An independent case study

Arm Holdings: the architecture tax on computing — re-rated for AI

A neutral, evidence-first reading of the company whose CPU designs sit inside almost every phone and a fast-growing share of AI data centers — assembled so you can reach your own conclusion.

47 sourcesAs of 6 June 20268 analysis sections

Arm designs the CPU architecture that powers ~99% of smartphones and licenses it to the world's chipmakers — a capital-light, ~98%-gross-margin franchise that the AI build-out has pushed into the data center, and, for the first time in 35 years, into selling its own chip.

The genuinely open question is not whether Arm is important — its architecture ships in ~99% of smartphones — but whether it can convert AI-era relevance into the growth its valuation demands without weakening the licensing ecosystem that made it ubiquitous. The evidence cuts both ways on every major question below. This study lays out both cases; the verdict is yours.

The decisive questions

Each links to the section that lays out the evidence on both sides.

Can the data-center engine outgrow mature mobile?

Arm says it reached ~50% of hyperscaler CPU compute, up from ~18% in two years, and data-center royalties more than doubled. But that is Arm's own metric, and smartphones — the royalty base — are flat to declining.

Does selling its own chip strengthen or threaten the franchise?

The March-2026 AGI CPU (lead customer Meta) captures far more value per system — but puts Arm in direct competition with the licensees whose royalties are its core business.

How real are RISC-V and disintermediation?

A royalty-free open ISA backed by Google, plus customers (Apple, Qualcomm, hyperscalers) designing their own cores — and a Qualcomm court win — all chip at Arm's pricing power. The ecosystem moat is deep but no longer unquestioned.

Is a ~190x forward P/E foresight or froth?

FY2026 revenue grew 23% to $4.92bn at ~98% gross margin — among the highest in large-cap tech. Yet the multiple prices years of flawless execution, and the average analyst target has sat well below the share price.

The growth that frames the debate

Total revenue by fiscal year (GAAP, US$bn). Three straight years above 20% growth — the bull case in one line, and the reason the valuation debate is so charged.

Arm total revenue (US$bn, FY ends 31 Mar)

⚖️

What reasonable people disagree about

Whether ~50% hyperscaler-compute share is durable share-taking or a flattering metric; whether the AGI CPU is a value-capture masterstroke or a channel-conflict risk to the royalty base; whether RISC-V is a low-end nuisance or a structural threat; and whether a ~190x forward P/E is justified by a one-of-a-kind franchise or simply priced for perfection. Informed observers land in different places — by design, this study does not pick for you.

How to read this

Eight sections, each built the same way: a neutral synthesis, a two-sided case-for / case-against ledger, dated quotes, framework visuals, and the sources used. Start with the question that interests you, or read in order from Overview.

🔍

Independent research artifact, not affiliated with or endorsed by Arm Holdings or SoftBank. All claims link to primary or reputable secondary sources fetched during the research run; where a figure is an estimate, the page says so. See Methodology & Limits.

Section 01

Company Overview & Timeline

Who Arm is, where it came from, and the milestones from a 1980s Cambridge processor to the architecture inside almost every smartphone.

5 sourcesAs of 6 Jun 2026

Arm sells architecture, not silicon: it licenses CPU designs and the Arm instruction set to the chipmakers who actually build and sell the chips. That model put Arm inside roughly 99% of smartphones and 350bn+ cumulative chips — and is now being tested as Arm pushes into the data center and, for the first time, sells its own chip.

From Cambridge to ubiquity

Three-plus decades of compounding the same idea: license the architecture, let others build the chips.

1985First ARM processor (ARM1) designed in Cambridge by Steve Furber & Sophie Wilson.
1990Spun out as Advanced RISC Machines — JV of Acorn, Apple and VLSI; chooses to license, not manufacture.
1998Dual IPO on the London Stock Exchange and NASDAQ.
2016SoftBank takes Arm private for ~$32bn — then the largest European tech acquisition.
2020–22Proposed $40bn sale to Nvidia announced (2020) and abandoned (2022) on antitrust grounds.
Sep 2023Returns to NASDAQ at a ~$54.5bn valuation; SoftBank retains ~90%.
Mar 2026Launches the Arm AGI CPU — its first in-house chip in 35 years — with Meta as lead customer.
FY2026Record revenue of $4.92bn (+23%); ~50% of hyperscaler CPU compute reported.

Arm traces to 1990, when it was spun out as Advanced RISC Machines — a joint venture of Acorn, Apple and VLSI — to commercialise a low-power RISC processor first designed in Cambridge in the 1980s [2][3]. Under early CEO Robin Saxby it chose to license its designs rather than manufacture, a model that scaled across the mobile era [2].

The ownership history matters to the investment case. Arm IPO'd in 1998, was taken private by SoftBank in 2016 for ~$32bn, saw a proposed $40bn sale to Nvidia collapse in 2022 on antitrust grounds, and returned to NASDAQ in September 2023 at a ~$54.5bn valuation [1][4]. SoftBank still owns roughly 87%, leaving a thin public float [1].

Today Arm is led by CEO Rene Haas (since 2022), employs ~8,300 people, and is headquartered in Cambridge, UK [1]. Its architecture is the common thread across Apple, Qualcomm, Nvidia, Amazon, MediaTek and Samsung silicon [3]. The open strategic question this study examines: whether Arm can ride the AI build-out into the data center — and even sell its own chips — without weakening the licensing ecosystem that made it ubiquitous.

Both sides of the ledger

Weigh these against each other — they are presented so you can reach your own conclusion, not to argue one way.

The case for

Three-decade, near-ubiquitous franchise: Arm IP is in ~99% of smartphones and 350bn+ chips shipped, an installed base and ecosystem that is extraordinarily hard to dislodge [5][23].
Capital-light, royalty-compounding model: every Arm-based chip a partner ships pays Arm a royalty, so Arm benefits from the whole industry's growth without building fabs [2][11].
Re-rated for the AI era: since its 2023 relisting Arm has posted three straight years of >20% revenue growth and pushed into data center, the fastest-growing slice of compute [25].

The case against

SoftBank controls ~87%, so public holders are minority passengers and the float is thin and volatile [1][45].
The core smartphone market is mature; future growth leans on newer, less-proven data-center and AI bets [40].
Its customers are also its rivals: Apple, Qualcomm and the hyperscalers can design their own cores, and some can move to royalty-free RISC-V [17].

Sources for this section

5 sources · en · tiers shown. Full bibliography on the Sources page.

T2 Arm Holdings — Wikipedia · en · neutral
T1 The official history of Arm — Arm Newsroom · en · supporting
T2 How Arm gained chip dominance with Apple, Nvidia, Amazon and Qualcomm — CNBC · en · supporting
T2 SoftBank to acquire ARM Holdings in $32bn deal — Financier Worldwide · en · neutral
T3 Arm Market Value, SoftBank's 87% Stake — TradingKey · en · critical

Section 02

Market & Industry Structure

Where Arm sits in the semiconductor value chain, the end markets it spans, and why the growth story moved from mobile to the data center.

6 sourcesAs of 6 Jun 2026

Arm's home market — mobile — is mature, so the growth story has moved to the data center, where Arm says its share of hyperscaler CPU compute reached ~50% in 2026, up from ~18% in mid-2024. The bull case is that AI made the CPU-architecture layer strategic; the caveat is that the 50% figure is Arm's own metric for hyperscaler compute, not the whole server market.

The data-center climb

Arm's reported share of hyperscaler CPU compute — its own metric, not whole-server-market share. The slope is the bull case; the denominator is the caveat. Hover for detail.

Arm share of hyperscaler CPU compute (%, reported)

In the value chain Arm sits above the chipmakers: it defines the instruction set and supplies CPU core designs; partners integrate those into chips; foundries such as TSMC manufacture them. That position lets Arm tax the industry's compute without the capital intensity of fabs [11].

By end market, Arm is overwhelmingly dominant in mobile (~99% of smartphones) and broadly used in IoT/embedded and automotive, where rising compute-per-device (ADAS, digital cockpits) lifts Arm content [5]. The decisive new front is the data center: AI servers pair GPUs with CPUs, and the leading custom server CPUs — AWS Graviton, Nvidia Grace/Vera, Microsoft Cobalt, Google Axion — are Arm-based [16].

Arm and outside commentators put its hyperscaler CPU-compute share at ~50% by Computex 2026, versus ~18% in mid-2024 — a steep climb that coincided with the AI-infrastructure boom [6][7][8]. Read carefully, this is share of *hyperscaler compute*, a flattering denominator; x86 still holds large swaths of enterprise and legacy server installs [8].

Both sides of the ledger

Weigh these against each other — they are presented so you can reach your own conclusion, not to argue one way.

The case for

The AI build-out turned the CPU-architecture layer into strategic real estate, and Arm is the default host CPU beside AI accelerators [16].
Data-center momentum is real and fast: Arm's reported hyperscaler-compute share roughly tripled to ~50% in about two years, and data-center royalties more than doubled year over year [6].
Beyond mobile, automotive and IoT add growing per-device content as cars and edge devices become computers [5].

The case against

The headline ~50% is Arm's own 'hyperscaler compute' metric, not whole-server-market share; x86 remains entrenched in enterprise and legacy workloads [8].
Mobile — still the royalty base — is a mature, unit-flat-to-declining market [40].
Much of the data-center 'Arm share' is hyperscalers' *own* custom chips, which is share Arm could later cede to in-house or RISC-V designs [17].

Sources for this section

6 sources · en · tiers shown. Full bibliography on the Sources page.

T3 ARM Holdings: Shaping the Future of Computing — Leverage Shares · en · supporting
T2 Arm posts record revenue, hyperscaler CPU share hits 50% — DCD · en · supporting
T2 Arm expects its data center CPU share to reach 50% — SemiWiki · en · supporting
T2 x86 Data Center Dominance Ends: Arm Crosses 50% Hyperscaler CPU Share — TechTimes · en · neutral
T2 Arm aims to capture 50% of data center CPU market — Tom's Hardware · en · neutral
T3 RISC-V 2026: How Open Chip Architecture Is Disrupting ARM and Intel — Programming Helper · en · critical

Section 03

Business Model & Unit Economics

License fees fund the IP; royalties compound on every Arm-based chip shipped. The growth lever is value captured per chip.

5 sourcesAs of 6 Jun 2026

The model is capital-light with ~98% gross margin: license fees fund development, royalties compound for years on the installed base. The growth lever is price per chip — newer Armv9 cores carry roughly double the royalty of Armv8, and Compute Subsystems (CSS) reportedly command north of 10% of chip value versus 1–3% traditionally.

FY2026 revenue mix

Two engines: recurring royalties on chips shipped and upfront license & other fees. In FY2026 they were close to balanced.

FY2026 revenue mix
Royalty — 53%
License & other — 47%

Royalty $2,613m · License & other $2,307m. Percentages are share of $4,920m total revenue.

Arm earns upfront license fees (estimated ~$1–10m) when a partner takes its IP, then royalties — historically ~1–2% of a chip's selling price — on every unit shipped, often for a decade or more [11]. Because the IP is reused across many chips, the model carries ~98% gross margin [25].

Licensing comes in tiers: a Technology License (a ready-made core, higher royalty), 'Built on Arm' tuning, and an Architecture License (ALA) that lets a partner design its own clean-sheet core implementing the Arm ISA (Apple, Qualcomm's Nuvia, the hyperscalers) — more design freedom, typically lower royalty [12].

The whole strategy is more value per chip. Armv9 carries roughly 2x the royalty rate of Armv8 and was ~25% of royalty revenue in early FY2025; Compute Subsystems (CSS) — pre-integrated, validated reference designs — reportedly command >10% per chip versus 1–3% traditionally, and the top Android vendors now ship CSS-based devices [13][14]. That rate expansion, not unit growth, is what lifted FY2026 royalties.

The same value-capture push has a sharp edge: Arm has explored charging device OEMs directly and raising terms, which critics say risks alienating the partners the model depends on [15].

Both sides of the ledger

Weigh these against each other — they are presented so you can reach your own conclusion, not to argue one way.

The case for

Royalty rate expansion is working: Armv9 (~2x royalty) and CSS (>10% per chip) lift revenue per chip even when unit growth is flat [13][14].
Capital-light economics: ~98% gross margin and royalties that compound on a 350bn-chip installed base for years after the license is sold [25][11].
Architecture licenses deepen lock-in: once Apple or a hyperscaler builds custom cores on the Arm ISA, the surrounding software ecosystem is hard to leave [12].

The case against

Raising prices and pushing direct-OEM licensing risks straining the ecosystem and inviting defections [15].
Architecture-license holders capture most of the design value and pay lower royalties — and, as the Qualcomm case showed, can fight Arm over terms [12][19].
Per-chip rate gains have a ceiling; at some point higher Arm royalties make royalty-free RISC-V look more attractive to big buyers [17].

Sources for this section

5 sources · en · tiers shown. Full bibliography on the Sources page.

T2 Arm and a Leg: Arm's Quest To Extract Their True Value — SemiAnalysis · en · neutral
T3 ARM Licensing vs Royalties Business Model Deep Dive — Screenwich · en · neutral
T2 Arm Q4 FY2025: Record Revenue as Armv9 and CSS Ramp Up — Futurum · en · supporting
T3 Arm Holdings and the Game-Changing Potential of Compute Subsystems — AInvest · en · supporting
T2 Arm Changes Business Model — OEM Partners Must Directly License From Arm — SemiAnalysis · en · critical

Section 04

Competitive Landscape & Positioning

Arm fights on three fronts: x86 in the data center, royalty-free RISC-V at the low end and rising, and disintermediation by its own licensees.

4 sourcesAs of 6 Jun 2026

Arm's competitive position is strongest where its ecosystem lock-in is deepest (mobile, embedded) and most contested where buyers are biggest and best-funded (data center). The two structural threats are RISC-V (royalty-free, Google-backed) and disintermediation — customers designing their own cores, now including Arm competing with them directly.

Five Forces on the CPU-architecture market

Click a force for the rated pressure and its sourced basis. Arm's position is strongest where ecosystem lock-in is deepest and most contested where buyers are largest.

CPU architecture & IP

Competitive rivalry — Medium. In mobile Arm is near-unchallenged (~99% of smartphones), but in data center it competes with x86 (Intel, AMD) and increasingly its own licensees' custom cores; competition is intensifying now that Arm itself enters the chip market.

Where Arm sits: openness vs. performance reach

A qualitative map (placements are judgments, not scores). Arm sits between fully-open RISC-V and the closed x86 duopoly while spanning the widest performance range. Hover a point for the basis.

ISA openness vs. performance reach

Hover a point to see the basis for its placement.

Against x86, Arm is winning share in the cloud: the marquee custom server CPUs are Arm-based, and Arm claims ~50% of hyperscaler CPU compute [16][8]. x86's defense is its enormous legacy software base and entrenched enterprise installs [8].

The structural threat is RISC-V — a royalty-free, open ISA that lets firms avoid Arm's fees and own their roadmap. It is entrenched in microcontrollers/IoT, Google has made it a Tier-1 Android architecture, and the first commercial RISC-V phones are expected late 2026 [17][18]. Skeptics note RISC-V still lacks Arm's mature software/tooling ecosystem at the high end, so near-term displacement is mostly low-end [17].

The newest dynamic is disintermediation cutting both ways. Architecture licensees (Apple, Qualcomm via Nuvia, the hyperscalers) design their own cores; the Arm v. Qualcomm suit — which Qualcomm won — affirmed that a licensee can carry custom Arm-compatible cores across an acquisition, limiting Arm's control [19]. And with its own AGI CPU, Arm now competes against some of those same customers (see Strategy & Moats).

Both sides of the ledger

Weigh these against each other — they are presented so you can reach your own conclusion, not to argue one way.

The case for

In mobile and embedded, Arm's ecosystem and software base make it effectively the default — switching costs are very high [16].
Arm is taking real data-center share from x86, the largest available pool of CPU value [8][6].
RISC-V still trails Arm's mature toolchain and software at the high end, slowing displacement beyond the low end [17].

The case against

RISC-V removes the royalty entirely and is backed by Google and hyperscalers seeking 'silicon sovereignty' — a direct attack on Arm's economics [17][18].
The Qualcomm loss showed Arm's leverage over how licensees reuse custom cores is weaker than assumed [19].
Arm's biggest customers are also its most capable potential competitors, and several now build their own silicon [16].

Sources for this section

4 sources · en · tiers shown. Full bibliography on the Sources page.

T3 Arm Is Quietly Becoming The CPU Backbone Of AI — 24/7 Wall St. · en · supporting
T2 Are open-source RISC-V CPUs a threat to ARM Holdings? — The Armchair Trader · en · critical
T3 The RISC-V Revolution: Open-Source Silicon Challenges ARM and x86 — FinancialContent · en · critical
T2 Qualcomm scores big win over Arm; court confirms Oryon/Nuvia cores licensed — Tom's Hardware · en · critical

Section 05

Strategy & Moats

The moat is architecture plus ecosystem lock-in. The strategy — raise royalty per chip, ride AI into the data center, and now sell its own chip — is also its biggest tension.

5 sourcesAs of 6 Jun 2026

The AGI CPU, launched March 2026 with Meta as lead customer, is the biggest strategic shift in Arm's history — and its biggest tension. Selling finished chips lets Arm capture far more value per system, but it puts Arm in direct competition with the licensees whose royalties are its core business.

Arm's durable advantage is architecture plus ecosystem: decades of software, tools and 22m+ developers built around the Arm ISA, reinforced by 350bn+ shipped chips [23]. That lock-in is what lets Arm raise per-chip royalties (Armv9, CSS) without mass defection — so far.

On 24 March 2026 Arm crossed a line it had avoided for 35 years and launched the Arm AGI CPU, its first in-house production chip, targeting agentic-AI data-center workloads — with Meta as lead customer and co-developer and OpenAI, Cerebras, Cloudflare, SAP and SK Telecom among launch partners [20][22]. Arm cited >$2bn of demand across FY27–FY28 (more than double prior guidance) and ~2x performance per rack versus x86 [23][22].

The strategic logic is value capture: a CPU sale is worth far more than a royalty, and Arm — paying itself no royalties — has a structural cost edge [21]. The risk is channel conflict: Arm now competes with Qualcomm, MediaTek, the hyperscalers and even partner Nvidia. Analysts call the conflict manageable while Arm stays in data-center CPUs (rivals are mostly mobile) but real if it expands [21][24].

Both sides of the ledger

Weigh these against each other — they are presented so you can reach your own conclusion, not to argue one way.

The case for

Deep, durable moat: the Arm ISA, 22m+ developers and decades of software/tooling make the ecosystem extremely sticky [23].
Moving up the stack multiplies value per system — a chip sale dwarfs a royalty, and Arm's no-royalty cost base lets it price aggressively [21].
Early validation: Meta as lead customer plus OpenAI/Cloudflare/SAP and >$2bn of stated AGI CPU demand suggest genuine pull, not just a reference design [20][23].

The case against

Channel conflict: selling chips pits Arm against the licensees whose royalties fund it — a risk to the core franchise if it widens [21][24].
Execution and capital intensity: building and supplying chips is a different, harder business than licensing IP, and Arm has already hit capacity limits [39].
The moat's edges are fraying — RISC-V at the low end, the Qualcomm ruling at the high end [17][19].

In their words

“Arm doesn't need to pay its own royalties and licensing fees, so it can likely afford to sell its AI accelerator chips at lower prices than its own chipmaking customers.”

The Motley Fool · investment analysis · Nov 2025 · source

Sources for this section

5 sources · en · tiers shown. Full bibliography on the Sources page.

T2 Arm launches its own CPU, with Meta as first customer — CNBC · en · neutral
T2 Arm Beats Estimates, but Its New Plan to Build Chips Is the Real Story — Motley Fool · en · critical
T3 Arm AGI CPU: The Licensing Giant Ships Its First Chip — MLQ.ai · en · supporting
T1 Arm delivers record-breaking quarter and full-year results — Arm Newsroom · en · supporting
T3 Arm Unveils First In-House Chip, the Arm AGI CPU — IndexBox · en · neutral

Section 06

Financials, Growth & Valuation

Record results and ~98% gross margins are not in dispute. The valuation — one of the richest multiples in large-cap tech — is.

7 sourcesAs of 6 Jun 2026

FY2026 revenue was $4.92bn (+23%) with ~98% gross margin and non-GAAP EPS of $1.77. Yet the stock has traded at a forward P/E around 190 — pricing in years of flawless data-center execution. The numbers are strong; the question is how much future is already in the price.

Revenue trajectory (fiscal years, US$bn)

Three consecutive years of >20% growthsince the 2023 relisting. Arm's fiscal year ends 31 March. Hover for the figure.

Total revenue (GAAP, US$bn)

For FY2026 (ended 31 March 2026) Arm reported revenue of $4.92bn, up 23%, split between royalty $2.61bn (+21%) and license & other $2.31bn (+25%) [25][26]. It was Arm's third consecutive year of >20% growth since relisting [25].

Profitability shows the two faces of the model. Non-GAAP operating margin was 43.0%, net income $1.89bn, diluted EPS $1.77. GAAP is far lower — operating margin 18.3%, net income $904m, EPS $0.85 — because heavy stock-based compensation and R&D hiring weigh on GAAP results [25]. Arm held ~$3.6bn of cash and investments [25].

Management set out a long-term ambition of ~$15bn AGI CPU revenue + ~$10bn IP revenue (~$25bn total) by ~FY2031 with >$9 EPS — the spine of the bull case [27]. Near term, it kept FY2027 at ~$10bn because AGI CPU supply is capped (only the first ~$1bn tranche secured), framing it as a supply, not demand, constraint [28].

The contested point is valuation. Arm has traded at a forward P/E near 190 and a price/sales in the 30–40x range, with valuation services flagging it as significantly overvalued and the average analyst target sitting well below the price [29][30][31]. Bulls say a unique, compounding royalty franchise at the center of AI deserves a premium; bears say the multiple leaves no room for error.

Both sides of the ledger

Weigh these against each other — they are presented so you can reach your own conclusion, not to argue one way.

The case for

Record, durable growth: $4.92bn revenue (+23%), three straight years of >20% growth, with both royalty and license lines accelerating [25].
Elite economics: ~98% gross margin, 43% non-GAAP operating margin, and royalties that compound on the installed base [25].
Large optionality: a ~$25bn revenue / >$9 EPS by ~FY2031 ambition anchored to data-center and AGI CPU demand [27].

The case against

Extreme valuation: a forward P/E near 190 and 30–40x sales price in years of perfect execution; the Street's mean target has sat ~38% below the price [29][31].
GAAP profitability is modest — 18.3% GAAP operating margin and $0.85 GAAP EPS — because stock-comp and R&D are large [25].
The long-term targets depend on an unproven chip business that is already supply-constrained [28][39].

In their words

“Arm delivered a third consecutive year of more than 20% revenue growth, driven by strong demand for the Arm compute platform.”

Rene Haas · CEO, Arm Holdings · May 2026 · source

Sources for this section

7 sources · en · tiers shown. Full bibliography on the Sources page.

T1 Arm delivers record FY2026 results and launches AGI data center CPU (6-K) — StockTitan · en · neutral
T1 Arm Holdings plc — Form 6-K, Q4/FY2026 results (Exhibit 99.2) — SEC · en · neutral
T2 Arm Q4 FY2026 slides: $15B chip revenue target, >$9 EPS by 2031 — Investing.com · en · supporting
T2 Earnings call transcript: Arm Holdings reports record Q4 FY2026 results — Investing.com · en · neutral
T2 ARM Holdings Forward PE Ratio — GuruFocus · en · critical
T2 Arm Holdings (ARM) Statistics & Valuation — StockAnalysis · en · neutral
T3 ARM Stock Is Up in 2026 but the Street Mean Target Sits 38% Below — TIKR · en · critical

Section 07 · Benchmarking

Peer Comparison

Arm against the chipmakers it enables and the design-IP peers it most resembles. Figures are most-recent-fiscal-year disclosures; margins are approximate.

6 peersAs of 6 Jun 2026

⚖️

Not a like-for-like

Arm sells IP, not chips, so revenue comparisons understate its strategic weight: it earns a few dollars where a chipmaker earns hundreds, yet sits inside their products. Read the table as scale and economics, not market share.

Company	Model	Revenue	Margin	Relationship to Arm
Arm	CPU architecture & IP licensing	$4.92bn (FY2026)	~98% gross / 43% non-GAAP op.	Licenses the ISA the others build on
Nvidia	AI GPUs & systems	~$130.5bn (FY2025)	~75% gross	Customer (Grace/Vera use Arm); ex-suitor
Qualcomm	Mobile/edge chips + licensing	~$44.3bn (FY2025)	~56% gross / ~30% non-GAAP NI	Major licensee and litigation adversary
AMD	x86 CPUs + GPUs	~$34.6bn (FY2025)	~50% gross	x86 rival in data center/PC
Intel	x86 IDM (designs + fabs)	~$52.9bn (FY2025)	~35% gross (turnaround)	x86 incumbent losing server share
Synopsys	EDA tools + design IP	~$7.05bn (FY2025)	~80% gross	Adjacent design-IP/tools peer
Cadence	EDA tools + design IP	~$5.29bn (FY2025)	~86% gross	Adjacent design-IP/tools peer

Revenue: Arm is tiny — by design

Arm earns a royalty where its customers earn a chip sale, so it is a fraction of their revenue while enabling much of it. Hover a bar for detail.

Most-recent-FY revenue (US$bn)

Nvidia

$130.5bn

Intel

$52.9bn

Qualcomm

$44.3bn

AMD

$34.6bn

Synopsys

$7.05bn

Cadence

$5.29bn

Arm

$4.92bn

Gross margin: the IP model tops the set

The flip side of small revenue is its ~98% gross margin — Arm's capital-light licensing model converts almost all revenue to gross profit, ahead even of the EDA peers.

Approximate gross margin (%)

Arm

98%

Cadence

86%

Synopsys

80%

Nvidia

75%

Qualcomm

56%

AMD

50%

Intel

35%

Sourced figures and the full competitive read are on the Competitive Landscape and Sources pages.

Section 08

Risks & Open Questions

Valuation, customer concentration, the Qualcomm precedent, China and Arm China, RISC-V, SoftBank control, and execution on the new chip business.

9 sourcesAs of 6 Jun 2026

None of Arm's risks is fatal alone, but they cluster around one theme: dependence on a handful of very powerful counterparties — its top three customers are 42% of revenue, SoftBank owns ~87%, ~a fifth-to-a-quarter of revenue runs through a now-independent Arm China, and its new chip depends on TSMC capacity.

Customer concentration. Arm's three largest customers were 42% of FY2026 revenue; the loss or weakness of any one would hit Arm harder than a diversified firm [38]. Those customers (Apple, Qualcomm, the hyperscalers, Nvidia) are also the parties most able to build their own cores or move to RISC-V.

Litigation / license precedent. Qualcomm won a 'complete victory' in the Nuvia dispute, with a Delaware judge dismissing Arm's last claim in October 2025; Arm said it would appeal and a Qualcomm counter-suit remained pending [42][43]. Analysts read it as constraining Arm's power to police how licensees reuse custom cores — material to the value-capture strategy [44].

China & Arm China. Roughly a fifth-to-a-quarter of Arm's revenue ties to China, much of it through Arm China (安谋科技), the joint venture Arm lost effective control of after CEO Allen Wu — voted out 7-1 by the board — refused to leave and used his hold on the company seal (chop), which under Chinese law binds the entity, to retain control and oust Arm-loyal executives; Arm today holds only a minority economic stake and limited operational control, concentrating geopolitical and export-control risk [41].

Execution, supply and the cycle. Arm shares fell ~7% on 7 May 2026 after it said it could not secure enough capacity to meet AGI CPU demand (the chip is made at foundries like TSMC) and kept its outlook conservative [39]. It also warned smartphone units could decline on a memory shortage — though the CFO sized a 20% unit drop as only a ~1–2% royalty hit [40]. And SoftBank's ~87% stake means a thin float, high volatility (the stock fell ~20% in one month), and a potential overhang if SoftBank sells [45][47].

Both sides of the ledger

Weigh these against each other — they are presented so you can reach your own conclusion, not to argue one way.

The case for

Risks are mostly about *concentration of upside* — Arm's customers are the strongest companies in tech, and switching off Arm is expensive and slow [38].
The smartphone-cycle risk is quantified and small: a ~20% unit drop is only a ~1–2% royalty hit, because Armv9/CSS rate gains dominate [40].
Most risks (RISC-V, litigation, China) are long-running and known, not sudden shocks; Arm has navigated them while compounding revenue [44].

The case against

Heavy concentration: top-3 customers are 42% of revenue and are also Arm's most capable rivals [38].
The Qualcomm precedent and RISC-V both chip at Arm's pricing power from opposite ends [42][17].
China exposure runs through an entity Arm no longer controls, and SoftBank's ~87% stake plus a thin float make the stock unusually volatile [41][47].

In their words

“Arm is nowhere without Taiwan.”

Rene Haas · CEO, Arm Holdings (at Computex 2026) · Jun 2026 · source

Sources for this section

9 sources · en · tiers shown. Full bibliography on the Sources page.

T1 Arm Holdings plc — Form 20-F (FY ended 31 Mar 2026) — SEC · en · critical
T2 Arm Holdings Posts Record Revenue but Stock Slides 7% — HeyGoTrade · en · critical
T2 Why is Arm Holdings stock sliding today? — Investing.com · en · critical
T2 The Semiconductor Heist Of The Century: Arm China Has Gone Rogue — SemiAnalysis · en · critical
T1 Qualcomm Achieves Complete Victory Over Arm in Litigation — Qualcomm · en · critical
T2 Judge dismisses Arm's last legal claim against Qualcomm — The Register · en · critical
T2 Litigation SITREP: The Ongoing Dispute Between Qualcomm and Arm — Futurum · en · neutral
T3 Why Arm Holdings Can Offer 25% Recovery in 2026 After Falling 24% — TIKR · en · supporting
T2 Why Arm Holdings Plunged Nearly 20% in December — Motley Fool · en · critical

Methodology & Limitations

How this was made — and where it may be wrong

An independent, point-in-time research artifact: the method, the frameworks, what's estimated vs. disclosed, and the known weaknesses.

As of 6 Jun 2026Independent · not affiliated

Method

Research proceeded by fan-out web search and direct fetching of primary and reputable secondary sources across eight question areas (overview, market, business model, competition, strategy, financials, peer comparison, and risks). Every URL cited was opened and read during the run; each claim was transcribed into a structured manifest tagging it with a source tier, a confidence level, and a stance, and an automated link checker validated every URL. The load-bearing figures here — Arm's FY2026 revenue of $4.92bn, the royalty/license split, the GAAP-versus-non-GAAP margin gap, the ~50% hyperscaler-compute share, customer concentration (42% from the top three), and the AGI CPU demand and supply commentary — rest on Arm's reported FY2026 results (filed on Form 6-K and summarized via a reporting mirror), its Form 20-F, the Arm newsroom, and the earnings call as reported. Because Arm is a UK/US company with English-language primary disclosure, no native-language pass beyond English was required.

Frameworks used

The analysis applies the Pyramid Principle for the answer-first Executive Summary; Porter's Five Forces for the CPU-architecture market, with each force rated against a sourced basis; a peer-comparables benchmark against Nvidia, Qualcomm, AMD, Intel, Synopsys and Cadence on revenue, margin and relationship to Arm; a value-chain framing to locate Arm above the chipmakers and foundries; and a qualitative two-axis positioning map (ISA openness vs. performance reach). A case-for / case-against ledger runs in every section so the bull and bear cases get equal scrutiny, and for each section a disconfirming search was run to surface the other side. A BCG portfolio matrix and a formal unit-economics waterfall were deliberately skipped: Arm does not break out product-line profitability cleanly enough to fill them honestly, and an empty framework is worse than none.

Disclosed vs. estimated

Disclosed, high-confidence figures — FY2026 revenue, royalty and license lines, GAAP and non-GAAP margins and EPS, cash, and the 42% top-three customer concentration — come from Arm's reported results and Form 20-F. Cross-company comparisons are directional: peer revenue and margins mix different fiscal-year ends and providers and are rounded. The ~50% hyperscaler-compute share, the ~18% mid-2024 starting point, the per-chip royalty rates for Armv9 (~2x v8) and CSS (>10%), the Armv9 share of royalties (~25%), and the ~$2bn AGI CPU demand are Arm's own figures or analyst characterizations, not independently audited; valuation multiples (a forward P/E near 190) and the Arm China revenue share (~a fifth to a quarter) are third-party estimates. Each is labeled in the prose where it carries weight.

⚠️

Where this case study may be wrong

Company-favorable metrics.The ~50% hyperscaler-compute share, per-chip royalty rates, and AGI CPU demand are Arm's own framing; the share in particular uses a flattering denominator (hyperscaler compute, not the whole server market).
Estimates. Valuation multiples, the Arm China revenue share, and peer margins are third-party or rounded figures, not Arm disclosures, and providers disagree.
Fast-moving and contested.The Qualcomm appeal and counter-suit, RISC-V adoption, and the stock's valuation move week to week; anything here can be stale soon after the as-of date.
Point-in-time. This is a snapshot as of 6 June 2026; figures go stale at the next earnings release.

Neutrality & independence

This is a compilation, not an argument. Every section pairs the case for and against with sourced evidence; the Executive Summary frames open questions rather than selling a verdict, and the Risks section stops short of a buy/sell call. The achieved evidence mix is disclosed for transparency — supporting 13 · critical 17 · neutral 17 citations. The Teardown is independent and not affiliated with, endorsed by, or sponsored by Arm Holdings or SoftBank. It is a point-in-time artifact as of 6 June 2026 and is not investment advice.

Full bibliography with tiers, stance, and language on the Sources page.

Bibliography

Sources

Every cited source was fetched during the research run. Tiers: 1 = primary/official, 2 = reputable press, 3 = tertiary/soft.

47 sourcesAll English-language

Tier 1: 9Tier 2: 25Tier 3: 13·Supporting: 13Critical: 17Neutral: 17

Overview & Timeline

[1]Arm Holdings — Wikipedia T2 neutral en
Arm Holdings is a Cambridge, UK semiconductor-design firm founded November 1990 as Advanced RISC Machines, a joint venture of Acorn, Apple and VLSI; it IPO'd in 1998, was taken private by SoftBank in 2016 for £24.3bn (~$32bn), saw a $40bn Nvidia acquisition collapse in 2022, and returned to NASDAQ in September 2023; SoftBank owns ~87%.
[2]The official history of Arm — Arm Newsroom T1 supporting en
Arm's official history: the ARM1 was designed by Steve Furber and Sophie Wilson and delivered in 1985; Arm was incorporated in 1990 with 12 engineers; Robin Saxby pioneered the IP-licensing model.
[3]How Arm gained chip dominance with Apple, Nvidia, Amazon and Qualcomm — CNBC T2 supporting en
Arm's architecture sits inside chips from Apple, Nvidia, AMD, Amazon and Qualcomm; its low-power designs became the default for mobile, giving it a near-ubiquitous position built over three decades.
[4]SoftBank to acquire ARM Holdings in $32bn deal — Financier Worldwide T2 neutral en
SoftBank agreed to acquire Arm in 2016 for ~$32bn, the largest-ever acquisition of a European technology company.
[5]Arm Market Value, SoftBank's 87% Stake — TradingKey T3 critical en
SoftBank's ~87% stake makes it Arm's controlling shareholder and the dominant beneficiary of the share-price rise; the thin float amplifies volatility, and SoftBank's own capital needs (e.g. AI investments) create a potential supply overhang if it sells down.

Market & Industry

[6]ARM Holdings: Shaping the Future of Computing — Leverage Shares T3 supporting en
Arm's architectures power approximately 99% of the world's smartphones plus a growing share of data center, automotive and IoT devices; content per device is rising as cars and edge devices become 'computers'.
[7]Arm posts record revenue, hyperscaler CPU share hits 50% — DCD T2 supporting en
Arm posted record FY2026 revenue and said its share of hyperscaler CPU compute reached ~50%, with data center royalties more than doubling year over year.
[8]Arm expects its data center CPU share to reach 50% — SemiWiki T2 supporting en
Arm publicly targeted ~50% of data center CPU sockets/sales; its hyperscaler share was roughly 18% in mid-2024 before the AI-infrastructure build-out accelerated adoption.
[9]x86 Data Center Dominance Ends: Arm Crosses 50% Hyperscaler CPU Share — TechTimes T2 neutral en
At Computex 2026 Arm said it had crossed 50% of hyperscaler CPU compute, framed by commentators as the end of x86's data-center dominance — though the figure is Arm's own and measures hyperscaler compute, not the whole server market.
[10]Arm aims to capture 50% of data center CPU market — Tom's Hardware T2 neutral en
Earlier coverage framed the 50% data-center figure as a stated aim/target rather than a settled fact, a reminder the milestone is Arm's goal and self-reported metric.
[11]RISC-V 2026: How Open Chip Architecture Is Disrupting ARM and Intel — Programming Helper T3 critical en
RISC-V is described as disrupting the Arm/Intel order as a 'third pillar' of computing, a structural challenge to Arm's position in the very compute markets it is trying to grow into. Such forecasts are promotional and penetration estimates vary widely.

Business Model

[12]Arm and a Leg: Arm's Quest To Extract Their True Value — SemiAnalysis T2 neutral en
Arm earns upfront license fees (estimated ~$1–10m) when it ships IP plus per-chip royalties (historically ~1–2% of chip ASP); the model is capital-light with near-100% gross margin, and Arm has pushed to capture more value per chip.
[13]ARM Licensing vs Royalties Business Model Deep Dive — Screenwich T3 neutral en
Arm licenses at three tiers — Technology License Agreements (a ready-made core, higher royalty), 'Built on Arm Cortex' (some tuning), and Architecture License Agreements (design your own core implementing the Arm ISA, lower royalty); TLAs carry higher royalty rates than ALAs.
[14]Arm Q4 FY2025: Record Revenue as Armv9 and CSS Ramp Up — Futurum T2 supporting en
Armv9 carries roughly double the royalty rate of Armv8 and was ~25% of royalty revenue in early FY2025; Compute Subsystems (CSS) — pre-integrated, validated reference designs — carry royalty rates reportedly north of 10% per chip versus 1–3% for traditional licensing.
[15]Arm Holdings and the Game-Changing Potential of Compute Subsystems — AInvest T3 supporting en
CSS is positioned as Arm's lever to raise per-chip economics; by FY2026 Arm reported a growing roster of CSS licenses across mobile, edge and cloud, with the top Android smartphone vendors shipping CSS-powered devices.
[16]Arm Changes Business Model — OEM Partners Must Directly License From Arm — SemiAnalysis T2 critical en
Arm has sought to change its licensing model — including reported plans to charge device OEMs directly rather than chipmakers — a move critics argue raises costs for partners and risks straining the ecosystem Arm depends on.

Competitive Landscape

[17]Arm Is Quietly Becoming The CPU Backbone Of AI — 24/7 Wall St. T3 supporting en
Arm-based CPUs (AWS Graviton, Nvidia Grace/Vera, Microsoft Cobalt, Google Axion) have become the default host CPU paired with AI accelerators, positioning Arm as the CPU backbone of AI data centers.
[18]Are open-source RISC-V CPUs a threat to ARM Holdings? — The Armchair Trader T2 critical en
RISC-V, a royalty-free open ISA, is a structural competitive threat: it lets firms avoid Arm's per-chip royalties and license fees and control their own roadmap, and is gaining in embedded/IoT and increasingly in higher-performance roles.
[19]The RISC-V Revolution: Open-Source Silicon Challenges ARM and x86 — FinancialContent T3 critical en
Early-2026 coverage framed RISC-V as a fast-rising 'third pillar' of computing, with Google making it a Tier-1 Android architecture and the first commercial RISC-V phones expected late 2026; some reports cite large RISC-V deployments saving up to ~50% of development cost. Penetration figures vary widely and many such reports are promotional.
[20]Qualcomm scores big win over Arm; court confirms Oryon/Nuvia cores licensed — Tom's Hardware T2 critical en
A US court rejected Arm's claims against Qualcomm, confirming Qualcomm can use the Oryon cores it acquired via Nuvia under Qualcomm's own Arm license — a setback that limits Arm's ability to police how licensees reuse custom designs.

Strategy & Moats

[21]Arm launches its own CPU, with Meta as first customer — CNBC T2 neutral en
On 24 March 2026 Arm launched the Arm AGI CPU, its first own data-center CPU, with Meta as lead customer and co-developer — a shift from pure IP licensing toward selling finished chips that compete in the same market as some of its licensees.
[22]Arm Beats Estimates, but Its New Plan to Build Chips Is the Real Story — Motley Fool T2 critical en
Arm's move into selling chips creates channel conflict: because Arm pays itself no royalties it could undercut licensees like Qualcomm and MediaTek on price; analysts argue the conflict is manageable near-term (rivals are mostly in smartphones) but real if Arm expands.
[23]Arm AGI CPU: The Licensing Giant Ships Its First Chip — MLQ.ai T3 supporting en
The Arm AGI CPU was described as production-ready with Meta as lead partner and OpenAI, Cerebras, Cloudflare, SAP and SK Telecom among launch partners; Arm framed it as targeting agentic-AI data-center workloads, citing ~2x performance per rack vs x86.
[24]Arm delivers record-breaking quarter and full-year results — Arm Newsroom T1 supporting en
Arm reported cumulative chips shipped above 350 billion and a developer ecosystem above 22 million; it said AGI CPU customer demand was more than $2bn across FYE27–FYE28 (more than double prior guidance) and cited up to $10bn-per-gigawatt capex-saving potential.
[25]Arm Unveils First In-House Chip, the Arm AGI CPU — IndexBox T3 neutral en
Commentators called the AGI CPU 'the biggest strategic shift in [Arm's] history,' noting shareholders had long asked why Arm did not sell finished chips itself — while flagging the implicit tension with licensees.

Financials & Valuation

[26]Arm delivers record FY2026 results and launches AGI data center CPU (6-K) — StockTitan T1 neutral en
FY2026 (ended 31 Mar 2026): total revenue $4,920m (+23%), royalty $2,613m (+21%), license/other $2,307m (+25%); GAAP operating income $900m (18.3% margin), GAAP net income $904m, GAAP diluted EPS $0.85; non-GAAP operating margin 43.0%, non-GAAP net income $1,889m, non-GAAP diluted EPS $1.77; cash & ST investments ~$3.6bn. Q1 FY2027 guidance: revenue $1.26bn ±$50m, non-GAAP EPS $0.40.
[27]Arm Holdings plc — Form 6-K, Q4/FY2026 results (Exhibit 99.2) — SEC T1 neutral en
Arm's Q4/FY2026 results were filed with the SEC on Form 6-K (exhibit 99.2), the primary disclosure for the revenue, margin and EPS figures cited here.
[28]Arm Q4 FY2026 slides: $15B chip revenue target, >$9 EPS by 2031 — Investing.com T2 supporting en
At Q4 FY2026 Arm laid out long-term targets of ~$15bn AGI CPU revenue and ~$10bn IP revenue by ~FY2031 (~$25bn total) with >$9 EPS — an aggressive ramp the bull case is built on and the bear case treats as unproven.
[29]Earnings call transcript: Arm Holdings reports record Q4 FY2026 results — Investing.com T2 neutral en
On the FY2026 earnings call management said FY2027 revenue is held at ~$10bn because AGI CPU supply (only the first ~$1bn tranche secured) caps near-term recognition despite demand, framing supply as a near-term overhang rather than a demand problem.
[30]ARM Holdings Forward PE Ratio — GuruFocus T2 critical en
As of mid-2026 Arm traded at extreme multiples — a forward P/E reported around 190 and trailing P/E far higher — with valuation services flagging the stock as significantly overvalued versus estimated fair value.
[31]Arm Holdings (ARM) Statistics & Valuation — StockAnalysis T2 neutral en
Arm's market capitalization in 2026 ran into the hundreds of billions of dollars on a small public float (SoftBank holds ~87%), producing very high price-to-sales and price-to-earnings ratios relative to semiconductor peers.
[32]ARM Stock Is Up in 2026 but the Street Mean Target Sits 38% Below — TIKR T3 critical en
After a large 2026 run-up, the average Wall Street price target sat well below the trading price (one analysis put the Street mean ~38% below), reflecting widespread analyst discomfort with the valuation even amid strong fundamentals.

Peer Comparison

[33]NVIDIA (NVDA) — Market capitalization — CompaniesMarketCap T3 critical en
Nvidia — Arm's largest data-center customer and former would-be acquirer — reported FY2025 revenue ~$130.5bn and by mid-2026 a market cap in the multi-trillion-dollar range, dwarfing Arm in absolute scale while depending on Arm IP for its Grace/Vera CPUs.
[34]Qualcomm Inc — Q4/FY2025 results (Exhibit 99.1) — SEC T1 neutral en
Qualcomm reported fiscal 2025 revenue ~$44.3bn with GAAP net income ~$5.5bn and non-GAAP ~$13.3bn (non-GAAP net margin ~30%); it is both a major Arm licensee and, via Oryon/Nuvia, an emerging Arm-architecture competitor.
[35]AMD Reports Fourth Quarter and Full Year 2025 Financial Results — AMD T1 neutral en
AMD reported full-year 2025 revenue ~$34.6bn with ~50% gross margin, operating income ~$3.7bn and net income ~$4.3bn; AMD is the x86 challenger whose share gains show Arm's data-center rise is partly at Intel's, not AMD's, expense.
[36]Intel Corp (INTC) Financials — StockTitan T3 neutral en
Intel — the incumbent x86 data-center leader Arm is taking share from — reported ~$52.9bn FY2025 revenue with depressed/negative operating margins amid its turnaround, a contrast to Arm's ~98% gross-margin licensing model.
[37]Synopsys records record-breaking full-year revenue — SiliconANGLE T2 supporting en
Synopsys — an EDA/IP peer that, like Arm, sells design tools and IP rather than chips — reported record FY2025 revenue ~$7.05bn (+~15%), illustrating the high-margin, recurring economics of the chip-design layer Arm sits adjacent to.
[38]Cadence Reports Q4 and FY2025 Financial Results — Cadence T1 neutral en
Cadence reported FY2025 revenue ~$5.29bn with non-GAAP operating margin ~44.6% — another high-margin design-IP/EDA comparable whose multiple, while rich, is below Arm's.

Risks & Open Questions

[39]Arm Holdings plc — Form 20-F (FY ended 31 Mar 2026) — SEC T1 critical en
Per Arm's FY2026 Form 20-F, its three largest customers accounted for 42% of total revenue in fiscal 2026, and Arm flags that adverse developments at a key customer would hit revenue more than for a diversified firm.
[40]Arm Holdings Posts Record Revenue but Stock Slides 7% — HeyGoTrade T2 critical en
Despite record FY2026 results, Arm shares fell ~7% on 7 May 2026 after it said it could not secure enough manufacturing capacity to meet AGI CPU demand and kept its ~$10bn FY2027 outlook — execution/supply risk on the new chip business.
[41]Why is Arm Holdings stock sliding today? — Investing.com T2 critical en
Management also warned smartphone unit growth could turn negative on a memory-chip shortage, though the CFO quantified a ~20% drop in smartphone units as only a ~1–2% hit to total royalties — the mature-mobile-market risk in context.
[42]The Semiconductor Heist Of The Century: Arm China Has Gone Rogue — SemiAnalysis T2 critical en
Arm China (安谋科技) — which sells Arm IP into China and has accounted for roughly a fifth to a quarter of Arm revenue — became effectively independent after a prolonged governance fight; Arm holds a minority economic stake and limited operational control, concentrating geopolitical and China-exposure risk.
[43]Qualcomm Achieves Complete Victory Over Arm in Litigation — Qualcomm T1 critical en
Qualcomm announced a 'complete victory' over Arm in September/October 2025 as the court confirmed neither Qualcomm nor Nuvia breached the Nuvia license; Arm said it would appeal, and Qualcomm's counter-suit remained pending — ongoing legal and licensing-enforcement risk.
[44]Judge dismisses Arm's last legal claim against Qualcomm — The Register T2 critical en
A Delaware judge dismissed Arm's last remaining claim against Qualcomm in October 2025, ending Arm's bid to force destruction of the disputed cores — narrowing Arm's leverage over how licensees reuse acquired custom designs.
[45]Litigation SITREP: The Ongoing Dispute Between Qualcomm and Arm — Futurum T2 neutral en
Independent analysis framed the Arm–Qualcomm dispute as high-stakes for the architecture-license model itself: a ruling for Qualcomm preserves licensees' ability to carry custom cores across acquisitions, constraining Arm's pricing/enforcement power.
[46]Why Arm Holdings Can Offer 25% Recovery in 2026 After Falling 24% — TIKR T3 supporting en
A bull/mitigant reading argues Arm's risks are surmountable: after falling ~24% the prior year, the stock had room to recover on durable royalty growth — framing the volatility and concentration as cyclical, not structural.
[47]Why Arm Holdings Plunged Nearly 20% in December — Motley Fool T2 critical en
Arm shares are highly volatile around the valuation debate — the stock fell nearly 20% in December (per Motley Fool) on AI-spending and multiple-compression worries, underscoring how sensitive the price is to sentiment shifts.

Cross-checked at build time by an automated link checker; a few primary sources (e.g. SEC filings) are bot-walled to automated fetchers and were verified manually. See Methodology & Limits.