ARM Holdings recently held its earnings conference call for the third quarter of fiscal year 2026. The company stated that its data center revenue is currently growing at a significantly faster rate than its other business segments. It is estimated that this segment now accounts for over 15% of total revenue and is approaching 20%. Within the next two to three years, the scale of the data center business is expected to match or even surpass that of the smartphone business, which currently represents approximately 40% to 45% of total revenue.

The company also noted that its research and development spending is currently growing faster than its revenue. In the short term, from the fourth quarter into the first quarter of next year, the increase in expenses is expected to be similar to the same period last year, maintaining roughly a low double-digit percentage increase sequentially. After the first quarter of next year, the growth rate of R&D spending is anticipated to slow compared to the current year.

Regarding performance guidance, the company's initial forecast for fiscal year 2026 was "at least 20%" growth. This has now been revised upward, with the median guidance reaching 22%, exceeding initial expectations. For fiscal year 2027, while the company is not providing formal full-year guidance yet, it believes that from a macro perspective, maintaining a 20% growth rate is very reasonable.

Addressing recent significant volatility in the secondary market for software stocks, ARM Holdings expressed that such short-term market fluctuations are common during periods of major technological transformation, as investors often become anxious about broad industry impacts. However, from the company's viewpoint, as a provider of intellectual property for physical chips, AI will not replace chips in the short term. On the contrary, the two are interdependent—any AI software ultimately relies on hardware to run. The deep integration of AI within enterprises is still in a very early stage. Even within ARM, while some AI has been introduced into systems like payroll, procurement, or SAP, it is far from being "transformative." This lag is primarily due to the high complexity of integrating large systems and altering software workflows. The company stated that the market is experiencing turbulence as it navigates the "final frontier" of this AI transformation, venturing into "uncharted waters." However, the core logic remains unchanged: global demand for computing power is immense, and providing that computing power is ARM's core mission. The company is more focused on the long-term opportunities in this marathon.

**Q&A Session**

**Q: What role do ARM's CPUs play in AI and cloud data centers, and how might this role evolve with the proliferation of AI Agents?** A: A significant shift is occurring in data centers, moving the focus from pure "training" to predominantly "inference." This evolution in workloads opens multiple technological pathways, with the rise of AI Agents being particularly crucial. Tasks such as when an Agent needs to interact with other Agents or manage specific workflows like service tickets are inherently well-suited for CPU processing due to CPUs' high energy efficiency, always-on capability, and extremely low response latency. We observe that data centers are already increasing CPU deployments to meet this demand. The key to solving these problems lies not only in CPU performance but also in the number of CPU cores. Given the strict power constraints within data centers, the energy efficiency of CPUs is paramount, creating a highly favorable tailwind for ARM. This trend is already evidenced: both hyperscale cloud providers and NVIDIA have significantly increased core counts in their latest-generation chips. We believe this trend of scaling up core counts will continue as AI workloads evolve further.

**Q: Regarding the resilience of FY27 royalty growth and potential risks from memory supply constraints impacting consumer electronics demand?** A: Regarding the impact of memory supply chain constraints, our view aligns with partners like MediaTek, anticipating a potential decline of around 15% in smartphone shipments next year. However, in-depth analysis reveals that manufacturers, when facing supply constraints, prioritize the premium and flagship markets. Since these segments are the primary application grounds for ARM's CSS and v9 architecture, which carry significantly higher royalty rates, this dynamic is actually beneficial for us. In contrast, supply chain pressure is mainly borne by the low-end market, which predominantly uses the v8 or older architectures and contributes minimal royalties. Our calculations indicate that even if total smartphone sales decline by 20% next year, the impact on our smartphone royalties would only be around 2% to 4%. For the entire group's business, the negative impact on total royalties would be merely 1% to 2%. A more critical support factor is that growth in cloud AI and infrastructure businesses continues to exceed expectations. This incremental growth is sufficient to offset potential risks from memory and the mobile segment. Therefore, we are very confident in next year's royalty revenue structure and are not concerned about substantive impacts from shipment fluctuations.

**Q: Is there a possibility that SoftBank might reduce its stake in ARM to raise funds for other investments, and what impact could this have on the company's share price?** A: There is much market speculation on this topic, even discussed on various forums, but I can share the outcome of our communication. I can quote him directly: he has absolutely no intention of selling a single share of ARM stock. This "not selling a single share" is meant absolutely, whether it's one, two, or three shares. He is extremely bullish on ARM's long-term prospects, aligning with my own stance, and plans to hold long-term. While there are many external rumors and articles about SoftBank needing to liquidate assets, based on my multiple direct conversations with him, I can clearly state that a stake reduction is not factual.

**Q: The company predicts a slowdown in royalty revenue growth. What are the specific trends behind this? Is it due to a high base from last year or other underlying factors?** A: Regarding next year's royalty revenue trend, the absolute amount remains robust. The potential slight impact of 1% to 2%, mentioned earlier due to memory shortages, is minor. The decline in the growth percentage is primarily due to a high base effect. Our performance last quarter exceeded expectations (originally expected 20% growth, actual growth was 27%, exceeding by approximately $30 million), and this quarter is expected to continue this strong momentum, objectively raising the comparable base. It is still too early to judge whether this stronger-than-expected growth momentum will fully carry over into next year. While the market热议 memory or even wafer supply shortages, compared to pure-play chip design companies, the substantive impact on ARM is relatively small. Current guidance remains around the absolute amount scale set at the beginning of the year. We will closely observe whether the recent strong growth momentum can be sustained and decide whether to revise expectations upward as next year's business progresses.

**Q: Revenue contribution from SoftBank increased from an expected $180 million to $200 million. What is the reason, and what is the future normalized level?** A: Regarding the growth contribution from SoftBank, the amount recognized last quarter was $178 million, which increased to approximately $200 million this quarter. This is not due to signing a new agreement but rather reflects the full quarterly impact of the agreement this quarter, whereas previous quarters did not cover a full three months. Looking ahead, we expect a quarterly contribution of $200 million to be the normalized run-rate for this business going forward. As for the data center business, management emphasized its strong doubling growth in the presentation. However, from a financial quantification perspective, the current focus is on confirming the rapid increase in its contribution proportion to overall revenue and the long-term cash flow stability brought by related agreements.

**Q: Can you quantify the specific amount of data center revenue?** A: For data center revenue, we typically provide detailed figures once a year. Looking back to the beginning of this year, this business's revenue share had just reached double digits. Given its current growth rate significantly outpaces other company segments, I estimate the share is now above 15% and moving towards 20%. From a longer-term perspective, as the CEO mentioned, within the next 2 to 3 years, you should see the scale of the data center business reach or even exceed that of the smartphone business. Currently, the smartphone business accounts for about 40% to 45% of total business, meaning the data center segment is set to become the company's most important growth engine and core revenue pillar.

**Q: Against the backdrop of declining smartphone sales, how will the higher royalties from the migration to the v9 architecture offset the volume loss?** A: Regarding v9 penetration in the smartphone market, the core logic lies in ARM's CSS strategy. In each smartphone refresh cycle, we launch a全新的 CSS product, and the Royalty Rates for each new generation typically achieve a year-on-year increase. Therefore, the smartphone market is currently undergoing a comprehensive transition to CSS, enabling us to gain stronger pricing power annually through year-on-year rate increases. In terms of specific financial impact, even if smartphone shipments experience a sharp 20% decline next year, the impact on our smartphone revenue would be at most between 4% and 6%. This already incorporates the higher per-unit Royalty Rates that take effect with new product shipments in the second half of the year, as stipulated in our contracts. This increase in "price" significantly smoothens the volatility in "volume."

**Q: Based on SoftBank's current AI roadmap and the approximately $200 million per quarter ARM receives from SoftBank for technology licensing and design services (NRE), can the market expect future collaboration on custom AI chips? What impact could this have on FY27?** A: Currently, we have no specific information to disclose externally. Regrettably, we cannot provide further details at this stage.

**Q: What is your view on ARM's current IP penetration rate in the AI data center semiconductor market, and how will this trend evolve over the next 3 to 5 years?** A: This is a very critical question. Over the next three years, the way data center chips are built will undergo a fundamental evolution. The current mainstream "traditional architecture" consists of CPUs and GPUs, each with distinct roles. However, as workloads shift towards "Agentic AI inference," CPUs will begin to take over many tasks previously handled by GPUs. This implies the market will require more CPU cores or more CPU-based custom chip solutions. Additionally, AI inference loads primarily consist of "pre-fill" and "decode" components. We will see innovative solutions targeting these specific stages. Simultaneously, this compute demand is migrating from data centers to smaller form-factor endpoint devices. In power-constrained "Physical AI" and edge device domains, the combination of IP and solutions will become more diverse. Since AI workloads will run on any hardware with computing capability in the future, and the vast majority of global computing platforms are already based on the ARM architecture, this massive installed base provides a significant opportunity for us to define the future compute landscape.

**Q: Regarding the progress of Compute Subsystems (CSS) licensing, its current share in royalty mix, and future outlook?** A: CSS business is progressing significantly, with 2 new license agreements added this quarter. Currently, 5 CSS-based chip solutions have achieved commercial shipment and are contributing royalty revenue, having a substantive impact on financial performance. The core driver for customers willing to pay higher licensing fees and royalties is that CSS can shorten chip design cycles by approximately half, an highly attractive value proposition in the competitive AI era. In terms of share, CSS's growth curve is very steep: last year its share in royalty revenue was just nearing 10%; this year it has steadily reached around 15%. Looking ahead 2 to 3 years, we expect this proportion could climb to over 50%. A key leading indicator is that all CSS customers whose contracts are expiring or facing product iterations have chosen to renew or upgrade to the next-generation CSS version. This fully demonstrates market recognition of the solution's system-level performance improvements and productivity optimizations. We expect this acceleration trend to continue as more customers adopt the solution to pursue faster time-to-market.

**Q: The company previously provided informal guidance of approximately 20% growth for FY26 and FY27. FY26 delivery is going well. Can management offer some preliminary outlook for the longer-term FY28?** A: For fiscal year 2026, as you mentioned, our initial guidance was "at least 20%," and we have now raised the median guidance to 22%, indeed exceeding the target. For fiscal year 2027, while we are not providing formal full-year guidance yet, from a macro perspective, maintaining a 20% growth rate is very reasonable, and we are not stepping back from this goal. As for fiscal year 2028, we have not outlined specific numbers yet. I suggest everyone stay tuned, as we are planning some potential new products and services. How these initiatives will specifically impact our financial performance is still under detailed internal calculation. We will provide updates on FY28 progress at a future point in time.

**Q: The guidance for next quarter's royalty growth is only "low double-digits." Is this already affected by memory supply constraints? Furthermore, facing current BOM cost pressures, are customers' willingness to adopt higher-priced CSS and v9 architecture declining?** A: Regarding the impact of CSS pricing on customer BOM costs, we currently see no negative signs whatsoever. The core logic is that the value derived from accelerating time-to-market far outweighs minor customer cost adjustments. As chip manufacturing advances to 3nm and 2nm, design complexity surges, and design windows become extremely compressed. For customers, missing the first few months of the shipment window or encountering any delays is devastating to profits. Therefore, the customer's decision point is "how to ensure profitability through CSS," not merely the IP cost. The guidance for next quarter's royalty growth slowing to "low double-digits" is influenced only very minimally by memory supply; it is not the primary driver. This is mainly due to seasonal factors and the high base from the same period last year—MediaTek had an atypical chip launch node in the same period last year, placing us in an exceptionally strong comparable period.

**Q: Regarding the trend of operating expenses and R&D investment. Currently, R&D investment growth is higher than revenue growth. Will this situation continue in FY27, or will we see R&D growth start to slow relative to revenue growth?** A: It is premature to discuss full-year details, but I can share current expectations. In the short term, from Q4 into Q1 of next year, the expense increase is expected to be flat year-on-year, maintaining roughly a low double-digit sequential growth rate. However, after Q1 next year, we expect the growth rate of R&D investment to slow compared to this year. This year we indeed experienced significant phased investment growth, but we do not anticipate such a large step-up growth recurring next year. As next year's business progresses, we will provide more specific details. From a macro modeling perspective, next year's investment pace is expected to stabilize.

**Q: How do you view the recent severe volatility in the secondary market for the software sector? Besides driving demand, how does AI internally affect ARM's business?** A: Regarding short-term market volatility, this is normal during periods of major technological transformation; investors often become anxious about broad industry impacts. However, from our perspective, as a provider of IP for physical chips, AI will not replace chips in the short term. On the contrary, the two are interdependent—any AI software ultimately must rely on hardware. The deep application of AI within enterprises is actually still in a very early stage. Even within ARM, while some AI has been introduced into systems like payroll, procurement, or SAP, it is far from being "transformative." This lag is primarily due to the极高 complexity of integrating large systems and changing software workflows. We are at the "final frontier" of this transformation, which promises unprecedented scale of productivity improvement, but everyone is striving to adapt. Look at the current scale of capital expenditures, for instance, Google's announcement of up to $180 billion in CapEx, which was almost the sum of several years of fab investments for the entire semiconductor industry in the past. We are in "uncharted waters," which explains the market's turbulence. But the core logic remains unchanged: global demand for computing power is immense, and providing that computing power is ARM's core mission. We focus more on the long-term opportunities in this marathon.

**Q: What is the impact of edge-side SRAM and new memory architectures on ARM's business? Also, regarding the transition from v8 to v9 architecture, what is the pace and forward-looking plan for ARM's power efficiency improvements?** A: Power efficiency is a core focus for ARM 24/7 because, as device sizes continue to shrink, battery life and thermal headroom are highly constraining physical limits. The challenge to power consumption is persistent and significant when we overlay AI computation onto devices that already need to drive displays, run applications, and perform voice recognition. Since ARM is already the de facto standard in the vast majority of mobile and edge platforms, we are in the best position to solve this "compute demand vs. power budget"矛盾, and this is where we invest significant effort. Regarding SRAM and new memory technologies, the CPU and memory have an inseparable symbiotic relationship; they must co-evolve in hardware design. We are deeply involved in research not limited to SRAM but also various alternative memory technologies to address the surging access demands of AI. From a broader perspective, the industry's greatest concern is not the absence of challenges, but the absence of "hard problems" to solve. Currently, every end application is being transformed by AI, and we believe these AI workloads will ultimately run on the ARM architecture. This extreme demand for compute power, energy efficiency, and memory innovation is driving our continued large-scale investment.