Arm Holdings plc

1Business snapshot

Arm designs the instruction set architecture and the processor designs that implement it, and licenses both to chipmakers. Customers pay an upfront licence fee — hundreds of thousands to hundreds of millions of dollars depending on scope and duration — then a royalty on every chip shipped, set as a percentage of the chip's selling price or a fixed fee per unit FY2026 20-F. Royalty is recognised in the quarter the licensee ships, on Arm's own estimate, and trued up when licensee reports arrive — so reported royalty lags the physical supply chain by roughly a quarter.

FY2026 revenue was $4,920m, up 23%: royalty $2,613m (+21%) and licence and other $2,307m (+25%), at a 98% gross margin and $900m of GAAP operating income FY2026 20-F. More than 350 billion Arm-based chips have shipped cumulatively. Arm has held above 99% share of mobile applications processors for years; that market is still 43% of royalty revenue. The business is organised around three AI domains — Edge AI (smartphones, IoT), Physical AI (automotive, robotics) and Cloud AI (data center, networking).

In March 2026 Arm did something structurally new: it launched the Arm AGI CPU, its own production data-center silicon, developed with Meta as lead partner 24 Mar 2026 press release. That moves Arm from selling design files to selling chips, in sockets where it previously collected only a royalty.

Arm employs 9,584 people across 19 countries, 84% in engineering. SoftBank Group owns 86.4% of the shares FY2026 20-F.

2Core debate

Is Arm a pricing machine — extracting more revenue each year from a socket it already owns, with a genuinely new and higher-value data-center pool on top — or is the headline growth rate partly manufactured by its controlling shareholder while the licensees it now competes with quietly fund the alternative, and the move into silicon dilutes the very margin structure the multiple is paid for?

The two sides do not disagree about the technology. They disagree about how much of FY2026's 23% growth was arm's-length, whether pricing power survives Arm competing with its own customers, and what a blended gross margin looks like once $2bn of wafers sits beside $5bn of IP.

3Base thesis

Arm gets paid more each year for the same chip, and has just opened two new ways to get paid for chips it previously only rented cores to. Unit growth is not available — mobile share is already above 99% — so the case must be, and is, a rate story: royalty per chip rises as Armv9 and Compute Subsystems displace older, cheaper content, while the data center adds sockets whose royalty is a multiple of a smartphone's. On top of that, the AGI CPU converts some of those sockets from a royalty into a whole-chip sale. Because Arm funds one core design and amortises it across every licensee, a fixed engineering base spread over a rising royalty stream should eventually restore operating margin.

Assumptions

1. Royalty per chip compounds above ~20% on a roughly flat unit base. Royalty rose 21% in FY2026 and 22% to $715m in Q1 FY2027, attributed to "an improved mix of products with higher royalty rates per chip, such as Armv9 technology" FY2026 20-F; 29 Jul 2026 letter.
2. The data center becomes a second royalty pool of comparable value. Data-center royalties more than doubled year over year in Q1 FY2027; cumulative Neoverse shipments passed 1.5 billion cores, the most recent 500 million in nine months against six years for the first billion 29 Jul 2026 letter.
3. The AGI CPU converts to shipped revenue. Stated customer demand rose from a "$1 billion opportunity" across FY2027–28 to "exceeds $2 billion" within one quarter, with manufacturing capacity secured for the $1bn case and initial product delivered 29 Jul 2026 letter.
4. The spending phase ends. R&D growth falls below revenue growth and GAAP operating margin recovers from single digits.
5. Switching costs hold. Every key mobile OS depends on Arm cores, so a customer seeking a cheaper royalty must move an ecosystem, not swap a core.

4Main bear case

The most likely way this loses money is not a bad quarter. It is that Arm's reported growth turns out to have been bought from its own parent, while the licensees it now competes with fund the alternative.

Two things happened in FY2026 that are easy to read past. First, revenue from a SoftBank Group affiliate under a consulting agreement was $704.4m against $145.5m a year earlier — $559m of the $913m total revenue increase, from an entity controlled by the 86.4% shareholder FY2026 20-F. Strip it out and FY2026 revenue grew about 9%, with third-party licence revenue down 9% to $1,298m (derived from disclosed figures). ACV decelerated from +22% at March 2026 to +13% at June 2026 6 May and 29 Jul 2026 letters. Second, Arm entered its customers' socket: Broadcom, Marvell, Qualcomm and the hyperscalers now buy IP from a competitor.

How it breaks. A licensee cannot punish Arm this quarter — royalties are contractual and switching is slow. It starts a RISC-V or in-house architecture programme that removes a socket in five years, invisible in the numbers until irreversible. Meanwhile the AGI CPU, if it works, pulls the 98% gross margin down, and equity compensation of $1,052m against $900m of GAAP operating income caps the operating leverage the thesis needs.

5Business quality

The moat is a software ecosystem, not a patent estate. Because the world's operating systems and applications are compiled for Arm, the cost of leaving is borne by the customer while the benefit of raising price accrues to Arm. That shows up as a 98% gross margin, near-zero marginal cost per chip, and a royalty stream that keeps paying on designs licensed a decade ago.

The quality is qualified in three ways. First, concentration: the top five customers — including Arm China and SoftBank Group — were 57% of FY2026 revenue, and Arm China alone 16% FY2026 20-F. Second, the largest single growth contributor in FY2026 was the parent. Third, and newest, Arm has voluntarily degraded its own structural position by competing with the customers who pay the royalties — a choice, not an accident. Capital intensity is still low but rising fast: capex went from $92m (FY2024) to $219m (FY2025) to $545m (FY2026), and property and equipment more than doubled to $772m FY2026 20-F.

6Key value drivers

• Royalty rate, not volume. Mobile share cannot grow; every incremental royalty dollar there must come from Armv9 and CSS mix.
• Data-center attach rate. A server CPU carries a far larger royalty than a smartphone SoC, and every AI accelerator needs a host CPU. Named programmes: NVIDIA Vera (full production), Google Axion, AWS Graviton5, Azure Cobalt 200, Qualcomm Dragonfly C1000 29 Jul 2026 letter.
• AGI CPU conversion. Turns dollars of royalty into thousands of dollars of silicon revenue per unit — at a much lower margin.
• ACV. The cleanest read on arm's-length licensing demand.
• R&D leverage. One design amortised across all licensees is the only route back to a normal GAAP margin.

7Financial profile

Source: FY2026 20-F. Revenue mix is roughly 53% royalty / 47% licence.

The balance sheet is unlevered and liquid: $3,888m of cash and short-term investments at 30 June 2026 against no borrowings; total equity $8,286m of $10,703m assets FY2026 20-F; 29 Jul 2026 letter. Trailing-twelve-month non-GAAP free cash flow was $1,397m at Q1 FY2027, against $597m a year earlier.

Two features distort the reported picture. Equity-settled share-based compensation of $1,052m exceeded GAAP operating income in FY2026 — the cost lands in the income statement rather than the share count, since diluted shares rose only from 1,044m to 1,068m over two years. And contract assets reached $977m current plus $320m non-current, of which $645.8m was unbilled revenue from the SoftBank affiliate.

Q1 FY2027: revenue $1,289m (+22%), royalty $715m (+22%), licence $574m (+23%), GAAP gross margin 97.2%, GAAP operating income $91m (7.1% margin, from 10.8%), non-GAAP operating margin 41.2%, net income $270m. Q2 FY2027 guidance: revenue $1.38bn ± $50m, non-GAAP diluted EPS $0.47 ± $0.04 29 Jul 2026 letter.

8Valuation

Arm is priced as an annuity with an option on the data center attached. The question is not whether the multiple is high — it is — but which earnings base it applies to. Three give very different answers: GAAP operating income of $900m (depressed by the R&D build and by equity comp that exceeds it); non-GAAP earnings, which add back roughly a quarter of revenue in stock compensation; and a normalised royalty stream that assumes the investment phase ends. The bull case implicitly capitalises the third. The growth rate being capitalised is also the reported 23%, not the ~9% arm's-length figure derived in §4 — the single largest valuation sensitivity in this file. And as AGI CPU silicon scales toward $2bn against ~$5bn of IP revenue, blended gross margin falls mechanically however well Arm executes.

9Management & capital allocation

Rene Haas has been CEO since February 2022, after running the IP Product Groups from 2017 and seven years at NVIDIA before that; Jason Child has been CFO since November 2022, previously CFO of Splunk, Groupon and Amazon International. Masayoshi Son chairs the board. The signature capital-allocation decision is the AGI CPU: rather than return cash, Arm is spending it on a business with structurally worse economics than the one it has, on the argument that the socket is worth more than the royalty. No dividend and no buyback; FY2026 financing outflow was almost entirely $529m of withholding tax on vested shares. The one divestiture was the Artisan foundation IP business, sold to Cadence in August 2025 for $127m.

Two governance facts deserve standing attention. Haas was appointed CEO of SoftBank Group International in April 2026 — described as limited and part-time — while remaining Arm's CEO and a SoftBank Group board member; the 20-F itself flags the competing demands and conflict risk. And Haas's FY2026 remuneration was $60.6m, of which $55.6m was share awards, in a year when equity compensation exceeded GAAP operating income.

Claims to verify over time. That the AGI CPU is incremental rather than a substitute for royalty revenue; that the SoftBank consulting arrangement is priced at arm's length; and that R&D growth is a phase rather than the new run rate.

10Risks ranked by damage

R1 — Arm now competes with the customers who pay its royalties

Mechanism

The AGI CPU and Arm's direct hyperscaler relationships put it in the same socket as Broadcom, Marvell, Qualcomm and other licensees. A licensee cannot punish Arm this quarter; it starts a RISC-V programme that removes a socket in five years.

Evidence today

Arm's own 20-F says customers may "seek alternative architectures", "withhold sensitive roadmap information" or "demand more favorable commercial terms".

What would confirm it

An architecture licensee announcing a non-Arm core for a segment where it currently ships Arm; renewals on visibly worse terms.

Impact on value

Permanent. The royalty base is the entire durable asset, and an architecture that leaves does not come back.

R2 — Arm cannot enforce the licence terms that make architecture licensing safe

Mechanism

If an architecture licensee can acquire a design house and keep its cores, an architecture licence becomes a cheap option to escape implementation royalty rates.

Evidence today

Arm sued Qualcomm and Nuvia to force destruction of cores developed under a terminated architecture licence. The jury found the technology licensed under Qualcomm's own ALA, and in September 2025 the court granted Qualcomm judgment as a matter of law. The appeal is pending; Qualcomm's counter-suit goes to trial in Q4 CY2026. Qualcomm was 9% of FY2026 revenue FY2026 20-F.

What would confirm it

Affirmation on appeal; an adverse Q4 CY2026 verdict; other licensees restructuring toward architecture-only terms.

Impact on value

The direct customer exposure is bounded; the price ceiling is not. Implementation royalties hold only while the cheaper legal route stays closed.

R3 — Most of FY2026's growth came from the controlling shareholder

Mechanism

A related-party consulting line inflates the growth rate the market capitalises, without proving third-party demand.

Evidence today

Consulting revenue $704.4m in FY2026 vs $145.5m in FY2025 — 14% of revenue and $559m of the $913m increase; $645.8m unbilled at 31 March 2026; third-party licence revenue down 9%; ex-consulting growth ≈ 9% (derived). ACV decelerated from +22% to +13% FY2026 20-F; 6 May and 29 Jul 2026 letters.

What would confirm it

The consulting line flattening while ACV growth stays in the low teens.

Impact on value

Less to cash flow than to the multiple — the market has been capitalising a 23% grower whose arm's-length business grew about 9%.

R4 — Production silicon destroys the financial model the multiple is paid for

Mechanism

Arm's 98% gross margin exists because it ships files, not wafers. Mixing $2bn of silicon into ~$5bn of IP revenue lowers blended margin arithmetically.

Evidence today

The 20-F warns of "margin compression, greater quarter-to-quarter revenue volatility, increased capital intensity and working capital requirements", plus inflexible foundry commitments and a longer cash conversion cycle. Capex already rose to $545m.

What would confirm it

The first quarter reporting inventory, wafer purchase commitments and a gross margin below the mid-90s.

Impact on value

Structural — and realised by the bull case working, not failing, which is why it is the easiest risk to underwrite away.

R5 — Stock compensation already consumes the operating profit

Mechanism

If equity comp scales with engineering headcount, R&D can decelerate and GAAP margin still not recover.

Evidence today

Equity-settled SBC of $1,052m against GAAP operating income of $900m in FY2026; a 41.2% non-GAAP operating margin in Q1 FY2027 was a 7.1% GAAP margin; headcount up from 8,330 to 9,584 FY2026 20-F; 29 Jul 2026 letter.

What would confirm it

GAAP operating margin staying in single digits as silicon cost of sales lands on top.

Impact on value

Caps the operating leverage assumption in §3.

R6 — The two largest revenue concentrations are entities Arm does not control

Mechanism

A sixth of revenue reaches Arm through a counterparty it cannot audit, in a jurisdiction that can close the channel.

Evidence today

Arm China — an independent distributor Arm neither owns nor manages — is the largest customer at 16% of FY2026 revenue; Arm relies on its self-reported figures to know what it is owed, and the credit-loss allowance against it reached $28.3m. PRC revenue 18% of total FY2026 20-F.

What would confirm it

A China export-control or end-use change touching the IPLA; a widening credit allowance; PRC revenue turning negative.

Impact on value

Bounded in any one year, structural over time.

11Metrics to monitor

12Thesis tripwires

13Open questions

1. What is Armv9's share of royalty revenue, and what is royalty per chip? Neither is disclosed — the central assumption in this file cannot be measured directly.
2. What gross margin does the AGI CPU carry? Not in sources; an external estimate of 40–50% exists (Morgan Stanley, 7 Apr 2026) but is unverified.
3. Is the SoftBank consulting revenue priced at arm's length, and what happens after the $300m fixed payment due in FY2027?
4. How large are the foundry and wafer commitments behind the ">$2bn" of AGI demand, and when do they reach the balance sheet?
5. Status and scope of the reported US FTC inquiry into Arm's licensing conduct (external, May 2026) — will it appear as a named proceeding in the next 20-F?
6. Are SoftBank's Arm shares pledged as collateral, and at what thresholds? Reported externally; not in Arm's filings.
7. No transcripts, investor-day materials or insider-trading data exist in the source set — management's long-term framing and analyst pushback are unobserved.

14Update log

15Source map