Report contents
Industry landscape
Where the company competes: one architecture, four very different markets
Arm Holdings licenses processor designs and the instruction set architecture they run on. Chip designers pay an upfront license fee for access to a core or a wider compute subsystem, then a royalty on each chip that ships, usually calculated as a percentage of the chip's selling price or a fixed amount per unit. Arm does not fabricate or sell silicon. Its 20-F for the fiscal year ended March 31, 2026 puts the company between the design and manufacturing layer on one side, where Synopsys and Cadence sell EDA tools and TSMC and Samsung run the fabs, and the chip companies and device makers on the other. The core market is global licensable CPU intellectual property. The specific arena is the instruction set and core designs that go into smartphone applications processors, data-center server CPUs, automotive electronics and embedded devices.
Geography matters less than the customer list suggests. Design wins are decided customer by customer, mostly with companies based in California, Taiwan, Korea and China, while the engineering base is spread across the UK, the US and India. Developing a core is a large fixed cost; licensing it to one more customer costs almost nothing. That asymmetry is the engine of the model, and it explains why a design that wins a socket can earn royalties on every unit for a decade.
The four end markets differ sharply in share, growth and competitive pressure. Smartphones are the mature base, where Arm's architecture sits in essentially all applications processors; Morningstar estimates the share at 99%. Growth there now comes from royalty rates, through the mix shift to the v9 architecture and to complete compute subsystems, since unit volumes are not expanding. Arm's management told investors that softer memory-driven handset demand could cut fiscal 2027 royalty revenue by 1% to 2%, which signals how little volume upside is left in that pool.
Data centers are the opposite case. x86 processors from Intel and AMD still account for most server CPU revenue, but Arm-based designs from Amazon Web Services, Ampere, Nvidia and Google are winning sockets, and Arm reported that data-center royalty revenue grew more than 100% year over year in the quarter ended December 2025. Automotive and embedded are smaller and more fragmented: Arm supplies cores for microcontrollers and driver-assistance chips, but competes there against in-house designs, Synopsys ARC cores and a growing set of RISC-V suppliers.
Concentration is high at the architecture level and much lower one layer down. Two instruction sets, Arm and x86, underlie most general-purpose processors sold today, with RISC-V as the credible open-standard challenger. Below that, the market for individual processor cores has more participants: Synopsys and Cadence license configurable cores, Imagination Technologies sells graphics IP, Ceva sells DSP cores, and SiFive and others license RISC-V cores.
That split decides the barrier assessment. In smartphone applications processors, the alternatives number fewer than five and each carries a decade of software and tooling investment that would have to be rebuilt, so entry barriers are high. In data-center servers the architecture is open to new entrants, but the software ecosystem is not: the reason Arm designs have gained traction is that customers with their own operating-system and compiler teams, such as the large cloud providers, can absorb the porting cost. In embedded and IoT the field is crowded and price-sensitive, and barriers are low.
The licensing line shows how lumpy the business is at the top. Arm reported $505 million of license revenue in the quarter ended December 2025, below consensus, and guided the following quarter above $750 million, according to Morningstar's read of the results. A handful of large agreements drives that swing, and each one is negotiated with a customer that has the technical capacity to evaluate an alternative.
Two developments would change this picture. The first is a flagship smartphone customer moving a high-volume line to RISC-V or to a fully in-house core; no public evidence shows that happening at scale, but the cost of trying has fallen. The second is Arm's ability to keep raising royalty rates as it bundles more into compute subsystems. If rate increases stall while handset units shrink, the smartphone base stops being a cash engine and becomes a slow decline, and the data-center franchise would have to carry the entire growth case on its own. Arm does not disclose revenue by end market in a way that lets an outside investor size that exposure precisely, so the durability of the smartphone royalty line remains the open question in how this market is defined.
Competitive advantage
Does the company have a moat? Yes, and it is deepest in phones
Overall rating: Fairly Strong. Two tests decide this: whether the market stays concentrated, and whether the leader keeps earning returns above its cost of capital. Arm passes both, with one qualification that keeps it out of the top band. Concentration is high at the layer where Arm operates. Two instruction sets, Arm and x86, carry most general-purpose processors sold today, and inside low-power computing Arm is the design that software is written for first. On returns, Arm reported gross margins above 92% in every fiscal year from 2022 through 2026 and generated $0.98 billion of free cash flow in fiscal 2026 on $4.92 billion of revenue. Those economics survived a handset downturn, a listing and a step-up in operating costs, which is what a barrier is supposed to produce. The qualification is that GAAP returns are less exceptional than the cash flow suggests, and the customers with the most to gain from an alternative are the ones with the engineering budget to build one.
Market position: dominant in phones, a challenger in servers
Market position: Strong. Arm's architecture sits in essentially every smartphone applications processor; Morningstar estimates the share at 99%. Arm is the largest vendor of licensable processor IP by revenue. Below that headline, the ranking depends on which end market you look at. In phone and tablet CPUs no competitor holds a meaningful share of core designs. In data-center server CPUs, x86 from Intel and AMD still carries most revenue, and Arm-based designs are the challenger. In microcontrollers and embedded, Arm is one of several suppliers, alongside Synopsys ARC cores, in-house designs and a growing list of RISC-V vendors.
Share stability: little public history, but no sign of erosion
Stability: Fairly Strong. The five-year test is hard to run here. Arm listed in September 2023, and no third party publishes a consistent five-year series for architecture share across phones, servers and embedded. The benchmark, where a five-year change above 5 percentage points suggests no barrier and under 2 points suggests a strong one, can only be applied to the phone market, and there the share has not moved. What is measurable is the direction of travel: revenue rose from $2.70 billion in fiscal 2022 to $4.92 billion in fiscal 2026, about 16% a year, while handset unit volumes were roughly flat over the same span. That combination points to higher royalty rates and a richer mix rather than to share loss. The server franchise went from negligible to meaningful in the same window. The rating stops short of Strong because the stability evidence is mostly Arm's own revenue and a single third-party phone estimate, and because the fiscal 2022 and 2023 figures predate the listing.
Profitability: 92% gross margins, ordinary GAAP returns
Profitability: Fairly Strong. Fiscal 2026 figures are calculated from reported revenue, gross profit, net income, equity and cash flow.
| Fiscal year (to March 31) | Revenue (USD bn) | Gross margin | Net margin | ROE | Free cash flow (USD bn) |
|---|---|---|---|---|---|
| 2022 | 2.70 | 91.9% | 20.3% | 15.5% | 0.38 |
| 2023 | 2.68 | 92.7% | 19.6% | 13.8% | 0.65 |
| 2024 | 3.23 | 92.8% | 9.5% | 6.5% | 0.95 |
| 2025 | 4.01 | 94.9% | 19.8% | 13.1% | 0.18 |
| 2026 | 4.92 | 92.5% | 18.3% | 10.9% | 0.98 |
Return on equity averaged about 12% over the five years, below the 15% level that would mark a clear competitive advantage. Two things explain the gap, and only one is cosmetic. The first is the equity base: the September 2023 listing enlarged shareholders' equity, and $2.75 billion of cash sat on the balance sheet at March 2026. Strip the cash out and invested capital was about $5.5 billion against $0.90 billion of net income, an estimated return near 16%. The second is cost. Gross profit of $4.55 billion in fiscal 2026 absorbed $3.64 billion of operating expense, overwhelmingly engineering, leaving operating income of $0.91 billion. Arm spends like a chip designer and collects like a licensor, and the gap between those two is where the GAAP margin goes.
Comparisons with Intel, Qualcomm or Nvidia are not like-for-like, because those companies carry wafers, inventory and fab capacity. Against the closest IP-licensing peers, Synopsys and Cadence, Arm's gross margin is higher and its revenue base smaller, so the edge comes from position in the design chain rather than from sheer size. The one place size is visible is reuse: the same core work is sold into phones, cars and servers.
Overall rating: Fairly Strong, carried by the phone franchise
Overall rating: Fairly Strong. The concentration test passes clearly in phones and only partly in servers and embedded. The profitability test passes on cash generation and gross margin and falls short on ten-year GAAP returns, which is partly an artifact of the listing, the cash pile and a heavy engineering budget, and partly a real cost of competing at the leading edge. Arm's returns exceed its cost of capital by a wide margin, but they rest on a customer base that includes some of the few firms capable of funding an alternative design. That combination fits a real but bounded moat rather than an impregnable one.
Where the advantage comes from and how long it can last: software locks, scale spreads
Two sources carry most of the weight: customer captivity built on software and design reuse, and scale economics in core development. Government protection, licenses or tariffs do not apply; Arm's barriers come from the private choices of its customers. Both sources are real, and both share a failure mode: they are strongest where the customer's software stack is standard and weakest where the customer writes its own.
Customer captivity: the software stack is the lock
Customer captivity: Fairly Strong. Software gets written once and reused across everything built on the same instruction set. Operating systems, compilers, driver stacks, middleware, debuggers and third-party libraries all target Arm, and every one of those layers has to be rebuilt for a different architecture. Inside a chip company, the lock is design reuse. A team that has shipped three systems-on-chip (SoCs) on Arm cores owns verification suites, physical-design flows, EDA scripts and engineers trained on the toolchain. Moving to a new architecture means carrying a first product with almost no reuse, on a schedule that is hard to promise to the customer buying the chip.
The asymmetry does the work. Migration cost is paid in full, in engineering hours, before any royalty saving arrives, while the saving is a future percentage of a chip price that may never ship in volume. That makes the calculation favor staying put even when a rival offers better terms.
Pricing power is the check on whether captivity is real. Arm grew revenue about 16% a year from fiscal 2022 to fiscal 2026 while phone unit volumes were flat, and it has pushed licensees toward the v9 architecture and toward complete compute subsystems, both of which carry higher royalties than the older core-by-core licensing. Revenue per chip rose without share loss in phones. That is captivity showing up in the P&L.
Where it breaks is instructive. The lock is weakest where the customer owns the software. Hyperscale cloud companies employ their own operating-system, compiler and firmware teams, which is why Arm's server entry depended on those customers absorbing the porting cost themselves. Architecture licensees, which design their own cores under an Arm license, already carry the engineering capability; their next step outward is shorter than for a merchant chip designer buying off-the-shelf cores. Arm does not disclose churn or per-unit royalty rates, so the strength of the lock can only be read indirectly, from pricing and from share over time.
Scale economics: one core, many licensees
Scale economics: Fairly Strong. A competitive CPU core takes years of engineering and a long verification cycle. The $3.64 billion of operating expense in fiscal 2026 is mostly that kind of work, and it is fixed. The marginal cost of licensing an existing design to one more customer is close to zero, which is why gross margin sits between 92% and 95% and why incremental revenue drops through to profit at a much higher rate than in chip manufacturing.
Reuse across markets amplifies the effect. The same core family, modified, goes into a phone, a car and a server. Each new end market raises the return on design work already paid for, without a matching increase in cost. That is why automotive and infrastructure matter more than their current revenue share implies: they spread the fixed base wider.
What matters is relative share within an end market, not absolute size. Software vendors and toolchain providers optimize first for the platform with the most sockets, and developers write for Arm in phones because that is where volume is. Each design win reinforces the next one. In a market where Arm starts near zero, the flywheel does not turn until a customer pays to port, which is exactly the hurdle the server business had to clear.
The limits are equally concrete. Scale protects markets Arm already leads; it does not block RISC-V, which has no licensor to recoup and can be adopted incrementally in a small microcontroller where software complexity is low. The largest customers can also fund an alternative out of one year's R&D budget. And fixed costs cut the other way when license revenue dips: the engineering base does not shrink with it. Free cash flow went from $0.95 billion in fiscal 2024 to $0.18 billion in fiscal 2025 and back to $0.98 billion in fiscal 2026, a reminder of how much the cash line depends on the timing of large license payments.
Durability: cheap substitutes, and a pricing strategy that invites them
The substitute threats are ranked by how fast they can bite. RISC-V is free and open, with no single owner to pay, and it has gained most ground in microcontrollers and among buyers who want to avoid dependence on one supplier. Its weakness is fragmentation and the absence of a vendor accountable for compatibility, which matters where certification and software support are expensive. That makes it a slow threat in phones and a faster one in embedded. x86 remains the incumbent in servers and PCs, and its power-efficiency disadvantage is the reason Arm got in at all.
The fastest-moving threat in dollar terms is in-house design. Every generation of proprietary core built by an architecture licensee reduces the value Arm adds per chip, and those designs do not need to beat Arm everywhere, only in the specific products where the customer controls the software.
Arm's own pricing strategy interacts with this. Value-based pricing and the sale of complete subsystems lift the royalty per chip, which raises revenue today and also raises the payoff to switching tomorrow: a higher royalty makes an alternative easier to justify. Selling more of the design also puts Arm closer to competing with the engineering teams it licenses to, which strengthens their incentive to look elsewhere.
The assessment would change on three observations. A top-tier phone customer shipping a high-volume line on a non-Arm core would undercut the captivity argument directly. A royalty increase followed by visible RISC-V adoption among mid-tier handset or microcontroller customers would show the pricing power has a ceiling. And a data-center franchise that stays confined to a handful of self-designing cloud companies would leave the growth case resting on the customers least likely to remain captive.
The moat is likely to hold in phones, automotive and embedded for the next five years, because rebuilding a software and qualification stack is expensive and no single buyer in those markets has both the scale and the software capability to walk away. The growth, however, is concentrated where the lock is weakest. How that resolves decides whether the next five years look like the last five.
Generated by DeepVal AI. Data and models may contain errors or be outdated. For research only; not personalized investment advice.
Continue reading the full report
- Comparison and summaryHow Arm Holdings plc American Depositary Shares compares with its closest rivals, and our moat rating
Free to sign up · 20 research points · unlocking this report uses 10
20 points cover 2 full reports. No card needed.
Research reflects its generation date. Preview the first two modules.
More on Arm Holdings plc American Depositary Shares
Arm Holdings plc American Depositary Shares
Business, moat and key questions
Open company page