Arm Holdings plc (ARM) - Deep Dive Research Report

Prepared: April 30, 2026

SECTION 1: WHAT THE COMPANY DOES

Arm Holdings designs processor architectures and licenses those designs to companies that make chips. It does not build a single factory, fab a single wafer, or assemble a single circuit board. Instead, it creates the intellectual property - the blueprints, the instruction sets, the verified hardware designs - that allow everyone from Apple to Amazon to Qualcomm to build their own semiconductors. Every chip that ships containing Arm technology pays Arm a royalty, typically a small percentage of the chip's selling price. That royalty, multiplied across more than 30 billion chips shipped per year across smartphones, laptops, servers, cars, and smart devices, is how the company earns its living.

The Founding Story

The story of Arm begins in a turkey barn in Cambridgeshire, England in 1990. Arm Holdings was founded as Advanced RISC Machines Ltd - a joint venture between three parties: Acorn Computers (the British microcomputer maker behind the BBC Micro), Apple Computer (which urgently needed a power-efficient processor for its Newton personal digital assistant), and VLSI Technology (a chipmaker that would manufacture the initial designs). Twelve founding engineers who had previously worked on Acorn's own processor designs relocated to a converted farm building in Swaffham Bulbeck, and set about designing processors that could compute on battery power.

The origin matters because it explains the core principle that has defined everything Arm has done since: energy efficiency above all else. In 1990, computing meant desktop PCs or mainframes plugged into walls. Acorn and Apple wanted something radically different - a chip that could run for hours on batteries, inside a device people could carry. The constraint was not raw performance; it was power consumption. That constraint forced the founding team to build from the ground up around the RISC (Reduced Instruction Set Computing) philosophy - simpler instructions, executed faster per watt.

Arm went public on the London Stock Exchange in 1998, was taken private by SoftBank in 2016 for $32 billion (the largest-ever acquisition of a European technology company at the time), and re-listed on Nasdaq in September 2023, pricing at $51 per share and raising approximately $4.87 billion. SoftBank retained approximately 90% ownership following the IPO, making Arm one of the most unusually concentrated public companies of its size.

The Core Value Proposition

Arm sits at a unique position in the technology supply chain. It is the only company that has built a single processor architecture that runs across virtually every computing category - from microcontrollers in industrial sensors drawing micro-amperes of current, to the M-series processors inside Apple MacBooks, to 192-core server CPUs inside Amazon data centers. That universality is not accidental. It is the product of three decades of deliberate ecosystem cultivation.

The value proposition for a chip designer licensing from Arm is fundamentally about reducing risk and time-to-market. Designing a processor architecture from scratch takes years and hundreds of millions of dollars in R&D. An alternative open-source architecture like RISC-V provides the instruction set for free but none of the surrounding software, tools, libraries, operating system support, or compiler optimization that decades of Arm deployment have accumulated. When a company licenses an Arm core, it gets not just the hardware design but access to a software ecosystem where millions of developers already write code, where every major operating system (Android, iOS, Windows, Linux) is already optimized, and where security certifications and supply chain qualifications already exist. That ecosystem moat - the accumulated software, tools, and developer knowledge - is what Arm has spent 35 years building and what any competitor would need to spend decades replicating.

How the Business Actually Works

When Arm develops a new processor design - say, the Cortex-X4 for high-end smartphones - it can license that design in two ways. Under an Architecture License, a customer like Apple or Qualcomm gets the instruction set specification and the freedom to design their own microarchitecture that is compatible with it. This is the highest-value license type: Apple's M-series and A-series chips, and Qualcomm's Oryon CPUs, are built this way. Under a Processor License (the more common variety), a customer gets Arm's ready-to-use processor design and integrates it directly into their chip. MediaTek and Samsung use many Arm processor licenses this way.

After the license is signed - which requires an upfront fee typically ranging from a few million to tens of millions of dollars - every chip shipped by that licensee that incorporates Arm technology generates a royalty. The royalty is a small percentage of the chip's average selling price, typically in the range of 1-2%, though this varies by product type. For a $500 smartphone chip, that might translate to $5-10 per chip. For a $10 IoT microcontroller, it might be $0.05.

The financial beauty of this model is its leverage. Arm's R&D expense creates a design once; that design generates royalties across billions of chips for a decade or more. The marginal cost of one additional royalty-bearing chip is essentially zero. This explains the company's extraordinary gross margin, which runs at approximately 97% - one of the highest of any company on earth.

A Concrete Example: What Arm Did for AWS

Amazon Web Services spent years paying Intel billions for x86 server CPUs. In 2018, AWS decided to build its own CPU - the Graviton series - using an Arm Architecture License. AWS's chip design team, starting from Arm's v8 architecture specification, created a custom processor optimized specifically for cloud workloads, with the right core counts, memory bandwidth, and security features for their data centers. The result was the Graviton line, with each generation dramatically outperforming Intel equivalents on performance-per-watt and performance-per-dollar. AWS told customers in 2023 that Graviton3 delivered 60% better energy efficiency than comparable x86 instances.

Every Graviton chip AWS ships pays Arm a royalty. Every time AWS runs an AI training job on a Graviton instance, Arm gets paid. AWS has gone from a customer of Intel to a customer of Arm, and the benefit for AWS - custom silicon optimized for their specific workloads - is so large that they, Microsoft (Cobalt), and Google (Axion) have each committed hundreds of engineers to building their own Arm-based server chips. Each of those programs is a direct revenue stream for Arm.

SECTION 2: BUSINESS SEGMENTS

Arm reports revenue under two financial line items: Royalty Revenue and License and Other Revenue. While these financial segments are straightforward, the company reorganized its operational structure in late 2025 into three customer-facing business units - Edge AI, Physical AI, and Cloud AI - reflecting where computing demand is migrating. Both lenses are needed to understand the business.

2.1 Royalty Revenue

Royalty revenue is the beating heart of Arm's economic engine. In FY2025, royalties generated approximately $2.17 billion, or 54% of total revenue. This stream is structurally recurring - it flows automatically from every chip shipped by every licensee that has signed an Arm agreement, with no additional sales effort required. Royalties grow in two ways: unit volume (more chips shipped) and average royalty rate (a richer mix of higher-value architectures and products).

The current royalty story is being driven by a powerful mix-shift dynamic. Arm's Armv9 architecture, which launched in commercial products in 2022, commands roughly double the royalty rate of its predecessor v8. In Q4 FY2025, Armv9 chips crossed 31% of total royalty revenue - the first quarter where Armv9 surpassed legacy v7 and older architectures. As the installed base of v8 chips (which dominate the current installed base) ages out and is replaced by v9, Arm collects a higher royalty on every replacement chip without needing to sell a single additional unit. This dynamic - architecture migration as a royalty rate escalator - has no direct equivalent in most other technology businesses.

On top of architecture migration, royalties are growing fastest in data centers. In Q3 FY2026 (ended December 2025), data center royalties doubled year-over-year. AWS Graviton5 (launched with 192 cores, doubled from Graviton4), NVIDIA's Vera CPU (88 Arm cores), Microsoft Cobalt 200 (132 cores), and Google Axion are all now shipping or in production. The data center royalty base that barely existed in FY2022 is now one of the fastest-growing segments Arm has ever seen.

Royalties span five end markets, each at a different stage of maturity:

Smartphones: Historically ~40-45% of total royalties, smartphones are Arm's largest single end market. Arm holds over 99% share in smartphone application processors - there is no meaningful competitive alternative. Every Android phone uses Arm; every iPhone uses Arm. The smartphone market grows units at sub-2% annually, but Arm's royalty revenue from smartphones has been growing at approximately 30% year-over-year through FY2026 due to the Armv9 migration and CSS (Compute Subsystems) adoption by flagship device makers.

Data Center/Cloud: The fastest-growing royalty segment. From near-zero five years ago, Arm-based chips are now shipping in approximately 50% of new server chips sent to hyperscalers. The royalty rate per data center chip is significantly higher than for a smartphone because the chips themselves sell for far more - a high-end server CPU sells for hundreds to thousands of dollars, making even a 1% royalty a substantial per-unit payment. Management guided in Q3 FY2026 that data center royalties could match or exceed smartphone royalties within 2-3 years.

Automotive: Approximately 10-15% of royalties today. Arm is embedded in infotainment systems, ADAS (Advanced Driver Assistance Systems), and increasingly in autonomous driving compute. Arm holds approximately 45% of the automotive semiconductor market. The royalty rate per chip is high because automotive chips are complex, safety-critical, and command high ASPs. Arm signed its first automotive CSS license in Q2 FY2026.

IoT and Embedded: The largest by unit volume but lowest per-unit royalty. Billions of Cortex-M microcontrollers ship annually in everything from smart meters to medical devices to industrial controllers. Low absolute royalties per chip but enormous volume creates a stable base.

Networking/Other: The smallest segment by royalty but with growing importance as AI drives massive investment in network infrastructure - switches, routers, DPUs (Data Processing Units).

2.2 License and Other Revenue

Licensing revenue is generated when companies pay upfront fees to gain access to Arm's latest IP, establish long-term licensing relationships, or expand the scope of their design rights. In FY2025, licensing generated approximately $1.84 billion, or 46% of total revenue. Unlike royalties (which flow from past design wins), licensing is a leading indicator - it reflects today's customer investment decisions, which will translate into royalty-bearing chips 2-5 years later.

Arm offers several licensing frameworks:

Arm Total Access (ATA): The premium tier. Customers pay a large upfront subscription for access to Arm's full IP portfolio - every processor core, every GPU, every NPU, every interface IP, and all future releases during the subscription term. This structure allows customers to experiment with new product categories without renegotiating individual licenses each time. As of Q4 FY2025, 44 companies had active ATA agreements.

Arm Flexible Access (AFA): A more accessible tier allowing design exploration before committing to royalty terms. Customers can prototype and develop using Arm IP, and negotiate royalty rates when a design moves toward production. 314 companies had active AFA agreements as of Q4 FY2025.

Compute Subsystems (CSS) Licenses: The highest-value licensing product Arm has ever created. CSS represents a significant strategic escalation - instead of licensing individual processor cores, Arm bundles cores, memory controllers, interconnects, power management IP, security features, and validated software into a pre-integrated subsystem. The customer gets a much faster path to a production chip (reportedly up to 12 months faster time-to-market), and Arm captures far more of the value in the chip than from a standalone core license. CSS licenses carry meaningfully higher royalty rates - management confirmed CSS delivers approximately double the royalty rate of a comparable Armv9 design. As of Q3 FY2026, Arm had 21 CSS licenses across 12 companies, with customers in mobile (Lumex CSS), data center (Neoverse CSS), and automotive (Zena CSS).

SoftBank/Strategic Agreements: A material but episodic portion of licensing revenue comes from Arm's parent company, SoftBank Group, for technology licensing and design services. In Q2 FY2026, SoftBank contributed $178 million to licensing revenue. These agreements are disclosed but can cause quarter-to-quarter volatility in the licensing line.

Licensing revenue is inherently lumpy - a few large agreements can swing a quarter significantly - which is why management has increasingly emphasized ACV (Annual Contract Value), a normalized measure of contracted licensing revenue, alongside reported licensing numbers.

2.3 Operational Business Units (Edge AI, Physical AI, Cloud AI)

In Q3 FY2026, Arm announced a reorganization into three customer-facing business units. This does not change the financial reporting structure (still Licensing and Royalty) but reflects where management is allocating engineering and commercial resources:

Edge AI covers smartphones, PCs, IoT, and wearables. This is the historical core - the Cortex-A/M/R families, Mali GPUs, and Ethos NPUs, plus the new CSS products for mobile (Lumex). Edge AI generates the majority of current royalty revenue. The unit's strategic thesis is that as AI inference migrates from the cloud to the device (for latency, privacy, and cost reasons), Arm's power-efficient architecture becomes more critical, not less.

Physical AI covers automotive, robotics, and industrial computing. This is Arm's highest-conviction growth bet for the decade beyond AI data centers. The unit includes the Zena Compute Subsystem for automotive, Arm's AE (Automotive Enhanced) technology series, and growing engagements with robotics companies. Key customer momentum: Rivian announced the first production vehicle using a custom Arm autonomy processor; Tesla Optimus uses an Arm-based AI processor.

Cloud AI covers data center infrastructure - server CPUs (Neoverse), CSS for infrastructure, networking chips, and now the AGI CPU. This is the unit that has generated the most investor excitement. It was the fastest-growing segment in FY2026, with royalties doubling year-over-year in Q3 FY2026, and management believes it has the potential to equal the smartphone royalty base within 2-3 years.

The three-unit structure signals where Arm believes the next decade of computing power will be consumed: at the edge on AI-capable devices, in physical systems (vehicles, robots) running real-time autonomy software, and in massive cloud clusters running AI inference.

SECTION 3: PRODUCTS AND BUSINESS DETAIL

The Cortex Family (Edge and Embedded Computing)

The Cortex family is Arm's workhorse - the product line that put Arm in essentially every smartphone, tablet, laptop, and embedded system on earth.

Cortex-A series are application processors - the high-performance cores that run operating systems like Android, iOS, and Linux. They range from mid-range Cortex-A55 (still shipping in hundreds of millions of affordable Android phones) to the premium Cortex-X series, which Arm develops specifically for flagship smartphone vendors. The Cortex-X4 inside current-generation Snapdragon, Exynos, and MediaTek Dimensity flagship SoCs is the pinnacle of Arm's mobile CPU performance. These designs are licensed to Qualcomm, Samsung, MediaTek, and others who integrate them into their SoC designs.

Cortex-M series are microcontrollers - tiny, ultra-low-power processors designed to run embedded software with minimal energy. A Cortex-M0+ core can run on microwatts and costs pennies to license. These ship in billions of units annually inside IoT sensors, medical devices, industrial controllers, automotive ECUs (Electronic Control Units), and smart home devices. The Cortex-M is the most widely deployed processor family in history by unit count.

Cortex-R series handle real-time control tasks requiring deterministic, guaranteed-latency responses - storage controllers, automotive safety systems, and communications baseband processors. This is a smaller volume market than A or M but commands premium licensing terms because of the safety-critical qualification requirements.

Neoverse Family (Data Center)

The Neoverse brand was launched in 2018 specifically to address the server and infrastructure market where x86 had dominated for 40 years. The product line includes multiple series optimized for different data center workloads:

Neoverse V series (V1, V2, V3): The performance-optimized tier designed for HPC (High-Performance Computing), AI inference, and enterprise workloads where raw throughput matters. AWS Graviton, Microsoft Cobalt, and Google Axion are all built on Neoverse architectures or CSS derivatives. The Neoverse V3 is the foundation for the recently announced AGI CPU.

Neoverse N series (N1, N2): Balanced performance-efficiency cores designed for cloud-native workloads, microservices, and general-purpose server applications. Many hyperscaler "capacity" instances run on N-series designs.

Neoverse E series: Efficiency-optimized for networking, telco, and edge infrastructure where density and power consumption per core are paramount.

As of Q2 FY2026, cumulative Neoverse CPU deployments surpassed 1 billion cores - a milestone that marks the definitive arrival of Arm in the data center.

Compute Subsystems (CSS)

CSS is the most strategically important product development in Arm's recent history. Traditional Arm licensing gives customers a validated processor core; the customer then has to design all the surrounding logic - memory subsystem, interconnect fabric, security engine, power management, and verification testbench - from scratch. This integration work can take 1-2 years and requires a large team of specialized engineers. It is the primary bottleneck slowing hyperscalers and chip startups from getting Arm-based silicon to market.

CSS eliminates that bottleneck by bundling everything into a pre-integrated, pre-verified subsystem. The customer receives not just CPU cores but a complete compute complex ready to connect to their custom accelerators, I/O, and memory. Arm guarantees the performance and area targets, provides the physical implementation collateral, and hands the customer a reference design that has already been taped out and validated on the intended process node.

The product lines within CSS are:

Lumex CSS (Mobile): Announced in Q2 FY2026, Lumex targets the premium smartphone market. OPPO and vivo committed to devices shipping that same quarter. CSS-based smartphones generate approximately double the royalty rate of a standalone Armv9-based design.

Neoverse Compute Subsystems (N2, V3): Pre-integrated data center subsystems targeting hyperscaler custom silicon. These are the products that AWS, Google, and Microsoft are building their next-generation chips on. The Neoverse CSS N2 targets balanced cloud performance; the V3 targets AI and HPC.

Zena CSS (Automotive): Arm's automotive CSS product, launched specifically for vehicles with advanced autonomy requirements. In Q2 FY2026, Arm signed its first automotive CSS license with "a leading global EV car manufacturer."

The CSS strategy has a compounding effect on Arm's financials. More CSS licenses mean more chips built from CSS blueprints. CSS chips carry double the royalty rate. CSS blueprints also include more Arm IP components, making each chip a larger licensing base. Management stated in Q3 FY2026 that CSS could represent over 50% of total royalties within 2 years, up from a low double-digit percentage in early 2026.

Mali GPU and Ethos NPU

Mali GPU is Arm's graphics processor IP, used primarily in smartphones and embedded devices where the chipmaker does not use a discrete GPU. Mali is the dominant GPU IP in mid-range and budget Android smartphones. It is not used by Apple (which has its own GPU) or Qualcomm (which uses its own Adreno GPU), but it is widely embedded in MediaTek SoCs and Samsung's mid-range Exynos chips. Mali generates licensing and royalty revenue as part of the broader Arm IP portfolio.

Ethos NPU is Arm's neural processing unit IP for on-device AI inference - running image recognition, voice commands, and other ML tasks directly on the device without cloud connectivity. As AI capabilities migrate from cloud to edge, Ethos becomes increasingly relevant for any chip designer building a device with local AI capabilities.

The AGI CPU (March 2026)

In March 2026, Arm announced the most significant strategic shift in the company's history: the launch of the Arm AGI CPU, a production silicon product designed and sold by Arm itself, not just licensed. This is the first time in Arm's 35-year history that it has entered the physical chip business.

The AGI CPU is a dual-chiplet design with 136 Neoverse V3 cores per processor, manufactured on TSMC's 3nm process. It targets the emerging category of agentic AI workloads - the orchestration layer in AI data centers that manages reasoning, memory, and agent coordination. Arm positions it specifically as the "CPU backbone" for AI infrastructure, designed to work alongside NVIDIA and other accelerators rather than replace them.

The technical specifications position it as directly competitive with high-end x86 server CPUs: 300-watt TDP, PCIe Gen6, CXL 3.0 support, 6GB/s memory bandwidth per core. Arm claims 2x performance-per-rack versus current x86 alternatives at a fraction of the power consumption.

The strategic rationale is significant. By entering silicon, Arm can capture the full economics of a chip sale (potentially hundreds of dollars per unit) rather than a 1-2% royalty. Lead partner is Meta, with more than 50 companies across hyperscale, silicon, memory, networking, and manufacturing committed as launch partners.

This move is not without risk to Arm's existing business - it puts Arm in direct competition with its largest licensees. Management has been careful to frame the AGI CPU as additive rather than cannibalistic, arguing that it serves a distinct workload category that hyperscalers want a purpose-built Arm-designed chip for, rather than building themselves. The market will validate or refute that framing over the next 2-3 years.

Arm China / Arm Technology (China) Co., Ltd.

Arm's operations in China are conducted through a joint venture known as Arm China (formally Anhui Zhongfu Xin'an Technology Co., Ltd.), in which Arm Holdings holds approximately 48% of the equity. Arm China operates as the exclusive distributor and licensor of Arm's technology within mainland China.

The JV structure was a condition of Arm's ability to operate in China, and it introduced significant complications. The most notable was a multi-year governance dispute with former Arm China CEO Allen Wu, who Arm attempted to remove in 2020 but who refused to step down, leading to a corporate standoff that was not fully resolved until 2023. The JV now operates under new leadership, but the structural complexity means Arm recognizes revenue from China at the JV level (based on licenses and royalties the JV collects from Chinese chip companies) rather than having full transparency into end-user customer activity.

China represented approximately $749 million, or 19% of Arm's total FY2025 revenue. Given US export restrictions on advanced semiconductor IP, Arm faces real constraints on which of its newest technologies it can license to Chinese customers, adding material uncertainty to this revenue line.

SECTION 4: CUSTOMERS

Who Buys

Arm's direct customers are semiconductor companies - the firms that design chips. The customer is never a phone user, never Amazon's retail business, and never a car driver. The customer is the engineering team at Qualcomm, Apple, MediaTek, Samsung, NVIDIA, Amazon's Annapurna Labs, Google's custom silicon team, or one of dozens of smaller chip designers building products for specific markets.

Within each of these companies, the decision to license Arm IP is made at the chip architecture level, typically by VP-level or above engineering leaders in collaboration with the company's product roadmap team. The decision is not made quarterly - it is made as part of a multi-year chip design cycle that typically runs 2-5 years from architecture decision to mass production. This means Arm's "sales cycle" is really measured in years, not months.

Named Major Customers

Apple: Arm's single most strategically important licensee. Apple holds an Architecture License, meaning it designs its own microarchitecture (the M-series for Mac, the A-series for iPhone/iPad) that is compatible with the Arm instruction set. Apple chips run at the absolute performance frontier of the ARM ecosystem. Apple does not disclose how much it pays Arm in royalties, but given that Apple ships ~230 million iPhones per year and its chips carry some of the highest ASPs in the industry, the implied royalty stream is substantial. Apple has been an Arm licensee since the founding joint venture in 1990.

Qualcomm: The dominant supplier of smartphone application processors for Android OEMs. Qualcomm's Snapdragon chips power Samsung Galaxy S flagships, most premium Android phones, and Windows-on-Arm PCs. Qualcomm also holds an Architecture License and designs the Oryon CPU microarchitecture. There is an ongoing legal dispute between Arm and Qualcomm regarding whether Qualcomm's acquisition of Nuvia (a chip startup that held its own Arm Architecture License) transferred the license rights to Qualcomm. This case has significant implications for how Architecture Licenses work industry-wide.

Amazon Web Services: Three generations of Graviton CPUs (now at Graviton5 with 192 cores) built on Neoverse CSS. AWS began aggressively pushing Graviton instances to customers as a cost and efficiency advantage. Every Graviton instance run by every AWS customer generates an Arm royalty.

Google: The Axion CPU, Google's first custom Arm-based server chip, launched in 2024 and delivers approximately 2x better price/performance than comparable x86 instances for certain workloads. Google is an architectural licensee.

Microsoft: Azure Cobalt 200, with 132 Arm cores, powers Azure services. Microsoft has committed to expanding Arm-based capacity across Azure.

NVIDIA: The Grace CPU uses 144 Arm Neoverse V2 cores and is part of NVIDIA's GB200 (Blackwell) system - the most in-demand piece of AI infrastructure hardware on earth today. NVIDIA's Vera CPU (88 Arm cores), the next-generation successor, is already shipping. Arm collects royalties on every Grace and Vera chip NVIDIA ships.

MediaTek, Samsung, Huawei (via Arm China): Volume-driving customers in the mobile segment. MediaTek's Dimensity series powers hundreds of millions of mid-to-premium Android phones. Samsung's Exynos chips are used in its own flagship devices in select markets.

Switching Costs

Switching costs in the Arm ecosystem are among the highest in all of enterprise technology. Consider what a company like Qualcomm would need to do to stop using Arm:

Design a completely new processor microarchitecture from scratch
Requalify every operating system, every compiler, every security library, and every application against the new architecture
Convince every app developer (22+ million developers build on Arm) to recompile or port their applications
Rebuild the entire software supply chain - debuggers, profilers, emulators, IDEs

Apple did most of this work over a decade to build its own microarchitecture on top of Arm's instruction set - and even Apple kept the Arm instruction set. The idea of a major chip company walking away from Arm's instruction set entirely is theoretically possible but practically requires a decade of parallel investment and would result in a product that runs no existing software.

For companies using Arm processor licenses (rather than architecture licenses), switching costs are even higher - the Arm-provided processor design is integrated into their SoC alongside custom accelerators, memory controllers, and I/O, requiring a full redesign of the chip if they switch to a different CPU design.

Contract Structures

Arm licenses are typically multi-year agreements. Total Access agreements run for multiple years (often 5+) and are renewed with updated terms. The royalty structure is governed by the license agreement and persists for the life of the product line - a chip design licensed in 2018 continues paying royalties on every unit shipped in 2026. This creates a long-duration revenue tail with no equivalent sales effort required.

The one area where Arm faces contract risk is Architecture Licenses - if a company with an architecture license creates a new microarchitecture and argues that it has "diverged" from Arm's ISA, they could potentially claim they owe reduced or no royalties. This is the heart of the Arm vs. Qualcomm/Nuvia dispute.

SECTION 5: COMPETITIVE LANDSCAPE

The Competitive Structure

Arm does not compete in the way that most technology companies compete. It does not compete with chip designers (its customers) for chip designs. It competes for the choice of processor architecture made by chip designers when planning a new product. The decision "should we build our chip on Arm, on RISC-V, or on something proprietary?" is the competitive battleground.

x86 (Intel and AMD)

Intel and AMD, the dominant x86 architecture providers, are Arm's primary competition in the server and PC markets. In mobile, x86 was effectively eliminated by the mid-2010s - Intel tried repeatedly to enter smartphones with Atom processors and gave up. In PCs, Arm-based laptops (Apple M-series, Qualcomm Snapdragon X Elite) have demonstrated that Arm can match or exceed x86 performance while dramatically reducing power consumption. Microsoft and Qualcomm are pushing Windows-on-Arm aggressively in the commercial PC market.

In servers, the dynamic is most dramatic. Arm's Neoverse-based chips went from essentially zero data center market share in 2019 to approximately 50% of new hyperscaler server chip shipments by 2025. Intel's share among hyperscalers who can design their own chips has collapsed - AWS, Google, and Microsoft are all reducing or eliminating Intel CPUs in favor of their own Arm-based designs. For enterprises buying off-the-shelf servers from Dell and HP, Intel and AMD still dominate. But the hyperscalers set the direction.

Intel's attempt to fight back includes investing in RISC-V, improving x86 power efficiency in its Lunar Lake and Arrow Lake architectures, and becoming a foundry that manufactures chips for others including some Arm-based designs. None of these moves directly threatens Arm's licensing model in the near term.

AMD does not compete with Arm as an IP licensor at all - AMD sells finished x86 CPUs and has no licensing business to speak of.

RISC-V

RISC-V is the most genuinely interesting competitive threat to Arm's long-term business model. RISC-V is an open-source instruction set architecture published under a free license. Any company can implement it without paying any licensing fee, ever.

The RISC-V ecosystem has grown rapidly - RISC-V International had over 4,500 members as of 2025, shipments are estimated at 20+ billion cumulative cores, and adoption is strongest in China (where the combination of free licensing and geopolitical desire to avoid US-controlled IP makes RISC-V very appealing), IoT/microcontrollers (where the Arm royalty on an ultra-cheap chip is disproportionate to its value), and specialized accelerators.

However, RISC-V has two serious limitations that make it a slow-moving threat to Arm's core business:

First, the software ecosystem. RISC-V has no Android support at the OS level, no iOS, no Windows, and limited high-performance compiler optimization compared to Arm. For any application that needs to run existing software, RISC-V is years behind Arm's ecosystem maturity.

Second, the IP quality gap. While RISC-V provides a free instruction set, it does not provide the high-quality processor implementations, GPU designs, NPU cores, security IP, memory controllers, and thousands of supporting IP blocks that Arm provides. A company using RISC-V for a complex SoC still needs to source most of this IP from elsewhere (Synopsys, Cadence, or custom design), erasing much of the cost savings from avoiding Arm royalties.

RISC-V is a real threat in IoT microcontrollers (where Cortex-M royalties are small but meaningful) and in China (where geopolitical pressure to avoid Arm is significant). In smartphones, servers, and premium devices, RISC-V is at minimum 5-10 years from challenging Arm.

Qualcomm's acquisition of Ventana Micro Systems (a high-performance RISC-V startup) in late 2025, reportedly providing architecture advantages for certain cloud-native workloads, is the most credible recent RISC-V competitive development in the data center space.

Imagination Technologies

Imagination Technologies licenses GPU IP and competes directly with Arm's Mali GPU in the mobile market. Imagination's PowerVR GPUs were used by Apple in iPhones until Apple switched to its own design in 2018. Today Imagination licenses to Chinese chip companies including Mediatek and various domestic Chinese fabless firms. Imagination is a small company relative to Arm and primarily competes in specific segments (GPU, AI accelerator IP) where Arm's Mali and Ethos compete. Arm wins on ecosystem breadth and the ability to bundle CPU+GPU+NPU under a single Total Access agreement.

Synopsys and Cadence (EDA + IP)

Synopsys and Cadence are primarily Electronic Design Automation (EDA) tool companies - they sell software that chip designers use to verify, simulate, and place-and-route chips. Both also sell semiconductor IP blocks including interface IP (PCIe, DDR, USB controllers) and some processor designs. Arm acquired Cadence's Artisan Foundation IP business in a reported transaction in April 2025 (strengthening its portfolio of physical IP). Synopsys and Cadence are not meaningfully competitive with Arm in the processor core market but compete in the broader IP licensing category.

CEVA

CEVA licenses DSP (Digital Signal Processor) cores and wireless connectivity IP - essentially the processing blocks used inside cellular modems and wireless chipsets. CEVA does not compete with Arm on general-purpose CPUs but competes in specific market segments like IoT connectivity and audio processing.

The Competitive Summary

Arm's dominant position rests on three structural advantages that are very difficult to replicate:

The software ecosystem: 22 million developers, every major OS, decades of compiler optimization
IP breadth: CPU + GPU + NPU + security + interface IP all pre-integrated and pre-validated
CSS: The Compute Subsystem products are a moat-within-a-moat - pre-integrated, validated subsystems that deliver faster time-to-market than anything a chip designer could build using alternative architectures

The genuine competitive threats are RISC-V in low-cost/China contexts and the theoretical possibility that major architecture licensees (Apple, Qualcomm) could transition to their own ISA over a very long time horizon. Neither is a near-term existential risk.

Competitor	Primary Market	Competitive Overlap	Arm's Advantage
x86 (Intel/AMD)	Servers, PCs	Direct in data center, PC	Power efficiency, custom silicon flexibility, 50% hyperscaler market share
RISC-V	IoT, China, specialized	CPU ISA across all segments	Software ecosystem, IP breadth, CSS pre-integration
Imagination Technologies	Mobile GPU IP	GPU IP licensing	Full-stack CPU+GPU+NPU bundling, Total Access
Synopsys/Cadence	Interface IP, EDA tools	Peripheral IP categories	CPU core dominance, ecosystem, CSS
CEVA	DSP, wireless	IoT, connectivity	General-purpose CPU reach

SECTION 6: INDUSTRY

What Drives Demand

Arm's revenue is a derivative of two things: how many chips get designed and manufactured in the world, and what those chips are worth. The semiconductor industry as a whole grows at roughly GDP+5% over the long run - chip content per device increases, and device penetration expands into new categories. But Arm's revenue growth can dramatically exceed industry growth when three specific conditions align:

Architecture migration (v8 to v9 doubles the royalty rate)
Market expansion (Arm entering data centers, automotive AI, PCs)
Product mix shift (CSS adoption amplifying per-chip economics)

All three are operating simultaneously in FY2025-FY2026.

Industry Size

The global semiconductor IP licensing market was valued at approximately $8.5-9.2 billion in 2026, growing at approximately 11-12% CAGR, and is projected to reach $13-19 billion by 2030-2032 depending on the source. Arm holds approximately 41% of this market by revenue, according to multiple market research firms, giving it a dominant position in its category.

The broader semiconductor market - in which Arm's IP is embedded but which it does not directly capture - was approximately $600+ billion in 2024 and is expected to approach $1 trillion by the end of the decade, driven primarily by AI compute infrastructure.

The AI Infrastructure Supercycle

The proximate driver of Arm's current growth acceleration is the AI infrastructure buildout. AI models require massive amounts of compute for training and inference, driving unprecedented capital investment in data center hardware by hyperscalers (AWS, Google, Microsoft, Meta) and sovereign AI initiatives. The Stargate initiative (a joint US government-private sector AI infrastructure program with projected investment in the hundreds of billions of dollars) explicitly names Arm as a critical infrastructure component.

What makes this tailwind particularly powerful for Arm is the energy efficiency argument. Running an AI data center at scale means consuming electricity on the scale of small cities. Every watt saved per server multiplied across millions of servers translates to hundreds of millions of dollars in electricity cost savings. Arm's architecture is inherently more power-efficient per computation than x86 for the workloads hyperscalers care about most. As power costs become the binding constraint on AI scaling, Arm's fundamental technical proposition grows more valuable.

Where Arm Sits in the Global Supply Chain

Arm's position is uniquely upstream. The supply chain for a smartphone runs roughly: Arm (IP) → Qualcomm/Apple/MediaTek (SoC designer) → TSMC/Samsung (fab) → assembly → device brand → carrier/retailer → consumer. Arm sits at the very top. It does not manufacture, does not assemble, and is not subject to supply chain disruptions that affect everyone downstream.

The implication: Arm is essentially immune to normal semiconductor supply chain risk (fab capacity, materials shortages, logistics). Its exposure is to design activity (will customers invest in new chip designs?) and to end-market demand (will chips actually ship in volume?). Both of those risks are real but operate on a 2-5 year lag from design activity.

Regulatory and Geopolitical Environment

The most significant regulatory overhang on Arm's business is US-China semiconductor export controls. US regulations restrict the export of advanced semiconductor IP to Chinese entities above certain performance thresholds. This affects Arm's ability to license its most advanced Neoverse designs to Chinese server chip companies. The restrictions have forced Arm to navigate what it can license through Arm China carefully, adding uncertainty to the ~19% of revenue from China.

SoftBank's attempt to sell Arm to NVIDIA in 2020 was blocked by regulators in the UK, US, EU, and China on competition grounds, demonstrating that Arm's dominant position in processor IP makes it a target for regulatory intervention in any M&A context.

The UK NSIA (National Security and Investment Act) and US CFIUS review processes would be triggered by any change-of-control transaction in Arm - SoftBank's ownership is unusual and potentially unstable over a multi-year horizon.

Cyclicality

Arm's business is not immune to semiconductor cycles, but it is significantly buffered by the royalty tail structure. When the chip industry enters a downcycle (as it did in 2023), shipment volumes decline and near-term royalties fall. But the license agreements already signed continue to generate royalty payments on whatever chips do ship, and the pipeline of design wins that will produce future royalties continues to develop. Arm does not have inventory risk, fab utilization risk, or working capital cycles in the way that a chip manufacturer does.

The licensing line is even less cyclical - customers sign multi-year Total Access agreements during both good and bad market conditions, driven by their 5-year chip roadmap planning rather than near-term market conditions.

SECTION 7: GROWTH TRIGGERS

All items are drawn directly from the four earnings calls cited. Concall dates: Q4 FY2025 (May 7 2025), Q1 FY2026 (Jul 30 2025), Q2 FY2026 (Nov 5 2025), Q3 FY2026 (Feb 4 2026)

CSS royalties expected to represent over 50% of total royalties within 2 years. In Q3 FY2026, management stated that CSS royalty contribution had grown from "just under double digits" to "well into double digits" quarter-over-quarter, and projected the 50%+ milestone. (Q3 FY2026, Feb 4 2026; first signaled Q2 FY2026)

"Our Compute Subsystems are beginning to take root in royalties. We went from just under double digits last quarter to well into double digits this quarter. And we think within two years that could be well upwards of 50% of our total royalties." — Rene Haas, Q3 FY2026 concall

Data center royalties expected to match or exceed smartphone royalties within 2-3 years. Data center royalties doubled year-over-year in Q3 FY2026. Arm's target is for Cloud AI to represent 40-45% of total royalties (equal to smartphones today), up from a low-teens percentage. (Q3 FY2026, Feb 4 2026)
Arm AGI CPU production silicon business targeting Meta and 50+ launch partners. Announced March 2026 (outside the concall window but referenced in Q3 FY2026 guidance as an upcoming catalyst), the AGI CPU marks Arm's entry into chip manufacturing with 136 Neoverse V3 cores on TSMC 3nm. This is a fundamentally new revenue stream with potentially far higher per-unit economics than royalties. (Q3 FY2026, Feb 4 2026 - guidance context)
Automotive CSS first license signed, automotive royalties expanding. In Q2 FY2026, Arm signed its first CSS license for automotive with "a leading global EV car manufacturer." This enables higher-royalty-rate automotive chips using the Zena CSS platform. (Q2 FY2026, Nov 5 2025)
Five CSS customers shipping CSS-based chips generating production royalties. As of Q3 FY2026, five companies had already shipped CSS-based chips generating royalties at double the rate of Armv9 standalone chips. CSS licenses grew from 13 (Q4 FY2025) to 19 (Q2 FY2026) to 21 (Q3 FY2026). (Repeated across Q1, Q2, Q3 FY2026)
Dream Big Semiconductor acquisition to accelerate networking/data center capability. In Q2 FY2026, Arm announced the acquisition of Dream Big Semiconductor to strengthen Ethernet and RDMA controller technology for data center networking - a segment growing rapidly as AI clusters require high-bandwidth interconnects. (Q2 FY2026, Nov 5 2025)
Lumex CSS mobile platform launched with immediate OPPO and vivo design wins. The Lumex smartphone CSS platform launched in Q2 FY2026 with immediate commitments from OPPO and vivo, representing volume Android OEM customers. Lumex delivers double the royalty of standalone Armv9 in the mobile segment. (Q2 FY2026, Nov 5 2025)
Arm PC expansion: new CSS licenses for PC segment. In Q1 FY2026, Arm signed a CSS license for PCs in addition to data center licenses, reflecting the Windows-on-Arm momentum behind Qualcomm's Snapdragon X Elite and anticipated follow-ons. (Q1 FY2026, Jul 30 2025)
Physical AI (robotics): Tesla Optimus and Rivian production vehicles on Arm. In Q3 FY2026, CEO Rene Haas cited Tesla Optimus (humanoid robot) using an Arm-based AI processor and Rivian's first production vehicle featuring a custom Arm autonomy processor as markers of the physical AI segment beginning to contribute royalties. (Q3 FY2026, Feb 4 2026)
Meta strategic partnership for AI efficiency across cloud and edge. Q2 FY2026 call noted an explicit strategic partnership with Meta to optimize AI workload efficiency, positioning Arm as the efficiency architecture of choice for Meta's massive AI infrastructure. (Q2 FY2026, Nov 5 2025)
Stargate: Five new Stargate sites announced in Q2 FY2026, each requiring Arm-based infrastructure. Arm cited Stargate AI infrastructure buildout as a "multi-hundred-billion-dollar opportunity" that directly benefits Arm's data center royalty line. (Q2 FY2026, Nov 5 2025)
Neoverse cumulative deployments surpassed 1 billion cores, growing installed base accelerating royalty compounding. Every billion cores deployed represents a growing base from which Arm collects royalties on future software stack optimizations and workload migrations. (Q2 FY2026, Nov 5 2025 - first cited)

Trigger	Timeline	Concall	Status
CSS > 50% of total royalties	Within 2 years	Q3 FY2026	Repeated (Q2, Q3)
Data center royalties = smartphone royalties	Within 2-3 years	Q3 FY2026	New at Q3
AGI CPU production silicon	H2 2026 production	Q3 FY2026	New at Q3
Automotive CSS: Zena licenses scaling	2026-2027	Q2 FY2026	New at Q2
Lumex CSS mobile OEM shipments	Shipping from Q2 FY2026	Q2 FY2026	New at Q2
PC CSS licenses signed	Ramping FY2026	Q1 FY2026	New at Q1
Physical AI royalties (Tesla, Rivian)	Current + accelerating	Q3 FY2026	New at Q3
Dream Big acquisition: networking IP	Integration in progress	Q2 FY2026	New at Q2

SECTION 8: KEY RISKS

1. China Revenue Concentration and Export Control Risk

China represents approximately 19-22% of Arm's total revenue, transacted through the Arm China JV in which Arm holds ~48% equity. The US export control regime restricts Arm from licensing its most advanced Neoverse and high-performance architecture IP to Chinese entities above certain performance thresholds. This caps the revenue Arm can extract from a market that historically grew at 17%+ annually.

The mechanism of harm is not just lost licensing revenue - it is that Chinese chip companies, cut off from Arm's most advanced IP, have strong incentive to invest in RISC-V alternatives. RISC-V designs compatible with Arm's instruction set are theoretically impossible (they would infringe Arm's architectural patents), but RISC-V as a clean-room alternative architecture with royalty-free licensing is exactly what Chinese policy wants to develop. If Chinese chipmakers successfully transition flagship products to RISC-V over the next decade, Arm's China revenue does not just stagnate - it potentially declines toward zero as license agreements expire and are not renewed.

2. SoftBank Overhang and Governance Risk

SoftBank owns approximately 90% of Arm. This creates several distinct risks:

SoftBank is a highly leveraged Japanese holding company with a history of dramatic investment cycles (Vision Fund boom-and-bust). If SoftBank faces a liquidity crisis, it could be forced to sell Arm shares on an accelerated timeline, creating substantial market price pressure and destabilizing the company's strategic direction.

Arm's CEO Rene Haas also recently assumed a role at SoftBank Group International, creating a potential conflict of interest where the CEO of a public company also serves a parent that owns 90% of it. Minority shareholders have limited recourse to challenge decisions that benefit SoftBank at the expense of broader shareholders.

Any strategic transaction involving SoftBank's stake - a secondary offering, a strategic sale, a merger - would trigger multi-jurisdictional regulatory review given Arm's monopolistic position in processor IP.

3. Armv9 / CSS Migration Headroom Is Not Infinite

The royalty rate escalation from v8 to v9 and from individual cores to CSS is the primary engine of Arm's current royalty revenue growth. But this escalation has a ceiling. Once the installed base of chips has fully migrated to Armv9/CSS, the mix-shift benefit disappears and royalty growth reverts to being driven by unit volume alone. Unit volume in mature markets (smartphones, PCs) grows at low single-digit percentages annually.

The timeline risk is that if data center and automotive royalties do not scale fast enough to replace the smartphone mix-shift benefit as that cycle matures, royalty growth could decelerate meaningfully in FY2027-FY2028. Management acknowledges this implicitly by highlighting data center and CSS as the next growth vectors - but those growth vectors need to scale on schedule.

4. Arm vs. Qualcomm / Architecture License Litigation

Arm filed suit against Qualcomm and its subsidiary Nuvia in 2022, arguing that when Qualcomm acquired Nuvia, Nuvia's Arm Architecture License did not automatically transfer. If Arm wins, Qualcomm could be required to pay significantly higher licensing fees retroactively and for future designs - or even stop shipping Snapdragon CPUs based on the Nuvia/Oryon architecture. If Qualcomm wins, it could embolden other Architecture Licensees to argue that their licenses transfer in M&A transactions, potentially weakening Arm's ability to renegotiate terms.

This case has sat in court for multiple years. Any outcome creates uncertainty for Arm's licensing revenue from one of its most important customers.

5. AGI CPU Cannibalization Risk

The March 2026 launch of the Arm AGI CPU - Arm's first production silicon product - places Arm in direct competition with hyperscaler custom silicon programs at AWS, Google, and Microsoft that are built on Arm architecture licenses. If a hyperscaler can buy a reference Arm-designed chip directly rather than paying engineers to design a custom chip, do they still need an Architecture License?

Management's answer is that the AGI CPU serves a distinct agentic AI workload category not served by existing designs, and that custom chips and the AGI CPU can coexist. This is plausible but not proven. The risk is that the AGI CPU business undercuts the licensing business rather than complementing it, leading to lower licensing revenue that isn't offset by silicon revenue at scale.

6. Share-Based Compensation Dilution

Arm's GAAP financials are significantly affected by share-based compensation (SBC). In FY2024, SBC was approximately $1.07 billion - more than total non-GAAP operating profit. While FY2024 was elevated due to IPO-related one-time grants, SBC remains structurally high as Arm competes for semiconductor engineering talent against Apple, NVIDIA, and Google.

SBC directly dilutes shareholders. Arm's share count has increased by approximately 1.82% in FY2025 alone, with minimal buyback activity (approximately $105 million in FY2024) to offset it. The arithmetic of compounding dilution at ~2% annually, with no meaningful buyback program, represents a quiet but real transfer of value from shareholders to employees.

"We don't have a buyback program in place and there are a number of factors that feed into that decision." — Jason Child, CFO (various calls)

7. Royalty Lag

Arm's royalty revenue lags chip design activity by 2-5 years. When Arm signs a license and helps a customer design a chip today, the royalties from that chip do not begin until the chip reaches mass production - which takes years. This lag means that any deterioration in the global chip design environment (capex cuts by hyperscalers, a severe economic downturn) would not immediately show in royalties, but would propagate into the royalty line 2-3 years later. Investors watching current royalty trends may be observing the outcome of design decisions made in 2021-2023.

SECTION 9: WALK THE TALK

Concall dates used: Q4 FY2025 (May 7, 2025), Q1 FY2026 (July 30, 2025), Q2 FY2026 (November 5, 2025), Q3 FY2026 (February 4, 2026). The most recent concall (Q3 FY2026, February 4, 2026) is 85 days before today's date (April 30, 2026), meeting the 90-day requirement.

From the Q4 FY2025 Call (May 7, 2025)

Management's most significant promise in Q4 FY2025 was the guidance for Q1 FY2026 revenue of $1.05 billion plus or minus $50 million. This was a notably cautious guidance, representing only 12% year-over-year growth against a prior year comparative of $925 million - the weakest sequential growth projection in several quarters.

"Given the uncertainty in the global trade and economic picture, we have lower visibility than is typical to start the year. As a result, we don't think it's prudent to issue full year guidance." — Jason Child, CFO, Q4 FY2025 concall

The decision to decline full-year guidance for FY2026, citing tariff uncertainty and macroeconomic volatility, was a clear exercise of conservatism. Management also flagged at this call that SoftBank's Masayoshi Son was "not interested in selling one share" of Arm, directly addressing an investor concern about the overhang.

From the Q1 FY2026 Call (July 30, 2025)

Arm reported Q1 FY2026 revenue of $1.053 billion, which landed at the midpoint of the guided $1.05 billion range (within $3 million of the midpoint). This was a precise execution against a guidance that many investors had thought might be conservative. Royalty revenue of $585 million was the real story - up 25% year-over-year and a record for the royalty line. CEO Rene Haas pointed directly to three drivers: Armv9 adoption, CSS double-royalty contribution, and data center growth.

"Our first generation of CSS is now in the market with five customers and is delivering double the royalty of Armv9. We signed three additional CSS licenses this quarter." — Rene Haas, Q1 FY2026 concall

Q2 guidance was set at $1.01-$1.11 billion - a wide range that acknowledged continued macro uncertainty. The midpoint of $1.06 billion was only modestly above Q1.

From the Q2 FY2026 Call (November 5, 2025)

Q2 FY2026 revenue of $1.14 billion obliterated the upper end of guidance ($1.11 billion) by $30 million. This was a significant beat driven primarily by licensing - $515 million in licensing revenue, up 56% year-over-year, powered by new ATA agreements and CSS expansion. CFO Jason Child directly acknowledged the beat: "We exceeded our guidance midpoint by $75 million."

The Q2 beat reversed the conservative narrative from Q4 FY2025. Management had guided cautiously; the business delivered clearly above the high end of the guided range. In hindsight, the conservatism in Q1 FY2026 guidance was appropriate given genuine tariff uncertainty, and the Q2 result suggested the underlying demand environment was stronger than management communicated in May 2025.

At Q2, management guided Q3 FY2026 revenue at $1.225 billion plus or minus $50 million.

From the Q3 FY2026 Call (February 4, 2026)

Q3 FY2026 revenue of $1.24 billion landed above the Q2 guidance midpoint of $1.225 billion by $15 million, and within the guided range. This was the fourth consecutive $1 billion-plus quarter. Data center royalties doubled year-over-year for the first time. CSS contribution grew "from just under double digits to well into double digits."

Management guided Q4 FY2026 at $1.47 billion - a significant sequential step-up implying approximately 18% year-over-year growth. This is the most aggressive single-quarter guidance Arm has issued.

"We expect revenue of $1.47 billion, plus or minus $50 million... with royalty growth in the low teens year over year and licensing growth in the high teens." — Jason Child, CFO, Q3 FY2026 concall

The Q4 FY2026 guidance would represent a ~$230 million sequential increase, partly reflecting a large SoftBank licensing payment expected in Q4. This follows a pattern where Q4 (the March quarter) historically sees elevated licensing revenue from SoftBank agreements and other large customers who sign in the fiscal year's final quarter.

Credibility Assessment

Looking across all four calls, Arm's management - led by CEO Rene Haas and CFO Jason Child - has demonstrated a consistent pattern of:

Guiding conservatively at the start of a period (Q4 FY2025's Q1 guide at $1.05 billion with wide range; declining full-year guidance)
Executing at or above the midpoint (Q1 actual: $1.053B; Q2 actual: $1.14B vs. guide of $1.11B; Q3 actual: $1.24B vs. guide of $1.225B)
Raising the narrative confidence each quarter as actual results validate the CSS/data center thesis

The one area where management's forward language warrants scrutiny is the long-duration predictions - "data center royalties matching smartphones in 2-3 years," "CSS above 50% of royalties in 2 years." These are aspirational targets, not quarterly guidance, and they have been repeated across multiple calls without a specific mechanism or timeline commitment. They represent the bull case thesis expressed as near-certainty, and the credibility of those predictions depends on execution over the next 8-12 quarters.

The decision to decline full-year FY2026 guidance in May 2025 due to tariff concerns was honest conservatism that the subsequent results vindicated. That type of intellectual honesty about uncertainty is a positive signal for management credibility.

What Was Guided	When	What Actually Happened
Q1 FY2026 revenue: $1.05B ±$50M	Q4 FY2025 call, May 7 2025	Q1 actual: $1.053B - hit midpoint
Q2 FY2026 revenue: $1.01-$1.11B	Q1 FY2026 call, July 30 2025	Q2 actual: $1.14B - exceeded upper end
Q3 FY2026 revenue: $1.225B ±$50M	Q2 FY2026 call, Nov 5 2025	Q3 actual: $1.24B - above midpoint
Q4 FY2026 revenue: $1.47B ±$50M	Q3 FY2026 call, Feb 4 2026	Not reported yet (reports May 6 2026)
Declined full-year FY2026 guidance	Q4 FY2025, citing tariff uncertainty	Prudent given subsequent volatility

SECTION 10: SHAREHOLDER FRIENDLINESS INDEX

Arm Holdings went public in September 2023 and has operated as a public company for fewer than three full fiscal years. Its shareholder return profile during this period is best described as minimal - the company has returned almost no capital to shareholders via dividends or buybacks, and share count has been increasing due to stock-based compensation.

Dividends: Arm pays no dividends and has made no indication it intends to do so. There have been no special dividends, no interim dividends, and no disclosed dividend policy. This is consistent with the company's growth-stage positioning - all available cash flow is directed toward R&D and strategic investment. Source: Arm's annual reports and investor relations disclosures; no dividend payments disclosed in any filing.

Share Buybacks: Arm's buyback activity has been token. In FY2024, approximately $105 million in buybacks were reported - a small fraction of the company's cash generation and far below the level needed to offset SBC dilution. CFO Jason Child has stated across multiple earnings calls that Arm does not have an active buyback program in place and cited "a number of factors" feeding into that decision. Given SoftBank's ~90% ownership concentration and the company's growth investment priorities, this stance is unlikely to change materially in the near term.

Share Dilution: The most honest accounting of shareholder capital allocation must acknowledge that Arm's employees and executives are being compensated partly through equity grants. In FY2024, total SBC was approximately $1.07 billion - elevated by IPO-related one-time grants that included a large RSU grant linked to the IPO event. In FY2025, SBC normalized but remained substantial. The net effect is that basic share count increased by approximately 1.82% in FY2025, and the diluted share count used for EPS calculations is meaningfully higher than the basic count. Arm's buybacks of ~$105 million do not come close to offsetting this dilution.

Capital Allocation Trajectory: The company has been investing heavily in R&D - the engineering headcount grew 17% year-over-year in FY2025, with the majority of new hires being engineers. The Dream Big Semiconductor acquisition (announced Q2 FY2026) and the AGI CPU silicon production program (announced March 2026) represent additional capital deployment into strategic expansion rather than shareholder returns.

Three-Year Summary Assessment: Arm has not returned meaningful capital to shareholders in the three years since IPO. The business model generates extraordinary gross margins and is beginning to generate strong operating cash flow. However, the capital priority order is: R&D investment, strategic M&A, working capital. Shareholders participate in value creation through share price appreciation rather than cash distributions. For investors expecting income or buyback-driven EPS accretion, Arm's capital allocation is a clear negative. For investors who believe the company is in a multi-year compounding growth phase where reinvestment creates more value than buybacks, the approach is defensible.

Source note: All shareholder return data sourced from Arm's FY2024 and FY2025 annual reports filed with the SEC, investor relations website, and concall disclosures. No dividend payments were found in any public Arm filing for FY2023-FY2025.

SECTION 11: SCENARIOS

Bull Case

In the bull case, the three vectors management has identified converge on schedule. CSS becomes the dominant design paradigm across all of Arm's markets - mobile, data center, automotive - within the 2-year window management has guided. Every major Android OEM ships CSS-based flagship phones by late 2026, doubling the royalty rate per handset relative to v8 legacy designs. Data center royalties hit the 40-45% of total revenue milestone as AWS Graviton5, Microsoft Cobalt, and Google Axion all scale aggressively through calendar 2026 and 2027, while the Stargate AI infrastructure buildout drives a genuine step-change in Arm-based server deployment.

The AGI CPU business defies the skeptics. Rather than cannibalizing the licensing model, it proves additive - hyperscalers use the Arm AGI CPU as a commodity orchestration layer alongside their custom silicon, generating a new revenue stream at silicon economics (hundreds of dollars per chip) rather than royalty economics (dollars per chip). The Physical AI business (automotive, robotics) begins generating royalties at scale as Rivian, Tesla Optimus, and a new wave of humanoid robotics manufacturers ramp production. Arm exits FY2027 with four meaningful royalty streams of roughly equal size - smartphones, data centers, CSS-amplified mobile/PC, and emerging physical AI.

In this world, Arm has successfully transitioned from a single-market (smartphone) royalty story into a genuinely diversified royalty and silicon company, with each of its three new business units generating material revenue and with CSS doubling per-chip economics across all of them.

Base Case

In the base case, Arm delivers approximately in line with current trajectory. CSS adoption continues but at a pace slightly slower than the "50% of royalties in 2 years" aspiration - it reaches 35-40% of royalties by FY2028. Data center royalties grow strongly but take 3-4 years rather than 2-3 to equal smartphones, partly because hyperscaler capex cycles are lumpy and partly because x86 retains meaningful presence in enterprise segments.

The AGI CPU generates initial revenues and validates the business model but faces the challenge of a company that has never managed a hardware product line navigating supply chain, customer support, and manufacturing relationships at scale. The transition to silicon is managed without fundamentally disrupting the licensing business.

China remains a managed risk rather than a crisis - the Arm China JV continues to operate, licensing to Chinese chip companies for consumer and automotive applications, while restrictions on the most advanced technology limit the upside. Full-year FY2026 (ending March 2026) comes in around $5.5-6 billion in revenue based on the current quarterly run-rate and Q4 guidance.

SoftBank gradually reduces its stake through secondary offerings over 2-3 years, which creates periodic stock price pressure but ultimately broadens the shareholder base and potentially removes an overhang.

Bear Case

In the bear case, several specific risks materialize in sequence. The Arm-Qualcomm litigation resolves unfavorably for Arm - either Arm loses the case (validating that Architecture Licenses transfer in M&A, weakening Arm's renegotiation leverage) or the dispute causes Qualcomm to begin a multi-year transition to proprietary architecture for next-generation Snapdragon designs. Qualcomm represents a substantial fraction of Arm's smartphone royalties - its potential exit from the Arm ecosystem is the single most damaging customer scenario Arm faces.

Simultaneously, US-China trade tensions escalate to the point where Arm must significantly curtail or terminate operations through Arm China, causing the ~$750 million per year China revenue to decline sharply over 18-24 months. Chinese chip companies accelerate RISC-V investment, and the ecosystem around RISC-V in China reaches a tipping point where domestic Chinese smartphone chips begin shipping on RISC-V architecture - first in budget phones, then gradually moving upmarket.

The AGI CPU launch disappoints. Hyperscalers who initially committed as launch partners find that the custom silicon programs they already have underway are more competitive than the reference Arm design, and AGI CPU orders are smaller than hoped. The market interprets this as evidence that Arm's entry into silicon was premature, and questions arise about whether the initiative has made relationships with licensees more cautious.

In this scenario, the CSS ramp also stalls - CSS adoption requires customers to redesign chips significantly, and in a downcycle where chip companies are cutting R&D budgets, the migration timeline slips. Royalty growth decelerates to high single digits. The gap between Arm's growth rate and its embedded valuation multiple becomes the defining investor concern.

Sources: