ARM(ARM.US)FY2026 Q3 Earnings Call: Data center business expected to rival smartphone business in size within the next 2 to 3 years

Recently, ARM (ARM.US) held a conference call to discuss its fiscal third quarter of 2026. The company stated that data center revenue is currently growing much faster than other business segments, accounting for over 15% of total revenue and approaching 20%. In the next 2 to 3 years, the data center business is expected to reach, or even surpass, the smartphone business. Currently, smartphones account for about 40% to 45% of the company’s total business.

ARM also indicated that R&D investment growth is outpacing revenue growth. In the short term, expenses from the fourth quarter through the first quarter of next year are expected to remain flat compared to the same period last year, with low double-digit quarter-over-quarter growth. After Q1 next year, R&D investment growth is expected to slow compared to this year.

For guidance, the company initially projected at least 20% growth for fiscal 2026, but has now raised the median guidance to 22%, exceeding expectations. For fiscal 2027, while a full-year official forecast is not yet provided, maintaining a 20% growth rate remains very reasonable from a macro perspective.

Regarding recent volatility in the secondary market software industry, ARM believes that short-term market fluctuations during major technological shifts are normal. Investors often feel anxious about broad industry impacts. From the company’s perspective, as a provider of IP for physical chips, AI will not replace chips in the short term; rather, they are “interdependent”—any AI software ultimately relies on hardware.

AI’s deep internal application within enterprises is still in its early stages. Even within ARM, AI has been introduced into payroll, procurement, or SAP systems, but not at a “transformational” level. This lag is mainly due to the high complexity of integrating large systems and changing software workflows.

The company states that at the “frontier” of this AI revolution, market turbulence is driven by the “unknown waters.” However, the core logic remains unchanged: global demand for computing power is enormous, and providing that computing power is ARM’s core mission. The company emphasizes long-term opportunities in this marathon.

Q&A Highlights

Q: What role does ARM’s CPU play in AI and cloud data centers? How will this role change with the proliferation of AI Agents?

A: Data centers are shifting significantly, moving from a focus on “training” to primarily “inference.” This evolution opens multiple technical paths, with AI Agents being particularly critical. When Agents need to interact with others or handle workflows like service tickets, these tasks are inherently well-suited for CPUs due to their high energy efficiency, always-on capability, and low latency.

We’ve observed that data centers are increasing CPU deployments to meet this demand. The key is not just CPU performance but also core count. Given strict power constraints within data centers, CPU energy efficiency is crucial—an advantage for ARM.

This trend is already evident: major cloud providers and NVIDIA’s latest chips have significantly increased core counts. We believe this core expansion will continue as AI workloads evolve.

Q: What is the elasticity of FY27 royalty growth, and what are the risks of demand decline due to storage supply constraints?

A: Regarding storage supply chain impacts, our view aligns with partners like MediaTek: next year, smartphone shipments may decline by about 15%. However, analysis shows manufacturers prioritize high-end and flagship markets (Premium/Flagship) when managing supply constraints. These markets generate higher royalty rates, which benefits us.

In contrast, lower-end markets, mainly using v8 or older architectures, contribute minimal royalties. Even if smartphone sales drop 20% next year, the impact on our smartphone royalties would be only about 2-4%. For the entire group, the negative effect on total royalties would be just 1-2%.

A key support is the continued strong growth in cloud AI and infrastructure, which offsets potential risks from storage and mobile markets. Therefore, we are confident in next year’s royalty revenue structure and are not worried about shipment fluctuations.

Q: Might SoftBank reduce its ARM stake to fund other investments? How would this affect the stock price?

A: There are many rumors and discussions about this, but I can share the outcome of our communication. I can directly quote: he has no intention of selling any ARM shares.

This “no sale” stance is absolute—whether it’s one, two, or three shares. He is extremely optimistic about ARM’s long-term prospects, sharing my view, and plans to hold long-term. Despite rumors and articles suggesting SoftBank needs liquidity, based on multiple direct conversations, I can confirm that a stake reduction is not happening.

Q: What are the specific trends behind the forecasted slowdown in royalty growth for FY26? Is it due to a high base last year or other deeper factors?

A: The absolute amount of royalty revenue next year remains solid. The impact of storage shortages might cause a slight 1-2% slowdown, but the main reason for the percentage decline is the high base effect. Last quarter, we exceeded expectations with 20% growth, actually reaching 27% (about $30 million above forecast), and this strong momentum is expected to continue this quarter, raising the comparison base.

Whether this exceptional growth will fully carry into next year is uncertain. While market chatter about storage and wafer shortages persists, their impact on ARM is relatively minor compared to pure chip design companies.

Our guidance remains near the initial absolute figures. We will monitor whether this strong growth persists and adjust expectations accordingly as the year progresses.

Q: SoftBank’s contribution increased from an expected $180 million to $200 million. Why, and what is the future normal level?

A: The increase from last quarter’s $178 million to about $200 million this quarter is not due to new agreements but reflects the full quarterly impact of existing agreements. Previous quarters did not cover the full three months.

Looking ahead, we expect a quarterly contribution of around $200 million to become the normal level for this business. Regarding data center, management emphasizes its doubling growth momentum, but financially, the focus remains on the rapid increase in its contribution to overall revenue and the long-term cash flow stability from related agreements.

Q: Can you quantify the specific amount of data center revenue?

A: We typically provide detailed data annually. Earlier this year, data center revenue accounted for a double-digit percentage of total revenue. Given its faster growth compared to other segments, we estimate it is now over 15% and approaching 20%.

Long-term, as the CEO mentioned, within 2-3 years, data center revenue could reach or surpass smartphone revenue, which currently accounts for about 40-45% of total revenue. This would make data centers the company’s primary growth engine and core income pillar.

Q: How will higher royalties from migrating to v9 architecture offset declining smartphone sales?

A: The key to v9’s penetration in smartphones is ARM’s CSS strategy. Each new smartphone generation introduces a new CSS product, with royalty rates typically increasing year-over-year. The industry is undergoing a comprehensive shift toward CSS, giving us stronger pricing power through rate increases.

Even if smartphone shipments decline 20% next year, the impact on revenue would be only 4-6%, as higher unit royalty rates from new products are already embedded in contracts. This “price” increase helps smooth out volume fluctuations.

Q: Considering SoftBank’s current AI roadmap and the ~$200 million quarterly licensing and design service fees (NRE) ARM receives from SoftBank, can we expect future collaboration on custom AI chips? How would this impact FY27?

A: We currently have no specific information to disclose. Unfortunately, we cannot provide further details at this stage.

Q: How do you view ARM’s IP penetration in the AI data center semiconductor market? How will this trend evolve over the next 3-5 years?

A: This is a critical question. Over the next three years, the construction of data center chips will fundamentally evolve. Currently, mainstream architectures are CPU and GPU, each serving their roles. But as workloads shift toward “Agentic AI inference,” CPUs will take over many tasks previously handled by GPUs, requiring more CPU cores or customized CPU-based solutions.

AI inference loads mainly consist of “pre-filling” and “decoding” stages, and we expect to see innovative solutions targeting these specific areas (similar to Groq’s approach). Additionally, this compute demand is migrating from data centers to smaller edge devices, where power constraints in “Physical AI” and edge computing will lead to more diverse IP and solution combinations.

Since AI workloads will run on any hardware with computing capability, and most global platforms are based on ARM architecture, this vast installed base offers huge opportunities to shape future compute paradigms.

Q: What is the progress of licensing for Compute Subsystems (CSS), its current share in royalty income, and future outlook?

A: CSS is progressing rapidly. This quarter, we added two new licensing agreements. Five CSS-based chip solutions are now commercialized and generating royalties, significantly impacting our financials. Customers are willing to pay higher licensing fees and royalties because CSS can cut chip design cycles by about half—an extremely attractive value in the competitive AI era.

In terms of share, CSS’s growth is steep: last year, it accounted for nearly 10% of royalty income; this year, it has risen to around 15%. Over the next 2-3 years, we expect this ratio to exceed 50%.

A key leading indicator is that all CSS customers with expiring contracts or product upgrades are renewing or upgrading to the next CSS version, demonstrating market recognition of the performance and productivity benefits. We anticipate this accelerating as more customers adopt CSS to speed up their product launches.

Q: Regarding the informal guidance of about 20% growth for FY26 and FY27, with FY26 performing well, can management provide some initial outlook for FY28?

A: For FY26, as mentioned, our initial guidance was “at least 20%”, now raised to 22%, exceeding expectations.

For FY27, while we do not yet provide a full-year official forecast, maintaining a 20% growth rate remains very reasonable from a macro perspective, and we do not intend to lower this target.

As for FY28, we have not yet outlined specific figures. I suggest staying tuned, as we are planning some potential new products and services. How these will impact our financials is still under detailed assessment. We will update everyone on FY28’s outlook at an appropriate future time.

Q: The next quarter’s royalty growth guidance is only “low double digits.” Has this been affected by storage supply constraints? Additionally, given current BOM cost pressures, are customers less willing to adopt high-priced CSS and v9 architectures?

A: Regarding the impact of CSS pricing on customer BOM costs, we see no negative signs. The value of accelerating product time-to-market far outweighs minor cost adjustments. As chip manufacturing advances to 3nm and 2nm nodes, design complexity increases sharply, and design windows are extremely compressed. Missing initial shipment windows or delays can be devastating for profits. Therefore, customers focus on “how to ensure profit margins through CSS,” not just IP costs.

The forecast of “low double-digit” royalty growth next quarter is not mainly driven by storage supply issues. It is influenced by seasonal factors and a high base last year—MediaTek had an atypical chip launch last year, creating an unusually strong comparison period.

Q: What are the trends in operating expenses and R&D investments? Will the current R&D growth outpace revenue in FY27, or will R&D growth start to slow relative to revenue?

A: It’s too early to specify full-year details, but I can share current expectations. In the short term, expenses from Q4 this year through Q1 next year are expected to grow at the same pace as last year, roughly low double digits quarter-over-quarter.

After Q1 next year, we anticipate R&D growth will slow compared to this year. We experienced significant phased investment increases this year, but such large jumps are unlikely again next year. As the business advances, we will provide more detailed updates. Overall, the investment pace is expected to stabilize.

Q: How do you view recent volatility in the software sector? Besides demand drivers, how is AI internally affecting ARM’s business?

A: Market fluctuations during major technological shifts are normal, and investors often feel anxious about broad industry impacts. From our perspective, as a provider of IP for physical chips, AI will not replace chips in the short term; rather, they are “interdependent”—all AI software ultimately depends on hardware.

AI’s deep enterprise applications are still in early stages. Even within ARM, AI has been introduced into payroll, procurement, or SAP systems, but not at a transformational level. The lag is mainly due to the high complexity of integrating large systems and changing workflows.

We are at the “ultimate frontier” of this revolution, which offers unprecedented productivity gains but also requires adaptation. For example, Google’s announced CapEx of $180 billion is nearly the total wafer fab investment of the entire semiconductor industry in previous years. This “unknown waters” explains market turbulence. The core logic remains: global demand for computing power is huge, and ARM’s mission is to supply that power—long-term opportunities are substantial.

Q: How will edge SRAM and new storage architectures impact ARM? Also, from v8 to v9, what is ARM’s pace and vision for improving power efficiency?

A: Power efficiency is a key focus for ARM 24/7. As device sizes shrink, battery life and heat dissipation become critical physical constraints. When AI compute is added to devices already running displays, applications, and voice recognition, power challenges intensify. Since ARM is the de facto standard in most mobile and edge platforms, we are well-positioned to address the “compute vs. power” dilemma, which is a major investment focus.

Regarding SRAM and new storage tech, CPU and storage are inseparable. Hardware design must evolve both together. We are deeply involved in research on SRAM and alternative storage solutions to meet AI’s increasing access demands.

From a broader perspective, the industry’s biggest challenge is not the absence of challenges but the absence of solvable problems. Every terminal application is being transformed by AI, and we believe these workloads will ultimately run on ARM architecture. The extreme demands for compute, energy efficiency, and storage drive our ongoing large-scale investments.