Palo Alto Networks (PANW.US) Q1 Earnings Call: Net addition of approximately 170 platformization deployments in the quarter! AI vulnerability discoveries approach 7 times last year's monthly average, enterprises begin tapping emergency budgets

date
15:34 19/11/2026
avatar
GMT Eight
Palo Alto Networks (PANW.US) disclosed details on its Q1 fiscal 2027 earnings call that reveal demand shifts more telling than revenue growth: enterprises have already tapped emergency budgets to test whether their systems can withstand attacks using frontier AI. Management observed that the number of critical and severe vulnerabilities discovered each month is currently close to 7 times last year's monthly average.
Cybersecurity supergiant Palo Alto Networks (PANW.US) disclosed details on its Q1 FY2027 earnings call that say more about shifting demand than revenue growth does: enterprises have already tapped emergency budgets to test with frontier AI whether their systems can withstand attacks. Management observed that the number of critical and severe vulnerabilities discovered each month is now nearly 7 times last year's monthly average. Palo Alto management said enterprises worldwide currently face a dual risk: leave vulnerabilities unaddressed and attackers may exploit them first; rush to install patches and production systems may be disrupted. That explains why customers are willing to buy virtual patches first to buy time for formal remediation, and then consolidate detection, protection and response tools previously scattered across multiple vendors. The quarter saw a net addition of approximately 170 platformization deployments, more than the sum of the first quarters of the prior two fiscal years. Customers who have adopted the platform contribute about two-thirds of next-generation security annual recurring revenue, with net revenue retention above 120%. The AI agent frenzy focused on highly efficient agentic AI workflows is changing the pricing logic of security software, but there is a time lag between traffic growth and revenue recognition. Agent traffic on Palo Alto Networks' secure access service edge (SASE) platform has reached about 9 times the year-ago level. Because customers overall remain within their purchased traffic quotas, management estimates it will take another 6 to 9 months before excess traffic begins to contribute incremental revenue. Prisma AIRS has a more direct business model: the more tokens enterprises use, the more tokens need to be inspected. Just five quarters after general availability, the product has more than 1,000 customers and annual recurring revenue approaching US$170 million. Management believes coding agent security will be the first to become a must-have. Enterprises already use coding tools widely, yet often cannot answer a basic question: when the tools create new agents on their own, who limits which systems those agents can access and which actions they can perform? The proposed all-stock acquisition of Applied Compute for US$4.5 billion targets precisely the areas that traditional security rules struggle to cover. Management argues for having the same team that post-trains models for an enterprise's business model also train a supervisory model that understands its intended behavior. In this way, the security system can not only block known attacks but also judge whether an agent has deviated from its authorized tasks. Management expects this business to contribute about US$1 billion in next-generation security annual recurring revenue by fiscal 2029. Compute will be deployed in phases as demand dictates, initially prioritizing cloud partners and existing contracts, with cumulative related capital expenditure through fiscal 2030 expected to be below US$3 billion. If the deal closes as expected in Q3 FY2027, it is expected to reduce operating margin by up to 1 percentage point in FY2027 and up to 2 percentage points in FY2028, returning to 29% to 30% in FY2029. The raised full-year guidance does not yet include this acquisition. Below is the full transcript of Palo Alto Networks' earnings call (AI-assisted translation; Palo Alto Networks is the Chinese name for Palo Alto Networks): Hamza Fodderwala, Senior Vice President and Head of Investor Relations, Palo Alto Networks: Hello everyone, and welcome to Palo Alto Networks' first quarter fiscal 2027 earnings conference call. I'm Hamza Fodderwala, Senior Vice President and Head of Investor Relations. Please note that it is 1:30 p.m. Pacific Time on Wednesday, November 18, 2026, and this call is being recorded. Joining me today to discuss first quarter results are Nikesh Arora, our Chairman and Chief Executive Officer, and Dipak Golechha, our Chief Financial Officer. Following prepared remarks, Lee Klarich, our Chief Product and Technology Officer, will join us for the Q&A session. You can find the press release and supplemental materials for this call on the company's investor relations website at investors.paloaltonetworks.com. Please click on the quarterly results link on the website to view the first quarter fiscal 2027 supplemental financial materials and earnings presentation. Today's call will contain forward-looking statements and projections regarding the company's business operations, financial performance and recent acquisitions. These statements made today involve a number of risks and uncertainties, and actual results may differ from those stated in the forward-looking statements. For a description of these risks and uncertainties, please refer to our press release and recent filings with the U.S. Securities and Exchange Commission. We undertake no obligation to update any forward-looking statements made on today's call. This presentation also contains non-GAAP financial measures and key metrics relating to the company's past performance and future expected performance. Non-GAAP financial measures are not a substitute for financial measures prepared in accordance with U.S. GAAP. The most directly comparable U.S. GAAP financial measures, as well as reconciliations between the two, are included in the press release and the appendix to the investor presentation. Unless otherwise specifically noted, all results and comparisons are on a year-over-year basis for the same fiscal period. I'll now turn the call over to Nikesh. Nikesh Arora, Chairman and Chief Executive Officer, Palo Alto Networks: Thank you, Hamza. Good afternoon, everyone. As you can see, we are off to a strong start in the first quarter of our fiscal year. All metrics exceeded company guidance. Next-generation security annual recurring revenue (NGS ARR) grew 66%, with the magnitude of the beat among the largest in company history. Total revenue grew 39% and remaining performance obligations (RPO) grew 35%, both accelerating from the prior quarter. Delivering these results in a seasonally lighter first fiscal quarter speaks to the strength of demand. As AI accelerates the speed of cyberattacks, enterprises are urgently replacing disconnected tools with a unified platform. This momentum gives us confidence to significantly raise our full-year guidance. Strong revenue performance also translated directly to profit. In the first quarter, our operating margin again exceeded 30%, and despite integrating a large acquisition, we achieved record operating profitability. Most exciting is that demand was strong across all platforms, with revenue growth accelerating in every one. Starting with our largest business, Network and AI Security, which grew 20%, with strong performance across secure access service edge (SASE) as well as software and hardware firewalls. Customers are moving away from point products. They recognize that relying on multiple vendors each defending the network perimeter in their own silos leaves blind spots in the two critical areas of visibility into network activity and enforcement of security policy. We are also seeing a fundamental change in the nature of the traffic traversing networks. Autonomous agents continuously connect to applications, application programming interfaces (APIs) and enterprise data, and these interactions all require real-time inspection. Agent traffic on the SASE platform has reached about 9 times the year-ago level. The traffic surge reflects enterprises putting AI into production. The need to protect this new attack surface also drove Prisma AIRS to nearly US$170 million in annual recurring revenue (ARR) within five quarters of general availability, making it the fastest-growing product in Palo Alto Networks history. Driven by XSIAM and observability, Cortex grew more than 50% in the first quarter. AI-driven attacks unfold at machine speed, and after-the-fact remediation spread across multiple data lakes is not an effective defense. We need solutions that can perform real-time analysis, provide observability, dynamically correlate different signals, and deliver correlated results and threat verdicts in real time. XSIAM now ingests more than 20 petabytes (PB) of telemetry per day, using AI to cut response times from days to less than 10 minutes. Turning to our identity security platform Idira, revenue grew 23% in the first quarter on a pro forma basis. Since completing the CyberArk acquisition earlier this year, we have been executing on the integration plan. In the first quarter, Idira added more than 100 net new customers from Palo Alto Networks' existing customer base, our best quarter of cross-selling between the two to date. The rise of AI agents brings an entirely new category of non-human identities that require real-time protection. Effectively governing these identities requires a complete view of their activity and the ability to immediately revoke access when suspicious behavior is detected. It is precisely to solve this problem that we launched Idira's Agentic Identity Provisioning service (Agentic IdP) in the first quarter and integrated it directly into Prisma AIRS' AI gateway to enforce agent permission policies in real time. Customers use our three platforms to solve the same core problem: shortening the time between discovering a security risk and eliminating it. Attacks move at machine speed, and point products simply cannot keep up. Enterprises are consolidating their security architecture onto unified platforms, and this shift continues to drive platformization. Yes, that word again, platformization. First quarter platformization deployments increased by a net approximately 170, a record, more than the sum of the first quarters of the past two fiscal years. Customers who have adopted platformization deployments now contribute two-thirds of our next-generation security annual recurring revenue, and these customers continue to expand purchases rapidly, with net revenue retention above 120%. Several of our largest deals this quarter demonstrated platformization momentum. In the first quarter, we signed an US$83 million agreement with a global financial institution. This customer consolidated network security onto our platform, adopted the Frontier virtual patching service, and expanded identity security purchases, signing the largest order in CyberArk history. A global consulting firm putting AI into production company-wide signed a US$46 million deal with us, including the largest order in Prisma AIRS history. Finally, we signed a US$20 million agreement with a global IT services enterprise to help it unify its security stack, reduce tool sprawl, and upgrade its defenses. The customer purchased Prisma AIRS and SASE to protect its AI deployments and network access for employees worldwide. Customers choosing us are making long-term decisions about their technology architecture. Every additional commitment from customers increases our responsibility. They want us to keep refining the products they rely on to operate today, and they also want us to address the next wave of change in advance. Six months ago, with the first release of models with cyber offense and defense capabilities such as Mythos, we predicted that equally powerful offensive AI capabilities would spread rapidly through the open-source ecosystem within three to six months, and that prediction has now come true. I want to be clear that the current situation is being driven by two distinct effects at the same time. First, the number of vulnerabilities discovered and the number of patches that need to be deployed have surged. The number of critical and severe vulnerabilities discovered is now nearly 7 times last year's monthly average. Vulnerabilities are being discovered so quickly and in such density that IT teams can hardly keep up. We believe many of the major service disruptions in recent weeks, including outages at large enterprises and government agencies, stem in part from unintended consequences of installing patches and changing configurations. IT teams struggle to patch these vulnerabilities quickly enough without disrupting production environments. The second effect is the rise of AI-assisted cyberattacks. Attackers are using open-weight models and new agent frameworks to discover and exploit vulnerabilities at unprecedented speed and scale. New agent frameworks such as ARTEX allow attackers to turn simple instructions into autonomous, multi-stage attacks, with just a few people compromising dozens of targets simultaneously. A few days ago, our Unit 42 team published research on a campaign in which a small hacker group linked to a foreign actor compromised 19 organizations simultaneously. Looking ahead, we expect cyber threats to remain elevated, but we believe enterprises have multiple ways to adapt. First, for at least the next 18 months, enterprises will still need to deploy a large number of patches, and the risk of collateral damage will persist. As customers get better at judging what to patch and when, we expect this collateral damage to gradually decline. Second, demand for virtual patching will grow across industries and software types. Enterprises can first establish reliable temporary protection for their infrastructure and then formally install patches once deployment schedules are coordinated. In the first quarter, demand for our advanced virtual patching service was strong, and with nearly 70,000 network security customers, we have a significant opportunity to offer this service to them. Third, for all enterprises, there is no perfect defense against attacks, and no organization can assume it will be spared. Attacks will happen, and defenders must be prepared to detect and respond to threats at unprecedented speed; piecemeal after-the-fact remediation is not enough. Therefore, we believe the most important of the three responses is the third: upgrading more than US$1 trillion of traditional security and IT infrastructure to real-time platforms. Despite intensifying threats, market demand for AI remains very strong. Every time inference costs fall, thousands of new agent workflows become viable. On this trend, industry token revenue could exceed US$1 trillion by 2029. The market is focused on the AI theme trade; we are focused on the inflection point in AI adoption, as the industry shifts from building AI infrastructure to enterprises actually using AI. Until recently, AI activity was concentrated mainly in consumer scenarios and cross-industry general applications such as coding, where the relevant tokens have little security protection. Looking ahead, we believe already-deployed AI infrastructure will also begin to expose vulnerabilities, and we have recently seen vulnerabilities at AI infrastructure providers being exploited. As coding tools increasingly execute tasks autonomously, insufficient security controls will also bring agent risk into enterprise environments. AI adoption and AI security must advance together. We are already helping customers close this gap with Prisma AIRS. Over the past five quarters, the product's customer count has exceeded 1,000, with annual recurring revenue approaching US$170 million. Today, Prisma AIRS provides comprehensive AI security capabilities across five areas, enabling enterprises to see, assess and govern all their AI deployments on a single platform. First is AI asset discovery and AI risk assessment. First, can you find every AI asset in your enterprise? Prisma AIRS can discover models and autonomous agents running anywhere, then continuously assess their risk across supply chain, security posture and automated AI red teaming. The next step is real-time policy enforcement through AI protection and AI governance. Can you bring all relevant traffic to a single policy enforcement point? With coordinated runtime security, identity security and AI gateway, customers can inspect prompts, block sensitive data leakage, protect agent identities, and enforce policies in real time on every interaction. This is the industry's first AI security platform. Tying these pieces together is AI observability. It captures and monitors every prompt, every response and every log, so that even at scale, enterprises can ensure all actions are traceable and accountable. Prisma AIRS now covers the important elements of AI security, but we expect that as models become more intelligent, manually defined static controls will eventually become ineffective. Therefore, AI security models need advanced reasoning capabilities comparable to frontier models. For people to trust large language models, we must be able to independently verify, explain and audit actual AI deployments. We believe trust and control are both critical to driving enterprise AI adoption. We have already seen that enterprise demand for autonomy and control is driving adoption of open-weight and open-source models. There are clear early signs of this shift. Over the past year, open-weight models' share of token traffic has risen from less than 5% to more than 60%. When enterprises put AI at the core of operations, they want to keep proprietary data, organizational judgment and competitive advantage firmly in-house. We expect that as open-weight and open-source models become widespread in enterprise applications, the future market opportunity for model inference services will reach hundreds of billions of dollars. Enterprises can both trust AI and control how it is deployed, training models themselves and running them on-premises or on sovereign technology stacks they control, which will accelerate enterprise AI adoption. This means models must be delivered securely, and where needed, in a way that safeguards enterprise autonomy, with control remaining fully in the hands of the enterprise. When we looked into training customized open-weight models ourselves, we searched across the AI ecosystem for partners with strong engineering capabilities and a proven track record of solving enterprise problems. We ultimately chose Applied Compute because no other team could match its technical depth in post-training open-weight models. We are bringing a world-class frontier AI research team into the company, and we expect researchers to reach 100 by year-end. Every enterprise has decades of proprietary data, internal operating procedures and organizational judgment that cannot be disclosed to public models. But until now, these enterprises have lacked a trusted partner to help them securely and autonomously post-train, run and supervise open-weight models. We believe Applied Compute is the right partner to pursue this opportunity with us. The team has already won over large enterprises and AI-native companies, reaching more than US$100 million in annualized revenue just 18 months after founding, a feat few companies can achieve. Applied Compute will help us advance in two directions. First, embedding AI into all security capabilities. We are training models for more than 35 specialized scenarios and delivering these capabilities to customers so they can use AI to fight AI. Second, we will help enterprises train open-weight models and provide model inference services in a secure, enterprise-controlled manner. Once open-weight models are deployed, we can deliver secure open-weight models to customers so enterprises can use AI safely. To support this work, we plan to build diversified global inference capacity exceeding 100 megawatts over the next three to five years. To that end, through fiscal 2030, we plan cumulative capital expenditure that is relatively limited, below US$3 billion, roughly equivalent to the free cash flow we generate in half a year. Combining their platform with our global go-to-market reach and the trust we have earned with CIOs around the world, we expect Applied Compute to continue scaling rapidly and to contribute more than US$1 billion to our next-generation security annual recurring revenue by fiscal 2029. In closing, the first quarter marked a decisive turning point for the cybersecurity industry. AI-driven attacks have eliminated any room for enterprises to improve gradually at a comfortable pace. Decade-old, isolated tools have proven unable to stop autonomous attacks, and US$1 trillion of traditional infrastructure urgently needs upgrading. At the same time, for AI to be used confidently across the enterprise, AI security must evolve from visibility into activity and policy enforcement to enabling enterprises to trust the models themselves. Without confidence that models behave reliably in production, enterprises will struggle to entrust mission-critical workflows to AI. Finally, as autonomy and control become increasingly important to enterprises, helping them adopt open-weight models securely will be a huge opportunity. For years to come, we will deliver this foundational capability alongside our core security platform, playing a central role in customers' security upgrades and AI transformation. I'll now ask Dipak to go deeper on the financial details and discuss the updated outlook. Dipak Golechha, Chief Financial Officer, Palo Alto Networks: Thank you, Nikesh. Good afternoon, everyone. We delivered record first quarter results, with key metrics measuring revenue and future revenue all accelerating. Remaining performance obligations (RPO) grew 35% year over year to US$21 billion, of which the portion expected to be recognized over the next 12 months, current RPO, grew 36% year over year to US$9.4 billion. Contract duration was flat year over year. Next-generation security annual recurring revenue (NGS ARR) reached US$9.73 billion, up 66% year over year, with net new NGS ARR more than double the year-ago period. NGS ARR beat guidance by a margin among the largest in company history on a quarterly basis. First quarter total revenue grew 39% year over year to US$3.44 billion. As announced last quarter, we now report revenue across three businesses: Network and AI Security, Cortex and Idira. First quarter revenue growth was indeed broad-based, with all three businesses accelerating. Network and AI Security is already substantial in scale, and after a strong fourth quarter, growth accelerated further in the first quarter. Revenue for this business grew 20% year over year to US$2.3 billion. Network security products grew strongly across all deployment forms. Starting with SASE, ARR grew 30% year over year and contributed more than US$100 million to net new NGS ARR. Demand was particularly strong for software-defined wide area network (SD-WAN) and secure browser, as customers continue to favor our integrated SASE platform over vendors offering only individual products. In firewalls, both hardware and software were strong, with booked orders over the past 12 months growing 28% year over year. Customers want unified visibility into network activity and unified enforcement, so they are consolidating vendors, and our market share continues to rise. Software firewall ARR accelerated for the third consecutive quarter, growing 34% year over year. The main driver was Prisma SASE, which remains our fastest-growing product, with ARR approaching US$170 million. Turning to Cortex, our security operations and observability platform. First quarter revenue grew 53% year over year, or 40% on a pro forma basis. As AI intensifies security threats, customers are prioritizing modernization of the security operations center (SOC), driving XSIAM ARR up more than 65% year over year to about US$800 million. In addition, our observability platform, recently relaunched as Cortex XCOR, continued strong growth, with ARR about four times the year-ago level. As we mentioned last quarter, this includes net new ARR of roughly nine figures from a large frontier AI lab customer migrating from its prior vendor to Chronosphere. Identity security platform Idira first quarter revenue grew 23% year over year on a pro forma basis. Our core sales team and the Idira team continue to work closely on go-to-market. In the first quarter, this collaboration added more than 100 net new Idira customers, all from our core business's existing customer base. This was the quarter with the most customer wins through cross-selling since the acquisition closed. We are also seeing significant synergies on the profit side. As a reminder, when we announced the acquisition, we set a target of having Idira's operating margin catch up to the core business within 18 to 24 months after closing. At the current pace, we will achieve this in about half the originally planned time. Performance was also broadly strong by region, with Americas revenue up 42% year over year, Europe, Middle East and Africa up 40%, and Japan and rest of Asia Pacific up 26%. We set records across revenue metrics in the first quarter while continuing to focus on profitable growth. Adhering to this principle further unlocked operating leverage year over year and strengthened our confidence in achieving full-year margin guidance targets. Our scale and diversified business model help mitigate gross margin pressure, with gross margin continuing to hold around 75%. First quarter total gross margin was 75.2%, down 170 basis points year over year. As discussed last quarter, we continue to actively manage multiple factors affecting gross margin in the near to medium term. First, hardware component costs remain elevated across the industry, and we are proactively leveraging our scale and relationships with strategic suppliers to manage this. Hardware accounts for about 10% of total revenue, but higher component costs are expected to affect gross margin throughout the fiscal year. Second, a higher mix of software-as-a-service (SaaS) products that are still early-stage and growing quickly also weighed on gross margin. This is particularly evident in Cortex. As we focus on expanding market share, both XSIAM and observability are growing rapidly. In addition, customer migration from Prisma to Cortex Cloud is still underway, temporarily weighing on that business's margin. Given platform migration and product mix changes, we expect Cortex gross margin to remain below the company average in the medium term, recovering above 70% over the long term. On operating margin, as the business scales and operating efficiency improves, we continue to offset higher cost of sales through operating expense leverage. Despite higher acquisition-related costs and gross margin pressure, first quarter non-GAAP operating margin still reached 30.7%, up 50 basis points year over year. We are also deploying AI in a targeted way across multiple parts of the company to improve efficiency. For example, we are using AI to assist with quoting, shortening time to close and improving sales efficiency. We are also using AI to improve procurement efficiency and have identified meaningful cost savings through supplier consolidation. On profit, non-GAAP diluted earnings per share were US$1.05, up 13% year over year and US$0.08 above guidance. On a U.S. GAAP basis, earnings per share were US$0.57, up 21% year over year. Improved profitability also drove cash generation in the first quarter, with adjusted free cash flow of US$1.9 billion and adjusted free cash flow margin of 56%. On capital allocation, we did not repurchase shares this quarter and will continue to evaluate repurchases opportunistically. Our existing share repurchase authorization is up to US$1 billion and expires in December 2027. The company's overall performance is directly attributable to our continued focus on innovation. Through internal investment and targeted acquisitions, we aim to stay ahead of technology inflection points and anticipate customer needs years in advance. As Nikesh just said, we are pleased to announce the proposed acquisition of Applied Compute, through which we hope to be first to deliver secure AI to customers. Total consideration is US$4.5 billion, all in stock, and the deal is expected to close in the third quarter of fiscal 2027. The acquisition includes Applied Compute's cash and prepaid compute purchase commitments, totaling about US$500 million. The company's growth has been impressive, reaching US$100 million in annualized revenue within 18 months of founding. From day one after closing, the acquisition will be accretive to our revenue growth rate. We expect the business to scale to contribute US$1 billion in incremental NGS ARR by fiscal 2029. Longer term, we believe this will become another multibillion-dollar ARR business for the company and will position us at the forefront of AI security. To provide customers with secure model post-training and inference services, we expect our compute needs to grow, but we still plan an asset-light model, maintaining strong free cash flow margins throughout the capital expenditure cycle and beyond. To meet compute needs, we will continue to purchase from cloud partners and gradually add our own dedicated compute. Most of this investment will be in graphics processing units (GPUs) and servers, with a smaller portion in facilities to house them. We intend to expand compute in phases based on demand we can already foresee. Overall, through fiscal 2030, we expect cumulative related capital expenditure below US$3 billion, roughly half of our expected free cash flow this fiscal year. Assuming the deal closes in the third quarter of fiscal 2027, we expect the transaction to dilute operating margin by up to 100 basis points in fiscal 2027 and up to 200 basis points in fiscal 2028, before returning to the current 29% to 30% level in fiscal 2029. Finally, from a free cash flow perspective, excluding data center capital expenditure, we expect free cash flow margin to remain above 38% in fiscal 2028 and beyond. After closing, we will provide more details and more precise modeling parameters. Turning to guidance. Guidance covers only Palo Alto Networks' existing business and does not include the pending Applied Compute acquisition. For the second quarter of fiscal 2027, we expect NGS ARR of US$10.20 billion to US$10.22 billion, growth of 61%. RPO of US$21.5 billion to US$21.6 billion, growth of 34% to 35%. Revenue of US$3.53 billion to US$3.55 billion, growth of 36% to 37%. Fully diluted share count is expected to be 837 million to 842 million shares, and non-GAAP diluted earnings per share of US$1.01 to US$1.03. For full fiscal 2027, we expect NGS ARR of US$11.475 billion to US$11.52 billion, growth of 26% to 27%. RPO of US$25.3 billion to US$25.5 billion, growth of 19% to 20%. Revenue of US$14.52 billion to US$14.57 billion, growth of 26% to 27%. Operating margin is expected to be 29.5%, non-GAAP diluted earnings per share of US$4.26 to US$4.30, fully diluted share count of 844 million to 847 million shares, and adjusted free cash flow margin of 38%. I'll now turn the call back to Hamza for the Q&A session. Below is the Q&A session of the earnings call (AI-assisted summary and translation): Hamza Fodderwala, Senior Vice President and Head of Investor Relations, Palo Alto Networks: Alright, thank you, Dipak. Our first question comes from Saket Kalia of Barclays, followed by Brian Essex of JPMorgan. Saket Kalia, Analyst, Barclays: Alright, great. Hi everyone, thanks for taking my question and congratulations on such a strong start to the new fiscal year. Great to see. Nikesh, you've talked about that US$1 trillion of technical debt for some time. Given the recent outages, it seems more visceral now. Based on your recent conversations with customers, as this new wave of cyber threats is clearly more active, where will CEOs and CIOs focus modernization efforts in 2027? Nikesh Arora, Chairman and Chief Executive Officer, Palo Alto Networks: Saket, thanks for the question. First, we're seeing strong customer interest in the Frontier AI Defense service. Nearly every company is running penetration tests and red teaming on its own infrastructure to make sure it uses these new AI capabilities to test its defenses before attackers do. As I said, the challenge is that so many new vulnerabilities are discovered every day that enterprises simply cannot patch them all in time. I think some enterprises are still caught up in this short-term crisis. Others are beginning to think about how this will affect their long-term architecture. You can see that in the platform consolidation we discussed. We completed 170 platform consolidations this quarter, meaning customers are saying, "Listen, I can't manage all this fragmented infrastructure. How do I bring it together?" The challenge is that whatever approach is chosen today takes three to six months, maybe even nine months, to implement, depending on how complex and legacy the existing environment is. Everyone is starting to make changes, and we will see more enterprises do so for the foreseeable future. Saket Kalia, Analyst, Barclays: Very helpful, congrats again. Hamza Fodderwala, Senior Vice President and Head of Investor Relations, Palo Alto Networks: Alright, thank you, Saket. Next we'll go to Brian Essex of JPMorgan, followed by Rob Owens of Piper Sandler. Brian Essex, Analyst, JPMorgan: Hi, thank you, Hamza, and thanks for taking my question. Nikesh, I find this acquisition very interesting. What conversations with customers prompted your interest in sovereign AI? Can you talk about the specifics of those conversations and how you plan to deliver AI autonomy? Nikesh Arora, Chairman and Chief Executive Officer, Palo Alto Networks: Brian, great question. We started by building Prisma AIRS. It lets you see what AI assets exist in your enterprise, presenting them to you one by one. That step is easy, basically deterministic. Assets are discrete objects that can be found. Then we thought, can we bring all traffic together to a single enforcement point and block malicious activity? We mapped the common methods for attacking large language models, such as prompt injection, and built corresponding protections. That step also went well. After the Hugging Face incident, we realized that stopping AI misbehavior requires more than deterministic controls, because such controls cannot anticipate every action an agent might take or every deception it might attempt. We realized that "using AI to fight AI" cannot just be a slogan. We really do need to train a large language model that can interpret, understand and stop AI misbehavior. Along these lines, we began to think about how we could train our own model. We looked at multiple companies to see who could do it, compared capabilities across the market, and Applied Compute stood out. We later concluded that in training models for enterprise applications, we also need to train a supervisory AI model. This is almost like having AI build AI, and then building another AI to supervise the first AI. These two things cannot be done separately. If I don't understand how someone else's application is supposed to run, I can't step in and judge what it is doing. Because I didn't train that application. Imagine if you could train your own application and have the same team train a model to supervise it. I think that is the future of AI security. From this perspective, to do model supervision well, we must get into the model training business. This is the direction we are betting on, and I'm glad the Applied Compute team shares this view. We have basically finalized and announced the deal today, everyone is excited, and I think the opportunity is very large. In fact, while I was sitting here, a CEO texted me. He met with them this morning and is very excited. Their original concern was how they could entrust large enterprise applications to a small team of only 100 people. Now, with Palo Alto Networks behind them, they feel comfortable treating this as a long-term partnership. I'm very excited about this. I hope we've bet in the right direction this time. If successful, I think its potential will far exceed what our existing cybersecurity business can achieve, and it could be larger than our business today. Brian Essex, Analyst, JPMorgan: Great. I have a follow-up, but I'll save it for later. Thank you. Hamza Fodderwala, Senior Vice President and Head of Investor Relations, Palo Alto Networks: Thank you, Brian. Next we'll go to Rob Owens of Piper Sandler, followed by Meta Marshall of Morgan Stanley. Rob, are you online? Alright, let's go to the next one. We'll take Meta Marshall of Morgan Stanley, followed by Patrick Colville of Scotiabank. Meta Marshall, Analyst, Morgan Stanley: Alright, thanks for taking my question. Nikesh, I wanted to ask about the mistakes that occurred as customers rolled out these measures, where certain virtual patches or accelerated patching clearly caused more service disruptions. How do you think customers are responding? Are they slowing down? Are they taking some kind of phased approach? Are services teams more involved? How do you think the increase in outages will change the patch deployment process? Lee Klarich, Chief Product and Technology Officer, Palo Alto Networks: Thank you, Nikesh. And thanks for the question. Let me first distinguish between the two things you mentioned. Patching too fast creates challenges and affects system operation, and virtual patching is precisely the solution to those problems. Stepping back, AI has discovered a large number of vulnerabilities, triggering a new wave of patching. The operational challenge for enterprises is then how quickly they can patch all affected systems. This puts pressure on operations teams. Some of the outages we've seen recently were precisely because enterprises were eager to move fast when actually installing patches. Virtual patching allows enterprises to safely deploy temporary protections, buying operations teams time to install formal patches. Ultimately, we can provide the real-time protection needed as soon as a vulnerability is discovered, giving enterprises time to install patches. For example, our service now includes thousands of virtual patches. We expect that number to reach tens of thousands by the end of this calendar year, and it is growing very fast to keep pace with vulnerability discovery. Meta Marshall, Analyst, Morgan Stanley: Alright, thank you. Hamza Fodderwala, Senior Vice President and Head of Investor Relations, Palo Alto Networks: Thank you, Meta. Next we'll go to Patrick Colville of Scotiabank, followed by Shaul Eyal of TD Cowen. Patrick Colville, Analyst, Scotiabank: Thanks, Hamza. Nikesh, and maybe Dipak as well. I want to follow up on the practical impact of heightened cyber risk. In recent conversations with some of your largest customers, it was very clear that demand for Frontier AI Defense is extremely strong. I'd like to hear how broadly you think this product applies. Is it applicable to all mid-sized and large enterprises? Could this category gradually become a fairly large market? Or is this a pain point specific to November 2026 that drives spending now and may stabilize over time? Thank you. Nikesh Arora, Chairman and Chief Executive Officer, Palo Alto Networks: The concept of vulnerability management has existed for a long time and is itself a market. In the past, it was roughly a group of professionals discovering vulnerabilities, systematically organizing those discoveries, and selling vulnerability intelligence feeds to other organizations so they could check their own infrastructure. Over the past few months, as frontier large language models demonstrated these capabilities, enterprises have been able to incorporate vulnerability detection into their regular workflows, checking their own infrastructure and specific configurations without outsourcing. Unfortunately, malicious attackers have acquired the same capabilities. Every enterprise is now racing to harden its network perimeter and infrastructure to ensure it is not breached. Vendors are also racing to close potential vulnerabilities. Models have advanced quickly over the past six months, and the latest models can now discover far more vulnerabilities than six months ago. I don't think model capabilities have peaked, and I don't expect them to peak over the next six to nine months either, meaning better models will keep emerging during that period. Smarter models can run deeper simulations and identify paths attackers might take. I think for the next 12 months, or roughly 12 to 18 months, enterprises will need to continuously use frontier AI for testing. That is why we call the product "continuous frontier AI defense," and we believe this demand will not fade anytime soon. From an industry landscape perspective, I believe continuous frontier AI defense will ultimately become a capability that platform vendors provide to every enterprise. For us, it can be delivered through the XSIAM platform or through the perimeter security platform, because our platform must continuously inspect customer infrastructure, stress-test it, and gradually become self-healing, rather than expecting customers to find problems first and then come to us to discuss how to fix them. You will see vendors across the industry integrate this capability into their products, but I think it will take two years. Before then, customers will continue to want or need to use open-weight models, open-source models or frontier large language models to test their infrastructure. Hamza Fodderwala, Senior Vice President and Head of Investor Relations, Palo Alto Networks: Alright, thank you, Patrick. Next we'll go to Shaul Eyal of TD Cowen, followed by Fatima Boolani of Citi. Shaul? Shaul Eyal, Analyst, TD Cowen: I'm here. Good afternoon, everyone, and congratulations. Firewall orders grew 28% year over year over the past 12 months. How much of that came from AI-related demand and how much from the equipment refresh cycle? What factors can support continued growth? Nikesh, to put it another way, this is a cyclical versus structural question. I mean, which parts of demand are cyclical and which are structural. Nikesh Arora