Arm Holdings: A Deep-Dive Value Analysis

Is revenue recurring, stable, and predictable? Royalty is good, license is weaker. In full-year FY2026, Arm's revenue was $4.92 billion, of which royalty was $2.613 billion and license and other was $2.307 billion; for FY2025 the figures were $4.007 billion, $2.168 billion, and $1.839 billion respectively. Royalty revenue is more recurring and rent-like; license revenue is clearly affected by the timing of large deals and is more volatile. Recent improvement in predictability comes from subscription-style licensing tools such as ACV, RPO, and Arm Total Access / Flexible Access: FY2026 Q4 ACV was $1.660 billion, up 22% year over year; RPO was $2.071 billion; there were 56 Arm Total Access licenses and 329 Flexible Access licenses. In other words, this business is no longer the traditional "one-time IP sale" but increasingly looks like "platform subscription plus a long tail of royalties."

The cost structure is beautiful, but transparency is not top-tier. FY2025 and FY2026 gross margins were roughly 97.0% and 97.5%, showing that this is a naturally high-margin intellectual-property business; what truly consumes profit is R&D and selling, general and administrative expense. FY2026 R&D expense was $2.776 billion and SG&A was $1.115 billion, together approaching 79% of full-year revenue; so Arm does not look like a "low-margin company" but rather one that keeps plowing most of its gross profit back into R&D and platform expansion. At the same time, the 20-F reminds investors that revenue recognition for long-term license contracts involves complex judgment, which Deloitte lists as a key audit matter; royalty revenue is also estimated based on customer shipments and trued up in the following quarter, so accounting profit and cash flow can swing significantly.

On dependencies, three things deserve special attention. First, customer concentration is not low. In FY2025 the top five customers together accounted for 56% of revenue, with the largest, Arm China, at 17%; in FY2023 that share once reached 24%, and although it is declining, it remains important. Second, Arm China is not directly controlled by Arm yet is the exclusive distribution channel for the China market; the 20-F discloses an initial agreement term running to 2048, and Arm China–related revenue accounted for roughly 17% / 21% / 24% of total revenue in FY2025 / FY2024 / FY2023. Third, royalty is closely tied to downstream chip shipments, so even though Arm is asset-light, it is not immune to the semiconductor cycle.

At the industry level, I define it as a "good company in a good industry," but not a "low-risk industry." WSTS expects the global semiconductor market to approach $975 billion in 2026, continuing strong growth over 2025; Deloitte likewise attributes 2025 semiconductor growth mainly to generative AI and data-center build-out, with the recovery in markets such as PCs and phones relatively muted. This aligns closely with Arm's exposure: cloud, data center, automotive, and edge AI are tailwinds, while smartphones remain a large but more mature pool.

Industry attractiveness score: 4/5. Business understandability score: 4.5/5. If the stock market closed for five years, I would be willing to own this business, provided the acquisition price is reasonable and the governance structure is acceptable. The question today is not "whether this business is worth owning," but "what price the public market has already put on it."

The Moat

Arm's moat is not a single moat but a set of mutually reinforcing ones.

The strongest layer is the network effect at the ecosystem and standards level. A CPU architecture is essentially the "common language" of software developers. The 20-F is clear about this: the architecture defines the executable software instructions and thereby forms a compatible software library; as of March 2025, cumulative Arm-based chips shipped exceeded 310 billion units, and the company states that its platform supports more than 20 million developers worldwide. In the official FY2026 Q4 materials, that developer figure has risen to over 22 million. The larger the ecosystem and the more mature the toolchain, the greater the friction of switching for customers — this is both a network effect and a switching cost.

The second layer is switching cost and software compatibility. Arm has held more than 99% share in the smartphone application-processor market for years, and in fiscal 2025 that market contributed roughly 46% of total royalty revenue. In markets such as phones, IoT, and consumer electronics, where software and development chains are highly mature, switching the ISA is not as simple as "swapping one IP"; it requires rebuilding the toolchain, developer adaptation, the verification system, and the production roadmap. For customers who have built up volume production and a software base, such a switch is not free.

The third layer is scale advantage and a neutral platform position. Unlike Intel or AMD, which compete directly with nearly all customers through their own branded CPUs, Arm has long played the role of a neutral platform partner "providing the compute foundation for everyone." This model lets it serve potentially competing customers at the same time — AWS, Google, Microsoft, NVIDIA, Qualcomm, and others. AWS officially describes Graviton as its family of Arm-instruction-set processors for EC2; Google's official materials for the Axion N4A claim up to 2x price-performance and 80% better performance-per-watt versus current x86 VMs; Microsoft explicitly states that its Cobalt 100 is built on the Arm Neoverse N2 design and targets a broad range of cloud workloads. In other words, Arm is no longer a "mobile-only" story but has become the general-purpose foundation for multiple hyperscalers.

The fourth layer is an "upgrade path to higher royalty rates." Armv9, CSS, and more integrated system-level solutions mean Arm is no longer merely "selling a CPU core" but is trying to raise its share of the value of the entire SoC. The FY2026 Q4 shareholder letter states plainly that royalty growth is driven jointly by higher royalty rates per chip (for example Armv9 and Arm CSS) and data-center deployment; in Q4 FY2026, royalty grew 11% year over year and license and other grew 29%. This shows the moat is not a static "collect rent while you sleep," but an active effort to convert "ecosystem position" into "a higher take rate."

But the moat is not without cracks.

The biggest crack is RISC-V. In its own 20-F, Arm lists RISC-V as one of the principal competing architectures and acknowledges that some customers also support RISC-V. External signs confirm this is not merely a paper risk: Google has publicly disclosed continued progress on Android support for RISC-V; Qualcomm announced long ago that it would bring a RISC-V-based wearable platform to the Wear OS ecosystem; and Reuters reported this year that China plans to issue policy promoting nationwide use of RISC-V chips to reduce dependence on Western technology. RISC-V is unlikely to shake Arm's dominance in high-value mainstream markets in the short term, but it will keep eroding the licensing pricing Arm can otherwise capture in "low-to-mid-end and specific vertical scenarios."

The second crack is customer in-house design and Arm's own strategic extension. Arm acknowledges in the 20-F that certain customers are already designing, or are in the process of designing, their own architectures, and that the company's own extension into CSS, chiplets, end-chip solutions, and more complete compute products will bring new competitive, brand, technology, regulatory, and financial risks. Simply put, the further it moves toward "more complete solutions," the more Arm can earn — but the more easily it shifts from "the platform that sells the water" to "half a player on the field," which would change an originally very elegant business model.

My judgment: the moat is still "stable and leaning toward expansion," but not as impregnable as most investors imagine. Brand advantage: medium. Cost advantage: weak to medium; the real advantage is not the lowest cost but the lowest total ecosystem cost. Scale advantage: strong. Network effect: strong. Switching cost: strong. Channel advantage: medium, especially the dependence on Arm China in China. Patent / license / regulatory barriers: medium. Data advantage: medium. Operating capability: medium to strong. Capital-allocation capability: not yet a moat.

Moat strength score: 4/5.

In an inflationary environment, Arm has some pricing power, particularly by lifting the royalty take rate per chip through Armv9/CSS; but it also admits that the annual price escalators on some long-term licenses may not outpace inflation. In a downturn, Arm would most likely preserve its gross margin and viability, but not necessarily the smoothness of its cash flow, because royalty depends on downstream shipments and license carries timing volatility. Its historically high gross margin is almost structural — but its high valuation is not.

Management and Capital Allocation

My overall view of management is "operationally credible, governance discounted, capital allocation still to be watched." Rene Haas has served as CEO since 2022; in the FY2025 annual report, the company disclosed that he holds related interests in 647,800 shares and that the CEO is required to hold Arm ADS equal to 1000% of base salary, with the annual report noting that the value of his unconditionally held shares already exceeds 5600% of his compensation. This shows the CEO's personal interest is not disconnected from the share price. The company also has a clawback policy, and the CEO's incentives carry clear long-term performance conditions and a relative-TSR constraint.

The problem is that the real control does not sit with management but with SoftBank. As of May 20, 2025, SoftBank Group beneficially owned roughly 87.1% of outstanding shares, making the company a "controlled company" under Nasdaq rules. The 20-F further discloses that SoftBank enjoys a series of special rights over board nominations, related-party transactions, information rights, consultation rights, and more; at the same time, financing arrangements in which SoftBank pledges Arm shares as collateral could, amid share-price volatility, bring risks of forced selling, additional collateral, or a change of control. This means that even if the CEO is honest, professional, and long-term oriented, minority shareholders still cannot treat Arm as a "governance-standardized, widely held company."

On capital allocation, there are two core observations. First, the company has not used excess cash for regular dividends or buybacks; from the FY2022–FY2026 cash-flow disclosures, cash has flowed mainly into R&D, PP&E / intangible-asset spending, net settlement of taxes related to equity incentives, and a small amount of ecosystem investment, and I have seen no evidence of regular dividends or systematic buybacks. For a company still in a high-growth phase, this is not necessarily bad; but it means the path to shareholder return depends almost entirely on higher per-share intrinsic value in the future rather than real, present cash return.

Second, equity incentives are heavy. Share-based compensation was $820 million in FY2025 and rose further to $1.052 billion in FY2026; the CEO's total single-year compensation in FY2025 was about $24.46 million, the vast majority from equity plans. More to the point, the share count is indeed rising: outstanding shares at year-end grew from 1.040 billion in FY2024 and 1.057 billion in FY2025 to 1.064 billion in FY2026. That is, Arm's "high growth" is not occurring under a completely frictionless, shareholder-friendly mechanism; existing shareholders keep paying a dilution cost for talent and expansion.

Is management candid? I lean toward "relatively candid." The 20-F does not dodge sensitive points such as customer concentration, Arm China being uncontrolled, SoftBank-related risk, the RISC-V threat, and the complexity of revenue recognition on long-term contracts, which is a plus. What truly keeps me reserved is not "whether risks are disclosed" but "whether capital discipline can maintain the past asset-light logic as Arm pushes further from platform licensing into in-house silicon." This only began in FY2026, and it is too early to draw a conclusion.

Management and capital-allocation score: 3/5.

Financial Quality and Owner Earnings

Facts first, then judgment.

The table below is compiled from Arm's F-1, the FY2025 Form 20-F, and the FY2026 Q4 / FY2026 shareholder-letter financial statements; all figures are in millions of U.S. dollars except per-share and percentage items. FCF uses the company's consistent long-disclosed definition: operating cash flow − purchases of property and equipment − purchases of intangible assets − payment of intangible-asset obligations; FY2022–FY2023 are primarily on a continuing-operations basis, and the FY2026 figures come from the May 6, 2026 quarterly update.

How should this table be read?

First, revenue growth is solid but not smooth. From FY2022 to FY2026, revenue rose from $2.703 billion to $4.92 billion, a four-year compound growth rate of roughly 16%; royalty grew more slowly than license, indicating that the high growth of the past two years came not only from "accumulated rent" but also from large license contracts and platform upgrades. For a value investor this is not bad, but it lowers your confidence in "stable predictability."

Second, the gross margin is very strong, yet the operating margin is not stable enough. The FY2024 operating margin was only 3.4%, recovered to 20.7% in FY2025, then fell back to 18.3% in FY2026. This shows that Arm's profit quality is not improving linearly but is highly affected by R&D investment, equity incentives, taxes, gains and losses on equity investments, and revenue timing. It is not the kind of "clearly stable, rent-collecting" software company.

Third, the balance sheet carries no high leverage, but operating cash flow swings a great deal. At the end of FY2026, cash and short-term investments stood at $3.601 billion against total liabilities of $2.417 billion, an overall net-cash position; this gives Arm strong, cycle-resistant staying power. The issue is cash-flow volatility: FY2024 OCF was $1.090 billion, fell to $397 million in FY2025, then rebounded to $1.524 billion in FY2026. The 20-F discloses that the sharp FY2025 decline in operating cash flow was driven mainly by post-IPO employer taxes related to equity incentives, the timing of receivable collections, and changes in other liabilities; FY2026 was lifted by improvements in working capital and the timing of tax payments. This shows that "the profit is not fake," but the cash flow is not stable and gets strongly distorted by working capital and equity incentives.

Fourth, this is not an inventory business, but receivables and contract assets deserve close watching. Arm has almost none of the inventory risk of a typical manufacturer; what truly matters are accounts receivable and contract assets. FY2025 net accounts receivable was $1.107 billion and contract assets were $988 million; FY2026 rose further to $1.300 billion and $1.297 billion. At the same time, the FY2026 cash-flow statement shows that accounts receivable and contract assets together consumed about $510 million of cash. This is a natural result of complex revenue recognition and the lengthening of contract terms, but for a conservative investor it means the conversion of "reported revenue" into "distributable cash" is not so charming.

Fifth, the rise in capital-spending intensity is an important change. Company-defined FCF for FY2023–FY2025 was $606 million, $907 million, and $99 million in turn; FY2026 returned to $882 million. On the surface the recovery looks fine, but FY2026 purchases of property and equipment jumped to $545 million, far above $219 million in FY2025 and $92 million in FY2024. Combined with management's announcement in the FY2026 shareholder letter of its first Arm-designed chip for the data center, one can reasonably infer that Arm is extending from a "pure licensing platform" toward "doing some silicon," which would raise its future revenue-capture ability but also push up capital intensity and weaken the elegance of the traditional asset-light model. I stress here: this is an inference, not an explicit, quantified disclosure from management.

Sixth, the accounting risk is not a "sign of fabrication" but a "complexity risk." I have seen no clear evidence of financial fabrication in the materials reviewed; but Deloitte lists the identification of performance obligations and the allocation of the transaction price in long-term revenue contracts as a key audit matter, and Arm explicitly states that royalty revenue must be estimated based on market and customer data and trued up in the following quarter. For a conservative investor, this kind of complexity itself warrants a higher discount rate, rather than being ignored simply because "the audit passed."

My conclusion on financial quality: Profit is broadly real, but "the cash owners can take away" is far less smooth than revenue and gross margin make it look. Arm is not a heavy-asset company that "needs more cash the more it grows," but it is no longer a near-pure software-like model that "replicates profit with almost no capital." Financial quality score: 3.5/5.

How to view owner earnings?

If you follow the company's definition exactly, FY2026 non-GAAP FCF was $882 million. Divided by the current market capitalization of roughly $269 billion, the FCF yield is about 0.33%, or about 305x FCF. That is already very expensive.

If you go a step more conservative, I would treat equity-incentive dilution as a real owner cost. From FY2025 to FY2026, shares outstanding grew from 1.057 billion to 1.064 billion, an increment of about 7 million shares. If you use a "replacement buyback cost" assumption clearly below the current share price — say $50 to $80 per share — then the "equivalent cash cost" required to keep the share count from expanding is roughly $350 million to $560 million. Subtracting that from FY2026's $882 million of FCF leaves conservative owner earnings of roughly $300 million to $500 million. This implies the current share price equates to about 540x to 900x conservative owner earnings. The $50–80 here is an assumption, meant to avoid using the overly high current price to make owner earnings look even more extreme; with a higher replacement price, the valuation would be even more expensive.

My owner-earnings judgment:

• If you ignore dilution, Arm's true distributable cash-flow ability is not bad, but it is extremely expensive.

• If you treat dilution as a cost, the company's "true distributable cash flow" to minority shareholders is not attractive today.

• This does not mean Arm is not a good company; it only means a good company and a good price are not the same thing.

Valuation and Margin of Safety

I use three methods: discounted owner earnings, relative valuation, and asset/liquidation value.

Discounted owner earnings. Here I use neither the most aggressive AI story nor the harshest method of "deducting all FY2026 SBC and all incremental working capital," which would distort the valuation downward. I adopt a framework closer to "what the market is willing to accept, while staying as conservative as possible":

• Conservative scenario: $882 million as the base owner-earnings proxy, growing 12% per year over the next ten years, a 10% discount rate, and 3% terminal growth.

• Base scenario: $1 to $1.2 billion as the base proxy, 18% growth over ten years, a 10% discount rate, and 4% terminal growth.

• Optimistic scenario: about $1.2 billion as the base proxy, 22% growth over ten years, a 9% discount rate, and 4.5% terminal growth. The core assumption of this framework is that Arm can truly convert AI, cloud, automotive, CSS, and a higher royalty mix into years of compounding owner-earnings growth; but I have not assumed sustained 30%+ growth over the very long term, nor have I applied an extreme terminal multiple. Under this framework, my approximate per-share value ranges are: conservative $25–35, base $45–70, optimistic $80–120. These figures are not "precise values" but "reasonable ranges based on explicit assumptions."

Relative valuation. Arm currently trades at a trailing P/E of about 302x, P/B of about 33x, P/FCF of about 279x, and EV/EBITDA of about 234x; the corresponding multiples for Cadence are roughly 81.8x, 14.8x, 67.7x, and 48.8x, and for Qualcomm roughly 22.1x, 7.9x, 17.1x, and 16.8x. Even allowing that Qualcomm is a downstream chip company and Cadence is a closer high-quality EDA/IP software-style company, Arm's valuation is still an order of magnitude expensive. More important, Arm's latest ROIC (third-party TTM basis) is about 14.3%, not meaningfully higher than Cadence's roughly 19.7%, yet it commands a far higher valuation. In other words, what the market pays Arm is not a "slightly higher" premium but a one-time payment for years of high-realization growth.

Asset/liquidation value. This company is not suited to the traditional "liquidation-discount" framework for judging cheapness, because the main value clearly does not lie in book assets but in the future royalty stream and the architecture ecosystem. Still, the asset method is useful because it tells you that "the downside protection is almost not on the balance sheet." At the end of FY2026, cash and short-term investments were $3.601 billion and total liabilities were $2.417 billion, leaving net cash of about $1.184 billion after covering all liabilities; total shareholders' equity was $8.286 billion, equivalent to a book value per share of less than $8. The current share price of $256.73 shows that this stock rests almost entirely on future earning power and the capital market's willingness to keep granting a high multiple, rather than on asset support.

Integrated valuation conclusion:

• Conservative intrinsic-value range: $25–40 per share

• Fair intrinsic-value range: $45–70 per share

• Optimistic intrinsic-value range: $80–120 per share

Relative to the current $256.73 share price, Arm is broadly in a state of significant premium:

• against the conservative range, the premium is extreme;

• against the fair range, the premium is roughly +267% to +471%;

• against the optimistic range, there is still a premium of about +114% to +221%. So my conclusion is clear: there is no margin of safety at the current price.

Ideal buy-price range: $30–50. Acceptable long-term holding range: $50–80. Clearly overvalued range: above $120. These three ranges are not market forecasts but capital-allocation discipline lines derived from the owner-earnings/DCF assumptions above. Different investors can use different parameters, but on a conservative basis it is hard to explain anything above $250 as a "value-investing price."

Margin-of-safety assessment: insufficient. The most fragile assumption in the valuation is not "whether Arm will grow" but "how long Arm must sustain extremely high growth to deserve the current price." If future growth falls short of expectations, if margins come under pressure from heavier capital and more silicon investment, or if the multiple the market is willing to grant compresses from over 200x FCF to a more normal 40–80x, then even if the company keeps growing, shareholders may still face years of low returns or even permanent capital loss. This is the classic "good company but bad price."

Risks, Comparisons, and Checklist

The most important risk is not share-price volatility but the sources of permanent capital loss:

First is competition and technology-substitution risk. Arm itself lists x86 and RISC-V as principal competitors in the 20-F and acknowledges that some customers also support RISC-V. Google's progress on Android on RISC-V, Qualcomm's bet on Wear OS / RISC-V, and China's policy push for RISC-V all show this threat is not fictional. If development toolchains, software adaptation, and performance-per-watt keep improving, RISC-V will first erode Arm's bargaining power in low-end MCUs, specific verticals, and geopolitically sensitive markets.

Second is customer and geographic concentration risk. In FY2025 the top five customers accounted for 56% of revenue and the largest, Arm China, for 17%; Arm China is not controlled by Arm yet plays an important commercial-channel role in the China region, and its receivables and information quality have had problems in the past. Although Arm China–related revenue is declining as a share, it remains a "single-point risk that cannot be ignored."

Third is business-model-evolution risk. What was originally most attractive about Arm was its extremely high gross margin, very low capital intensity, broad neutral platform, and long-tail royalty. The new change in FY2026 is the launch of its first Arm-designed data-center chip, accompanied by a sharp rise in PP&E spending. If this brings higher value capture, today's high valuation will look less absurd; but if it draws Arm deeper down a heavier-asset, higher-competition, lower-return path, then what the market grants today is a "software premium" while what the company delivers in the future could be a "semiconductor hardware return."

Fourth is governance and control risk. SoftBank's 87.1% stake, board-nomination rights, related-party-transaction arrangements, and financing activity using Arm shares as collateral mean minority shareholders have very limited influence at key moments. For a value investor, this will not necessarily ruin the investment, but it should materially raise your price requirement.

Fifth is overvaluation risk. This risk needs no special trigger event. Simply having the market gradually re-rate Arm from an "AI-core-platform narrative stock" to a "high-quality but still-realizing IP company" would be enough to cause a clear drawdown. Arm's current FCF yield is far below the S&P 500 earnings yield and below the U.S. 10-year Treasury yield; this means investors buying now receive not a "current cash return" but a promise of extreme optimism about future growth.

The strongest counterargument is: Arm is indeed an excellent company, but it is shifting from "a very good IP platform" to "the imagined super-core of AI infrastructure," and the share price has already priced in this second identity. In other words, the easy mistake is not to view Arm too negatively but to view it as too perfect: to assume it will keep lifting royalty into the AI era, successfully enter the data center and more system-level value layers, avoid erosion by RISC-V and customer in-house design, and at the same time maintain a software-company-like return on capital. If even one or two of these four things fail to hold simultaneously, today's valuation will look too high.

Which facts would overturn my investment judgment? If the following facts appear over the next two to three years, I would admit I "was previously too conservative":

• data-center / cloud-AI royalty keeps growing strongly and rises markedly as a share of the revenue mix;

• Armv9/CSS/in-house silicon meaningfully raises value capture per chip, while the company's capital-expenditure rate falls back again;

• equity incentives decline as a share of revenue and share dilution slows, while per-share owner earnings step up clearly;

• Arm, without sacrificing its neutral platform position, truly converts demand from 2 billion+ customers into high-return revenue.

Conversely, if the following occur, I would consider the bullish logic broken:

• RISC-V penetration in Android, edge devices, automotive, or the China market is clearly faster than expected;

• Arm China's collections, audit information, or commercial relationships deteriorate again;

• capital expenditure, SBC, and share count keep rising but do not translate into higher per-share owner earnings;

• data-center/AI growth is very high, but the royalty take rate and earnings quality do not improve to a comparable degree;

• SoftBank-related transactions or governance arrangements clearly disadvantage minority shareholders.

How does it compare with other opportunities? If you compare it with the closest listed high-quality software/IP comparable, Cadence trades at a far lower valuation than Arm despite a higher ROIC; if you compare it with the S&P 500, the index's current earnings yield is about 3.13% while Arm, on FY2026 company-style FCF, is only about 0.33%; and against the U.S. 10-year Treasury, the latter yields about 4.67%. So from the current starting point, buying Arm is not clearly better than buying the index, let alone better than the certainty of a high-grade bond's return. The only reasonable reason to buy Arm is if you believe its long-term owner earnings will far exceed my base assumptions above, and that the market will remain willing to grant it a high premium. For a balanced-to-conservative investor, I do not think this is the best use of capital.

A compressed checklist follows. These are judgments, not facts in themselves, but each rests on the data cited above.

Open questions and limitations:

• The most recent complete statutory annual report is still the FY2025 Form 20-F; for FY2026 I rely mainly on the May 6, 2026 quarterly update and condensed statements, and have not yet seen the formal full-year FY2026 20-F.

• Among listed comparables, none is a pure ISA/IP/licensing company structurally identical to Arm; Cadence, Synopsys, and Qualcomm can only provide reference points for valuation and return on capital, not a one-to-one match.

• The conservative estimate of owner earnings carries explicit assumptions in its treatment of SBC, maintenance capital expenditure, and the "normalization" of working capital, and should not be read as a single precise value.

Final Investment Conclusion

【Final rating】 Watch

【One-line investment thesis】 Arm is worth studying as an "excellent company," but not worth buying at today's price, which "prepays almost a decade of success."

【Core bullish reasons】

• High-quality business model: high-margin IP/ISA licensing plus a long tail of royalty, with built-in long-term compounding potential.

• Strong ecosystem: more than 310 billion to 350 billion Arm-based chips, more than 20 million to 22 million developers, and an extremely strong smartphone position.

• Clear AI/cloud/automotive tailwinds: AWS Graviton, Google Axion, Microsoft Cobalt, and others are all strengthening Arm's cloud position.

• Solid FY2026 data: revenue of $4.92 billion, net income of $904 million, and operating cash flow of $1.524 billion.

• Strong balance sheet: cash and short-term investments of $3.601 billion at the end of FY2026, an overall net-cash position with strong cycle resistance.

【Core bearish reasons】

• Extreme valuation: about 302x P/E, 279x P/FCF, and 234x EV/EBITDA currently, leaving almost no room for error.

• Very low current cash return: on FY2026 FCF, the equity FCF yield is only about 0.33%.

• Customer and Arm China risk is real: the top five customers account for 56%, and Arm China is still 17% of revenue and uncontrolled.

• Equity incentives are heavy and the share count is still rising, continuously diluting minority shareholders.

• The business model is getting heavier: extending into CSS/system-level solutions/in-house silicon may weaken the original asset-light character.

【Key assumptions】

• AI and data-center adoption can keep converting into a higher royalty take rate, not merely high license growth.

• Armv9 / CSS / platform solutions can expand economic-interest capture.

• RISC-V will not quickly break open the high-value mainstream market.

• The Arm China relationship stays controllable, with stable collections and reliable information quality.

• In-house silicon will not materially impair return on capital.

【Fair buy price】 $30–50 per share. Basis: under a conservative-to-base owner-earnings/DCF valuation framework, only this range begins to reflect compensation for governance, dilution, competition, and financial volatility.

【Target holding period】 Ten years or more. But the premise is not "a long horizon justifies a high purchase price"; it is that "a good business should also be held for a long time at a reasonable price."

【Expected annualized return】

• Conservative scenario: -8% to -12%

• Base scenario: -1% to +3%

• Optimistic scenario: +4% to +8%

These return estimates apply to buying at the current price and holding for the long term, and assume strong owner-earnings growth over the next ten years but a reversion of the valuation toward a more normal range. They are not price forecasts but a "return-risk profile at the current price."

【Maximum loss risk】 A 60% to 80% permanent capital loss is not an exaggeration. The worst case is not company bankruptcy but the simultaneous occurrence of slower-than-expected AI realization, erosion by RISC-V / customer in-house design, rising capital intensity, and a large reversion of the valuation multiple. For such a high starting valuation, all it takes is for "good" to fall short of what the market originally imagined, and shareholders may go years without a reasonable return.

【Tracking metrics】

• Royalty revenue growth

• The revenue/royalty mix of Armv9 and CSS

• The share and growth of data-center and cloud-AI royalty

• ACV, RPO, and the number of Total Access / Flexible Access

• Arm China revenue share, receivables, and collection quality

• SBC as a share of revenue and changes in shares outstanding

• PP&E / intangible-asset spending as a share of revenue

• The match between operating cash flow and net income

• Changes in the top-five-customer share

• Changes in SoftBank's stake and governance arrangements

【Signals that trigger reassessment】

• RISC-V accelerates commercialization in Android / wearables / automotive / the China market

• Arm China's information, collections, or compliance run into problems again

• The data-center silicon business lifts capital expenditure without bringing high returns

• Dilution keeps rising while per-share owner earnings does not improve in step

• The company loses its neutral platform character and its customer relationships deteriorate

• SoftBank-related arrangements clearly disadvantage minority shareholders

【Final recommendation】 If your framework is "buy a stock the way you would buy an entire company," then what most deserves respect about Arm today is its business quality, and what most needs restraint is the purchase price. My recommendation is not to dismiss Arm but to place it on a high-priority watch list: keep tracking the quality of its AI monetization, the degree of dilution, its capital-expenditure path, and the RISC-V competitive landscape; until there is a larger margin of safety, do not treat it as a "good investment" just because it is a good company, a hot company, and a rising company.

This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.