction in our workforce, and Mark will take you through the details. This is the right decision and direction for Rackspace, and we are managing it with the care and the respect our Rackers have earned. An integrated go-to-market strategy removes the fragmentation that can slow execution and strengthens the accountability our customers expect from a single operator end-to-end. The result is a company that is growing with discipline, investing in what matters exiting what does not and operating with the cost efficiency that long-term performance requires. We are not restructuring for growth alone. We are building a company that earns the right to grow by operating well. Before I turn to the specifics of the agreement, I want to take a moment to recognize the team at AMD. This partnership with more than a commercial arrangement. It reflects a shared belief in where our governed enterprise AI should look like and who should operate it. We are grateful for the confidence AMD has placed in Rackspace, and we look forward to building this together. The definitive agreement establishes AMD as a strategic technology partner at the silicon layer of Rackspace's governed AI stack. The agreement supports phased deployment of 30 megawatts of AMD AI compute capacity across Rackspace data centers with Rackspace functioning as the operator layer through which it is delivered. AMD selected Rackspace for this partnership because it speaks to what differentiates us. We have a global data center footprint with available capacity to support deployment, including the 30 megawatts contemplated under this agreement, which is committed and will be deployed in phases from late 2026 through 2029. We bring more than 2 decades of operating regulated mission-critical workloads in health care and financial services, where we are already strong. We bring deep operational expertise in managed infrastructure at enterprise scale. And we bring a governed operator-led model. We do not simply resell compute we operate it and remain accountable for the outcome. That combination is difficult to assemble and it is what makes Rackspace the right partner to bring AMD Instinct into regulated enterprise production. Initial deployments will be established across key markets with AMD Instinct MI355X and MI350P GPUs and AMD EPYC CPUs available for deployment across our data center footprint. The deployment model is capital efficient, leveraging existing infrastructure ordered upgrades and data center consolidation. We expect the initial deployment to commence in late 2026 and the balance of the contemplated 30 megawatts to be deployed in phases through 2028. We believe the demand environment supports this trajectory. We are engaged in active commercial conversations across health care, financial services, public sector and energy weighted towards our existing enterprise customers where adoption cycles are shortened and cost is already established. Our near-term pipeline is anchored in this installed base and our intent is to match initial deployments to identify customer demand. Both Rackspace and AMD are committing dedicated sales and engineering resources to joint customer engagement. This is a go-to-market partnership, not a supplier arrangement. This agreement accelerates 4 integrated capabilities. Enterprise AI cloud, our fully managed private and hybrid AI environment built on AMD Instinct accelerators with 1 operator accountable across the stack. Enterprise Inference Engine, a context-aware influence front time that retains domain knowledge, session history and enterprise-specific data context across queries with Rackspace owning the SLA. Inference as a Service, dedicated managed AMD Instinct compute as a governed alternative to commodity GPU rental and Bare Metal AMD Instinct for training and inference workloads requiring deterministic dedicated performance. Strategic focus requires specificity. Rackspace has a clear path to win in regulated industries, health care, financial services and sovereign cloud as the government operator of Enterprise AI, and in private cloud and governed infrastructure environment. These are areas defined by our ability to deliver simplicity, accountability for outcomes and speed of execution at production scale. Our credibility is demonstrated through what we already operate. Health care environments, including EPYC at scale, sovereign cloud deployments in the U.S. and U.K., strategic partnerships with Palantir and Uni4 both building towards the government enterprise AI platform and now anchored by the AMD definitive agreement that commits the compute foundation we need all of it. With that, I will turn it over to Mark for additional financial context.

Thank you, Gajen. Let me provide context on the financial dimensions of this agreement. The definitive agreement establishes a base commercial framework governing 30 megawatts of AMD compute deployment commencing late 2026 and scaling through 2028. Deployment authorizations are executed in tranches, providing both parties visibility into the economics of deployment at scale. As we scale, we will provide additional transparency around key operating metrics. We've identified multiple sources of financing who are supportive of this initiative, and we have confidence in our ability to secure adequate financing for initial deployments near term. We currently estimate that our first deployment will be approximately $50 million to $100 million of CapEx. As Gajen outlined, integrating our go-to-market focus is the alignment of our structure to our strategy, and that alignment has a financial dimension. In connection with this transition, we announced a workforce realignment plan that includes a reduction of up to 15% of our global workforce. This realignment is predominantly driven by the company's strategic decision to deemphasize certain legacy service delivery functions, primarily within its public cloud business unit and geographic rationalizations in favor of redeploying resources towards this Enterprise AI build-out. We expect to incur onetime charges of approximately $14 million to $19 million in 2026. Following full implementation, we expect to realize approximately $75 million to $85 million in annualized run rate savings. A significant portion of those savings will be reinvested into our highest growth capabilities including forward deployed engineering, AI solutions delivery and enterprise AI infrastructure build-out. This is a deliberate reallocation of capital from offerings that are not aligned to our strategic priorities, towards the government enterprise AI platform we are building. We view this as a time-limited cost with a clear and measurable return. I'll return the call to Gajen.

Thank you, Mark. Let me close with this. Over 2 decades, Rackspace has earned the trust of the world's most demanding regulated enterprises operating in environments where security, compliance, resilience and accountability are nonnegotiable. That institutional capability is not assembled overnight. It is not replicated by operators whose accountability ends at the infrastructure perimeter. The announcements we are making today reflect the convergence of a defined category, a unified company structure to capture it and committed infrastructure to execute. We have sharpened our strategic focus. We are going to align how we operate to where we win. We have secured the first infrastructure commitment through our agreement with AMD. Our infrastructure, combined with our forward deployed engineers enables Rackspace to be the operator of government enterprise AI from silicon to outcomes. The demand pipeline is active and advancing. None of this comes without difficult decisions. Reducing our workforce affects real people who have contributed to building this company, and we do not take that lightly. We have an obligation to concentrate our people, capital and energy where we have the greatest path to succeed, and we are confident today's announcements position Rackspace to deliver for our customers, our rackers, and our shareholders. Rackspace is the government operator for enterprise AI accountable from silicon to outcomes, operating at production grade, built for regulated industries where it matters most. One operator, full accountability. Thank you for joining us today. Back to you, Sagar.

Thank you, Gajen. Before we open the line, I want to focus our Q&A on today's announcement. We ask that participants limit to one question per caller. Broader financial results and guidance will be addressed at our second quarter earnings call. If you have any follow-up questions after today's call, please reach out directly at ir.rackspace.com. Operator, please go ahead and open the line for Q&A.

[Operator Instructions] Our first question coming from the line of Kevin McVeigh with UBS.

Congratulations on formalizing AMD. Really, really terrific context that you folks are able to offer. I guess just to follow up on that a little bit. Mark, I think you talked in $50 million to $100 million of initial CapEx. Any sense of when that's going to start to come in? And then if you're able to maybe reconcile that to the annualized run rate savings. And I know it's probably relatively abstract. But any way to dimensionalize what that 30 megawatts could mean from a cash flow perspective, EBITDA revenue? Just a lot of really, really good momentum? Just trying to frame it a little bit more in terms of impact on the model.

So Mark, let me go and Mark, you can chime in. Kevin, first and foremost, thanks for the question. Thanks for joining us. And again, I couldn't be more excited about announcing this agreement with AMD and also a massive thank you to AMD for their collaboration as we went through this process. To answer your question, I think the way we've structured this, Kevin, again, going back to sort of who we are and how we operate. This is about how do we run AI in production, right, in enterprises and regulated enterprises. And so the way we have approached it is sort of through 3 different vectors, they're important before markets into his piece. One vector is our customers themselves, sort of the customers we serve today and the customers that AMD has that we collaborate on together to go build the demand side or to capture the demand side is probably more appropriate. The second 1 is the type of workload. And that matters because in production or in inference, Customers are going to run across high-end GPUs as well as CPUs. And so understanding the type of workload and how to deliver that in the most efficient manner becomes really important. And then third 1 is supply chain, right? And so when we think about the opportunity itself, it's less about demand and more about, I think, the type of demand and the and the supply chain that enables us to deliver the compute that is needed across that demand. So that then provides the context, I think, Mark, to kind of answer the question.

Yes. Thanks, Gajen, and thanks, Kevin. Yes. So as you can imagine, I mean, today, we're not going to be providing specific revenue guidance around the full 30 or that singular deployment, right? I think you can as you can imagine, you could look out a public data right now and see sort of well-established industry reference points around both what GPU as a service pricing in bare metal pricing? And just sort of keep in mind what this could mean for Rackspace, right? We're going to be playing in not just Bare Metal GPU service market here, but we're really moving up stack to enterprise AI. So from a margin accretion and cash flow perspective, you'd be looking at a little bit higher throughput there. And then as we previously called out, 30 megawatts is existing capacity, existing power, right? So we'd be getting a nice fixed cost leverage, fixed cost or, if you will, related to those 30 megawatts. And just from a deployment perspective, depending on supply chains and timing, it is our intent to start receiving GPUs in the fourth quarter. I'd say no material impact to the financial statements this year, but certainly hit the ground running for next year.

Our next question coming from the line of David Paige with RBC Capital.

Congrats on getting this deal signed. You mentioned that the pipeline is the demand pipeline is very strong and active, so I was wondering maybe you could flesh that out a little bit more? And then maybe a quick follow-up. I know you said 30 -- the initial 30 million-megawatt footprint. Could you give us a better sense on maybe after 2028, I know it's far off from now, but how you see the business evolving through that.

Thank you, David. With regards to the demand side, if you look at our customer base, it's predominantly health care, financial services, energy and government. And picking health care, as an example, the demand is being driven by -- even within health care, if I said the provider as a specific subsegment. The demand is driven by clinical use cases, which are predominantly inference driven. And then there are, what I would say, R&D requirements, which would be a mix of compute -- sorry, training and inference. And what we are seeing is that as the regulated customers begin to embrace AI and start to put it into production, there's very little capability in the market. But us and even -- I might even go as far as to say that we might be the only 1 that is looking at providing enterprise-grade government AI for these customers to run in a way where it is governed with data sovereignty and residency, which is critically important for these industries, David. So that's where the demand is coming from. So think of it as production demand, primarily driven by inference as well as some training. And then there is also then our partner, AMD, who, again, they have their own set of customers coming to want to use their specific compute. And so that's another vector of demand that's coming in, which is why I feel that when you look at it through the lens of demand, that's less of the challenge. It's really about -- when you look at the supply side in terms of the compute, if you think about the networking, the memory, et cetera, it's just really trying to land the right type of compute environment and then ensuring that we can get it deployed within a reasonable time frame. So that's sort of the balance that we're working our way through. And Mark, I'll let you pick up the 30 megawatts. Actually, I can answer it. On the 30 megawatts, a great question. I think, look, the way we have approached this, David, is to be thoughtful about how we ramp up the compute, right? I think the -- as you can imagine, the market is significantly dynamic. One thing that's happening is that we went from top and Maxine to Token efficiency in a 3-month window. And so I believe that it's our responsibility to deliver the most efficient token for the type of workload that's coming through. And to me, having both the customer work to understanding, having the partners like Palantir and Uni4 on the platform layer as well as then having an orchestration layer that is model agnostic and an inference layer that is context aware really allows us to manage a workload through the process to the most efficient token, if you will, or whether we are describing it. And so I think once we get to sort of consuming this available capacity, if you will, once you become -- once we start to utilize that, we certainly have visibility to incremental compute. Again, keep in mind, we are inference not training. Therefore, the type of compute we need is different in terms of power, capacity, density, cooling, et cetera. So there is a lot more availability and we should be able to ramp up as and when that demand is needed.

And there are no further questions in the queue at this time. Ladies and gentlemen, that does conclude our conference call for today. Thank you for your participation, and you may now disconnect.