Arm Holdings Royalty Revenue From AI Chips Has Become the Primary Growth Driver and Compute Subsystem Licensing Is Expanding the Model
Arm Holdings reported royalty revenue of $1.1 billion in fiscal Q4 2026 (the quarter ended March 2026), a 46 percent year-over-year increase that reflected the accelerating shipment of AI chips — including Nvidia’s Blackwell B200 GPU family, Apple’s M4 processor series, Qualcomm’s Snapdragon X Elite, and Amazon’s Graviton 4 data center processor — that use ARM-designed processor cores and pay Arm a per-chip royalty each time a chip ships at commercial volume. Arm Holdings’ investor relations filings for FY2026 show total annual revenue reaching $4.7 billion (up 47 percent from FY2025’s $3.2 billion), with royalty revenue growing faster than licensing revenue for the second consecutive year — a mix shift that is commercially significant because royalties scale with semiconductor unit shipments and average selling prices without requiring Arm to add customers or renegotiate contracts. The royalty acceleration reflects two compounding factors: the transition from ARM v8 architecture to ARM v9 architecture (which carries a higher royalty rate per chip than v8) is now approximately 70 percent complete across major chip designs, and the average selling price of ARM-based chips is increasing as AI processing requirements push chip designers toward higher compute, higher memory bandwidth designs that carry larger royalty obligations. Arm does not disclose per-chip royalty rates publicly, but management has indicated that the blended royalty rate improvement from v8 to v9 is in the mid-single-digit percentage range as a fraction of chip ASP — a modest per-chip improvement that compounds into material revenue growth at the billions of chips that ARM-architecture designs ship annually. Qualcomm’s Snapdragon X commercial traction in AI PCs exemplifies the royalty revenue dynamic: each Snapdragon X Elite shipped generates an ARM v9 royalty payment at a higher rate than the Snapdragon 8 Gen series it replaces for comparable market segments, so Qualcomm’s AI PC market share growth directly expands Arm’s royalty revenue without requiring any change to the licensing arrangement between the two companies.
Arm’s Compute Sub-System (CSS) licensing program represents the structural innovation in Arm’s business model that has the most significant long-term implications for the company’s revenue per customer relationship. Traditional ARM licensing involves a chip designer licensing the ARM instruction set architecture (ISA) or a specific CPU IP core design, then integrating that core with other IP (GPU, memory controller, interconnect) to build a complete chip. Arm’s CSS program packages pre-integrated compute cluster designs — a CPU IP core, memory subsystem, interconnect, and coherency fabric that are already validated to work together — that chip designers can license as a complete module and integrate into a larger SoC without performing the architectural integration work themselves. CSS licensing carries a materially higher annual license fee than individual IP licensing, because the value delivered is not just the CPU IP but the engineering work of making a validated, production-ready compute cluster. Major semiconductor companies that have adopted CSS include MediaTek (for its Dimensity AI chiplets), Marvell Technology (for its data center interconnect and infrastructure processors), and several smartphone chip vendors that use CSS to accelerate their design cycles. The CSS model is Arm’s response to the competitive pressure from RISC-V — an open-source instruction set architecture that chip designers can use without royalty payments. RISC-V adoption has grown from negligible to approximately 3 to 4 percent of addressable embedded and microcontroller chip designs by 2026, but has not yet penetrated the high-performance computing segments (smartphones, data center CPUs, AI accelerators) where Arm earns the majority of its royalty revenue. TSMC’s 2nm N2 process node expansion is relevant to Arm’s royalty trajectory because TSMC N2 is being used for ARM v9-based designs from Apple (M5), Qualcomm (Snapdragon X successor), and MediaTek — each additional N2 tapeout at TSMC typically represents an ARM v9-based design that will generate royalties at the higher v9 rate.
How Arm’s Data Center Revenue Has Grown Beyond AWS Graviton
Arm’s data center CPU revenue was initially concentrated in Amazon Web Services, which began deploying its custom Graviton ARM-based server processors in 2018 and has expanded Graviton to power approximately 40 percent of AWS EC2 compute capacity by 2026. The data center story has broadened since 2023-2024 in ways that are materially increasing Arm’s data center royalty exposure: Microsoft Azure has deployed Arm-based Cobalt 100 processors in its cloud infrastructure, Google has deployed custom ARM-based Axion processors for internal and customer workloads, and Ampere Computing (an ARM-based server CPU startup backed by Oracle) has captured meaningful share in hyperscale-adjacent deployments. The Nvidia GH200 and GB200 superchips — the configurations that hyperscalers are buying at scale for AI training and inference — integrate Nvidia’s Grace CPU (ARM Neoverse V2 core design) with an H200 or B200 GPU on a high-bandwidth NVLink interconnect, making every AI accelerator sale at the datacenter hyperscaler level simultaneously an ARM v9 royalty event. Arm management has not disclosed the Grace royalty rate specifically, but the Neoverse V2 license agreement was disclosed as a multi-hundred-million-dollar arrangement at the time of the Nvidia-Arm acquisition attempt (2020-2022), and the per-chip royalty on each GH200/GB200 sold represents a significantly higher absolute royalty than a standard server CPU because the overall chip selling price is dramatically higher. The $700+ billion in cloud capex commitments from Amazon, Microsoft, and Google for AI infrastructure translates directly into a sustained multi-year demand signal for ARM-based data center processors — both hyperscaler-custom (Graviton, Cobalt, Axion) and standard Arm Neoverse-based server CPUs from Ampere and other vendors — that Arm’s management has explicitly cited as a visibility improvement in royalty revenue forecasting. The earlier trajectory of ARM’s data center penetration through AWS Graviton established the initial data center royalty stream that Arm’s FY2026 results now show has expanded to multiple hyperscaler-custom designs and Nvidia’s AI accelerator superchip configurations.
What Arm’s Stock Performance and Valuation Reflect About the Royalty Model’s Growth Ceiling
Arm Holdings stock (NASDAQ: ARM) traded near $200 per share by mid-2026, approximately four times its $51 IPO price from September 2023, giving the company a market capitalization of approximately $170 billion on roughly $4.7 billion in FY2026 revenue — a revenue multiple of approximately 36 times that reflects investor expectations of sustained high-rate royalty revenue growth as AI chip designs proliferate and ARM v9 adoption nears completion. The valuation premium relative to standard semiconductor companies (which trade at 5 to 8 times revenue) reflects Arm’s royalty model characteristic: revenue grows with each new ARM-based chip that ships globally without requiring Arm to add manufacturing capacity, hire proportional headcount, or incur proportional cost of goods sold. Arm’s operating margin profile — approximately 45 percent in FY2026 — is structurally above semiconductor manufacturers’ margins because royalties are essentially pure margin above the fixed costs of maintaining the ISA and supporting licensees’ design implementations. The growth ceiling question is the central investor debate: how much higher can royalty revenue grow if ARM already powers approximately 99 percent of smartphones, approximately 40 to 60 percent of data center CPUs by unit count, and a growing share of AI accelerators? Arm’s answer is that royalty revenue per ARM chip shipped has not yet peaked, because the shift from v8 to v9 architecture and the shift from general-purpose CPUs to AI-specific compute designs (higher ASP, higher royalty absolute value) will continue to expand revenue per unit even as total unit growth moderates from peak smartphone growth rates. IDC’s semiconductor market research for 2026 projects AI-accelerated chip demand to sustain ARM royalty revenue growth above 20 percent annually through 2027, with the primary risk factor being an acceleration of RISC-V adoption in edge AI and IoT applications that could reduce Arm’s royalty exposure in embedded markets while leaving the high-value data center and mobile segments intact. Financial Times technology coverage through Q2 2026 characterizes Arm as the most structurally advantaged company in the AI infrastructure supply chain because its revenue scales with AI chip deployment volume across every chip vendor rather than being concentrated in a single product line, hardware generation, or customer relationship — a diversification that Nvidia, TSMC, and ASML each lack in their respective portions of the AI chip supply chain.
