Arm Holdings

Phase A — Understand the business

Company Overview

What it is. Arm Holdings plc is a UK-domiciled (Cambridge, England & Wales) semiconductor IP company — it designs CPU architectures and processor cores and licenses them; it has historically manufactured nothing. It lists as ADRs on Nasdaq (ticker ARM, one ADS = one ordinary share) and files as a foreign private issuer on Form 20-F, not a 10-KForm 10-KA company’s audited annual report to the US regulator. The most complete thing it publishes.. SoftBank Group is the controlling shareholder at ~86.4% of issued share capital, so the public float is ~14% — Arm is a "controlled company" under Nasdaq rules.

How it makes money — two-component model:

1. License & other revenue — upfront/over-time fees for the right to use Arm IP (architecture licenses, core licenses, CSS, design services, tools, support). FY26: $2,307M (47% of revenue).
2. Royalty revenue — per-chip royalties recognized when a licensee ships a chip containing Arm IP, accrued from industry shipment estimates. FY26: $2,613M (53% of revenue). Royalties are the annuity: today's licenses seed tomorrow's royalty stream, and royalties keep paying for the 10–20-year life of a chip design.

Scale. Total FY26 revenue $4,920M (+23% YoY). The architecture is ubiquitous — effectively 100% of smartphones, plus deep penetration in IoT, autos, networking, and a fast-rising data-center share. Mobile applications processors are ~43% of royalty revenue, the mature base; data center and AI are the growth vector.

Customers / suppliers / competitors. Customers are the entire fabless + IDM + hyperscaler world: Apple, Qualcomm, MediaTek, Samsung, NVIDIA, AMD, Amazon (Graviton), Google, Microsoft. "Suppliers" is nearly n/a — Arm is an asset-light IP firm; its true input is R&D engineering talent (~84% of employees are in engineering ). Competitors: the open-source RISC-V ISA (the existential long-term threat), x86 (Intel/AMD) in PCs/servers, and increasingly its own licensees' in-house designs.

Contract structure / key terms. Licenses split into over-time vs point-in-time recognition (FY26 license & other: $1,080M over-time, $1,227M point-in-time ), which makes license revenue lumpy quarter-to-quarter. Royalties accrue on shipment with a one-quarter lag and estimation judgment. Remaining performance obligations (backlog): $2,071.4M, ~28% expected to convert within 12 months — a visibility anchor but small relative to a ~$440B market capMarket capitalisationThe share price multiplied by the number of shares. What the market says the equity is worth..

Verdict (Lens 1): A rare asset-light, ~98%-gross-margin toll-road on global compute, with an annuity royalty tail and a controlling-shareholder overhang. The model is among the best in semiconductors; the question every other lens probes is whether the price and the strategy pivot respect that.

Supply Chain

Arm sits unusually high in the stack — it has almost no physical supply chain of its own, which is the point. Map it:

Upstream (inputs to Arm): EDA tools (Synopsys, Cadence) and IP building blocks; foundry process design kits (TSMC, Samsung, Intel Foundry) that Arm co-optimizes cores against; and human capital (the binding input). No wafers, no fabs, no materials.

Arm (the node): Designs ISAs (Armv8, Armv9), cores (Cortex-A/N/M, Neoverse), and now Compute Subsystems (CSS) — pre-integrated, foundry-optimized core complexes — and a first-party AGI CPU (136-core, data-center).

Downstream (Arm → end customer):

• Licensees / chip designers: Apple, Qualcomm, MediaTek, NVIDIA, AMD, Amazon, Google, Samsung, Broadcom — they embed Arm cores into SoCs.
• Foundries: TSMC, Samsung Foundry, Intel Foundry physically print those SoCs.
• OEMs / hyperscalers: Apple (iPhone), Samsung, Xiaomi (handsets); AWS/Microsoft/Google/Meta/Oracle (data-center buyers of Arm-based server CPUs); auto Tier-1s.
• End customer: every smartphone owner, every cloud workload, increasingly every AI inference rack.

Named chokepoints & single-source dependencies:

• Arm China is a structural chokepoint — Arm's largest single customer (~16% of revenue ) is a separately-controlled entity Arm does not consolidate, through which essentially all PRC licensing flows. PRC = ~18% of total revenue FY26.
• The AGI CPU exposes Arm to TSMC for the first time — to sell a physical chip, Arm must buy advanced-node wafer capacity. CEO Haas flagged the bottleneck explicitly: AGI CPU demand doubled to >$20B in six weeks but supply is only locked for the first ~$1B. Arm has become a price-taker for the very capacity (TSMC leading-edge) that NVIDIA, Apple, and AMD also fight for.

Verdict (Lens 2): As a pure-IP firm Arm had no supply chain to break — its FY26 strategic pivot manufactures one. The single most important new dependency on the map is TSMC advanced-node allocation, and the chronic one is the un-consolidated Arm China conduit. Names present, chokepoints marked — lens passes.

Competitive Advantages (moats)

The moat, ranked by durability:

1. The ecosystem / installed base (strongest). >350 billion Arm-based chips shipped cumulatively; the entire mobile software stack, toolchains, and developer mindshare are Arm-native. Switching an ISA is a multi-year, multi-billion-dollar software-rewrite decision. This is a genuine network-effect + switching-cost moat.
2. Architecture-license lock-in + rate escalation. Arm doesn't just license cores; it licenses the architecture, and it has engineered a royalty-rate escalator into each generation: Armv8 ≈ 2.5–3.5% of chip ASP, Armv9 ≈ 5%, CSS ≈ 10% of ASP. So Arm grows royalties even on flat unit volumes by migrating the base up the value stack. Armv9 is only ~25% of royalty sales today against a 60–70% long-term attach target — years of mix-shift runwayRunwayHow long the cash lasts at the current rate of spending. It shortens the moment spending rises, which is why a figure taken from a quiet quarter flatters..
3. R&D scale & talent density. ~84% of staff in engineering; FY26 R&D $2,776M (56% of revenue). The cost to replicate Arm's core-design cadence from scratch is prohibitive.
4. Standards gravity. Being the default ISA means foundries co-optimize for Arm and software ships Arm-first — a self-reinforcing position.

Bargaining power. Over suppliers: near-total (Arm buys little). Over customers: high but eroding at the top end — the very largest customers (Apple, Qualcomm, NVIDIA, the hyperscalers) hold architecture licenses that let them build custom cores, capping Arm's pricing leverage over its biggest accounts and giving them a credible RISC-V/in-house exit threat.

Where the moat is thinner than it looks (developed in Lenses 12–13): (a) RISC-V is free and improving; (b) the Qualcomm ruling proved Arm cannot easily weaponize ALA termination to claw back a licensee's custom work; (c) the AGI-CPU pivot converts moat-protecting neutrality into competitive threat.

Verdict (Lens 3): A wide, real moat anchored in ecosystem + a clever per-generation rate escalator — but it is a moat over the licensing business, and management is now spending some of that goodwill to enter the chip business, where the moat is much narrower.

Segments

The research-layer our figures is empty, so the breakouts below are from the 20-F's disaggregation and geographic notes. Arm reports by revenue type and by geography, not by product division.

By revenue type ($M):

Of license & other, related-party (Arm China et al.) is $1,009M of $2,307M; of royalty, related-party is $490M of $2,613M. Royalty acceleration is mix-driven, not volume-driven — "improved mix of products with higher royalty rates per chip, such as Armv9" — exactly the rate-escalator thesis. Data-center royalty "more than doubled year-over-year" in Q4.

By geography ($M, by customer HQ):

The signal: Japan revenue nearly tripled (FY25→FY26) and is up ~7x off FY24 — almost certainly SoftBank-adjacent / AI-datacenter (e.g., SoftBank's own AI ambitions, Rapidus, and AGI-CPU lead activity). It is now Arm's #2–3 geography from a rounding error two years ago. Worth watching whether it is durable customer demand or related-party concentration.

Verdict (Lens 4): Royalty growth is high-quality (rate/mix, not units); the Japan surge is the most interesting under-discussed line; and the segment data confirms the bull engine — value-stack migration — is real in the actuals, not just the deck.

Phase B — Measure performance

Earnings Result (latest print — Q4/FY26, reported 6 May 2026)

Full-year FY26 (ended 31 Mar 2026) ]:

Q4 FY26 specifically: Revenue $1.49B (+20%), a record and above guidance midpoint. Licensing $819M (+29%); Royalty $671M (+11%), with data-center royalty more than doubling. Record non-GAAP EPS $0.60.

Beat/miss & reaction: The print beat on revenue and EPS — and the stock fell ~7.25% to $220.10 on 7 May 2026. Two reasons: (1) smartphone unit growth guided negative on a memory-chip shortage rippling through the handset supply chain; (2) the AGI-CPU opportunity was huge in demand ($20B+) but capped in supply ($1B locked), so the near-term revenue print didn't move enough to satisfy a stock priced for perfection.

Margin read: Gross margin is structurally ~97–98% (pure IP). The real operating story is R&D rising to 56% of revenue — Arm is deliberately spending its operating leverage to fund the AGI-CPU roadmap. GAAP operating margin fell (21%→18%) even as revenue grew 23%, because R&D grew faster. That is a choice, and the bull/bear hinges on whether it pays off.

Balance-sheet flags: Cash & equivalents $2,751M + $850M short-term investments ≈ $3.6B, effectively net-cash, no meaningful debt. Accounts receivable $1,331M (up from $1,127M) — rising with revenue, watch the royalty-estimation accrual (Lens 10). FY26 had a $253M tax expense vs a benefit in FY24–FY25, which masked some of the pre-tax improvement at the net line.

Anomaly vs own history: Beating and dropping 7% is the tell of an expectations problem, not a business problem — the recurring FY26 pattern (see Lens 8).

Verdict (Lens 5): A genuinely strong, accelerating, ~98%-gross-margin print — undercut near-term by a memory-driven smartphone air-pocket and a supply-capped (not demand-capped) AGI ramp. The business is firing; the stock reaction shows the bar is the risk.

Earnings Calls (sentiment trend)

No transcripts on disk; this lens is -grounded from FY26 call coverage and CEO interviews.

What management is focused on (FY26 arc): The narrative has shifted hard from "we license IP" to "we are becoming a CPU maker." CEO Rene Haas's recurring 2026 themes:

• AGI CPU / first-party silicon — the dominant new message. "Core count grows faster than chip count" in the AI-agent era; targets largest CPU share by 2030, $25B total revenue by FY31.
• x86 displacement / efficiency — explicitly framing Intel as "historic" and pitching Arm power-efficiency for AI data centers.
• CSS & Armv9 rate migration — the steady drumbeat that royalty per chip keeps climbing.

Tone shift over time: Through FY25 the tone was "neutral IP partner riding everyone's growth." Through FY26 it became assertive, competitive, almost combative toward x86 and implicitly toward its own licensees. Confidence is high and the TAM claims are escalating (Bernstein lifted Arm's own framing to a $223B CPU TAM by 2030 ).

What they stopped saying: Less emphasis on "neutral Switzerland of silicon" — the positioning that historically reassured Apple/Qualcomm/NVIDIA. That omission is the channel-conflict risk surfacing in the language.

Verdict (Lens 6): Sentiment is euphoric and the strategy is bold — but the rhetorical pivot from neutral arms-dealer to competitor is the single most important qualitative change, and it cuts both ways.

Comps

The comp verdict: On any peer set Arm is the most expensive name by a wide margin — ~200x non-GAAP vs a ~24x semis industry average, and roughly 3–4x the EDA licensors that share its model. Even versus NVIDIA (faster absolute growth, fatter dollar profits), Arm trades at a multiple premium. The market is paying for optionality (AGI CPU + Armv9 rate ramp), not current earnings. Dividend yield: 0% — Arm pays no dividend. 5-yr average ROE: n/a (only a ~2.5-yr public history since the Sept-2023 IPO; ROE is volatile given the FY24 near-breakeven and IPO-related SBC).

Stock-Price Catalysts (>5% moves, last ~5 years / since IPO)

Arm has only traded publicly since 14 Sept 2023 (IPO at $51, popped ~25% day one; SoftBank retained ~90.6%, now 86.4%). The >5%-move catalyst pattern:

• Earnings prints dominate. Q4-FY24 (May 2024): +~11% on +47% revenue / EPS up ~18x. The stock routinely gaps ±10–25% on quarterly royalty/guidance surprises.
• Guidance > the print. Q4-FY26 (May 2026): beat and fell ~7% on a soft smartphone unit guide — proof the market trades Arm on forward narrative, not trailing beats.
• AI-CPU / product news. Mar 2026: +16% the day the AGI-CPU revenue windfall was disclosed ("significant shift"); repeated 5%+ up-days through Jun 2026 on AGI-CPU customer adds and TAM raises (e.g., +4.9% on 20 Jun, +5.7% testing ~$478).
• Analyst re-ratings. Bernstein → $500 target (from $300, +67%) on 17 Jun 2026 supercharged a multi-day rally; an Apr-2026 downgrade on "AI-CPU roadmap costs and growth fears" cut the other way.
• SoftBank / macro / AI-complex beta. Arm trades as a high-beta AI proxy; SoftBank's own AI/OpenAI news and the broader AI-capex narrative move it.

What the market actually reacts to: (1) royalty/guidance trajectory, (2) AGI-CPU TAM and customer wins, (3) the gap between euphoric sell-side targets and a fundamentally rich multiple. It does not reward trailing beats — the bar is always the next leg of the AI story.

Verdict (Lens 8): A momentum/narrative stock. Catalysts are earnings guidance, AGI-CPU milestones, and analyst-target leapfrogging. That makes it a story stock with a fundamental floor far below spot — fragile to any narrative crack.

Phase C — Judge people & books

Management

CEO — Rene Haas (CEO since Feb 2022).

1. Track record: Has delivered three consecutive years of 20%+ revenue growth ($3.2B→$4.0B→$4.9B), took Arm through a successful, large IPO (Sept 2023, the biggest of that year), and architected the value-stack strategy (Armv9 rate escalator, CSS, now AGI CPU). Operationally credible; a genuine product/strategy CEO, ex-NVIDIA, deep semis pedigree.
2. Tenure & skin in the game: ~4 years in seat. Direct insider ownership is small in % terms because SoftBank owns ~86.4% — the alignment that matters is with SoftBank (Masayoshi Son), not with minority holders. our figures not present — individual exec holdings n/a.
3. Capital allocation: Reinvestment-heavy — R&D scaled to 56% of revenue to fund the AGI-CPU roadmap; no dividend, no material buyback. Net cash ~$3.6B sits as dry powder. The bet is plow-it-into-R&D-and-first-party-silicon; ROE/ROIC trend is noisy off the FY24 near-breakeven and IPO SBC.
4. Red flags: (a) SBC is enormous — gross share-based comp $1,052M in FY26, larger than GAAP operating income of $900M; non-GAAP EPS ($1.77) flatters the GAAP ($0.85) by ~2x largely on SBC add-backs. (b) Controlling-shareholder governance — SoftBank's interests (it is also a customer-adjacent party via Arm China and the broader SoftBank AI complex) can diverge from minorities; related-party revenue is ~30% of the total. (c) The AGI-CPU pivot risks the licensee relationships that are the business.
5. Archetype: Professional manager operating a founder-controlled (SoftBank) asset — empowered to be bold precisely because a single 86% owner backs the long game. Implication: long-horizon strategic bets get funded, but minority holders are along for SoftBank's ride.

Verdict (Lens 9): A strong, credible operator executing an aggressive, coherent strategy — with two governance caveats that the valuation does not discount: SBC that consumes most of GAAP profit, and an 86%-owner whose agenda sets the direction.

Forensic Red Flags

Forensic lens — every figure labeled.

Income statement.

• Revenue recognition judgment (royalty accrual): Royalties are booked on estimated licensee shipments with a one-quarter lag and "judgment for several key attributes, including industry estimates of expected shipments". This is an inherently estimate-laden line — not fraud, but the single softest accounting spot; a true-up risk if shipment estimates overshoot.
• License lumpiness: $1,227M point-in-time vs $1,080M over-time license revenue; "fluctuation in timing and size of multiple high-value license agreements" means quarters can be steered by deal timing — watch for pull-forwards.
• SBC flattering non-GAAP: Gross SBC $1,052M (FY26) vs GAAP operating income $900M. Non-GAAP EPS $1.77 is ~2x GAAP $0.85 — the gap is mostly SBC and amortization add-backs. Anyone valuing Arm on the ~200x non-GAAP multiple is already ignoring a billion dollars of real economic dilutionDilutionIssuing new shares, so each existing share owns a smaller slice of the same company..

Balance sheet.

• Receivables outpacing? Mild. AR rose $1,127M→$1,331M (+18%) vs revenue +23% — receivables grew slower than revenue, which is reassuring (not a red flag this year).
• No debt, ~$3.6B cash — clean.
• Related-party balances: ~30% of revenue is related-party (Arm China + SoftBank affiliates); the un-consolidated Arm China relationship is the standing structural complexity (see below).

Cash flow vs earnings: FCFFree cash flowCash left after paying to run and maintain the business. Unlike profit, it is hard to flatter with accounting choices. not in research-layer (our figures empty); as a near-zero-capex IP firm, FCFFree cash flowCash left after paying to run and maintain the business. Unlike profit, it is hard to flatter with accounting choices. should track net income closely — but the SBC add-back inflates operating cash flow relative to true owner earnings. n/a for the exact FY26 FCFFree cash flowCash left after paying to run and maintain the business. Unlike profit, it is hard to flatter with accounting choices. figure (would need the cash-flow statement; not extracted).

Regulatory findings (required sub-section). Read regulatory/regulatory-findings.md.

• SEC Litigation Releases / AAERs: None. 0 SEC findings naming Arm Holdings in the 2021–2026 search window, verified via SEC EDGAR EFTS (LR + AAER).
• Non-SEC enforcement (web): An active U.S. FTC antitrust investigation is reportedly examining whether Arm leverages architectural dominance to degrade or deny design licenses to third-party chip customers as it expands into proprietary silicon. This is material and thesis-relevant (it directly targets the channel-conflict strategy). No monetary penalty or consent decree reported as of this writing.
• Item 3 / Legal Proceedings (10-K equiv. — here, 20-F risk factors): The 20-F discloses ongoing litigation with Qualcomm and Nuvia as a material legal matter. Outcome (web, below): Arm lost the core ALA claims; Qualcomm's countersuit is pending (trial expected Mar 2026).
• Summary: No accounting-fraud findings (SEC clean). The live regulatory/legal risks are an FTC antitrust probe of the licensing-vs-first-party conflict and the Qualcomm/Nuvia litigation — both, both strategic rather than forensic-accounting in nature.

Verdict (Lens 10): Books are clean of enforcement history; the genuine forensic caution is the estimate-driven royalty accrual and SBC > GAAP operating income (a quality-of-earnings flag for anyone leaning on non-GAAP). The bigger risks are regulatory/legal, not accounting.

Phase D — Project & stress-test

Forward Projection (FY27 / FY28 / FY29 non-GAAP EPS)

Anchor (actuals + guidance): FY26 revenue $4,920M, non-GAAP EPS $1.77. Q1-FY27 guided: revenue ~$1.26B (+~20%), non-GAAP EPS $0.40 ±$0.04. Management frames FY27 royalty growth ~20%/qtr and licensing ~60% H2-weighted.

Build (every input labeled; output ):

• Base FY27 non-GAAP EPS ≈ $2.15 — squarely in line with the Street consensus ~$2.17. Sanity-check passes.
• Bull FY27 ≈ $2.35–2.50.
• Bear FY27 ≈ $1.85–1.95.

The valuation math that matters: At ~$439 and base FY27 EPS ~$2.15, ARM trades at ~205x forward non-GAAP. To "grow into" even a still-rich 50x by FY29, EPS must roughly quadruple from here and the multiple must compress 75% — both have to go right. Even Bernstein's bullish $500 implies the market cap ($530B) is ~110x FY27 non-GAAP. The Street's mean target sits ~38% below the spot price — i.e., the median analyst thinks the stock is ~38% too high even while the marginal buyer keeps bidding.

Forecast logging: Per --watchlist rules, no our model create in this unattended breadth pass (only log a Brier forecast on a genuinely committed base case). If promoted to a thesis, the natural scoreable forecast would be: "ARM FY27 (ending Mar-2027) non-GAAP EPS ≥ $2.15, p≈0.55, resolves 2027-03-31." Not logged here.

Bull vs Bear

Bull case. Arm is the toll-road on all of computing, and the toll-per-vehicle is rising on its own engineering cadence (Armv8→v9→CSS lifts ASP royalty from ~3% → 5% → 10% ) with Armv9 only ~25% penetrated against a 60–70% target — so royalties compound even if unit volumes stall. Layer on a second engine: the AGI CPU turns Arm from a ~3-cents-on-the-dollar IP licensor into a seller of whole chips, with disclosed demand already >$20B and marquee customers (Meta lead; OpenAI, Oracle, ByteDance, Cerebras, Cloudflare, SK Telecom). Power-efficiency is the structural tailwind: AI data centers are power-bound, and Arm's perf/watt is its weapon against x86. ~98% gross margins, net cash, an installed base of >350B chips, and a controlling owner (SoftBank) funding a decade-long bet. If AGI CPU even half-works, Arm's revenue path to $25B by FY31 re-rates the whole story.

Bear case (2–3 permanent-impairment risks).

1. Valuation is the risk. ~200x non-GAAP, ~180x GAAP, ~90x sales, Street mean ~38% below spot. Even flawless execution can lose money if the multiple normalizes toward the ~24x semis average or even the ~50x EDA-peer range. This is the dominant near-term risk and it does not require the business to stumble.
2. The moat-eroding pivot. By building first-party silicon, Arm attacks the neutrality that made Apple/Qualcomm/NVIDIA/AWS comfortable standardizing on it — accelerating the RISC-V migration and inviting an FTC antitrust action. The Qualcomm/Nuvia loss proved Arm cannot use ALA-termination to police licensees, structurally weakening its enforcement leverage.
3. Concentration + geopolitics. Top-3 customers = 42% of revenue, Arm China alone ~16%; PRC ~18% is hostage to US/China export-control escalation. A single large-customer in-housing event or a China clampdown dents the royalty base.

Pre-mortem (18 months out, thesis broke — what happened?): AGI-CPU supply stayed capped (TSMC allocation lost to NVIDIA/Apple), so the second engine slipped to FY29; the memory-shortage smartphone air-pocket dragged two quarters of royalty growth below 20%; an FTC action or a high-profile RISC-V defection (a top hyperscaler moving a flagship line off Arm) cracked the narrative — and a story stock at 200x re-rated to 80x, a ~55% drawdown, with earnings still growing.

Are multiples too high? Yes, on any historical or peer frame — the only justification is the optionality value of the AGI-CPU TAM, which is real but unproven and years from materiality.

Contrarian view (what the market refuses to see): Bulls treat the AGI-CPU pivot as pure upside. The contrarian read is that the pivot is partly defensive and partly self-cannibalizing — Arm is moving into chips because the licensing model's pricing power over its biggest customers has peaked (they all hold architecture licenses and can build their own), and in doing so it converts its best customers into competitors and regulators' targets. The thing that makes the bull deck exciting is the same thing that threatens the annuity.

Devil's Advocate (short-seller)

Dismantling the bull case.

• What structurally breaks the model: Arm's revenue is a derivative of other people's chip shipments and license appetite. Its two biggest growth claims (Armv9 attach to 60–70%; AGI-CPU to $25B) are both multi-year, partly-out-of-Arm's-control bets. If hyperscaler custom-silicon increasingly uses architecture licenses (lowest Arm take) or pivots to RISC-V to dodge both cost and the new channel conflict, Arm's effective royalty rate stops climbing — and the entire bull thesis is "the rate keeps climbing."
• Concentration: Top-3 = 42%, Arm China ~16%. Lose or in-house one flagship account and a quarter's growth evaporates. The customers most able to leave (Apple, Qualcomm, the hyperscalers) are exactly the ones with architecture licenses and RISC-V teams.
• The moat is weaker than bulls think: The Qualcomm/Nuvia ruling (Sept 2025, Arm's last claim dismissed; Arm appealing) is a precedent that a licensee can acquire a company, fold its custom cores into an existing Arm license, and Arm cannot force destruction of the work. That is a material crack in the enforcement moat, established in court.
• Most dangerous competitor bulls underestimate: Not Intel — RISC-V. It's free, royalty-free, improving fast, and every Arm price increase or channel-conflict move is a marketing campaign for it. Qualcomm, NVIDIA, and the hyperscalers all fund RISC-V optionality.
• Worst capital-allocation / incentive issues: SBC $1,052M > GAAP operating income $900M — shareholders are funding ~$1B/yr of dilution while the company touts non-GAAP EPS that adds it back; an 86%-owner (SoftBank) sets strategy and harvests the upside; no dividend/buyback to offset dilution.
• What must hold for today's price: ~22%+ revenue CAGR for years, Armv9/CSS rate escalation uninterrupted, AGI-CPU reaching ~$10B+ by ~FY30, no major RISC-V defection, no adverse FTC action, no China shock — and the multiple staying north of 100x. That is a lot of "ands."
• If growth disappoints 20–30%: A story stock at ~200x non-GAAP that prints +12% instead of +22% likely halves or worse — multiple compression dominates the math.
• Single permanent-impairment scenario & plausibility: A top-2 hyperscaler publicly standardizes a flagship AI/server line on RISC-V (citing cost + Arm's competitive pivot), validating the open ISA at scale. Plausibility: low-to-moderate over 18 months, moderate-to-high over 5 years — and it would permanently reset Arm's terminal royalty-rate assumptions.

Verdict (Lens 13): The short case isn't "the business is bad" — it's "the price assumes a flawless decade while management lights a channel-conflict fire and the courts just weakened the moat." It's a valuation + strategy-risk short, not a fraud short.

Management Questions (ordered by information value)

1. AGI-CPU demand is >$20B but supply is locked for ~$1B — what specific TSMC/foundry allocation have you secured for FY27–FY29, and what is the realistic shipping revenue, not the order book?
2. As you sell first-party chips, how do you keep Apple, Qualcomm, NVIDIA, and the hyperscalers from accelerating RISC-V or in-house designs — what concretely changes in your licensing terms or conduct to preserve neutrality?
3. What is the status and scope of the FTC inquiry into design-license access, and what behavioral remedies would you accept before they're imposed?
4. Armv9 is ~25% of royalties vs a 60–70% target — what is the realistic annual attach-rate trajectory, and what stalls it?
5. Gross SBC ($1,052M) exceeds GAAP operating income — when does SBC fall below, say, 10% of revenue, and why should investors value the stock on non-GAAP EPS that adds it back?
6. The Qualcomm/Nuvia ruling went against you and you're appealing — if the appeal fails, what does it change about how you can enforce architecture licenses against acquirers?
7. Japan revenue tripled to $825M — how much is durable third-party demand versus SoftBank-related, and is it a concentration risk?
8. With Arm China ~16% of revenue and un-consolidated, what is your actual control over that revenue, and your exposure if US/PRC export controls tighten?
9. Smartphone units are guiding negative on the memory shortage — how deep and how long, and what's the royalty-revenue sensitivity?
10. The AGI-CPU competes with your customers' chips — will you guarantee licensees equal/earlier access to the IP inside your own products, and how is that policed?
11. Capital allocation: net cash ~$3.6B, no dividend, no buyback — at what point do you return capital, or is it all earmarked for first-party silicon?
12. CSS is ~20% of royalties at ~10% ASP — what's the ceiling on CSS attach, and does it cannibalize higher-margin pure-IP licensing?
13. What is the gross-margin and operating-margin profile of the chip business versus the ~98%-GM IP business, and how does mixing them change consolidated margins?
14. SoftBank owns ~86.4% — what governance protections exist for minority holders against related-party dealings, and is there a float-increase plan?
15. If RISC-V reached parity in your core data-center segment by 2030, what is your defensive plan beyond price and ecosystem inertia?