Arm Holdings: AGI CPU Demand Doubles in Six Weeks, Data Center Royalties on Track to Double Again
Q4 Fiscal Year 2026 Earnings Call — May 6, 2026
Arm Holdings closed its fiscal year 2026 with a record quarter and delivered the kind of forward guidance that reframes the company's growth trajectory. Revenue of $1.49 billion in Q4 grew 20% year-on-year, capping a full-year revenue figure of $4.92 billion — the third consecutive year of more than 20% growth since the IPO. But the headline number was secondary to two developments that carry far greater weight for the long-term investment thesis: the AGI CPU demand pipeline has more than doubled in just six weeks since launch, and data center royalty revenue is growing at triple-digit rates year-on-year with no sign of deceleration.
AGI CPU Demand Doubles Before a Single Production Chip Ships
When Arm unveiled its AGI CPU at the Arm Everywhere event in late March, the company disclosed roughly $1 billion in customer demand to support a supply chain it had already secured. As of the May 6 earnings call, that figure has jumped to over $2 billion across fiscal 2027 and 2028 — without any revision to the revenue guidance. CEO Rene Haas was direct: "The number that we talked about at end of March was supply in place to support $1 billion of demand, and that includes memory, that includes wafers, that includes packaging, that includes access to test equipment. So for the $2 billion, we are now in the process of securing supply to support that."
CFO Jason Child confirmed the near-term revenue target remains approximately $90 million in Q4 of fiscal 2027, the first quarter in which production chip revenues are expected. The mismatch between demand and supply is a near-term constraint, not a demand problem. Child noted that by Q3 of this fiscal year, the company expects to provide firmer guidance on Q4 delivery volumes and initial views on fiscal 2028. The $15 billion AGI CPU revenue target by fiscal year 2031, first articulated at the March event, remains intact.
Meta is the lead partner and co-developer of the AGI CPU, with a multi-generation roadmap aimed at supporting what Haas described as "personal super intelligence for more than 3 billion users." Additional named customers include OpenAI, Cerebras, Rebellions and Positron — all deploying the Arm AGI CPU as the host processor alongside accelerator-based systems. SAP has committed to migrating core database and business application workloads to Arm, beginning with AWS Graviton and expanding to the AGI CPU. Cloudflare will deploy Arm across its global network.
Data Center Royalties: Doubling Again
Data center royalty revenue has more than doubled year-on-year in fiscal 2026, and Haas stated explicitly he expects it to double again in fiscal 2027. This is not purely a custom silicon story. Child pointed to broad hyperscaler adoption — Google's TPU 8t and 8i replacing x86 host processors with custom Arm Axion CPUs, delivering an 80% performance improvement. AWS continues scaling Graviton alongside Trainium and Nitro. Microsoft is deploying Cobalt across Azure. NVIDIA's Vera CPU, built on Arm architecture, now ships as a standalone 256-chip rack designed specifically for agentic AI orchestration adjacent to Vera Rubin accelerator systems. Child summarized it plainly: "The three largest GPU providers across NVIDIA with Vera, Google pairing TPUs with Axion 2, and AWS with Graviton — all of those partners are now all in on Arm."
Arm's share of compute at the top hyperscalers has reached approximately 50% market share, and the company holds close to 100% market share in data center networking chips, particularly DPUs and SmartNICs. The structural argument underpinning continued royalty growth is that as agentic AI workloads expand, the CPU's role in orchestrating tasks, managing memory, enforcing security and coordinating around accelerators becomes disproportionately important. Haas cited an internal analysis suggesting data centers will require more than four times today's CPU capacity by 2030, implying a CPU market opportunity exceeding $100 billion — a figure AMD subsequently referenced at a higher level of $120 billion in its own commentary the following day.
The Silicon Business: Margin Structure and Ecosystem Risk
Arm is entering the merchant silicon market for the first time, and questions about ecosystem conflict were inevitable. Haas addressed this directly, noting that every major licensee — including AWS, Google, Microsoft and NVIDIA — was approached before the March announcement and every single one agreed to publicly endorse the strategy. More than 50 partners participated in the Arm Everywhere event. "At the end of the day, we are responding to customer demand in a market," Haas said. "We don't take it for granted. We're very, very appreciative of it."
The gross margin profile on first-generation AGI CPU products has been previously disclosed at approximately 30% plus. Child elaborated on the economics, noting that the most expensive element — the compute die — is effectively the CSS technology already developed within the IP business, making the incremental cost structure for the chip business unusually lean. "The incremental costs and OpEx that we're having to add to the chip business is not that significant. It's a team that's in the dozens of people, not hundreds." He confirmed the chip business is expected to be operating profit positive in fiscal 2028, scaling toward a 35% operating margin at the $15 billion revenue level by fiscal 2031, compared to approximately 65% for the IP business at that scale.
The Core Count Thesis and CPU-to-GPU Ratios
A substantive exchange during Q&A clarified how Arm is thinking about the structural CPU demand expansion. Rather than framing the opportunity through a fixed CPU-to-GPU chip ratio, Haas argued the more useful lens is core count. "As I mentioned earlier, could I see those core counts doubling or quadrupling over the next number of years? Absolutely." The Arm AGI CPU ships with 136 cores; NVIDIA's Vera ships with 88. Both are purpose-built for agentic orchestration workloads that benefit from running discrete jobs on single, independent cores. "It's actually pretty good for a single core design to handle that as opposed to having multi-cores having to run that all together in unison." Higher core counts drive higher ASPs, which is the mechanism behind the $100 billion-plus TAM projections regardless of absolute chip unit volumes. Haas added: "I'm actually confident that by the end of the decade, I believe the largest market share by CPU type will be Arm."
Royalty Outlook and Licensing Cadence for Fiscal 2027
For Q1 fiscal 2027, Arm guided to revenue of $1.26 billion, plus or minus $50 million, with both royalty and licensing revenue expected to grow approximately 20% year-on-year. Non-GAAP EPS guidance is $0.40 plus or minus $0.04. Child noted that licensing revenue will remain back-half weighted at roughly 60/40, consistent with prior years. The SoftBank technology licensing and design services arrangement contributed $200 million in Q4, flat sequentially, and remains a recurring element of the licensing line.
Annual contract value, the metric Arm uses to smooth licensing volatility, grew 22% year-on-year, above the company's stated long-term expectation. Child indicated that while the near-term licensing growth rate is running in the 20% range driven by the AI investment cycle, the long-term floor should be viewed as at least 10% annually. On the royalty side, smartphone unit volumes turned modestly negative in the most recent quarter, but higher ARMv9 penetration and compute subsystem attach rates in premium handsets are more than compensating. Automotive remains a double-digit growth contributor with no signs of deceleration.
OpEx Trajectory and Path to Incremental Margin
Non-GAAP operating expenses in Q4 came in at $734 million, approximately $10 million below guidance, and grew 30% year-on-year, reflecting heavy R&D investment. Child guided Q1 operating expenses to approximately $760 million and indicated that OpEx will step up modestly each quarter through fiscal 2027, growing sequentially by a few percent per quarter. Critically, he noted that by year-end, revenue growth will outpace expense growth, restoring incremental margin expansion. "We will be growing expenses less than revenue by the time we end the year," he said. The longer-term EPS target of more than $9 by fiscal year 2031 was reiterated, underpinned by the combined $25 billion revenue target across AGI CPU and IP businesses.
Arm Holdings plc Deep Dive
Introduction: The Architectural Fulcrum of AI
The semiconductor industry is undergoing a structural realignment, shifting from pure transistor scaling to architectural efficiency as the primary vector of performance. At the exact center of this realignment sits Arm Holdings plc. For over three decades, the company was primarily known as the architect of the mobile revolution, providing the energy-efficient instruction set that powered virtually every smartphone on the planet. Today, that narrative is severely outdated. Arm has aggressively repositioned itself as the foundational compute layer for the entire artificial intelligence infrastructure complex. As power constraints become the ultimate limiting factor for data center expansion, Arm's historical obsession with performance-per-watt has transitioned from a mobile-specific advantage to a hyperscaler imperative. The company is currently executing one of the most ambitious business model evolutions in the history of the semiconductor industry, transitioning from a passive licensor of intellectual property to an active vendor of pre-integrated subsystems and production silicon.
Business Model Evolution: From IP Licensing to Silicon Reality
Understanding Arm requires understanding its complex, multi-layered monetization engine, which has historically relied on two distinct revenue streams: an upfront licensing fee for access to its architecture or core designs, and a trailing per-chip royalty based on the final selling price of the semiconductor. Historically, this model was volume-dependent and yielded low single-digit royalty rates, typically ranging between 1 percent and 3 percent for the ubiquitous Armv8 architecture. Over the last two years, management has orchestrated a masterful value-capture pivot. The primary driver of this pivot is the transition to the Armv9 architecture, which effectively doubles the royalty rate to roughly 5 percent per chip. However, the true economic engine of the modern Arm is the Compute Subsystems platform. Rather than selling bare intellectual property blocks, Arm now packages compute, memory, and security elements into pre-validated blueprints. These Compute Subsystems command royalty rates north of 10 percent and already account for nearly 20 percent of total royalties, pulling forward revenues by accelerating customer design cycles.
The most radical transformation of the business model arrived in March 2026, when Arm fundamentally crossed the Rubicon from intellectual property vendor to fabless semiconductor company. With the launch of the Arm AGI CPU, co-developed with Meta Platforms, Arm is now delivering actual production silicon manufactured on a 3-nanometer process. This is a monumental paradigm shift. By selling finished server chips directly to data centers and original design manufacturers, Arm is expanding its total addressable market from a $3 billion royalty pool to a $100 billion data center silicon market. While this introduces new supply chain risks and capital expenditure requirements, the unit economics and gross profit dollars generated by physical silicon sales dwarf those of traditional licensing. Arm is systematically moving up the value chain, ensuring it captures a far greater share of the economics its architecture generates.
Customer Ecosystem, Competitors, and Supplier Dynamics
Arm's customer base represents a veritable who's who of global technology leadership. In the mobile and edge compute markets, the company remains deeply tethered to Apple, Qualcomm, and MediaTek. In the high-performance computing tier, its customer profile has shifted toward the major cloud hyperscalers, all of which are actively designing custom Arm-based silicon. Amazon Web Services with its Graviton processors, Google with Axion, Microsoft with Cobalt, and Nvidia with the Grace central processing unit represent the bleeding edge of Arm's enterprise adoption. Meta has further cemented this ecosystem by partnering directly with Arm as the lead co-developer for the new AGI CPU. On the supply side, Arm's foray into physical silicon makes it heavily dependent on Taiwan Semiconductor Manufacturing Company, binding Arm to the same geographic and capacity constraints faced by its fabless peers.
The competitive landscape is bifurcated. In the traditional data center space, Arm is waging an aggressive and highly successful war against the legacy x86 duopoly of Intel and Advanced Micro Devices. The x86 architecture, burdened by decades of legacy instructions, simply cannot match the energy efficiency of modern Arm Neoverse designs. However, Arm's aggressive push to extract higher royalties has spawned a potent counter-reaction, elevating open-source RISC-V architecture from an academic curiosity to a Tier-1 competitor. As Arm attempts to squeeze more economics out of its ecosystem, it is inadvertently incentivizing its largest customers to fund its open-source rival.
Market Share and Competitive Moats
Arm operates as a functional monopoly in the mobile and edge computing markets, boasting an impregnable market share of roughly 99 percent in smartphones. While mobile volumes have largely saturated, Arm's revenue in this segment continues to grow due to the structural royalty rate expansion driven by Armv9 upgrades. The more consequential market share battle is occurring in the data center. As of early 2026, Arm-based processors are expected to reach a 50 percent share among top hyperscaler internal deployments. This footprint is reflected in Arm's financial results, with data center royalty revenues growing by more than 100 percent year-over-year in the third quarter of fiscal 2026, putting the segment on track to surpass mobile as the company's largest revenue contributor.
The company's competitive moat is exceptionally wide, predicated on a network effect that is nearly impossible to replicate from a standing start. There are currently over 22 million software developers globally writing code optimized for the Arm architecture. Furthermore, the introduction of Compute Subsystems acts as a significant moat-deepener. By utilizing pre-validated Compute Subsystems, semiconductor companies can reduce their time-to-market by 12 to 18 months and save tens of millions in non-recurring engineering costs. Once a customer integrates a Compute Subsystem into their silicon roadmap, the switching costs become astronomically high, virtually locking them into the Arm ecosystem for future product generations.
Industry Opportunities and the Agentic AI Compute Pivot
The prevailing narrative surrounding artificial intelligence infrastructure has historically centered entirely on parallel processing accelerators. However, as the industry transitions from the brute-force training of large language models toward the deployment of reasoning-based, agentic artificial intelligence, the infrastructure bottleneck is shifting. Agentic artificial intelligence requires systems that can reason, plan, execute multi-step workflows, and constantly move massive datasets. This coordination layer is entirely dependent on central processing units. While the accelerators generate the tokens, the central processing units must orchestrate the data flow. Industry projections suggest that modern agentic computing will require more than four times the current central processing unit capacity per gigawatt of data center power.
This dynamic plays perfectly into Arm's core engineering philosophy. Data centers are hitting hard physical limits regarding power consumption and heat dissipation. Because Arm processors can deliver more than double the performance per rack compared to legacy x86 systems within the same thermal envelope, hyperscalers are aggressively pivoting their capital expenditure budgets toward Arm-based infrastructure. The launch of the Arm AGI CPU is a direct response to this exact industry dynamic, providing a 136-core powerhouse designed specifically to handle the "data dump truck" logistics required by agentic artificial intelligence clusters. The power-constrained nature of global computing is the most significant structural tailwind in Arm's history.
Threats and Disruptive Entrants: The RISC-V Rebellion
While x86 is the incumbent Arm is displacing, RISC-V is the disruptive entrant that poses an existential, long-term threat to Arm's terminal value. Driven by a collective industry desire for architectural sovereignty and an escape from Arm's rising royalty rates, the open-source RISC-V instruction set has achieved an estimated 25 percent global market penetration by early 2026. Initially confined to low-end microcontrollers and embedded internet-of-things devices, RISC-V is now moving aggressively up the compute stack into high-performance domains, directly encroaching on Arm's most lucrative territories.
This threat was violently accelerated by Arm's own legal strategies. Following a bitter, high-profile legal dispute over licensing terms stemming from Qualcomm's acquisition of Nuvia—a lawsuit that a U.S. District Court conclusively resolved in Qualcomm's favor in late 2025—the relationship between Arm and one of its largest customers remains deeply fractured. In direct response to Arm's attempt to cancel its architectural license, Qualcomm executed a $2.4 billion acquisition of Ventana Micro Systems, a leading designer of data center-class RISC-V cores. Simultaneously, Meta acquired RISC-V specialist Rivos. These are not speculative venture investments; they are calculated, multi-billion-dollar strategic maneuvers by Big Tech to build viable, royalty-free, Arm-independent silicon roadmaps. Furthermore, massive sovereign investments from China and India into RISC-V, aimed at circumventing Western technological hegemony, ensure that the open-source architecture will possess the capital and engineering resources necessary to rival Arm's performance in the coming decade.
Management Track Record: Flawless Execution Amidst Transition
Under the leadership of Chief Executive Officer Rene Haas, Arm's management team has executed with clinical precision. Since navigating the company back to the public markets in late 2023, management has completely overhauled the investment narrative. They successfully convinced the institutional market that Arm is an artificial intelligence infrastructure play rather than a stagnant smartphone derivative, and the financial results have thoroughly validated that thesis. The company has delivered four consecutive quarters of revenue exceeding $1 billion, culminating in a massive top-line beat of $1.49 billion in the fourth quarter of fiscal 2026. Licensing revenues have surged, providing a highly reliable leading indicator for royalty growth two to three years into the future.
More impressively, management has successfully navigated the tightrope of aggressive monetization without immediately breaking the ecosystem. Driving the transition from standard licensing to Compute Subsystems required immense commercial leverage, and introducing first-party production silicon risked alienating the very semiconductor partners that build upon Arm. Yet, by securing Meta as a co-developer for the AGI CPU, Haas effectively neutralized potential backlash, framing the physical silicon business as a necessary evolution to meet hyperscaler demand rather than a hostile encroachment on its own partners. The track record over the last three years is characterized by disciplined capital allocation, precise market timing, and an unapologetic focus on value capture.
The Scorecard
Arm Holdings represents one of the most compelling structural margin expansion stories in the semiconductor sector. The company has successfully decoupled its revenue growth from underlying device volume growth through the dual levers of higher-royalty Armv9 adoption and the explosive uptake of Compute Subsystems. Furthermore, the bold pivot into selling physical data center silicon vastly expands its total addressable market, positioning the company to capture the lucrative hardware economics of the ongoing build-out in agentic artificial intelligence infrastructure. Near-term and medium-term financial visibility is exceptionally high, backed by surging licensing pipelines and dominant market share in both the mobile and power-constrained data center arenas.
Conversely, the long-term terminal value is clouded by the rapidly accelerating adoption of RISC-V. Arm's aggressive tactics to extract higher tolls from its ecosystem have inadvertently unified its largest customers—most notably Qualcomm and Meta—into funding a highly capitalized, open-source alternative. While switching architectures remains a monumental engineering task, the semiconductor industry is clearly constructing an escape hatch to break the Arm monopoly. Investors must weigh the certainty of massive, near-term royalty compounding and data center share gains against the long-tail risk of an industry-wide migration toward architectural sovereignty.