Your first question comes from the line of Tim Long with Barclays.

Two questions, if I could. First, on the AI front, Francois, thanks for detailing the wins in the different solutions there. Just curious, I think last quarter, you gave a little benchmark number for revenues. Are you seeing and expecting the revenues for your AI-related direct-related business to grow in line with that customer base? Or should we think that 50% growth is kind of a ballpark number there? And then the second one, maybe for Cooper on gross margin. It sounds like in the near term, the component costs are not as bad as expected, but you still seem to be building in a buffer for next year. So could you maybe talk about maybe near-term gross margin related easing and maybe potential for pressure, I guess, it's still a pretty dynamic environment.

Thank you, Tim. I'll start on your -- on the first part of your question. On AI, I wouldn't necessarily approximate the number -- the growth in the number of customers with the growth of revenue. We did give an indicator of that last quarter that we have passed the $50 million revenue mark. That particular number, we don't intend to update every quarter. We may come back to that at the end of the year. But I can tell you that the momentum in direct AI use cases is very strong. We've seen AI traffic generally increase with our customers. But specifically, AI data delivery continues to see strong momentum and a gain in customers. We also had a very strong quarter in AI security. The number of AI security customers actually grew 100% just in the quarter. driven by a number of customers wanting best-in-class security in front of their AI model and their AI applications. So we're very excited about the momentum that we're seeing there. In addition, Tim, I would just remind you as well that our AI opportunity isn't just these direct AI use cases. There is an indirect opportunity that manifests itself in the amount of traffic that we're seeing with our customers on our ADC systems. And where you see that is the amount of expansion we're getting on ADC systems, the growth that we have seen this year in hardware. This quarter, I think hardware growth was 32% year-on-year. But if you look at even the first 3 quarters of the year, the hardware growth is above 30%. And some of that expansion is because our customers are now scaling inference for AI in their enterprise environment, and we are in front of that traffic. So 2 categories of opportunities, and we're seeing strong momentum with both today.

Yes. And Tim, thanks on the GM question. So yes, we saw the favorable result in Q3 and a little bit better outlook in Q4, and it was really tied to 2 dynamics. The first is that we've seen a stronger adoption of some of our more high-performance appliances that carry a better gross margin profile. And then we're sourcing some of these components around memory and SSDs from multiple vendors as part of our strategy to get in front of any potential supply chain risk on these components. And we had just a better outcome in terms of delivery from some vendors that carry a little bit lower cost profile on those components for Q3 and Q4, and that's really what drove the improved outlook. So we're not updating the guidance at this time for next year, just given how dynamic the environment is, but that's something we'll continue to monitor.

Your next question comes from the line of Joseph Cardoso with JPMorgan.

Maybe for my first one, can you just touch on just maybe going back to the mix impact on the high-performance systems that you're seeing here? What are you seeing that's driving customers to kind of move mix up into these higher-performance systems? And then the second aspect there is more how sustainable that mix up is into the high-performance systems? And then I have a follow-up.

Yes. I think it's a lot of the dynamics that we've been talking about over the last several quarters, and we talk about our hardware business and so much of the growth is coming from expansion use cases and digital sovereignty. And then the big one is really the traffic we're seeing from AI use cases. And I think that it's really driving a need for customers to quickly provision higher performance units in support of that expansion initiative. And those higher-performing units have higher gross margins, but that's a trend that we think will continue. I think it's still relatively early. It's one of the dynamics that's really been driving the overall growth we've been seeing in our systems business.

Got it. And then maybe just quickly on an update on how you're thinking about the trends between software and system revenues and like maybe more specifically there, as we think about software accelerating, obviously, still doing high-single digits versus the long-term guide of more into that double-digit type of framework. I guess how are you guys thinking about potentially seeing an inflection upcoming? Is this still more in fiscal '27, better visibility there? Any kind of comments that you can provide around the trajectory of software here?

Yes, I'll provide a couple of thoughts. So first off, just right now, our pipeline looks really good. So we feel good about how we're positioned for the remainder of this year in terms of software. We're seeing strong trends on our Distributed Cloud business as well. And we've talked about in the past that our SaaS and Managed Service business has been a little bit of a headwind to growth in prior years because we were undertaking a transition on some legacy offerings, and we're effectively through that transition now. And so we anticipate that, that is going to be a growth driver to our software business next year, along with a much stronger renewal cohort on these multiyear agreements that we've been booking over time. And so we've got a really good base as we head into FY '27. We've got pretty good visibility as to the utilization that we're seeing from customers across those software opportunities, and that gives us a good window into the expansion that we would expect to see for next year. So it all adds up to what we think is a pretty compelling growth opportunity, getting back to a solid double-digit growth outlook for next year.

Your next question comes from the line of George Notter with Wolfe Research.

I guess I wanted to ask about progress with the iSeries refresh. I'm just curious like what your updated comments are and the pacing of the refresh. Obviously, the end of the support is the end of this year. I'm just curious about how much installed base is left, how you see that kind of playing out in terms of the migration of iSeries to rSeries? And then also just the impact of Mythos and Frontier AI models. I assume that's going to help kind of pull forward that refresh, but I'm wondering if you guys are seeing it.

Yes, I'll start, and then I'll let Francois follow up on the Mythos effect. So yes, it's going really well. When we talk about refresh, we've talked a lot about the dynamic we've been seeing with this refresh that we've likened it to a refresh plus because of the expansion that we've been seeing at the time of refresh. But just in terms of refreshing the existing iSeries base, I would say it's been very orderly and in line with our expectations. It's been very consistent quarter-to-quarter. And so as you noted, there's an end of software support date in our Q2 of '27, which we think will be a bit of a driver for acceleration of that activity over the next few quarters. But the thing that's been more interesting to us is that rate of expansion we're seeing at that time of refresh has continued to improve. And then, of course, the business we're seeing outside of refresh has also been very strong. So we're seeing really good growth in terms of expansion, both at the time of refresh and capacity expansion outside of the refresh. And then we're seeing new projects around data center modernization, digital sovereignty and then, again, performance associated with AI-driven workloads.

And then, George, on the Mythos effect, I think we're seeing that with our customers in a couple of ways. The first is we're seeing a number of customers really focus on addressing technology debt, and that includes avoiding to have aged estates. And that is actually a tailwind to the refresh motion because some customers that may have had aged estates and may have been dragging their feet to refresh those estates are actually taking that very seriously. We're also seeing customers that are focusing a lot on remediating vulnerabilities that may exist in their environment and in some cases, have stopped development of new applications and new features to go and address vulnerabilities. But the other effect for F5 is what's happening in runtime security. And so what customers realize now is that if you don't have the time, if you don't have the same window that you used to have, to be able to patch your applications, having the path to the applications be secure is incredibly important. And that's precisely where F5 is positioned in runtime security in front of applications. And we're seeing that. We had a very strong security quarter. Just to give you a manifestation of that, we just launched a quarter ago our AI-powered web application firewall on distributed cloud. And the uptake of that has been extraordinary. It's been the fastest-growing uptake of any product we've launched really ever with hundreds of customers already running on this product in the first few months of the technology. And the vast majority of them actually using the product in blocking mode to stop potential AI-powered attacks. And this product has already stopped potential zero-day attacks that would have come on to these customers. So we see a lot of energy with AI-powered security, and that's why we're continuing to make these investments in WAF, API security to drive that. Even beyond just AI-powered security, we're also seeing traction with securing AI itself. And I mentioned the number of AI security customers for F5 have grown 100% this quarter, largely because there is also a realization in the market that best-in-class security is also needed for AI models. You have probably followed the last week's saga of OpenAI and Hugging Face. And that just reminded all of us of the importance of having best-in-class security for AI models because the security that's provided by the frontier model providers is inadequate and we have invested in best-in-class security, and we're seeing that traction now. I think I gave the example in the script, but one of the major sports league in the U.S. just recently put in production AI-enabled applications that have sensitive data around player health or team strategies and needed to have best-in-class security for these AI models and chose F5 to provide that. So the effects of Mythos are actually wide ranging. We're seeing that in refresh for customers that want to kind of remediate tech debt. We're seeing that in more energy in AI-powered security in front of applications. And we're also seeing it in terms of the momentum we're seeing in AI security now to secure AI models.

Your next question comes from the line of Matt Hedberg with RBC Capital Markets.

Congrats on the quarter. Great to see. My first question is kind of a follow-up to a couple of prior questions. Obviously, you're seeing strong refresh activity and Cooper, I think in a prior question, you talked about seeing strong capacity expansion outside of the refresh and obviously, a lot of new business opportunities here in AI. I guess like philosophically speaking, growth from this cycle will peak at some point. But how should we think about growth post this cycle? Do you think it will be a shallower kind of trough than prior -- post prior cycles?

Yes. I would say that is our belief for a couple of reasons. One is just that in prior cycles, we didn't really have this kind of contribution to sales coming from new projects, whether that was digital sovereignty or AI-driven performance needs. The customer takeout opportunity -- or sorry, competitive takeout opportunity has been stronger in this period as well compared to where we were at the last time we had a product cycle. And so those are all opportunities for us to continue to drive healthy growth. And then we think that it's kind of a bit of a wildcard, but post-quantum crypto is on the minds of a lot of our customers, and that potentially could drive an earlier adoption of the next product cycle. It's really too early for us to try to handicap that. But I think that those are all dynamics that give us confidence that we're going to continue to see healthy growth from this business over the next several years.

That's great to hear. And then I've been really fascinated with sort of your AI factory success and being part of the NVIDIA reference architecture certainly seems to be helping you. Francois, in your prepared remarks, you talked about token costs. And I guess I'm wondering, could you talk a little bit about the ROI and maybe how it's added to growth when you're seeing customers see higher GPU utilization and optimize token spend?

Yes, Matt. So we -- I think as I shared before, we -- the -- our traffic management software, when paired with NVIDIA DPUs really on a number of tests has shown the ability to generate 40%, anywhere between 30% to 50% more tokens for a given GPU infrastructure. We think that value proposition is compelling. And here's what we're seeing in the early market work that our teams have done with customers. Where we are -- actually, interestingly, where we are seeing traction now is more with sovereign AI customers. So this is typically telcos outside of the U.S. that are implementing AI factories and are starting to put these AI factories in production, and they are quite cost conscious and really want to get the most amount of tokens for the infrastructure that they're building. And so we think that will continue. That said, those sovereign AI factories are generally relative to the Neocloud and Neoscalers, they're not the largest builds of infrastructure. With the Neoscalers, what we're seeing today is most of them, their business model is GPU as a service and offering GPU as a service typically to hyperscalers. And so the Neoscalers themselves are not concerned with how many tokens they're generating per GPU given that they're renting on a GPU unit. We think that is going to change. We think the market is going to evolve and that the economics of tokens are going to become important for Neoscalers as they go into the enterprise market or even as they sell more and more to AI native companies. So we think that market is going to evolve over time and Neoscalers will have -- will care more about the tokenomics, sorry, which is really the value add that F5 provides.

Your next question comes from the line of Simon Leopold with Raymond James.

I guess I'm trying to understand why systems are continuing to outgrow software. And I guess a couple of theories I wanted to throw out there. One is just a total cost of ownership for systems maybe being less costly than the TCO on the software solution or whether it's a possibility that your customers are having trouble getting the servers to run software on. We've heard that from others or if there are other reasons for this growth difference.

No, it's a really good question. Look, in terms of software, the growth that we're seeing this year is actually very much in line with our expectations, the expectations we had set at the beginning of the year. I think we had shared that the renewal cohort coming from the 2023 year was not a very strong renewal cohort. And so we expected growth this year that was in the mid-single digits. And I think we will do that, maybe a little better. But that's been the expectation. And conversely, as Cooper shared earlier, we expect software growth to reaccelerate next year in part because of a much stronger renewal cohort from 2024. Now -- so as it relates to hardware, there are a number of dynamics that are playing out that are really causing our customers to want to put more hardware in. One, of course, is we are in a refresh cycle. And so of course, that provides strength to hardware bookings. But there are other structural shifts we're seeing that are driving customers to want higher capacity hardware, higher throughput for higher performance. And those are -- those dynamics include some -- basically, sovereignty is a big part of it. So we're seeing the customers -- we started to see this trend in Europe a few quarters ago, but now it's expanded to the Middle East, it's expanded to Asia, where customers are repatriating workloads from global hyperscalers into local alternatives. In fact, there is a forecast out by one of the analysts, I think, this quarter that there's an expectation now that 75% of enterprises internationally will repatriate workloads into local alternatives inside the next 3 to 5 years. So we're seeing that phenomenon. And sovereignty also implies often building on their own infrastructure stack in their private data centers. There's also a dynamic around competitive displacement. We continue to win share from our competitors in hardware because we have placed bets on hardware innovation when they did not. And today, we find ourselves with hardware that has better throughput, higher performance, higher capacity, less power consumption. And those benefits are translating into customer wins. And then there's another dynamic around hardware, which is consolidation when customers have used multiple point solutions and want to consolidate on F5. So all of these dynamics are really driving the growth that we're seeing on hardware. And many of these elements are structural in nature, and we think will continue, which is the reason our Investor Day a month ago, we changed our long-term guidance on hardware, which has been in the mid-single-digit decline, and we confirm that we expected hardware to be growing for the next few years.

Your next question comes from the line of James Fish with Piper Sandler.

Nice quarter. Maybe just specifically, what are you seeing on the inbound interest in pipeline build, specifically on AI Guardrails as of late? And are you seeing any customers specifically budget for AI agent security as sort of this new line item or is it kind of being bundled into that broader application security platform? And I just have a follow-up on the billings.

It's interesting. It varies from customer to customer. What we're seeing right now is in financial services -- financial services and then other organizations that are quite advanced or sophisticated in their use of AI or their care for safety, we are seeing that they are budgeting specifically for securing AI. And that includes securing AI models, securing AI agents, governance around AI, discovery of what their employees are using for AI. It's a number of functions. And we have invested quite rapidly in this area around testing and penetration of AI models, securing AI models in production. We made an acquisition this quarter of a company called SurePath AI that complemented our capabilities with Discovery. And we launched our AI security platform because we're seeing customers increasingly wanting more and more of these point solutions combined into a single platform. And we think that's where the market is headed, just like the WAAP market consolidated a few years ago with WAF, DDoS, API and Bot Defense in a single platform, we see the same consolidation of functionality happening with AI security platforms. So that's where we're seeing very strong interest. And I shared earlier that the number of customers in that area doubled this quarter and we expect that momentum to continue to grow with the investments we're making in this area.

And then Cooper, just for you. Obviously, good top line quarter, but actually, if I look at billings here, another exceptional quarter again. Are you seeing product backlog build and that explains some of this? Or is it more so the Services lags here just has yet to flow into kind of the revenue line?

Yes, thanks. So no, it doesn't have anything to do with backlog because we wouldn't have billings yet. We don't bill until we ship product. It's more to do with, as you said, the lag factor on the Service revenue and then we had a strong bookings quarter with new FCPs and so there's a portion of the FCP booking that gets deferred. So you get 2/3-or-so gets recognized upfront and the remainder gets deferred. And so that drove some strength in the deferred revenue -- on the deferred subscription revenue, which fed into that billings improvement.

Your next question comes from the line of Meta Marshall with Morgan Stanley.

Great. Maybe first for Francois. Just in the 3 AI use cases, data delivery, runtime security and load balancing, noted the commentary on the growth in customers on the AI security side. But just are there any developments in terms of kind of deal sizes or just how you would size any of those opportunities versus particularly as runtime security and load balancing mature? And then maybe just a follow-up question for Cooper. Just if there's any kind of contribution to the quarter we should be mindful of from pricing, if you could just outline that?

Thank you, Meta. Look, the deal sizes are -- it's interesting because there's a wide divergence. They could be as small as hundreds of Ks to be as large as we've done some 8-figure deals in this area of AI. I think in security, as we move more towards platform over time and we consolidate these functionalities, deal sizes will obviously grow. In AI data delivery, we are already seeing substantial deals because it's all about -- AI data delivery, it's all about getting data faster to models in training or for inference. So customers are highly sensitive to latency, want very high throughput. And generally, that means they want high-capacity hardware to perform that, and that leads to large deal sizes. In AI factory load balancing, the deal sizes are initially smaller when you have factory build-outs of just a few hundred GPUs. I think as more of the -- we get into more of the larger AI factories, we should see greater deal size there. Overall, look, I think our AI revenues are growing quite rapidly, and we may be able to touch on this more in -- at the end of the year once we have a year's worth of data to share with you.

And then in terms of price realization, so there's -- we saw a modest contribution from price realization that was tied more to our discount governance. We've been pretty effective this year, continuing to improve our discounting over time. The adjustments that we've made to pass through some of the costs on components will be more of a contributor to our FY '27 revenue, but nothing that really hit our revenue in Q3.

Your next question comes from the line of Tal Liani with Bank of America.

It's Tomer Zilberman on for Tal tonight. Just wanted to ask on a geographic basis, if we look at your Americas revenue, it accelerated from the last few quarters of low- to mid-single digits, I think you said to 11% on the call. Just wanted to ask what the trends you were -- you saw in the geo? Was it more of a catch-up because of the last few quarters of underinvestment, or what did you see there?

Yes. I'll take that one, Tomer. So really, we've seen good strength across all theaters from a bookings perspective. So it is more to do with the timing of shipments. And Francois noted that last quarter on the call because the growth rate was down a bit from Americas, and we said that the underlying demand is actually quite strong, and we thought that, that would play out over time as we were able to ship more of those orders. In this environment, there is a bit of a lag on shipping time from when we book the orders. And so we saw that come through in the reported revenue in Q3. You'll also notice this quarter, APAC is down 11% on a revenue-reported basis. That is also not indicative of the underlying demand strength that we're seeing in APAC. So I think just in this environment where there's a little bit more of a lag on the shipping timing, you're going to get more of that variability in the growth rates. But over a longer period of time, multiple quarters, you start to see that growth rate start to normalize. But broad answer to your question, the underlying demand growth has been healthy across all of our theaters.

Got it. And maybe as a follow-up and a similar question to touch on the subscription line. Last quarter, it was, I think, flat on a dollar basis sequentially, and you talked about a weak renewal base, which I think you also mentioned in the prepared remarks today. So if I look at '23 as a weak renewal base, I think in aggregate is what you mentioned, what drove the stronger performance this quarter?

Yes. So when we talk about the weak renewal base, we're largely talking about the full year. So we had flattish software revenue in FY '22 -- sorry, FY '23 against FY '22. And so as a result, the base that we had to renew for this year was more modest than it typically would be, and it's going to be much stronger in our FY '27. But from a sequential basis, you still get that variability. And so we did see some strength in our subscription business this quarter tied to a stronger cohort and also good expansion in the quarter. But overall, the growth rate has been a little bit more orderly in terms of the quarter-to-quarter variability in the growth rate this year than what we've seen in past years. Now next year, we think that the growth rates will be stronger in the second half of the year. So it's just one of these things where we're always tracking what that base is that's coming up for renewal, what the expansion opportunities are and then trying to give some visibility if there's any unusual patterns that we see. But for this year, it's been more muted, the variability in the growth rates.

Your next question comes from the line of Michael Ng with Goldman Sachs.

I just had a few follow-ups on kind of implications for 2027. Maybe first on the systems side, just given all the momentum that you have there in terms of refresh and the expanding use cases, do you see upside to the prior outlook of mid- to high-single digits for 2027? Just would love any general thoughts there. And I certainly hear you on the reiterated gross margin outlook for 2027. Just would love some commentary on whether you are seeing some better operating leverage now but are kind of reserving some margin in the case that there's some outside commodity cost inflation? Just trying to understand like what's kind of embedded in the reiterated 80% to 82% gross margin guide for next year.

Michael, thank you. Let me start, and then Cooper will address the gross margin question. On the revenue expectations for next year, as you know, we've given guidance of upper single-digit growth for '27 and actually beyond at our Investor Day about a month ago. We're not updating that guidance today. We -- generally, we feel good about the -- what we're seeing in the marketplace, the demand we're seeing, both for hardware and our software. But beyond the immediate demand, what we are most pleased about is the structural demand drivers that we're seeing. Our -- we continue to drive competitive wins. We continue to drive consolidation onto F5. We have a number of customers that were using point solutions from cloud providers or other players in the space and our whole investment into a platform strategy with our application delivery and security platform, we're seeing real momentum with the ADSP platform and the real effect of that in consolidation. We just launched F5 Insight also this quarter, which really glues elements of our platform together. We're seeing very strong adoption of F5 Insight among our customers, which tells us that they're seeing significant value in our platform. And of course, we're seeing growth in AI use cases across data delivery and security. And so you take all of these drivers, I think we feel good about the guidance we've given for the next 3 years. And we'll talk specifically about 2027 on our October call.

Yes. And then in terms of the gross margin, so we didn't reiterate our gross margin for next year. We just are not updating at this time. I know it's a little bit of a subtle nuance. But really, I think the way to think of it is we're encouraged with what we saw in Q3 and in Q4, but we are cognizant that things are fluid. And transparently, we are getting mixed signals. So we've seen a lot of signs of some stability in terms of memory pricing this quarter. But then we're -- we also hear other signals that it still could be more variable. And so -- and then there's -- looking at the rest of our BOM as well and other components that we're sourcing, and just as early as it is right now, we felt like it was too early to make an adjustment to our gross margin forecast for next year. But that said, I think our manufacturing team has done an outstanding job in terms of doing -- making advanced purchases on some of these more scarce components, diversifying the supply base. And so we feel like we've got a pretty good handle on the supply that we need for our revenue for next year, and we think we'll be in a position to give a better update on the gross margin guidance in October.

Your next question comes from the line of Jeffrey Hopson with Needham.

We're seeing with AI workloads, some of a recent push to moving things on-prem for security or sovereign reasons. Just curious in those more regulated industries that are on-prem kind of where are we in that adoption of AI and where are we with newer AI companies starting to consider on-prem deployments?

Well, I think -- well, in regulated industries, frankly, it varies from an industry to the next and from one customer to the next. I think in financial services, we're seeing rapid adoption of AI. We're also starting to see that in health care, and we're seeing our telco customers also regulated, but leveraging AI to significantly improve productivity and operations and the experience of their customers. In terms of what stack are people using and what models people are using, what I think we're seeing already is very strong signs that both enterprises, government agencies and telcos are all going to use a mix of models. They are going to use closed models from frontier AI providers and open weight and/or open source models. They are going to consume public AI from the public cloud, and they are going to build their own stack, their own infrastructure on-premise to run their models. And they're going to want to basically find the best model for each task and ideally the cheapest model for each task. And that multiplicity of models and types of infrastructure is exactly what plays to the strength of F5 because we have built the hybrid multi-cloud company precisely to serve and enable that choice for customers, just in the same way that we predicted several years ago that our customers would end up wanting to have the choice in infrastructure environments and deploy applications in multiple infrastructure environments, and we designed our company to meet that moment. We think the same thing is happening with AI, where customers want the best model, the best environment for each task, and they're going to want a partner that can deliver these models and secure all these models across any environment and in any cloud. And that is the company that we've built. That is the company we're building for AI. And we are, I believe, meeting the moment, and we're very excited about the ways in which different customers are going to deploy AI and the way we're going to enable that choice for them.

Got it. And maybe just a follow-up on agentic workloads across apps. Is there any way to think about how they may be similar or different than traditional workloads and the resulting impacts on F5 then?

So agents are fundamentally just regular applications with the exception that they're invoking models. They're invoking models for inference. And agents also have the ability to make subsequent calls inside of a data center or an environment. And so what we really see is agents are increasing data center traffic or hybrid multi-cloud traffic inside these organizational environments. And so for F5, what that fundamentally means is sitting in front of more of this traffic that's going to APIs, that's going to traditional applications, securing that traffic, but then also securing and delivering traffic to these AI models as well, whether they're frontier models or models that are fundamentally set up inside of a local data center or in a local environment for an organization.

Thank you, Kunal. And Jeff, that was Kunal, our Chief Product Officer.

Your final question comes from the line of Amit Daryanani with Evercore.

Francois, you folks saw obviously some very impressive growth on the systems side, and you talked about multiple factors that are sort of helping you folks out in that space right now. I think what investors are really going to struggle and trying to understand is how many of these vectors are cyclical versus secular? And so just from your perspective, qualitatively or quantitatively, how do you think about this 32% growth and how much of this do you think is cyclical versus secular as you look forward?

Well, clearly, Amit, we think a portion of it is cyclical. If we didn't think that we would be guiding 30% hardware growth for the next 3 years. And this is not what we -- as you know, that's not what we guided at Investor Day. We shared we expected hardware to be in the single-digit growth to mid-single-digit growth going forward. So what's cyclical, obviously, is that we are in the middle of a refresh cycle with customers. And that, by definition, is cyclical. But there are a number of more secular shifts that have layered on to that and that are driving hardware growth in where you're seeing it. And those secular drivers we've talked about, it's the move now outside Europe, Middle East and Asia for sovereignty that creates reinvestment in data centers. If there was a phase where customers were thinking about cloud first, anytime they were thinking about new applications, now we're in a phase where customers think about sovereignty first, and it means that they're reinvesting in data centers or building resilience between their cloud infrastructure and their data center infrastructure. That's a driver. A second driver, which is difficult to measure today, but I think we'll get more refined about it in the future and we'll get more intelligent about it in the future is the amount of AI traffic that is now flowing on on-premise infrastructure and how much that is driving customers to either need more units from F5 or higher capacity units from F5. But that is clearly also a driver. And then there are things that are specific to our competitive position, specifically what we're doing with platform adoption, what we're doing with competitive takeouts that are also exacerbating this trend. And so all of that is -- it's not trivial to completely separate and quantify for you what's cyclical, what's these other drivers because sometimes these other drivers also manifest themselves in a refresh motion, meaning our teams are speaking with customers about a refresh. And rather than refreshing just the capacity they have, we're seeing very strong expansion at refresh that is driven by new use cases. So those -- all of these factors kind of come together and manifest themselves in the numbers. But I think you can fairly think about it as there's clearly a refresh portion that is cyclical, and there are these 3 or 4 drivers that we think are durable. And that is why we have changed our view on the trajectory of hardware, not just for this year, but for the next few years.

That concludes our question-and-answer session. I will now turn the call back to Suzanne DuLong for closing remarks.

Thank you, Tiffany. Thanks, everybody, for joining us. We look forward to seeing many of you out and about at conferences during the quarter. Please reach out with any questions.

Ladies and gentlemen, this concludes today's call. Thank you all for joining. You may now disconnect.