Arm (ARM.US) Q2 Earnings Call: Confidence in AGI CPU Commercialization "Strongest to Date"; Data Center Royalties to Maintain Triple-Digit Growth for Years to Come

Arm (ARM.US) Q2 Earnings Call: Confidence in AGI CPU Commercialization "Strongest to Date"; Data Center Royalties to Maintain Triple-Digit Growth for Years to Come British chip design company Arm (ARM.US) released its fiscal 2027 second-quarter earnings report, for the period ended September 30, 2026, after the U.S. market close on Wednesday. The report showed Arm's Q2 revenue rose 27% year-over-year to $1.443 billion, beating analysts' expectations of $1.38 billion and marking the best second quarter in its history. This was Arm's fifth consecutive quarter of year-over-year revenue growth exceeding 20%. On profitability, adjusted net income was $566 million versus analysts' expectation of $514.2 million; adjusted operating income was $638 million versus analysts' expectation of $575.7 million; adjusted earnings per share were $0.52 versus analysts' expectation of $0.47. Driven by the full-scale explosion in artificial intelligence (AI) computing demand, Arm's licensing and other revenue for the quarter reached $773 million, up 50% year-over-year; royalty revenue was $670 million, up 8% year-over-year, with data center royalty revenue doubling year-over-year. The strong growth in licensing revenue mainly came from increased customer demand for Arm's intellectual property (IP), as well as the signing of multiple large licensing agreements and changes in the timing of revenue recognition. The company disclosed that annualized contract value (ACV) grew 12% year-over-year to $1.79 billion. Arm said that as customers integrate higher-performance CPUs, GPUs and NPUs into chip designs, system complexity and power constraints continue to rise, driving customers to increase their investment in Arm technology. It is understood that Arm's share in the data center market is rising rapidly. The share of servers based on the Arm architecture among leading hyperscale cloud providers has climbed from about 15% in calendar 2024 to more than 50% in calendar 2025, and the company expects that share to exceed 60% by the end of 2026. Royalty revenue mainly depends on actual shipments of chips that use Arm technology. Arm said royalty revenue growth this quarter was driven by increased adoption of higher-royalty-rate technologies such as the Armv9 architecture and Arm Compute Subsystems (CSS), as well as expanded deployment of data center chips, but was partly offset by the impact of rising memory prices on the smartphone market. This divergence also constitutes an important highlight of the quarter's results: customers' upfront investment in Arm technology is accelerating, but licensing revenue growth has not yet fully translated into synchronized royalty revenue growth. Arm management believes the strong momentum in the licensing business is an important leading indicator of future royalty revenue growth, and that royalty revenue is expected to accelerate as new designs enter mass production. However, the report also showed hidden concerns. Affected by memory price increases squeezing the smartphone market, phone-related business came under pressure. Free cash flow for the quarter was $110 million, down 73% year-over-year (free cash flow over the past 12 months was $1.096 billion, remaining stable). On a U.S. generally accepted accounting principles (GAAP) basis, the company's net income for the quarter was $130 million, down 45% year-over-year, mainly because the year-ago period included a one-time gain of $131 million from a business disposal. On guidance, Arm expects third-quarter revenue of $1.54 billion (plus or minus $50 million), slightly above the analysts' average estimate of $1.53 billion but below some analysts' forecast of $1.7 billion; it expects adjusted earnings per share of $0.56 (plus or minus $0.04). The core focus of this earnings call was the commercialization outlook for Arm's self-developed chip business, the Arm AGI CPU, and the explanation for royalty revenue coming in below expectations. CEO Rene Haas said demand has continued to strengthen since the Arm AGI CPU was launched in March, the company has secured purchase orders from multiple customers extending into calendar 2027, and has signed supply agreements with supply chain companies. "I can tell you today that confidence (in achieving the $2 billion target) is the strongest to date, absolutely stronger than 90 days ago," he said, with first commercial deliveries still on track for the fourth quarter of fiscal 2027; Meta is the lead customer, "but there will be multiple customers beyond Meta." Addressing questions about supply assurance, he stressed multi-year partnerships with Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR, Micron, Samsung, SK and other partners, viewing it as a "5-to-10-year endeavor." Regarding royalty revenue coming in below expectations, management attributed it to memory price increases suppressing mid- and low-end smartphone shipments, with edge AI royalties roughly flat year-over-year; however, data center royalties doubled year-over-year and physical AI royalties grew 50%, and the company expects data center royalties to maintain triple-digit growth for years to come. Licensing business hit a record, with long-term agreement signings accelerating. Licensing and other revenue in the second fiscal quarter was $773 million, a record for a second fiscal quarter and about $100 million above the company's own expectations, due to rising interest from customers seeking long-term licenses to Arm technology. During the quarter, it signed 3 new Arm Total Access (ATA) agreements (with total active ATA subscriptions exceeding 60, including a first-time signing by a major customer), and signed 4 new Compute Subsystem (CSS) licenses (two for data center chips and two for smartphones). Among these, the technology licensing and design services agreement with SoftBank contributed $209 million, up 17% year-over-year, and the company expects this revenue to remain near this level in the remaining quarters of the year. Annualized contract value (ACV) grew 12% year-over-year. CFO Jason Child called the licensing pipeline "surprisingly strong" and noted that licensing beats typically design wins and more royalties 12 to 24 months later. Royalty structure diverged, with data centers doubling and phones under pressure. Data center royalties doubled year-over-year, driven by volume ramps on Google Axion, AWS Graviton, Microsoft Cobalt and NVIDIA platforms; Rene Haas said that beyond share gains, he also sees a trend of head-node CPUs connecting accelerators shifting from x86 to ArmAlphabet Inc. Class C has announced that TPU v8 will switch to Axion for connectivity, and NVIDIA Vera Rubin also adopts the Arm architecture. Physical AI royalties grew 50% year-over-year, driven by advanced driver assistance and in-vehicle infotainment; the company also launched the Arm Total Design program for physical AI, bringing together more than 80 companies. Edge AI royalties were roughly flat year-over-year: memory price increases suppressed mid- and low-end smartphone shipments, but higher royalty rates from Armv9 (penetration already above 50%) and CSS partly offset this headwind. For next year, management expects smartphone industry shipments to be roughly flat (IDC forecasts about -1%), but with v9 and CSS penetration, the company's mobile business royalties are still expected to achieve double-digit-plus growth. In the quarter, partners shipped more than 10 billion Arm-based chips in a single quarter for the first time, an increase of 1.5 billion year-over-year. AGI CPU commercialization path, OCI to offer instances in 2027. Oracle Corporation Cloud Infrastructure (OCI) currently offers NVIDIA Vera CPU based on Arm, and plans to offer Arm AGI CPU instances in 2027; Super Micro Computer, Inc. (Supermicro), Lenovo and ASRock are building full racks for the product and deploying them to data centers such as OCI. Rene Haas said agentic workloads are inherently CPU-boundeach agent running a specific task typically occupies one CPU core in the form of a virtual machine, which will drive CPUs to evolve from 128 cores to 256 cores and beyond 500 cores, thereby raising average chip prices. CFO Jason Child added that the company's confidence in achieving up to $2 billion in AGI CPU revenue in fiscal 2027-2028 has strengthened; the original plan was to reach about $2 billion the following year and about $5 billion the year after that, with the next update to be given at next year's Arm Everywhere event; gross margin for the first-generation AGI CPU is expected to be in the 30% to 40% range, and the next-generation product is expected to move toward the market standard of above 50%. Responding to x86 competition. An analyst asked about AMD claiming at its earnings call that the Venice CPU offers the best performance per watt. Rene Haas responded that independent testing by Signal65 at the OCP Summit showed that the Arm AGI CPU delivers about twice the performance per rack of AMD EPYC 9755, and that the test compared against the generation before Venice, "we are very confident in these numbers." He also stressed that the differentiation of the Arm model lies in allowing customers to highly customize around their own systems"what end customers can do, standard products can never achieve." The following is the detailed content of Arm's second-quarter earnings call. I. Opening Remarks and Management Commentary Host: Hello everyone, thank you for your patience. Welcome to Arm's fiscal 2027 second-quarter webcast earnings call. Please note that today's call is being recorded. I will now hand the call over to today's first speakerIan Thornton, Vice President of Investor Relations. Please go ahead. Ian Thornton (VP of Investor Relations): Thank you very much, and welcome to our fiscal 2027 second-quarter earnings call. Joining us today are Arm CEO Rene Haas and CFO Jason Child. Today's call contains forward-looking information about the company and its financial performance. These statements represent our best judgment at present, but our business faces numerous risks and uncertainties, and actual results may differ materially. Important risk factors that could affect our business and future financial performance are described in our 20-F annual report filed with the U.S. Securities and Exchange Commission. Arm undertakes no obligation to update any forward-looking statements. We will also reference non-GAAP financial measures; reconciliations of these measures to the most directly comparable GAAP measures, as well as discussion of certain forward-looking non-GAAP measures that cannot be reconciled without unreasonable effort, are included in our shareholder letter and supplemental financial information. Earnings materials are available at investors.arm.com. I will now hand the call over to Rene. Rene Haas (CEO): Thank you Ian, and good afternoon everyone. Arm delivered a record second quarter: revenue reached $1,440 million, up 27% year-over-year, our fifth consecutive quarter of growth above 20%. Licensing and other revenue grew 50% to $773 million, driven by partners increasing their investment in the Arm compute platform. Strong licensing momentum is an important leading indicator of future royaltiesas these new designs enter mass production, we expect royalties to accelerate. Second-quarter royalty revenue grew 8% to $670 million, with data center royalties doubling year-over-year. Non-GAAP earnings per share grew 33% to $0.52, even as we continued to increase R&D investment. AI is driving unprecedented compute demand across cloud, endpoint devices and autonomous systems, and Arm is the compute platform of choice. Today, our business is driven by two high-growth areas: first, the IP businesspartners are increasing their investment in the Arm compute platform and deploying a record number of Arm-based chips; second, our new chip businessdemand for the Arm AGI CPU is extremely strong. Both rest on the same Arm compute platform and software ecosystem. We clearly see these two growth engines firing simultaneously in data centers: partners are expanding deployment of Arm-based infrastructure while demand for Arm's self-developed chips is strong. Oracle Corporation Cloud Infrastructure (OCI) is a prime example of the two businesses reinforcing each other. Today, OCI offers NVIDIA Vera CPU based on Arm; it also plans to offer Arm AGI CPU instances in 2027. Since the Arm AGI CPU launched in March, customer demand has continued to strengthen. We have secured purchase orders from multiple customers extending into calendar 2027, and first commercial deliveries remain on track for the fourth quarter of fiscal 2027. Over the past 90 days, with added supply assurance and a broader customer mix, our confidence in achieving up to $2 billion in Arm AGI CPU revenue in fiscal 2027 and 2028 has further strengthened. We remain focused on expanding capacity to meet growing customer demand. Power availability makes performance and energy efficiency even more important. Morgan Stanley forecasts a global data center power shortfall of 38 gigawatts by 2028. Arm is helping customers get more useful compute from existing power to address this constraint. At the OCP Summit, independent testing by Signal65 showed that the Arm AGI CPU delivers about twice the performance per rack of AMD EPYC 9755. Improvements like this can halve the projected global power shortfall. Our data center royalties doubled year-over-year, and Arm-based servers' share among leading hyperscalers has more than tripledfrom 15% in calendar 2024 to more than 50% in calendar 2025. Based on expected volume ramps on Google Axion, AWS Graviton, Microsoft Cobalt and NVIDIA platforms, we now expect that share to exceed 60% by the end of calendar 2026. Beyond data centers, Arm's role in edge AI is also rising as AI agents become more capable and more continuously online. Arm-based PCs now account for 20% of global shipments. Alphabet Inc. Class C recently launched Googlebooka new category of AI-native laptops built for Gemini intelligence, with Arm as the primary compute platform for many of the models. The NVIDIA RTX Spark systems now on the market further extend Arm into local AI, content creation and gaming. In smartphones, we recently launched CSS for Mobile 2, improving performance on agentic AI workloads by up to 70% through SME2. Our first AI-native Mali GPU can achieve up to 4x the frame rate at the same sustained power. These capabilities let more AI and richer experiences run directly on-device without sacrificing battery life. Arm is the platform of choice for physical AI: royalty revenue grew 50% year-over-year, driven by advanced driver assistance and in-vehicle infotainment. We also launched the Arm Total Design program for physical AI, bringing together more than 80 companies to accelerate development and deployment of complete physical AI systems. In the quarter, our partners shipped more than 10 billion Arm-based chips in a single quarter for the first time, an increase of 1.5 billion year-over-year, backed by a software ecosystem of more than 30 million developers. Demand for Arm technology has never been strongerAI is driving more complex chip designs and greater demand for energy-efficient compute. Our IP business generates licensing revenue when designs start and royalties when they enter production; now, the Arm AGI CPU, as a new direct chip business built on the same compute platform and software ecosystem, gives Arm two strong growth engines, which we expect to drive faster revenue growth in the future. I will now hand it over to Jason. Jason Child (CFO): Thank you Rene. We delivered another strong quarter: total revenue grew 27% year-over-year to $1,440 million, a record for a second fiscal quarter and the second-highest quarterly revenue in our history, reflecting continued business strength. Licensing and other revenue was $773 million, up 50% year-over-year, also a record for a second fiscal quarter and the second-highest ever. Licensing revenue came in about $100 million above our expectations, due to rising interest from customers seeking long-term commitments to license Arm technology. In the quarter, we signed 3 new Arm Total Access (ATA) agreements, with total active ATA subscriptions exceeding 60, including a major customer signing an ATA license for the first time. We also signed 4 new Compute Subsystem (CSS) licenses, two for data center chips and two for smartphones. For existing customers, these agreements expand and deepen their rights to access our future technology roadmap, laying an important foundation for Arm's future royalty growth. In addition, we see ATA and CSS becoming a path to faster adoption for new customerscustomers that previously had limited use of Arm technology, several of which signed ATA or CSS licenses this quarter. Of the $773 million in licensing revenue, our technology licensing and design services agreement with SoftBank contributed $209 million, up 17% year-over-year. We expect the quarterly run rate to remain near this level in the remaining quarters of the year. Annualized contract value (ACV) grew 12% year-over-year. As we have discussed, licensing revenue can fluctuate significantly quarter to quarter due to the timing and accounting of large agreements, so ACV remains an effective measure of underlying licensing trends. Now turning to royalties. Royalty revenue was $670 million, up 8% year-over-year, also our highest ever for Q2. Cloud AI remains the largest growth driver: data center royalties again doubled year-over-year, with Arm-based CPUs ramping in AI infrastructure and general-purpose cloud workloads. Physical AI royalty revenue grew 50% year-over-year, driven by increased adoption of Arm technology in automotive and Siasun Robot&Automation. Edge AI royalty revenue was broadly flat year-over-year. Higher memory prices continued to suppress smartphone shipments, especially mid- and low-end models. This headwind was partly offset by higher royalty rates from adoption of the latest Armv9 and CSS technologies. As smartphone market conditions normalize, we expect edge AI royalty revenue to reaccelerate. As Rene discussed, customer demand for the Arm AGI CPU remains very strong. Last quarter, we said our confidence in achieving more than $1 billion in revenue in fiscal 2027 and 2028 had strengthened. Over the past 90 days, we signed purchase orders with multiple customers and supply agreements with supply chain companies, which further strengthened our confidence in achieving up to $2 billion in revenue. We will provide an update when we report third-quarter resultsvisibility will be better then. Now turning to operating expenses and profitability. Non-GAAP operating expenses were $774 million, up 19% year-over-year, as we continued to expand engineering capacity and invest in next-generation platforms, compute subsystems and the Arm AGI CPU product family. Non-GAAP operating income was $638 million, up 37% year-over-year. Non-GAAP operating margin was 44.2%, up 300 basis points year-over-year, with revenue growth outpacing operating expense growth. Non-GAAP earnings per share were $0.52, up 33% year-over-year. Importantly, even at this high level of investment, we generated $1.1 billion in free cash flow over the past 12 months, giving us the ability to continue investing in long-term growth. Now turning to guidance. For the third fiscal quarter, we expect revenue of $1.54 billion, plus or minus $50 million. The midpoint implies year-over-year growth of about 24%. We expect licensing and other revenue to continue growing strongly, up slightly more than 50% year-over-year, supported by strong customer demand and pipeline visibility. On royalties, we expect continued strength in cloud AI and physical AI, offset by continued pressure in edge AI from weak smartphone demand and a tough year-over-year comparison. As a result, we expect royalty revenue to grow low double digits sequentially and low single digits year-over-year. We expect non-GAAP operating expenses of about $825 million and non-GAAP earnings per share of $0.56, plus or minus $0.04. Looking further out, we continue to see strong customer demand across all business lines, and combined with the expansion of our technology portfolio and deepening strategic customer relationships, we are confident in our long-term growth prospects. I will now hand the call back to the host for the Q&A session. II. Q&A Session Host: Our first question today comes from Gary Mobley of StoneX Group. Gary Mobley (StoneX Group): I think everyone's main focus is the Arm AGI CPU. You are clearly signaling revenue closer to $2 billion over the five quarters ending with fiscal 2028. I want to know what gives you that confidence, especially in locking in more supply with different partners? I assume most of the initial revenue will support Meta, but can you talk about whether you have signed binding contracts with other customers? Rene Haas (CEO): Okay, thanks Gary. I'll take this one, and Jason can add if I miss anything. On the ramp of this product, we manage supply and demand expectations carefully every quarter. Demand for the Arm AGI CPU is extremely strong overallthat's no secretand what really drives it is everything we're seeing in agentic workloads, where the CPU is central. Demand today is even stronger than what we talked about at the Arm Everywhere event in March. Since then, demand from multiple customersMeta is the lead customerhas also exceeded our expectations. That said, over the past several quarters we have been working very hard to lock in everything needed in the supply chain to meet this demandacross multiple areas: wafer supply, substrates, test equipment, memory and so on. We want to be fully confident when we signal to the outside world that "we can meet additional demand with supply over the coming quarters." So at Arm Everywhere day, we talked about $1 billion of visibility, and we quickly saw that become a number of about $2 billion. I can tell you today that confidence regarding locked-in supply is the strongest to date, absolutely stronger than 90 days agoand that covers all the areas mentioned above. But we didn't start this yesterday or last quarter; we've actually been working on it for several quarters, and we have very good partners supporting us. As you mentioned, Meta is the lead customer, but there will be multiple customers beyond Meta. Jason, anything to add? If not, thanks Gary for the question. Gary Mobley (StoneX Group): Can I just... Host: Please go ahead, Gary. Gary Mobley (StoneX Group): Sorry, I just wanted a quick follow-up. You gave enough information in the shareholder letter to calculate the year-over-year change in per-chip royalty rate, and it looks like it's trending up at least mid-single digits. I want to know whether this is mainly a mix-driven anomaly or whether it should be viewed as a long-term trend of per-chip royalty rate growth? Jason Child (CFO): Okay. There is a mix element, but also, we shipped 10 billion chips this quarter, so the average rate fluctuates by quarter. I'd say mobile average rates are definitely lower than servers and physical AI, because chip specifications and value differ, so that has some effect. I wouldn't recommend using the overall average across 10 billion chips as the best way to forecast the future. I think you should look at it split across the three business units: mobile growing fairly flat; servers growing very strong double digits, with core counts increasing and more CSS revenue flowing in this quarterthat trend will continue, which is exactly why we have high confidence in triple-digit growth (in data center royalties) over the coming years. Same with physical AI, which we said grew 50% year-over-year, and it will also provide some tailwind to overall royalty rates and overall royalty growth. But again, I'd recommend looking at it across these three categories rather than averaging everything together. Rene Haas (CEO): Gary, let me add one more granular point on data center growth. I think there are two drivers of royalties independent of the Arm AGI CPU: first, Arm gaining share, which I mentioned earlier and which we now think will reach 60% by year-end. That share gain comes both from shipment growth and from more design activity on platforms like Graviton, Cobalt and Axion driven by agentic AI. Beyond that, we also see head-node CPUs connecting accelerators shifting from x86 to Arm. We already know about Vera Rubin, and Alphabet Inc. Class C has also announced that TPU v8 will switch from x86 to Axion for connectivity. So back to Jason's point, this is why we have high confidence in triple-digit growth in data center royalties over the coming years. Host: The next question comes from Sebastien Naji of William Blair. Sebastien Naji (William Blair Research): I want to ask how Arm will beat x86 in AI data centers. At the OCP Summit, your agentic AI benchmark showed the AGI CPU delivers twice the performance per rack of x86 competitors. But last night on its earnings call, AMD said its Venice CPU offers the best performance per watt. Rene, can you talk about where you see the most meaningful opportunity for Arm CPUs versus x86 in data centers? Rene Haas (CEO): Okay. I'll answer from two angles. First, from the standpoint of Amazon.com, Inc., Alphabet Inc. Class C or Microsoft Corporationthey build custom CPUs for their own implementations, racks and systems. In these designs, especially when the end customer is designing the SoC, they have enormous freedom to innovate around their own systemsvarious trade-offs within the SoC around data movement, DPUs and storage. At that level, you almost have to measure performance per rack rather than per chip. And at the chip level, Arm not only gives them the most efficient CPU cores on the planet, but also lets them design the most efficient SoC and ultimately the most efficient system. We've seen how quickly these customers have shifted designs from standard x86 to Arm. As for the Arm AGI CPU benchmark we ranthat was done by Signal65 at OCPit compared against the generation before Venice. We are very confident in those numbers, and very confident in the choices customers make after head-to-head comparisons with that design. Venice is a good product, but we are equally confident in the demand for the Arm AGI CPU and our future roadmap. In summary, there are two dimensions: first, the energy efficiency, performance per dollar and performance per watt of the AGI CPU; second, at the customization level, what end customers can do, standard products can never achievethat is the beauty of our business model. Host: The next question comes from Krish Sankar of TD Cowen. Sreekrishnan Sankarnarayanan (TD Cowen Research): Rene or Jason, you launched the AGI CPU reference server design at OCP. I want to know, are you working with ODMs or server manufacturers? Is the target enterprise private cloud? Along those lines, regarding the fiscal 2031 target of $15 billion in AGI CPU revenue, what are the biggest drivers? Is it increasing core counts, or higher ASPs versus the first-generation product? Rene Haas (CEO): Okay, thanks for your question. On our partnersit's Super Micro Computer, Inc. (Supermicro), Lenovo and ASRock. They build full racks for us and then deploy them to data centers like OCI. They are our build partners. As for the revenue outlook, let Jason talk about where and when the specific numbers land. What is clear is that core counts on these SoCs will definitely accelerate. That is largely driven by agentic workloadsspecifically, take an agent running a specific task; it typically runs as a virtual machine on one CPU core, and software managing across single or multiple cores is quite complex, so running on a single core is the better choice. So do I see CPUs going from 128 cores to 256 cores and then beyond 500 cores? Absolutely yes. As you can imagine, chips with more cores will definitely have higher average selling prices. But I think we will see core counts move up faster, which will help grow revenue. Jason, anything to add on the numbers? Jason Child (CFO): Okay. We said at the Arm Everywhere event last March that by the end of fiscal 2027 and in fiscal 2028, we had $1 billion of visibility. Of course, we have now raised our confidence to achieving $2 billion. As for the year after that, we haven't updated it, but the original plan was about $2 billion (per year) the following year and about $5 billion the year after that. So if we succeed in achieving more than $2 billion, you can expect there may be upside to the numbers after that. We plan to provide updated targets at the next Arm Everywhere eventthe timing is still being set, likely next spring or the first half of the year. Rene Haas (CEO): Let me add one point: when we held Arm Everywhere day in March, the general market view was that this was a $60 billion market and wouldn't grow much. But our own analysis reached a different conclusion, and the number we gave that day was $100 billion by 2030. I think that number has since been raised by various partiesLisa (note: likely referring to AMD CEO Lisa Su) even said more than $200 billion. I think that's roughly right, and maybe even an underestimate, who knows? What is clear is that the compute demand of agentic AI tasks is CPU-bound. The more agentic workloads running, the more CPUs are needed. So this will drive higher-core-count devices, which is very favorable for usbecause at the core level, what matters is performance per watt, having the most efficient cores on the planet, and that is what we excel at. Host: The next question comes from Mehdi Hosseini of Susquehanna. Mehdi Hosseini (Susquehanna Financial Group Research): I have two follow-ups. Back to the updated $2 billion AGI chip revenue targetshould we assume any revenue contribution in the September or December quarter of this fiscal year (fiscal 2027)? Jason Child (CFO): Please assume our shipments begin in the fourth fiscal quarter. If there is some in the third fiscal quarter, we will let you know, but even if there is, it won't be much. We are really targeting the fourth fiscal quarter. Mehdi Hosseini (Susquehanna Financial Group Research): Okay. Just a quick follow-up: as a percentage of royalty revenue, is it fair to think data centers approach one-third (30% to 35%) this fiscal year? And potentially rise toward about 45% next fiscal year (fiscal 2028)? Rene Haas (CEO): Sorry, you're asking... Jason Child (CFO): As a percentage of total royalties? Rene Haas (CEO): Data centers as a percentage of total royalty revenue, is that the question? Mehdi Hosseini (Susquehanna Financial Group Research): Yes. Is it fair to assume about one-third in fiscal 2027, and then another step up toward 45% in fiscal 2028? Jason Child (CFO): OkayI mean, we look at the businesses separately. Edge AI royalties will be fairly flat this quarter, and we said we'd start to see some recovery in the fourth fiscal quarter. When modeling, you can think of it this way: edge business growing high single digits, data centers growing double digitswe said last year data centers were roughly close to about 20% of total royaltiesand physical AI growing in the range of about 20%-plus to 30%-plus. You should build your model along those lines. I haven't specifically calculated the percentage. I don't think data centers will reach one-third this year, but it will get there before long. I'm just not sure it will get there by year-end. Host: The next question comes from Simon Leopold of Raymond James. Simon Leopold (Raymond James Research): I want to ask about customer concentration in the cloud and AI portfoliocan you give us a better quantitative sense? For example, can we guess that about 60% of this royalty segment comes from 3 customers? Any quantitative information would help. Rene Haas (CEO): I can say thisand Jason can add: Amazon.com, Inc., Alphabet Inc. Class C and Microsoft Corporation are the largest contributors, and then NVIDIA is now ramping rapidly. These four are the customers contributing the vast majority of royalties within data centers. Host: The next question comes from Harlan Sur of JPMorgan. Harlan Sur (JPMorgan Research): You've had a strong start with your first-generation AGI CPU, and I believe customers are already looking forward to the next-generation platform. Can you update us on second-generation AGI progress? I recall the plan was to tape out around the end of this year or early next year, with volume in calendar 2028. Perhaps equally important: as you continue adding capabilities to the compute chip portfolio, does it make sense to add more networking and high-performance I/O capabilities? We increasingly see compute and networking tightly coupled, jointly determining system performance. Will the team add more networking capabilities to the portfolio? Rene Haas (CEO): Okay, thank you Harlan, all good questions. Without disclosing the precise timeline for follow-on chips, you can extrapolate from this market's roughly 18-to-24-month production cadence. The first generation is entering volume ramp, and you can work out the approximate launch timing for the second generation yourself. As for other things that could be added to the "toolbox" to improve performancewhether broadly connectivity, specific I/O, networking, or other aspects of the systemwe are absolutely looking at all of them. We develop these products with full-rack system performance in mind. The chip, the CPU, is an important part of that, but as you ask, how to move data into and out of the system is a major topic as important as any other area. We have a lot of work underway in all these areas. There is nothing to announce on today's call, but we are fully aware that this is an area where we can create even greater differentiation. Host: The next question comes from Lee Simpson of Morgan Stanley. Lee Simpson (Morgan Stanley Research): If I may, two quick questions. First for Jason: $773 million in licensing revenue is very impressive, congratulations to the team. You talked about Total Access and also CSS agreementsparticularly smartphones and data centers. Were there any architecture licenses in there? And can we view $770 million-plus as a new structural level? A follow-up for Rene: along the lines of the earlier questions, as higher-ASP, higher-core-count CPUs emerge in the future, will the dollar value per core also rise? In other words, will those associated capabilities mentioned in the previous questionnetworking, I/Opush up dollar value per core? Jason Child (CFO): Okay, thanks Lee. First on licensing. What we're mainly talking about now is ATA and CSS, which account for the vast majority of our future sales. Architecture licenses have become very, very rare. So you can assume it's mainly ATA and CSSthe absolute majority of future sales. As for the absolute amount, $773 million is indeed a great quarter. But if you look closely, I said it will go higher next quarter and step up again in the fourth fiscal quarter. We haven't talked about next yearwhether this will become the new normal every quarter, I don't want to say that. But I will say the pipeline is surprisingly strong, stronger than we expected at the start of the year. It's a bit like the past few years: there are deals in the pipeline that are not renewalsthey are new deals that are hard to predict in advance. We are in that same situation again. Interestingly, at the start of the year we thought this year would be mainly driven by royalties, and that licensing growth related to AI investment had largely passed. But the reality is that with AI, ongoing workload changes, and now everyone dealing with agentic workloads, demand for additional licenses, additional product variants and different technologies has not slowed at all. That is especially exciting. As you know, any licensing beatespecially ATA and CSSis highly likely to design wins and very likely into more royalties 12, 18, 24 months later. That lays a good foundation for next year and beyond. As for next year, I'll need a few more quarters before I can talk about it. But the pipeline does look very strong. Even with the memory issue creating a tough year-over-year comparison for mobilethose effects largely end in the third fiscal quarter we're now in, and starting January/February next year we begin to lap those headwinds and the year-over-year comparison becomes easier. Overall, we think the setup is very good for the rest of this year and into next year. Rene Haas (CEO): Let me add a point on pricing. This is a nuanced issue: since we launched the Arm AGI CPU, memory prices have clearly risen a lot; data center construction is also becoming more complexmemory prices, labor and so on. Given these external factors, we want to be restrained on pricing and not push too hard; but it's fair to say that on these larger, higher-core-count chips, we may have some pricing flexibility to use. We will obviously weigh all factors, but there should indeed be some upside. Host: The next question comes from Vivek Arya of Bank of America Corp Securities. Vivek Arya (Bank of America Corp Securities Research): One each. Jason, assuming smartphone shipments decline double digits again next year, do you think royalties would perform similarly to this year? Or are there other dynamics at play? Rene, on the AGI platform: assuming we accept its technical advantages, my question is whether Arm has enough ecosystem scale to lock in supply? For example, when AMD reported yesterday it raised its server CPU business by several billion dollars, which shows they have the scale to secure incremental supply. How do you respond to thatgiven your confidence in the product's technical advantages, does Arm have enough scale to secure supply and become a meaningful player in the CPU market? Rene Haas (CEO): Okay, thank you Vivek. I'll answer the second part of your question first, and then Jason can address the smartphone scenario you described. Your question is fair. My view is: the Arm AGI CPU is our first product. AMD has been doing this for many years, has many SKUs, buys large amounts of capacity from Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR, and has deep relationships built over decades. This is our first product. But then again, this is not our first time working with partners like Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR, nor our first time working with memory makers like Micron, Samsung and SKthey have known us for years, even decades, and we have been sharing our long-term plans with them very seriously. I think supply will be tight across the industry. As Jason mentioned, we will talk more about the outlook in the coming quarters. But we will work very, very hard to lock in the supply needed to meet expansion demand, because we think that demand is very, very, very large. Our partners know this too. We have been communicating with them not on a weekly or monthly basis, but for possibly more than a yearsince we began planning this platform. We view it as a long-term endeavor, not a one-year single project, but a 5-to-10-year endeavor that requires partnerships across the entire