The Multiple Halved, and So Did the Reason to Pay It

Results vs. Consensus

Arm reported its fiscal second quarter, ended September 30, 2026, after the close on November 4, with the call at 17:00 Eastern the same day. On the headline it was a clean double beat and the company's second-highest revenue quarter ever. Beneath the headline it produced the widest gap between the two revenue lines' growth rates in our coverage, 50% against 8%, and the line that lagged is the one we pay a premium multiple for.

Q2 FYE27 Scorecard

Year-over-Year Comparison

Sequential Comparison

Sequential figures for the first quarter are derived from the six-month columns of this quarter's filing less the three-month columns, and tie exactly to the first-quarter report in every case.

Quality of Beat

• Revenue: Fully organic. There was no acquisition contribution to revenue, no currency effect disclosed, and no one-time item in the revenue lines. The composition is the issue, not the quality. The $103 million license overshoot came from three new Arm Total Access agreements and four new compute subsystem licenses, including one major customer signing its first Total Access deal, which is real contracted business rather than a pull-forward of an existing renewal. Against that, the CFO declined to call $773 million a structural level, and annualised contract value grew only 11.9%, so the recognised line is running well ahead of the committed-fee base.
• Margins: The non-GAAP margin beat is high quality and now has a consistent pattern behind it. Operating expense of $774 million came in $6 million under guide, the fifth consecutive quarter below plan in our coverage, and grew 19.4% against revenue up 27.1%. Non-GAAP operating margin of 44.2% is the highest of our coverage period outside the seasonally peak fourth quarter. The offset is entirely in GAAP: share-based compensation grew 41.9% to $376 million, or 26.1% of revenue, which pushed GAAP operating margin down 50 basis points even as non-GAAP margin rose 310.
• EPS: Operational, not below-the-line. Non-GAAP EPS of $0.52 is carried by the $171 million increase in non-GAAP operating income; interest income contributed $29 million against $27 million, the non-GAAP tax rate of 14.9% is in line with the 15.0% half-year rate, and the diluted share count rose less than 1%. This is the cleanest EPS beat of the coverage period, and the exact opposite of last quarter, when GAAP net income "doubled" on equity-investment gains and a tax benefit.

Revenue assessment

Total revenue of $1,443 million is the company's second-highest quarter ever, behind the $1,490 million of the March quarter, and its fifth consecutive quarter of growth above 20%. We verified that streak against our own series: 34.5%, 26%, 20.1%, 22.4% and now 27.1%. On a trailing-twelve-month basis revenue of $5,464 million grew 23.8%, so the franchise is compounding at better than 20% with no sign of fatigue in the aggregate. What has changed is where the compounding comes from. Trailing-twelve-month license revenue grew 32.2% while trailing royalty revenue grew 16.8%, and the gap is widening in the wrong direction for a company valued as an annuity.

Margin assessment

Our central bear point for the first three quarters of coverage was an expense base outgrowing a record revenue line. That point is now comprehensively answered. Operating expense has come in below guide in all five quarters of our coverage, the guided path to roughly $3.20 billion for fiscal 2027 implies roughly 18% growth against fiscal 2026's 33%, and the resulting 44.2% operating margin is being delivered while research and development still grows 17%. The only genuine margin caution is the one management cannot adjust away: the 40%-gross-margin silicon business begins shipping in the fourth quarter, and on our estimates it drags blended non-GAAP gross margin from 97.9% toward roughly 96.5% this fiscal year and toward the low 80s once the full "up to $2 billion" frame is recognised.

Earnings and cash assessment

The GAAP headline needs correcting before it is used. GAAP net income of $130 million against $238 million will be reported as a 45% collapse. It is not one. The year-ago base contains a $131 million gain on the Artisan disposal, visible in that quarter's reconciliation inside $128 million of other non-operating income. Excluding it, prior-year GAAP net income was $107 million and this quarter is up 21.5%. The real GAAP deterioration is narrower and sits in share-based compensation. The cash statement is the genuinely poor part of the quarter: free cash flow of $110 million converted at 19% of non-GAAP net income, trailing-twelve-month free cash flow fell $301 million in three months to $1,096 million, and the contract-asset build that we reported as resolved in August has returned, consuming $203 million of cash in the quarter and $153 million across the half.

Revenue Line Performance

Royalty

Royalty revenue of $670 million is the central fact of this quarter and the reason a double beat still traded lower after hours. It grew 8.1%, fell 6.3% sequentially, came in roughly $30 million below the growth rate management guided to in July and roughly $35 million below where the Street sat. Management broke the line into its three business units, and the decomposition is clear even without dollar sizing.

Assessment: Edge AI is the majority of the royalty base and it went flat, which is enough on its own to explain the quarter. The concerning part is not the print but the forward path. The December quarter is guided to low single digits year over year, which on our arithmetic puts nine-month royalty growth near 10.5%. Hitting the high-teens fiscal 2027 royalty frame management offered in July would now require about 36% growth in the fourth quarter against a $671 million base. That frame is gone, and management did not replace it with another.

License and other

License and other revenue of $773 million grew 50.1% to a second-quarter record. The sources are specific and contracted: three new Arm Total Access agreements signed in the quarter, taking active subscriptions above 60, including one major customer's first Total Access deal, plus four new compute subsystem licenses split two for data centre chips and two for smartphones. The SoftBank agreement contributed $209 million of the total, up 17.5%, and the CFO expects that run rate to hold for the rest of the fiscal year. Management guided license revenue up a little over 50% again in the December quarter and said it will "step up again" in the March quarter.

"These were incremental and new deals that are kind of hard to forecast. We are in that same situation again." — Jason Child, CFO

Assessment: This is Arm's best second-quarter license print and its second-best license quarter, behind the $819 million of the March quarter, and the forward signal is real, because Total Access and compute subsystem agreements are the leading indicator of royalties a year or two out. Two things stop us from capitalising it. First, management will not call it structural, and "hard to forecast" is its own caveat. Second, annualised contract value grew only 11.9%, so the committed-fee base is not confirming the recognised line. Trailing-twelve-month license revenue now runs at 1.49 times annualised contract value, against 1.39 times three months ago and 1.26 times a year ago. Either the metric understates the book, or recognition is running ahead of commitments.

Key Operating Metrics

Key Topics & Management Commentary

Overall Management Tone: Confident and tightly scripted on the two growth stories, and notably unwilling to engage on the one that went wrong. The royalty shortfall was framed as a memory-pricing and comparison issue in a short prepared-remarks passage, and in Q&A it came back only as a handset-unit and comparison story, never against the guide it missed, while the license overshoot, the silicon revenue frame and the hyperscaler share gain each received extended treatment. Where management was pressed for a number it did not want to give, on cloud customer concentration and on the next silicon generation's schedule, it answered with names and analogies rather than figures. The posture was the most promotional of our coverage period, and it was deployed in the quarter with the least favourable underlying mix.

1. Royalty growth halves, and the escalator stops offsetting units

One quarter ago we wrote that the rate escalator had passed its hardest test, because smartphone royalties grew double digits into a double-digit unit decline. This quarter the same test produced the opposite result. Edge AI royalty was flat year over year. The rate gains from Armv9 and compute subsystems are still accruing, but they no longer exceed the unit and mix damage from memory-driven handset compression.

"In edge AI, royalty revenue was roughly flat year-on-year. Higher memory prices continue to pressure smartphone unit sales, particularly in the mid and lower-tier segments. That headwind was offset by the benefit we continue to see from higher royalty rates as our latest Armv9 and CSS technologies are adopted. As smartphone market conditions normalize, we expect edge AI royalty revenue to reaccelerate." — Jason Child, CFO

The word doing the work in that passage is "offset". Last quarter management said the rate benefit was outrunning the unit headwind and projected double-digit smartphone royalty growth. This quarter it merely cancelled it. There is also a disclosed reason the escalator should weaken from here: management said Armv9 penetration is "now getting over 50%", which it characterised as "a little less of a tail end" of upgrade benefit. The remaining rate lever is compute subsystem adoption and next-generation compute subsystem pricing, not the architecture transition that drove the last three years.

Assessment: This is the first quarter in which Bull 1, the pillar the whole thesis rests on, failed its own test rather than passing it. We are moving it from on track to at risk. The mechanism is intact and the rate per chip is still rising, by an amount an analyst calculated on the call as at least mid-single digits, which management did not dispute, although the CFO cautioned against using the blended average to forecast. But a rate escalator that only cancels a unit decline is a defensive asset, not a 20%-compounder, and the premium multiple was underwritten on the latter.

2. The fiscal 2027 royalty frame disappears rather than being reset

In July the CFO walked the full-year royalty growth frame down from around 20% to "closer to the high teens", in a Q&A answer rather than prepared remarks. We named grading that frame as the third item on our watch list. It is now arithmetically unreachable, and nobody restated it.

Management's forward royalty commentary beyond the December quarter came business unit by business unit. Data centre royalties are expected to keep "growing triple digits over the next number of years". For edge, the memory headwind "does end mostly in this quarter that we're now in", recovery starts in the March quarter, the comparisons get easier from January and February, and mobile royalty growth should be "at least being in the kind of double digits plus" next fiscal year. Those are statements about individual units, offered in answers to questions about units and mix, and none of them is a full-year royalty frame.

Assessment: A company that guides one quarter forward, cut its full-year royalty frame in July and has now let the reduced frame lapse without restating it is asking investors to model the most important line in the business from nothing. The fiscal 2028 and 2029 claim of 20%-plus royalty growth, described in July as "very much still intact", went unmentioned on this call. We treat it as withdrawn until it is repeated.

3. Second-quarter-record licensing, and why a $103 million overshoot is not a run rate

The license line did the opposite of the royalty line, and management's explanation of why is the most useful disclosure on the call. The overshoot was not a timing shift in a known renewal. It was incremental demand that was not in the plan when the fiscal year started, and management said the plan had assumed the opposite mix.

"What is interesting is, when we started this year, we thought it was going to be much more around royalties and that a lot of the license growth related to all the AI investments, we thought maybe we kind of had passed a lot of that. What is interesting is, it seems to be that with AI and the consistent variability, everyone is trying to address all the agentic workloads, we are seeing that the need for additional license and additional product variations and different technologies is not slowing down at all." — Jason Child, CFO

Management also set out the conversion mechanism explicitly: "any license beats, especially with ATAs and CSSs, that is very highly likely to turn into design wins and very likely to turn into even more royalties, probably 12-18 months, 24 months down the road." Architecture licenses, which historically produced the largest and least repeatable deals, were described as "very, very few and far between", so the mix is now almost entirely Total Access subscriptions and compute subsystems.

Assessment: This is a genuinely good forward signal, and it is the strongest argument available against our Hold. Subscription and compute subsystem licensing is higher quality than one-off architecture deals, and a strong license book today is royalty revenue in fiscal 2028 and 2029. The reason we do not capitalise it yet is that the mix shifted toward licensing because royalty demand came in short, not in addition to it, and management explicitly refused to call $773 million a new level. We will know within two quarters whether the book converts.

4. Arm AGI CPU: the frame doubles, with Meta as chief customer

The silicon business is the clearest positive development of the quarter. The revenue frame moved from "more than $1 billion" across fiscal 2027 and 2028 to "up to $2 billion", and for the first time it is backed by signed purchase orders rather than a demand pipeline.

"Last quarter, we said that we had increased confidence in achieving more than $1 billion across fiscal year 2027 and 2028. In the past 90 days, we have signed purchase orders from multiple customers and signed supply agreements with companies in our supply chain, giving us increased confidence in achieving up to $2 billion across fiscal 2027 and 2028." — Jason Child, CFO

The specifics improved on every axis we asked about in August. Purchase orders extend through calendar 2027. Supply agreements are signed with companies in the supply chain, which the CEO said spans wafers, substrates, testers and memory. First commercial deliveries remain on track for the March quarter. Oracle Cloud Infrastructure plans to offer Arm AGI CPU instances in 2027, which gives the product a named cloud channel alongside the direct customers. Supermicro, Lenovo and ASRock build the reference racks. And the CEO called Meta, already named as lead partner in May, the chief customer.

"Demand is stronger today than we talked about when we did the Arm Everywhere event in March, and subsequently, demand from multiple customers, Meta being the chief customer, has also been beyond what we have expected." — Rene Haas, CEO

Assessment: Doubling a revenue frame in 90 days while converting demand indications into purchase orders and supply agreements is the single best execution datapoint of our coverage period, and it answers the supply question we flagged in August. The counterweight is concentration: a $2 billion programme anchored on one named customer, with first revenue still a quarter away, is not yet a diversified business.

5. The silicon gross margin band widens downward

Last quarter management put the first generation at "high 30% range, maybe low 40s percent" and the 50% target at "over the next couple of years". This quarter both ends moved, and in opposite directions.

"Yeah, maybe just to reiterate what we said was for this first version, we're expecting to be somewhere in the high 30%s to the low 40% gross margin. For the first version, we expect it to be probably a little lower, then as we develop more expertise, probably rely on doing a little more work ourselves and maybe less from partners over time, which others in the market do. Over time, we'll get ourselves to more of the market standard of 50%+, but I don't expect that to happen on the first version. That's probably going to happen more on the next version. So, for this version, I would stick to the kind of 30%-40% range, then we'll obviously give you updates on a quarterly basis as we start shipping in Q4." — Jason Child, CFO

The answer begins at "high 30%s to the low 40%" and ends, four sentences later, at "the kind of 30%-40% range" with the first version "probably a little lower" than that. The 50%-plus target has moved from a two-year in-sourcing programme to "the next version", which is nearer in time but conditional on a product that has no announced schedule. The CEO separately said memory pricing "has obviously increased a lot since we announced the Arm AGI CPU" and that data centres are "getting more and more complicated to build", while arguing there is "probably some pricing elasticity" and "headroom" to offset it on higher-core-count parts.

Assessment: Take the blended arithmetic seriously, because the reporting will not show it. On our fiscal 2027 estimate of roughly $5.94 billion with about $150 million of silicon at 35% to 40%, blended non-GAAP gross margin lands near 96.5% against 97.9% today. Recognise the full $2 billion across fiscal 2027 and 2028 and the blend moves toward the low 80s. Arm will not break silicon out as a separate line until it exceeds 10% of revenue, expected in fiscal 2028, so the entire dilution transition happens inside lines investors cannot decompose.

6. Data centre royalties double a fifth consecutive quarter, still unsized

Cloud AI remains the strongest part of the business and the hardest to model. Data centre royalty revenue doubled year over year for a fifth consecutive quarter, and the share data management disclosed is the most striking it has given.

"Our data center royalties doubled year-over-year, and Arm-based server market share at the top hyperscalers has more than tripled from 15% in calendar 2024 to more than 50% in calendar 2025. Based upon expected ramps across Google Axion, AWS Graviton, Microsoft Cobalt, and NVIDIA platforms, we now expect that share to exceed 60% by the end of calendar 2026." — Rene Haas, CEO

The CEO also returned to a second, distinct vector that management has cited since May: the conversion of head-node CPUs attached to accelerators from x86 to Arm. He cited NVIDIA's Vera Rubin platform and, more usefully, Google's announced conversion of x86 to Axion for the hosts connecting to its eighth-generation TPU. Supporting third-party datapoints in the shareholder letter include Amazon reporting Graviton revenue commitments up more than 20 times year over year, and NVIDIA's Vera CPU now generally available on Oracle Cloud Infrastructure and in production with OpenAI. Management expects data centre royalties to grow triple digits "over the next number of years".

Assessment: Bull 2 stays on track and is the most durable part of the investment case. Going from 15% to more than 50% of top-hyperscaler server share in one calendar year is a share shift with few precedents in semiconductors, and the head-node conversion opens a socket category that was not in the original thesis. The disclosure complaint is now five quarters old: management has said "doubled" five times without ever publishing the base in dollars. When asked this quarter whether data centre is approaching a third of total royalties, the CFO offered only a rough share: "kind of approaching 20%-ish of total royalties" last year, and "I do not think data center will get quite to a third, but it will sometime soon."

7. Operating leverage compounds, and opex lands under guide in every quarter we have covered

The bear point that drove our initiation Hold is now the most clearly resolved part of the thesis. Operating expense of $774 million came in $6 million under guide, grew 19.4% against revenue up 27.1%, and produced a 44.2% non-GAAP operating margin.

"Non-GAAP operating expense was $774 million, up 19% year-on-year, as we continued to expand our engineering capacity and invest in next-generation platforms, compute subsystems, and our Arm AGI CPU product family. Non-GAAP operating income was $638 million, up 37% year-on-year. Non-GAAP operating margin was 44.2%, up 300 basis points from a year ago, as revenue outpaced operating expense growth." — Jason Child, CFO

The pattern now spans all five quarters we have covered: operating expense has come in under guide every time. In the two most recent, the roughly $760 million guide for the June quarter produced a $733 million print, and July's roughly $780 million guide produced $774 million. The December guide of roughly $825 million is in line with the mid-single-digit sequential path management committed to in July. Stacking the guided quarters puts fiscal 2027 operating expense near $3.20 billion, which is roughly 18% growth against fiscal 2026's 33%.

Assessment: Bear 2 stays contained, and this is now a company that has stopped surprising to the upside on spend while still growing research and development 17%. The unresolved offset is share-based compensation, which grew 41.9% to 26.1% of revenue and is the sole reason GAAP operating margin fell while non-GAAP margin rose 310 basis points. Diluted shares rose less than 1% because of withholding, so the dilution is being paid for in cash: $407 million of withholding tax on vested shares across the half.

8. The cash flow reversal reverses

In August we reported that the cash conversion problem flagged at initiation "did not just improve, it inverted", and we moved the bear point from emerging to contained on the strength of a single quarter. That was premature.

The half is fine: free cash flow of $775 million grew 38.1%, and conversion of 74% is in line with the prior year's 71%. The quarter is poor, and the trailing-twelve-month figure is the honest summary, because it fell $301 million in three months as this quarter's $110 million replaced last year's $411 million in the window. The driver is the same balance-sheet line as at initiation. Contract assets, current and non-current, rose $153 million across the half to $1,450 million, and contract liabilities fell $171 million, a combined $324 million working-capital drag. Related-party contract assets rose $40 million sequentially to $617 million after one quarter of decline.

Assessment: We are moving Bear 3 back from contained to emerging, and we should say plainly that calling it contained after one quarter was a process error on our part rather than new information. The lesson is specific to this company: Arm's cash conversion swings on contract-asset timing with a controlling-shareholder counterparty, and single-quarter readings in either direction are not evidence. The trailing-twelve-month series is the only reliable frame, and on that basis free cash flow has gone from 134% growth to 2.1% growth in one quarter.

9. The SoftBank agreement stops being a growth prop

For four quarters the related-party licence agreement with the controlling shareholder inflated reported growth. This quarter it did the reverse, and the reversal is large.

"Of the $773 million in license revenue, our agreement with SoftBank for technology licensing and design services contributed $209 million, up 17% year-on-year. We expect the quarterly run rate for the rest of the year to remain around this level." — Jason Child, CFO

Every growth measure is now better excluding the agreement than including it, which has not been true in any prior quarter of our coverage. The gap on total revenue was negative 14.2 points in the December 2025 quarter and negative 3.8 points in June; it is positive 1.8 points now. Revenue from external customers grew 42.9%, which is the cleanest read on third-party demand available and the highest of the coverage period.

Assessment: The growth leg of Bear 1 is now fully resolved, and we are moving the tag from materialising to emerging. What keeps it above contained is unchanged and specific: five quarters into coverage there is still no disclosed term, scope, pricing basis, renewal mechanic or recognition pattern for an agreement running at roughly $820 million a year, and the related-party contract-asset balance rose again this quarter. Disclosure of those terms was an upgrade trigger at initiation and it remains the single cheapest thing management could do to improve the quality of reported revenue.

10. Annualised contract value decelerates a third consecutive step

Annualised contract value is the only contracted-revenue metric Arm still publishes, after remaining performance obligations and the licence counts were withdrawn in the June quarter. It grew 11.9% to $1,790 million, which is the third consecutive step down from a 28% plateau.

Management addressed the metric only to reaffirm its usefulness: "license revenue can vary significantly from quarter to quarter based on the timing and accounting treatment of large agreements, so ACV remains a useful measure of the underlying licensing trend." That framing is harder to accept in a quarter when the recognised line grew 50% and the measure of the underlying trend grew 12%.

Assessment: Bear 5 stays at materialising. Arm has removed two disclosures and is now asking investors to rely exclusively on a third that is decelerating, diverging from recognised revenue and narrated as stable. The positive reading is available and we should state it: if the metric understates a book being signed in multi-year Total Access form, the divergence is a timing artifact of the annualisation convention. We cannot test that, which is the point.

11. A $249 million acquisition nobody mentioned

The cash flow statement shows $249 million of "payments for acquisitions, net of cash acquired" in the quarter, against nothing in the comparable quarter and nothing in the prior half. Goodwill rose $92 million and net intangible assets rose $155 million across the half, which together account for the cash. The non-GAAP reconciliation carries $12 million of acquisition-related expenses, the largest such charge in our coverage period, against $1 million in each of the December and March quarters.

The target is not named in the shareholder letter. The transaction is not mentioned in the letter's four pages of narrative, not in either prepared remarks section, and not raised by any of the nine analysts who asked questions. For a company with $11.4 billion of total assets, a quarter in which it deployed $249 million of cash on an unnamed acquisition and $105 million more on equity investments across the half passed entirely without comment.

Assessment: This is not large enough to change the thesis and it is not evidence of anything improper. It is evidence about the disclosure culture, which is the subject of two of our five bear points. A single sentence naming the target and its purpose would have cost nothing. Its absence, in the same quarter that a backlog metric stayed withdrawn and a customer-concentration question went without a figure, is a pattern rather than an oversight.

12. Physical AI, edge devices and ten billion chips in a quarter

The third royalty pool produced its first disclosed growth rate. Physical AI royalty revenue grew 50% year over year on advanced driver assistance and in-vehicle infotainment adoption, and management launched Arm Total Design for Physical AI with more than 80 participating companies. The supporting datapoint in the letter is specific: Tensor's Level 4 personal Robocar integrates more than 400 Arm CPU cores.

On the edge side, Arm-based PCs have reached roughly 20% of global shipments, Google introduced a category of AI-native laptops with Arm as the primary compute platform for many models, and NVIDIA's RTX Spark systems are now available. In smartphones, Arm CSS for Mobile 2 lifts agentic AI workload performance by up to 70% through SME2, with OPPO and vivo named as adopters, and the first AI-native Mali GPU delivers up to four times the frame rate at the same sustained power, with Xiaomi adopting. Across everything, partners reported more than 10 billion Arm-based chips shipped in a single quarter for the first time, up 1.5 billion year over year, against an ecosystem of more than 30 million developers.

Assessment: Bull 3 is confirmed, and for the first time one leg of it has a published growth rate rather than an adjective. A 50% royalty growth rate in physical AI is a real third pool, and the automotive content story of more cores per vehicle is the same rate-escalator mechanism operating in a market where units are not falling. The edge-device evidence is strong on design wins and silent on economics: Arm has not sized the PC royalty pool, and the 20% shipment-share figure is a unit statistic, not revenue.

Guidance & Outlook

The December quarter guide of $1.54 billion would be a record quarterly revenue, above the $1,490 million printed in March, and 24% above the year-ago quarter. The composition continues the swap this quarter began. License revenue is guided up a little over 50% again, which implies roughly $770 million to $790 million, and the CFO added that it "is going to step up again" in the March quarter. Royalty revenue is guided to low teens sequentially and low single digits year over year, which implies roughly $750 million to $775 million against a $737 million base, and management attributed the deceleration to continued edge weakness plus "a tough year-on-year comp", which is accurate: the December 2025 quarter was the strongest royalty growth of fiscal 2026 at 27%.

Implied quarter-over-quarter ramp: Hitting our fiscal 2027 revenue estimate of roughly $5.94 billion requires about $1.67 billion in the March quarter, up 8% sequentially from the December guide and 12% year over year. That split needs royalty near $790 million, which assumes the edge recovery management has guided to the March quarter actually arrives, plus license near $880 million including the first Arm AGI CPU commercial deliveries. The royalty half of that is the load-bearing assumption in the whole model.

Street at: No independent pre-print consensus for the December quarter was available to us at publication. The $1.54 billion midpoint sits 24% above the year-ago print, and sell-side price target revisions published on the reaction day cut targets without changing ratings, which suggests the Street has trimmed royalty estimates rather than its structural view.

Guidance style: Arm guides one quarter forward, gives full-year direction only as verbal frames, cut its fiscal 2027 royalty frame in July and has now let it lapse without restating it. On the quarterly guide itself the company is reliably conservative: revenue has cleared the midpoint in all five quarters of our coverage, EPS has landed at or above the top of the range in three of five, and operating expense has come in under guide in all five. The problem is not the quality of the one-quarter guide. It is that one quarter is all there is.

Analyst Q&A Highlights

Whether a $773 million licensing quarter represents a new structural level

The most consequential exchange of the call asked management to commit to the $773 million license figure as a base rather than a spike, and specifically whether architecture license agreements, historically the largest and lumpiest deals, were inside it. The answer removed the lumpiest category from consideration, confirmed sequential step-ups in both remaining quarters of the fiscal year, and then declined to extend the commitment into next year.

Q: "$773 million for licensing really stands out. Congratulations to everyone there. You did talk about Arm Total Access, you did talk about CSS deals, particularly smartphones, and I think data center you mentioned. But do we have any sort of architectural license agreements in there? Could we look at $770 million plus as a new level, a structural level?"
— Lee Simpson, Morgan Stanley

A: "Architecture licenses are very, very few and far between. I would think of it as really being ATA and CSS. That is the bulk of most everything we are selling going forward. In terms of absolute magnitude of dollars, yeah, $773 million was a great quarter. If you, I think, kind of work out the details, I said it was going to be increasing and going up from there in this next quarter, and even in Q4, it is going to step up again. We have not talked about next year, whether or not that is the new number every quarter. I do not know if I would say that, but I will say the pipeline has been surprisingly strong, stronger than we expected when we started the year."
— Jason Child, CFO

Assessment: This is the exchange that decides how to value the quarter, and management answered it honestly rather than conveniently. Removing architecture licenses from the mix genuinely improves the quality of the line, because Total Access and compute subsystem deals are subscriptions and design commitments rather than one-off transfers. But "I do not know if I would say that" is a refusal, and it is the right one. The license line is guided higher for two more quarters and then becomes an open question, which is exactly how we are modelling it.

What changed in 90 days to double the silicon revenue frame

The opening question of the call asked what specifically underwrote the move from a $1 billion frame to $2 billion, and whether additional customers were contractually committed beyond the anchor. The answer was the most specific supply-chain disclosure management has given on this programme, and it confirmed Meta as the chief customer.

Q: "I think on the top of the mind for everybody is Arm AGI CPU, and you are obviously signaling something closer to $2 billion in revenue over the five-quarter period up to the end of fiscal year 2028. I am wondering what is driving your confidence into that with respect to securing more supply with your different partners. I would assume most of the revenue initially goes to supporting Meta, but maybe if you can talk about whether or not you have signed and committed contractually with additional customers."
— Gary Mobley, StoneX Group

A: "Demand is stronger today than we talked about when we did the Arm Everywhere event in March, and subsequently, demand from multiple customers, Meta being the chief customer, has also been beyond what we have expected. That being said, we have been working very hard over the last few quarters on securing what is necessary in the supply chain to address that demand. There are multiple components to it. There is wafer supply, there are substrates, there are testers, there is memory, et cetera... I can tell you today that the confidence is the strongest it has been to date, and it is definitely stronger than it was 90 days ago in terms of securing that supply."
— Rene Haas, CEO

Assessment: Management answered the supply question we set as a watch item in August, and answered it well. Purchase orders through calendar 2027 and signed supply agreements are a materially higher standard of evidence than the demand pipeline cited three months ago. The question's second half went partly unanswered: "multiple customers beyond Meta" is a count without names or a share, so the concentration of a $2 billion programme remains unquantified at the point where it starts generating revenue.

Whether Arm has the scale to secure supply as a merchant CPU vendor

The sharpest challenge of the call set Arm's supply position against a competitor that had raised its server CPU outlook by billions of dollars the previous evening, implying that incumbency in foundry and memory relationships is itself the barrier. Management did not dispute the premise.

Q: "Rene, on the AGI platform, let's assume we believe in the technical merits. My question is, does Arm have the scale in the ecosystem to secure supply? Because, for example, yesterday when AMD reported, they raised their server CPU business, right, by several billions of dollars. So, that tells me that they have the scale to go and get incremental supply, right? So, how are you dealing with that question about does Arm have enough scale to go and get the supply to be a meaningful player in the CPU markets, right, given your belief in the technical merits of the product?"
— Vivek Arya, Bank of America Securities

A: "The way I think about it is that the Arm AGI CPU is our first product. AMD's been at it for a long time. They've got a lot of SKUs. They buy a lot of capacity from TSMC. They have a lot of relationships that they've built over decades. This is our first product. That said, it's not our first relationship with folks like TSMC. It's not our first relationship with memory vendors such as Micron, Samsung, and SK... I think supply for the industry is going to be tight... It's a 5-10-year endeavor that's going to require relationships across all the supply chain, and we're very, very serious about making sure we get all that done."
— Rene Haas, CEO

Assessment: The candour is creditable and the content is a warning. "This is our first product" and "supply for the industry is going to be tight" are the two most important sentences on the silicon programme, and they sit against a revenue frame that just doubled. The $2 billion figure is capped by "up to" for a reason. An investor underwriting the silicon business should treat the supply commitment as the binding variable and the demand book as the comfortable one, which is the inverse of how the programme was framed at launch.

The royalty profile if handset units stay weak through next year

A direct question asked whether another year of double-digit smartphone unit declines would reproduce this year's royalty profile. The answer contained the call's only forward growth figure for the mobile royalty line, and it was narrower in scope than the frame it replaced.

Q: "Jason, let's say smartphone units are down again, double-digit next year. Do you think royalties will kind of be similarly profiled to what we saw this year, or are there other dynamics at play?"
— Vivek Arya, Bank of America Securities

A: "Regarding the mobile business, when we look at some of the external forecasts, I think IDC has units in mobile growing basically flat. I think it's minus 1% next year... There's already been kind of the hit and the reduction due to the memory constraints, and I think the expectations are those constraints stay relatively consistent, which basically means flat growth, not further deterioration. And then, because we continue to have more v9 adoption, we're now getting over 50%, so it's a little less of a tail end. But CSS adoption continues to grow... All that means that we will grow the overall business much faster than the market. Think of it at least being in the kind of double digits plus in growth rate for next year."
— Jason Child, CFO

Assessment: Two things in this answer matter more than the headline. First, the premise was rejected rather than answered: management's base case is flat units next year, not another double-digit decline, so the "double digits plus" growth figure is conditioned on a recovery that has not happened. Second, and more consequentially, the admission that Armv9 is "now getting over 50%" and therefore "a little less of a tail end" is an acknowledgement from management that the architecture-transition rate lever is maturing. That lever is Bull 1. Its replacement, compute subsystem adoption at higher rates, is real but smaller.

The gross margin of the silicon ramp as input costs rise

The final question of the call asked how Arm offsets the higher supply costs implied by securing more capacity, and whether pricing can absorb it. The answer is where the first-generation gross margin band slipped.

Q: "You talked about securing more supply on AGI, and I assume that comes at a higher cost. Can you just kind of talk us through the ability to offset that and how we should think about this, the gross margin puts and takes as you start to ship?"
— Joe Quatrochi, Wells Fargo

A: "Yeah, maybe just to reiterate what we said was for this first version, we're expecting to be somewhere in the high 30%s to the low 40% gross margin. For the first version, we expect it to be probably a little lower, then as we develop more expertise, probably rely on doing a little more work ourselves and maybe less from partners over time... Over time, we'll get ourselves to more of the market standard of 50%+, but I don't expect that to happen on the first version. That's probably going to happen more on the next version. So, for this version, I would stick to the kind of 30%-40% range, then we'll obviously give you updates on a quarterly basis as we start shipping in Q4."
— Jason Child, CFO

Assessment: The band moved twice inside one answer, from "high 30%s to the low 40%" to "probably a little lower" to "the kind of 30%-40% range". That is a downward revision delivered as a reiteration: the guidance was characterised as unchanged while the words describing it got softer. Against 97.9% for the intellectual property business, the difference between 40% and 35% on $2 billion of revenue is roughly $100 million of gross profit, which is material against our fiscal 2027 net income estimate.

Customer concentration inside the cloud royalty line

An analyst asked for a quantification of concentration in the fastest-growing royalty pool, offering a specific hypothesis to react to. Management answered with four names and no number, and the CFO's entire contribution to the exchange was one word.

Q: "I wanted to ask about basically the cloud and AI group of products and how you might frame the customer concentration there to help us give us a better appreciation. For example, could we guess that maybe 60% of these royalties are coming from three customers? Any kind of quantification would be helpful to understand that."
— Simon Leopold, Raymond James

A: "I think we can say, and Jason, chime in on this thing, Amazon, Google, and Microsoft are the largest contributors, and then you've now got NVIDIA ramping up quickly. They are the four guys inside the data center that command the majority of the royalties."
— Rene Haas, CEO

Assessment: The question asked for a percentage and offered one as a prompt. The answer named four companies and used the word "majority", which is the least precise quantifier available. It compounds the absence of any dollar sizing of the data centre royalty base. The practical consequence for a model is that the fastest-growing, highest-rate royalty pool in the company cannot be stress-tested against the loss or insourcing decision of any single customer.

How the rack benchmark stands against the newest competing part

A question noted that the competitor whose part Arm benchmarked against had claimed best-in-class performance per watt for its newest generation the previous evening, and asked where Arm actually wins. The answer contained a material qualification of the benchmark Arm had promoted in the shareholder letter and the prepared remarks.

Q: "At OCP, you introduced agentic AI benchmarks that showed your Arm AGI CPU can deliver twice the performance per rack versus a competing x86 solution. Last night on its earnings call, AMD said that its Venice CPU offers the best performance per watt. Maybe, Rene, could you talk about where you see the most meaningful opportunity for the Arm CPU to win in the data center versus x86?"
— Sebastien Naji, William Blair

A: "The Arm AGI CPU benchmark that we ran, that was done through Signal65 at OCP, was looking at the previous generation to Venice, and we are very confident in those numbers, and we're very confident in terms of the choices that customers make vis-à-vis head-to-head against that design. Venice is a great product, but we're very confident in terms of the demand we have for our product, Arm AGI CPU, and the roadmap going forward."
— Rene Haas, CEO

Assessment: The two-times-performance-per-rack claim appears once in the shareholder letter and once in prepared remarks. Both name the comparator part, the AMD EPYC 9755, but neither says that part is a generation behind AMD's newest. That label only emerged because an analyst had read a competitor's call. That does not make the claim wrong, and the custom-silicon argument that followed it is the genuinely durable competitive point, since hyperscalers designing their own system-on-chip can make trade-offs a merchant part cannot. But leading with a prior-generation comparison without flagging it as one is a presentation choice, and the power-shortfall framing built on top of it should be discounted accordingly.

The revenue ladder beyond fiscal 2028 and the fiscal 2031 frame

An analyst asked what drives the fiscal 2031 silicon revenue figure, core count or average selling price, and in answering, management laid out a multi-year revenue ladder it has not previously articulated while declining to attach years to it or to restate the 2031 number.

Q: "Along the same path on FY 2031, $15 billion AGI CPU revenues, which is the biggest factor? Is it core count increase or is it ASP increasing from the current version 1?"
— Krish Sankar, TD Cowen

A: "Yeah. No, we said back at the Arm Everywhere event last March that we had the line of sight to $1 billion for this kind of end of 2027 and FY 2028. Of course, now we are increasing our confidence to deliver $2 billion. The following year, we have not made any updates, but I think the original plan was that we were going to get to about $2 billion year after that, and then we were going to go up to about $5 billion year after that. To the extent that we are successful in delivering over $2 billion, you should expect that we will probably have some upsides after that. That is something that we will plan on providing in terms of updated possible targets when we have the next Arm Everywhere event."
— Jason Child, CFO

Assessment: As in July, the $15 billion fiscal 2031 figure was raised by an analyst rather than by management, and management did not restate it. What replaced it this time is more useful and less committing: a $2 billion, then $2 billion, then $5 billion ladder with no years attached and no path to $15 billion. The CEO separately endorsed a CPU market of "$200 billion or more" by 2030 against Arm's own published $100 billion, which widens the gap between the market opportunity management will describe and the company revenue it will commit to. The next event, guided to spring or the first half of next year, is where that gap either closes or becomes a permanent feature.

What They're NOT Saying

1. Any fiscal 2027 royalty growth frame: The high-teens figure given in July is unreachable on the guided path, and no replacement was offered in prepared remarks or in Q&A. For a company that guides one quarter forward, abandoning a full-year frame for its most important line without restating it leaves the line unmodellable beyond December.
2. The fiscal 2028 and 2029 royalty claim: Described in July as "very much still intact" at 20% or better. Not mentioned on this call at all, in a quarter when royalty growth fell to 8%.
3. Data centre royalty revenue in dollars: Five consecutive quarters of "doubled" with no dollar base ever published. Asked this quarter whether data centre is approaching a third of total royalties, the CFO offered only a rough share: "approaching 20%-ish" last year and not "quite to a third" yet.
4. Cloud royalty customer concentration: Asked directly, with a specific percentage offered as a prompt. Answered with four company names and the word "majority", and no figure.
5. The $249 million acquisition: Paid in cash this quarter, generating $92 million of goodwill, $155 million of intangibles and $12 million of acquisition expenses. The target is not named anywhere in the shareholder letter and the transaction was not mentioned on the call.
6. Any term of the SoftBank agreement, five quarters into our coverage: No duration, no scope, no pricing basis, no renewal mechanics, no recognition pattern, on roughly $820 million of annualised revenue from the controlling shareholder. This has been an explicit upgrade trigger since initiation.
7. Remaining performance obligations: Not reinstated, not mentioned, and not asked about, one quarter after withdrawal. Annualised contract value is now the sole contracted-revenue disclosure and it decelerated again.
8. The second-generation Arm AGI CPU schedule: Declined. The analyst offered a specific tape-out window; the CEO replied that the market runs on an 18-to-24-month production cadence and said "you can kind of do the math", which is a cadence, not a date.
9. Where Arm AGI CPU revenue is reported before fiscal 2028: Still unresolved. Management confirmed shipments start in the March quarter and that there will be nothing material in December, but not which line the revenue lands in. The same quarter carries a guided license step-up, so the two will be indistinguishable.
10. The $15 billion fiscal 2031 figure: Raised by an analyst, as in July, and not restated by management. The ladder offered instead carries no years and stops at $5 billion.
11. Any capital expenditure guidance: Still none, now in the quarter before silicon shipments begin. Capital intensity happened to fall this quarter, which is favourable, but it remains unguided and unexplained either way.

Market Reaction

• Pre-print setup: Shares closed at $191.80 on November 4, entering the print up 75.5% year to date and 19.3% over trailing twelve months, but down 36.7% over the trailing 30 days from $302.90 and 56.4% below the record close of $439.46 set on June 18. The 52-week closing range was $104.55 to $439.46. The options market implied a move of roughly 8.9%.
• After-hours November 4: Down approximately 1.8% after the release and the call. That is the market's verdict on the print itself, and it was close to neutral.
• Reaction session November 5: Opened at $172.05, a 10.3% gap down, traded between $163.13 and $175.83, and closed at $166.72, down 13.1% or $25.08. Volume of 15.2 million shares ran 2.9 times the 30-day average of 5.3 million.
• Market and peer context: The S&P 500 fell 4.3% in the same session and the Nasdaq Composite fell 5.5% into correction territory, with all eleven S&P sectors lower. AI-complex peers including NVIDIA, AMD and CoreWeave fell sharply in the same session, and ARM's 13.1% decline was broadly in line with that group.
• Since initiation: ARM has returned 5.4% from the $158.25 initiation reaction close of November 6, 2025, against 9.5% for the S&P 500 over the same period, so the shares now trail the index since coverage began. Measured instead from our February 5, 2026 note, the baseline our August report used, ARM is up 50.4% against the index's 8.2%.

The after-hours move, small as it was, is consistent with what changed. A double beat that still traded lower says the market was positioned for the recurring line and got the lumpy one. The December royalty guide of low single digits is the specific sentence that did the damage, because it converts a single soft quarter into a two-quarter trend and removes the possibility that the shortfall was a timing artifact. The 10.3% opening gap the next morning belongs to the selloff rather than to the print. Sell-side price target revisions published during the session cut numbers without changing stances, which is the signature of an estimate problem rather than a thesis problem.

The pre-print setup amplified rather than cushioned the move, which is the reverse of the dynamic we described in August. Three months ago a stock that had halved from its June peak carried a mechanical reflex to buy a beat, and it gapped up 14.7% before fading. This time a further 36.7% decline over 30 days had not restored any willingness to absorb bad news on the royalty line, because the thirty-day drawdown was itself the AI complex de-rating that continued into the reaction session. Positioning in this name remains a bet on sector flows at least as much as on the franchise, and this session was the clearest demonstration of that yet.

The one durable consequence is for us rather than for the company. Our Hold has been wrong on price for three consecutive quarters: at the August report, measured from our February note, the shares were up 151% against 15% for the S&P 500. As of this session it is no longer wrong on price measured from where coverage actually began: the shares have returned 5.4% against the index's 9.5% since the November 2025 initiation. We note that without satisfaction, because the de-rating was delivered by a market-wide repricing of AI infrastructure rather than by the disclosure and valuation arguments we actually made.

Street Perspective

Debate: Is the royalty deceleration cyclical or structural?

Bull view: The bull case on the Street is that this is entirely cyclical and already dated. Memory pricing compressed mid and lower-tier handset units, that headwind laps in January and February, the December comparison is the hardest of the year at 27%, and the two structurally growing pools are compounding fast: data centre doubling for a fifth consecutive quarter and physical AI at 50%. On this reading the December quarter is the trough, the March quarter reaccelerates on the edge recovery management has guided to, and a strong license book converts to royalties through fiscal 2028 and 2029.

Bear view: The bear camp argues that the cyclical story is doing too much work. Edge AI is the majority of the royalty base, Armv9 penetration has passed 50% so the architecture upgrade lever is maturing by management's own admission, and the full-year royalty frame has been cut once and then left unrestated in two quarters. A line that grew 27%, then 11%, then 22%, then 8%, and is guided to low single digits, is not obviously an annuity at all. The bears also note that smartphone unit forecasts for next year are flat at best, so the recovery is a comparison effect rather than demand growth.

Our take: The bulls have the better of the near-term mechanics and the bears have the better of the multiple. Most of this deceleration is identifiable, time-boxed and attributable to memory pricing, and we expect reacceleration. What we are not willing to do is treat the pre-deceleration growth rate as the normal state, because the rate lever that produced it is maturing and management, having cut its full-year number once, has now declined to recommit to one. We model fiscal 2027 royalty growth near 12% and fiscal 2028 in the mid teens, below both the abandoned frame and the bull case.

Debate: Does a 50% license quarter deserve a royalty multiple?

Bull view: The optimistic framing is that the distinction is artificial. Total Access and compute subsystem licensing is subscription revenue on a multi-year term, architecture licenses have all but disappeared from the mix, and every license signed is a design commitment that becomes royalty revenue in 12 to 24 months. On that logic a 50% license quarter is simply the royalty line arriving early, deserves the same multiple, and the pipeline that management calls surprisingly strong is the most valuable disclosure of the quarter.

Bear view: The skeptical view is that the metric Arm itself publishes to normalise this line grew 12%, not 50%, and that recognised license revenue now runs at 1.49 times annualised contract value against 1.26 times a year ago. The line is also 27% related-party, management explicitly declined to call the level structural, and the deals were described as incremental and hard to forecast. A revenue stream whose own company will not forecast it beyond two quarters does not earn an annuity multiple.

Our take: We are closer to the bears, with a concession. The shift from architecture licenses to subscriptions and compute subsystems is a genuine quality improvement and we are raising our fiscal 2027 license estimate materially because of it. But the widening gap between recognised license revenue and the committed-fee base is the kind of divergence that resolves one of two ways, and Arm withdrew the disclosure that would have let us tell which. Until annualised contract value growth reaccelerates toward the recognised line, or remaining performance obligations come back, we will value the license line at a discount to the royalty line.

Debate: Is roughly 79 times forward earnings now the entry point?

Bull view: The bull argument is that the de-rating has done the work the fundamentals did not need to. The multiple has fallen from about 139 times to about 79 times forward earnings in one quarter while the earnings estimate went up, the silicon revenue frame doubled, hyperscaler share tripled and the related-party quality concern resolved. A company compounding revenue above 20% with a 44% operating margin, a doubling silicon business and a 60% share of top-hyperscaler servers at 79 times is, on this view, the cheapest Arm has been relative to its own trajectory since it listed.

Bear view: The bear camp counters that 79 times forward earnings and a 0.61% free cash flow yield is not cheap in absolute terms, that the de-rating was delivered by a sector-wide repricing rather than by anything improving, and that three of the five things we cannot see are unchanged or worse. Free cash flow conversion collapsed to 19%, trailing free cash flow growth fell to 2%, the contract-asset build returned, a backlog metric is still missing, and the annuity line is guided to low single digits. Paying 79 times for a business whose recurring line is guided to low single digits and whose growth line will not be forecast is paying a growth multiple for a transition.

Our take: We published an explicit revisit trigger in August at "nearer 90 times forward earnings". It has fired, with room to spare, and we have looked. What we found is that the multiple roughly halved and so did the growth rate of the asset the multiple was paying for, which is close to a wash. The honest statement is that ARM is the least expensive it has been on our framework since our February note, and that we need one quarter of evidence on the royalty line rather than one quarter of price action to move. That is a Hold at higher conviction, not an upgrade.

Model Update Needed

Valuation impact: We continue to publish no price target on ARM, for the reason given at initiation and unchanged since: the range of outcomes turns on disclosure that does not exist. At the November 5 close of $166.72 the shares carry a market capitalisation near $179.9 billion against $3,621 million of cash and short-term investments, for an enterprise value near $176.3 billion. That is roughly 79 times our fiscal 2027 non-GAAP EPS estimate of $2.10, about 29.7 times our fiscal 2027 revenue estimate, and a 0.61% trailing free cash flow yield. The comparable figures at our August note were 139 times, 50 times and 0.47%, and at initiation 93 times, 34 times and 0.63%. The multiple has returned to roughly where coverage began while our earnings estimate has risen 24% over the same period.

Thesis Scorecard Post-Earnings

Overall: The thesis did not weaken or strengthen so much as change composition, and in a direction that makes it harder to underwrite. On the positive side, the two bear points that drove our initiation Hold on quality of earnings continued to resolve: operating expense has come in under guide in all five quarters of our coverage, and the related-party growth concern inverted outright, with every ex-agreement growth measure now better than reported. On the negative side, the pillar everything rests on moved to at risk for the first time, and a bear point we declared contained in August reverted within a quarter, which is as much a comment on our process as on the company. Net of all of it, the business is better run and more concentrated in its two fastest-growing pools than we underwrote, while the recurring line that justified the multiple is growing at a third of its recent rate and will not be forecast beyond December.

Action: Hold, maintained, conviction raised from 5 to 6. We published three revisit triggers in August. The first, disclosure of the SoftBank agreement's terms, did not arrive for a fifth quarter. The second, reinstatement of a contracted-revenue metric, did not arrive and the remaining metric decelerated again. The third, a de-rating toward 90 times forward earnings, fired: the shares closed at roughly 79 times our raised fiscal 2027 estimate, and they now trail the index since coverage began. We have revisited, as we said we would, and we are not upgrading on price alone, because the asset the price was paying for weakened by about as much as the price fell. For existing holders the operating trajectory, the expense discipline and the silicon order book all justify staying. For new capital, the next upgrade is contingent on one specific, gradeable event rather than three: evidence in the March quarter that edge royalty revenue reaccelerates as management has now guided twice, which would re-establish Bull 1 and put the royalty line back on a path we can model. Absent that, 79 times forward earnings for a business whose recurring line is guided to low single digits is not an entry we will take, and we would reconsider the rating in the other direction if the March quarter misses on royalty again.