Arm Delivers the Operating Leverage It Promised and Misses on the Line That Justifies the Multiple

Results vs. Consensus

Arm reported its fiscal fourth quarter and full fiscal year, both ended March 31, 2026, after the close on May 6. It was a beat on both headline lines and a beat against the company's own guide on all three guided metrics. It was also, underneath, a quarter in which the composition of the beat inverted relative to what anyone expected going in.

Q4 FYE26 Scorecard

Year-over-Year Comparison

Sequential Comparison

The sequential table is where the shape of the quarter is legible. Licensing added $314 million and royalties gave back $66 million. Operating expense grew 2.5% against revenue up 20.0%, the first quarter of the fiscal year in which revenue meaningfully outgrew cost, and the operating margin snapped back 840 basis points as a result. Capital expenditure fell 58.7%. Free cash flow still went down.

Full Year FYE26

Q1 through Q4 revenue of $1,053M, $1,135M, $1,242M and $1,490M sums to $4,920 million, which is the reported full-year figure. Every line above traces to the income statement in the Form 6-K rather than to figures quoted verbally on the call.

Revenue

Total revenue of $1,490 million was a record and grew 20.1%, capping a fiscal year at $4,920 million and a third consecutive year above 20% growth. The two lines inside it moved in opposite directions and for reasons that matter more than the total.

The composition question we raised at initiation has both improved and hardened. It improved because the SoftBank contribution stayed flat at $200 million while the licence line grew 62% sequentially, so the parent's share of the licence line fell from 40% to 24% and its share of total revenue fell from 16% to 13.4%. It hardened because the reported revenue-from-external-customers line, which strips every related party including Arm China, grew 5.9% in the quarter and 7.4% for the fiscal year. Related parties supplied 30.5% of FY26 revenue against 20.5% in FY25. That line requires no estimate and no assumption: it is printed on the face of the income statement, and it is the cleanest available measure of how fast Arm is growing with people who do not sit on its share register.

Assessment: Two revenue lines with different quality are still being reported as one growth rate, and this quarter a third framing became available. On reported revenue Arm grew 20%. On revenue from third parties it grew 6%. Neither number is wrong and neither is sufficient alone, but a market paying 45.6 times enterprise value to revenue is paying the first number.

Margins

The margin story reversed this quarter and it is the single most important thing that happened. Non-GAAP operating expense of $734 million came in $11 million below the ~$745 million guide, following a $4 million underrun in Q3. Sequentially it grew 2.5% against revenue up 20.0%, a 17.5-point gap. Operating margin also expanded sequentially in Q2, by 200 basis points on a 3.1-point gap, so this is not the first quarter of leverage. It is the first on a scale that changes the full-year arithmetic, and the 49.1% operating margin is the highest of FYE26 by eight points.

The full-year picture is still one of investment outrunning revenue: non-GAAP opex grew 32.6% against revenue up 22.8%, and non-GAAP R&D grew 42.6%. Note also that engineer headcount grew 16.1% to 8,058 while non-GAAP R&D grew 42.6%, so cost per engineer is rising far faster than the engineer count. That is consistent with silicon design work, which carries tooling, mask and EDA costs that headcount does not capture.

Assessment: The FY26 numbers show the investment cycle at its peak and the Q4 numbers show it cresting. What converts this from a data point into a thesis change is the Q1 guide, which asks for only 3.5% sequential opex growth against the low-double-digit step-up management had committed to in February. Management chose to under-deliver on its own spending plan, and said so explicitly. That is the strongest signal in the release.

Earnings and Cash

Cash is where the quarter is weakest, and the weakness is now three quarters old. Operating cash flow of $260 million was flat year over year against non-GAAP net income up 9.8%. Free cash flow of $152 million fell 6.7% and converted at 23.7% of non-GAAP net income, against 37.0% in Q3 and 27.9% in the year-ago quarter. It is the lowest conversion rate of the fiscal year.

Two working-capital items account for it. Accounts receivable consumed $324 million, and contract assets consumed a further $171 million. Contract assets from related parties finished the year at $646 million against $152 million twelve months earlier, a $494 million build, of which $147 million arrived in this quarter alone. Receivables from related parties, by contrast, fell to $270 million from $279 million, so the third-party receivable build is genuine collection timing rather than parent-company exposure.

The full-year figure looks transformational and mostly is not. FY26 free cash flow of $882 million against $99 million is a 791% increase, but the FY25 base was distorted by the reversal of an IPO share-vesting payroll-tax benefit that management itself called out at the time. On conversion, FY26 ran at 46.7% of non-GAAP net income. That is a normal number for a company investing hard, and it is well below where a 98%-gross-margin licensing business should sit.

Assessment: Capital intensity got better this quarter and cash conversion got worse. Capital expenditure of $74 million was flat year over year and down 58.7% sequentially, which materially softens the "capex is running away" concern we raised in February. But the contract-asset build kept going, and $646 million of unbilled revenue owed by the controlling shareholder is now sitting on the balance sheet against a stream whose contract terms have never been disclosed. Trailing free cash flow of $882 million against a $227.8 billion market capitalisation is a 0.39% yield.

Revenue Line Performance

Arm does not report profit by segment. It reports two revenue lines and describes its end markets qualitatively under three AI headings introduced last quarter: Edge AI, Physical AI and Cloud AI. It attaches dollars to none of them. What follows uses the disclosed revenue lines, with end-market colour from the shareholder letter and the call.

Royalty

Royalties of $671 million grew 10.5%, a record for a fiscal fourth quarter and the weakest growth rate of the fiscal year: the four quarters ran +25%, +21%, +27% and +11% as the company reports them. Sequentially the line fell $66 million from Q3's record $737 million. Management attributed the deceleration to a comparison effect rather than a demand change, isolating a prior-year product ramp that landed unusually early.

"Royalty revenue grew 11% year-on-year to $671 million, our highest ever figure for Q4 royalty revenue. The biggest contribution to royalty revenue growth was from Cloud AI. Data center royalty revenue continues to more than double year-on-year, and we see no break in this momentum."
— Jason Child, CFO

The underlying mechanism is unchanged and remains the strongest part of the business. Arm is being paid more per chip rather than for more chips, through Armv9 penetration and compute subsystem adoption into higher-end devices. The shareholder letter states that Arm-based CPU compute is now "about 50% share among top hyperscalers", and management said data center networking silicon, specifically DPUs and SmartNICs, runs at close to 100% Arm share. The claim that carries the most forward weight came in Q&A: Neoverse-based royalties doubled year over year and management expects them to double again in FY27.

Assessment: An 11% royalty quarter with a credible comparison explanation is not a broken thesis, and the full year at +20.5% cleared the "north of 20%" commitment management made in February. But it cleared it by half a point, it came in below the "low teens" management guided for the quarter, and it missed the Street. Royalties are the line that earns the premium multiple, and this is the first quarter of our coverage in which they did not carry the print. The FY27 guide asks for a return to roughly 20% immediately, in Q1. That is now the gradeable number.

License and other

Licence and other revenue of $819 million grew 29.2% year over year and 62.2% sequentially, and was the reason the quarter beat. At initiation we calculated that the Q4 guide required licensing near $733 million if royalties held flat. Royalties did not hold flat, they fell $66 million, and licensing still delivered $86 million above that requirement. Total Access licences rose by six in the quarter to 56, the largest quarterly addition of the fiscal year and up from 44 a year ago; Flexible Access customers rose by eleven to 329.

"Of the $819 million of license revenue, our agreement with SoftBank for technology licensing and design services contributed $200 million, flat with the prior quarter. As always, licensing revenue varies quarter-to-quarter due to timing and size of high-value deals."
— Jason Child, CFO

Annualised contract value, the metric management prefers as the underlying licensing indicator, grew 21.6% to $1,660 million. That is a deceleration from three consecutive quarters at 28% and only 2.5% sequential growth. Remaining performance obligations fell again, to $2,071 million, down 3.6% sequentially and 7% year over year. For the first time the letter offered an explanation, attributing the RPO decline to "improvements in the timing of revenue conversion".

Assessment: This is the best licence quarter Arm has reported and the composition is genuinely better than Q3's. Six new Total Access licences and two next-generation CSS agreements are third-party demand, not parent-company demand, and they are what a 62% sequential move should be made of. The offset is that both contracted-revenue metrics moved the wrong way in the same quarter. ACV decelerating six points while RPO declines for a second consecutive quarter is harder to dismiss as recognition timing than either fact alone.

Key Operating Metrics

Assessment: The licence-count metrics are unambiguously good and are the best third-party demand evidence in the release. Six Total Access additions in a quarter, against two in each of the prior two quarters, is the kind of number that shows up in royalties three years out. The contracted-revenue metrics are the counterweight, and this is the quarter in which they stopped disagreeing with each other in a comfortable direction: ACV came down toward RPO rather than RPO coming up toward ACV.

Key Topics & Management Commentary

Overall Management Tone: Confident and, for the first time in our coverage, willing to be pinned down on numbers that constrain rather than flatter. Management volunteered the opex walk-down, quantified the chip business's cost structure and long-run margin, and restated a five-year revenue and earnings frame on an earnings call rather than reserving it for an investor day. Where the posture thinned was on the royalty deceleration, which was explained by a comparison effect and then moved past quickly, and on the supply constraint, which was described as a work in progress without a date.

1. The royalty engine slowed to its weakest growth of the fiscal year

Royalty growth of 10.5% is the number the market cared about and the one that got the least airtime. It follows +25%, +21% and +27% in the preceding three quarters and it undershot the "low teens" management guided for the quarter three months earlier. Against the Street's $694.93 million it missed by 3.4%. Management's explanation is the same one it gave in advance: a prior-year smartphone chip ramp landed unusually early, creating a hard comparison.

"As indicated by our guidance, we're expecting that to get back to the kind of 20% range by Q1."
— Jason Child, CFO

The explanation is checkable and it is consistent with what was said in February. It is also now a promise with a two-month fuse, because the Q1 FYE27 guide assumes royalties immediately recover to roughly 20% growth from 10.5%.

Assessment: One soft quarter with a pre-announced comparison effect is not a thesis break, and the full year still cleared 20%. But the rate escalator was our first bull pillar precisely because it was supposed to be insensitive to unit cycles, and this quarter it decelerated by 16 points while management described the mobile market as flat to slightly negative. The recovery to 20% in Q1 is the cleanest test the thesis has faced.

2. Operating leverage arrived a quarter earlier than committed

In February management told analysts to expect a low-double-digit sequential increase in operating expense from Q4 into Q1, mirroring the prior year, with moderation only afterwards. The Q1 FYE27 guide of approximately $760 million against $734 million delivered is a 3.5% step. Management addressed the change directly and unprompted, then extended it across the year.

"I know we initially had thought maybe the OpEx would step up a little more into -- from Q4 into Q1 as we just guided to, we're expecting that to be a little less than we thought before. ... we will be growing expenses less than revenue by the time we end the year"
— Jason Child, CFO

Applied to the guided base, a "few percent" sequential path from $760 million lands FY27 operating expense near $3.2 billion, roughly 18% growth against FY26's 32.6%. That is a halving of the growth rate, and it is the first time management has framed the spend as bounded rather than as an open-ended function of opportunity.

Assessment: This is the most consequential disclosure of the quarter for the investment case, and it is the direct answer to the bear point we opened at initiation. Two consecutive guide underruns plus a walked-down forward step is a pattern rather than a data point. The caveat is that Q1 FYE27's implied operating margin of roughly 38.0% is still below the 39.1% Arm posted in Q1 FYE26, and it is guided on roughly 20% more revenue, so the leverage is a second-half story even on management's own numbers.

3. Arm AGI CPU: demand doubled in six weeks, supply did not

The launch product from the March 24 event went from $1 billion of stated demand to more than $2 billion across FY27 and FY28 in six weeks. Arm did not raise its revenue outlook, because it cannot yet build the second billion.

"We now have line of sight to more than $2 billion of demand across fiscal '27 and '28. However, we are maintaining our outlook of $1 billion while we pursue supply chain capacity, and we still expect the first revenues from production chip sales to land in the fourth quarter of this fiscal year."
— Jason Child, CFO

The CEO described what the constraint actually consists of, which is more useful than the headline: "the number that we talked about at end of March was supply in place to support $1 billion of demand, and that includes memory, that includes wafers, that includes packaging, that includes access to test equipment. So for the $2 billion, we are now in the process of securing supply to support that." Named demand includes SAP moving core database and business application workloads to Arm, Cloudflare deploying across its global network, and design wins at F5 and SK Telecom, with Cerebras, OpenAI, Positron, Rebellions and Verda on the AGI CPU specifically. Racks are orderable from Supermicro, Lenovo, Quanta and ASRock.

Assessment: Doubling demand for an unlaunched product in six weeks is a genuine positive and the customer list is not speculative. It is also the reason the shares fell. A company whose entire historical appeal was capital-light royalty capture has just discovered that its new business is gated by memory, wafers, packaging and test capacity, which is a set of constraints Arm has never had to manage and does not control. The revenue outlook being unchanged at $1 billion while demand doubles is honest and it is also a demonstration of exactly how the new model differs from the old one.

4. The economics of selling silicon, finally quantified

The chiplet and complete-SoC ambition that was named once in the Q3 call and never revisited is now a shipping product line with a cost structure. An analyst put the first-generation gross margin at "about 30% plus" citing the March event, and management did not correct it. Management then gave the operating-expense side and the long-run target.

"It's a team that's in the -- probably in the dozens of people, not hundreds. And so you can assume that it's operating profit positive next year. ... I think we said the IP business probably gets to something like a 65% operating margin business or EBITDA margin business and that the chip business is probably in the 35-ish percent range."
— Jason Child, CFO

The reason the incremental cost is low is structural and credible: the most expensive part of designing the chip is the compute die, which is effectively the compute subsystem Arm already builds and sells as IP. Arm is monetising an existing engineering asset a second way rather than standing up a parallel design organisation.

Assessment: The disclosure is good and the argument is sound, and it still dilutes. A business at roughly 30% gross margin and 35% long-run operating margin, bolted onto one at 98.3% and 65%, lowers the blended multiple the whole company deserves as it scales. Management's own FYE31 frame has chips at $15 billion and IP at $10 billion, which means the majority of the revenue five years out carries the lower structure. At initiation we said we would want to see the SoC ambition clarified without a material rise in capital intensity. Capital intensity behaved this quarter. The margin structure is the part that did not.

5. Data center: 50% hyperscaler share and a second doubling promised

Data center royalty revenue more than doubled year over year for a third consecutive quarter and was the largest single contributor to royalty growth. The shareholder letter puts Arm's share of CPU compute at about 50% among top hyperscalers. Asked directly whether the pace continues into FY27, the CEO committed.

"The royalties associated with customers building chips based on Neoverse, those royalties have doubled year-on-year. And I'm looking at Jason here, I expect they're going to double year-on-year again in this year."
— Rene Haas, CEO

The supporting evidence is again shipping silicon rather than design wins. Google announced TPU8t and TPU8i, both replacing x86 host processors with Arm-based Axion CPUs; NVIDIA announced Vera and a standalone rack integrating 256 of them; Microsoft is expanding Cobalt supply across Azure regions; AWS disclosed that its custom silicon business including Graviton is running above $20 billion annually and growing triple digits.

Assessment: This remains the most credible part of the bull case and it is corroborated by third-party product announcements rather than by Arm's own adjectives. It also remains unsized. Arm discloses data center revenue once a year, and the FY26 disclosure did not appear in this letter. A business described as about to become the company's largest, doubling annually, and driving the whole royalty line is being underwritten on a growth adjective.

6. The SoftBank agreement held at exactly $200 million and disclosed nothing further

The related-party stream came in at precisely the run rate the CFO named in February, flat sequentially, and no further detail was given. No contract term, no total value, no recognition pattern, no renewal mechanics. Three quarters into the agreement, an outsider still cannot test the durability claim management made at initiation.

What the balance sheet shows in the meantime is that the revenue is being recognised well ahead of billing. Contract assets from related parties finished the year at $646 million against $152 million twelve months earlier, a build of $494 million against roughly $578 million of contribution management has quantified across three quarters.

Assessment: The concentration ratio improved this quarter for the right reason, which is that third-party licensing grew rather than that the parent shrank. At 13.4% of revenue the stream is less dominant than the 16% we flagged in February. But the disclosure gap is now three quarters old, it was our first named upgrade trigger, and it did not close. That most of the associated revenue is still sitting on the balance sheet as an unbilled receivable from the controlling shareholder makes the absent contract terms more material, not less.

7. The FYE31 frame: $25 billion of revenue and more than $9 of EPS

Management closed the prepared remarks by restating the March event's five-year target on an earnings call, and attached an earnings number to it for the first time in this forum.

"For those of you that missed our March Arm Everywhere AGI CPU event, we expect that by FYE31, we will be generating $15 billion in AGI CPU revenue and $10 billion in IP revenue for a total of $25 billion. We expect this to translate to more than $9 in EPS."
— Jason Child, CFO

Against the May 7 close of $213.31, more than $9 of FYE31 earnings is 23.7 times a terminal year five years out. Against FY26 revenue of $4.92 billion, $25 billion implies a 38% compound growth rate for five years, with the majority of the increment coming from a business that has not shipped a unit and is currently supply-capped at $1 billion.

Assessment: Publishing a five-year EPS number is a real commitment and management deserves credit for making it on the record. It is also the frame the market is now pricing, and it does most of the work in justifying the current multiple. The useful discipline is to ask what happens if the chip business lands at half the target: $7.5 billion of chips plus $10 billion of IP is $17.5 billion, and the earnings number falls further than proportionally because the IP half carries the higher margin. The frame is an argument for owning the stock at some price. It is not, by itself, an argument for this one.

8. RPO fell again, and ACV growth came down to meet it

Remaining performance obligations of $2,071 million declined 3.6% sequentially and 7% year over year, a second consecutive sequential fall. In February we said that another decline while ACV kept compounding at 28% would stop being explicable as recognition timing. What happened instead is that ACV growth came down six points toward RPO, decelerating to 21.6% from three straight quarters the company reported at 28%.

The letter offered an explanation for the first time, attributing the RPO decline to "improvements in the timing of revenue conversion", which is to say Arm is recognising contracted revenue faster than it is signing replacement backlog. That is a coherent story and it is exactly what a large, front-loaded related-party recognition would do to the ratio.

Assessment: A written explanation is better than the silence we got in February. But the two metrics converging downward is a worse signal than one metric diverging upward, because the convergence removes the interpretive escape route. Arm's forward-visibility disclosure now consists of one decelerating metric and one declining one, and no analyst asked about either on this call.

9. Smartphone units turned negative and royalties grew anyway

The memory-shortage thesis that dominated Q3's Q&A arrived in the numbers this quarter. Management said mobile unit growth turned negative and expects the market to stay flat to slightly negative, while maintaining that Arm's exposure is concentrated where the damage is not.

"We're going to continue to see very flattish, maybe slightly negative numbers for the overall market. The lower end of the market is probably going to be where most of that impact is. So it doesn't have too much impact on us."
— Jason Child, CFO

The prepared remarks were consistent, describing smartphone revenues as continuing to deliver growth despite end-market weakness, driven by Armv9 and compute subsystem penetration into higher-end devices. What did not appear was any restatement of the numeric sensitivity, which management gave two different versions of on the Q3 call and has still not reconciled.

Assessment: This is the rate-over-units argument being tested in real conditions for the first time, and it passed on direction: units fell and smartphone royalty revenue grew. The test is incomplete because the quarter's overall royalty growth was still the weakest of the year, so we cannot yet separate how much of the deceleration was the MediaTek comparison and how much was the handset market. Q1 will separate them, because the comparison effect does not repeat and the memory constraint does.

10. Cash conversion was the worst quarter of the fiscal year

Free cash flow of $152 million against $641 million of non-GAAP net income is 23.7% conversion, below Q3's 37.0% and below the year-ago quarter's 27.9%. Neither the prepared remarks nor any question addressed it. The full-year headline of $882 million against $99 million is arithmetically true and analytically misleading, because the FY25 comparison base was suppressed by the reversal of an IPO-related payroll-tax benefit.

The offsetting good news is on the investing line. Purchases of property and equipment were $74 million, flat with the year-ago quarter and down from $179 million in Q3. The FY26 total of $545 million against $219 million is still a 149% increase, but the run rate cooled sharply in the quarter that Arm launched its silicon product, which is not the sequence a capital-intensity bear would predict.

Assessment: Capital intensity looked like the emerging risk in February and it behaved this quarter. The contract-asset build did not. A company recognising revenue $494 million ahead of billing against its controlling shareholder, while reporting its lowest quarterly cash conversion of the year, has a working-capital question rather than a cash-generation question. Those become the same question if the billing does not follow.

Guidance & Outlook

The Q1 guide of $1,260 million is 19.7% growth against the $1,053 million Arm reported in Q1 FYE26, and it landed ahead of where the Street sat. It is also a 15.4% sequential decline, which is normal for Arm's fiscal first quarter and was not the source of the market's disappointment. Management guided both revenue lines to roughly 20% year-over-year growth in the quarter, which for royalties means an immediate recovery from 10.5%.

The full-year framing came in Q&A rather than in the letter. Royalty growth was put at "roughly 20-ish percent for the year" with each quarter "relatively close to the 20s percent range". Licensing was framed the same way for FY27, with a second-half weighting of roughly 60/40, and a long-term expectation well below that: "the long-term target is probably in the high single-digit, low double-digit range", with at least 10% described as the floor. Applying the guided opex path to the guided revenue path produces FY27 operating expense near $3.2 billion, growth of roughly 18% against revenue growth of roughly 19%.

Implied quarter-over-quarter ramp: at $1,260 million in Q1 and roughly 19% full-year growth, the remaining three quarters need to average about $1,530 million, exiting the year above the $1,490 million just delivered. With licensing weighted 60/40 to the second half, most of that step sits in Q3 and Q4.

Street at: the Q1 revenue guide came in above consensus, and at least 14 brokerages raised price targets following the print. Consensus entering the quarter had royalties at $694.93 million and licensing at $769.66 million, so the Street's FY27 mix assumption was too royalty-heavy and will need to shift.

Guidance style: conservative and, this quarter, deliberately so on cost. Arm has now beaten its own revenue midpoint and come in below its own opex guide for two consecutive quarters, and management pre-emptively lowered the sequential opex step it had signalled three months earlier. The pattern across our coverage is a company that guides revenue tightly and spends less than it says it will.

Analyst Q&A Highlights

Where the incremental AGI CPU demand came from, and whether supply can serve it

The opening question was the one the whole call turned on: demand for an unshipped product doubled in six weeks, and the revenue outlook did not move. Management split the answer, describing the demand as a mix of existing launch customers raising forecasts and new customers arriving, then handing the supply half to the CFO, who declined to change the number.

Q: "You're now saying that you've got over $2 billion of demand for the next 2 fiscal years. I was just hoping you could give us a bit more detail on how that additional demand has transpired over the last 6 weeks or perhaps type of customer is some of the greater demand with the launch customers you highlighted to us in San Francisco? Is it all new interest, new customers?"
— Andrew Gardiner, Citi

A: "So the number that we talked about at end of March was supply in place to support $1 billion of demand, and that includes memory, that includes wafers, that includes packaging, that includes access to test equipment. So for the $2 billion, we are now in the process of securing supply to support that. And the teams are working around the clock to make sure we can find the right answers for our customers."
— Rene Haas, CEO

Assessment: The most informative exchange of the call, and the reason a double beat sold off. Management answered the demand question fully and the supply question honestly, and the honest answer is that half the demand has no capacity behind it and no date attached. The CFO's follow-on framing was that roughly $90 million of chip revenue lands in Q4 FYE27 with a firmer estimate to come at the Q3 results, so the first real checkpoint on this business is nine months away.

Royalty puts and takes for the coming year

The second question went straight at the quarter's weakest line and asked management to decompose the outlook between data center strength and consumer softness. The answer conceded the mobile market is turning negative and argued the mix protects Arm.

Q: "Maybe first, just as we think about, I guess, like the royalty rate growth for 1Q and then just thinking about the full year, is there any help that you can provide just kind of the puts and takes there? Clearly, data center, very strong, accelerating. But then how do you kind of think about consumer electronics, smartphones, et cetera?"
— Joseph Quatrochi, Wells Fargo

A: "We're going to continue to see very flattish, maybe slightly negative numbers for the overall market. The lower end of the market is probably going to be where most of that impact is. So it doesn't have too much impact on us. And then, of course, any sort of negative impact we do see, we expect that to be more than offset by demand in Cloud AI or specifically in the data center."
— Jason Child, CFO

Assessment: A direct answer that restates the rate-over-units argument without re-quantifying it. The claim that data center more than offsets mobile weakness is the whole royalty thesis in one sentence, and this quarter it was not true: royalties decelerated 16 points while data center doubled. Management's position is that the gap was a comparison effect and that Q1 returns to 20%. That is the test.

Arm's natural share of a data center CPU market everyone is now sizing above $100 billion

A pointed challenge on TAM arithmetic: a competitor had just claimed a larger market and half of it, incumbents will not vanish, and hyperscalers build their own. The question was where Arm's share actually comes from. The CEO acknowledged the arithmetic problem directly before making the argument.

Q: "So I'm just trying to understand what is Arm's natural kind of niche in the market? Like which of these parties conceptually can you take share from to grow your business?"
— Vivek Arya, Bank of America Securities

A: "As far as the market share numbers, AMD has 50, Intel has 50 and we have 50. So you add up to some crazy number. ... And I'm actually confident that by the end of the decade, I believe the largest market share by CPU type will be Arm."
— Rene Haas, CEO

Assessment: The self-aware opening is the tell that management knows the TAM claims circulating are not mutually consistent. The substantive answer is that Arm participates twice, through hyperscaler custom silicon that already attaches to the dominant accelerators and through customers who will not design their own chip. Both are real. Neither resolves the question of whose share falls, which is the question that was asked.

The shape of FY27 and whether the spending step-up still happens

This exchange produced the quarter's most consequential commitment, and it came as a routine modelling question about phasing. Management used it to walk down the operating expense trajectory it had set in February and to put a date on when leverage appears.

Q: "And maybe a quick follow-up, just looking at FY '27 for the year. Jason, could you just give us some help in terms of the shape of how you expect royalty revenues to kind of flow through the year? And maybe some commentary around the OpEx as well would be really helpful."
— Timm Schulze-Melander, Rothschild & Co

A: "And now I would expect that the license -- or I'm sorry, that the OpEx growth is going to grow sequentially every quarter by, call it, a few percent. And the overall expense will still kind of be -- will show incremental margin improvement throughout the year, and we will be growing expenses less than revenue by the time we end the year"
— Jason Child, CFO

Assessment: In February the same question produced a commitment only that the rate of increase would moderate. This time it produced a direction, a cadence and a deadline. It is the answer we said we wanted and it is materially better than the one we underwrote. The hedge is "by the time we end the year", which concedes that the first half still spends faster than it earns.

Whether selling silicon puts Arm into conflict with the licensees who pay its royalties

The structural objection to the whole strategy, asked plainly. Management's answer was procedural rather than economic: it went to the ecosystem first and asked permission.

Q: "One of the questions that is top of mind for many investors is how the shift in your business model is impacting some of your existing IP customers who also sell Arm-based CPUs in some form. Could you maybe just give us a sense of how your larger customers have reacted since your announcement in March and how you think about managing that potential tension between your product and IP businesses?"
— Sebastien Naji, William Blair

A: "And we asked for their support of the strategy at the Arm Everywhere Day. And every single partner we asked said yes. We had probably over 50 different partners, the names I just gave you, all of them. ... And at the end of the day, we are responding to customer demand in a market, and we see this today because we're sold out and we've got people looking for more products, there's demand for these products."
— Rene Haas, CEO

Assessment: Fifty partners publicly endorsing a strategy that competes with several of them is a genuine signal, and "we're sold out" is the strongest single phrase in the answer. The unanswered part is what happens in the second and third generations, when the silicon product is no longer supply-capped and is competing for the same sockets as licensee-designed parts. Consent obtained at launch is not a permanent condition.

The cost structure of the chip business and when it becomes accretive

The most useful modelling question of the call, and the one that produced the numbers the market now has to weigh against a 98% gross margin. Management gave the headcount, the timing of profitability, and the long-run margin split between the two businesses.

Q: "I think at the event, you told us gross margin profile on these first-gen products is about 30% plus. I know it's embedded in your total cost structure. But Jason, can you just give us a rough sense on the OpEx attributed to supporting your chip business this year and next year? ... as you drive your chip revenues higher, like what -- when can it drive accretion to the earnings power?"
— Harlan Sur, JPMorgan

A: "So the incremental costs and OpEx that we're having to add to the chip business is not that significant. It's a team that's in the -- probably in the dozens of people, not hundreds. And so you can assume that it's operating profit positive next year."
— Jason Child, CFO

Assessment: The question embedded a roughly 30% gross margin for the first generation and management did not dispute it, which is how that number became public on an earnings call. The operating-cost answer is more favourable than expected and rests on a real argument, that the expensive compute die is already built for the IP business. Both facts together say the chip business is cheap to run and structurally lower-margin than what it sits next to. Investors should model a blended margin that declines as the mix shifts, which is the opposite of the direction a software-like multiple assumes.

Whether the CPU-to-GPU ratio is the right way to size the opportunity

Two separate questioners pressed on the same modelling framework, trying to convert the agentic-AI narrative into a countable ratio. Management rejected the chip-count framing and redirected to core count and average selling price.

Q: "I just wanted to go back to the ratio of CPUs to GPUs, particularly as it relates to orchestration. ... are we looking at this as each agentic flow requires one core? Or could we look at this differently and say that there is an average number of Arm instructions that are needed to orchestrate each token generated by the accelerator?"
— Multiple analysts incl. Lee Simpson, Morgan Stanley, Krish Sankar, TD Cowen

A: "So our viewpoint is very much one of more cores is better. And that's why I think you're going to see increasingly larger core counts in these CPU chips. So you'll see more CPUs cores being shipped. You may not see 3x the chips, but the chips will be more expensive, which is why when people look at $100 billion TAM in 5 years or $125 billion, whatever the number is, largely, it's going to be driven by the fact that these CPU chips are going to have lots of CPU cores, which will drive ASP up."
— Rene Haas, CEO

Assessment: This is the rate-escalator argument transplanted from smartphones to the data center, and it is the same mechanism that has worked for Arm in mobile: more content per socket at a higher price, rather than more sockets. It is also the reason the company is comfortable with a TAM number that its own unit forecasts cannot support, since the growth is priced per core rather than per chip. The framework is coherent. It is not yet visible in a disclosed data center revenue line.

What They're NOT Saying

1. Data center revenue in dollars, for a second straight year. This is the annual disclosure quarter, the business is described as about to become the largest at Arm, and the letter does not size it. Management said Neoverse royalties doubled and will double again; nobody asked for the base.
2. The SoftBank agreement's terms. Three quarters in, the only disclosed fact is a quarterly dollar figure given verbally. No term, no total value, no recognition pattern, no renewal mechanics, and no explanation of the $494 million related-party contract-asset build that accompanies it.
3. Why cash conversion fell to the lowest level of the fiscal year. Free cash flow declined 6.7% against non-GAAP net income up 9.8%, and neither document mentions it. The full-year 791% increase was foregrounded instead, against a base management itself had described as distorted.
4. Any capital expenditure guidance. FY26 capex rose 148.9% to $545 million and property and equipment more than doubled. A company that has just launched a silicon product and is scrambling for wafer, memory, packaging and test capacity gave no forward capital plan.
5. The memory sensitivity, still unreconciled. Two different numbers were given on the Q3 call for the same calculation. This quarter mobile units actually turned negative and no updated sensitivity was offered.
6. What happens to the $2 billion of AGI CPU demand that has no supply. Management said it is "in the process of securing" capacity. There is no date, no named partner, and no statement of what happens to the unserved orders if capacity does not arrive.
7. The gross margin on the chip business, from management's own mouth. "About 30% plus" entered the record as an analyst's characterisation of a March event. Management neither confirmed nor corrected it, and the letter does not address it.
8. Any FY27 free cash flow or conversion framework. Revenue, operating expense and EPS were all framed for the year. Cash was not.

Market Reaction

• Pre-print setup: ARM closed at $237.30 on May 6, an all-time high and the top of a 52-week closing range of $104.55 to $237.30. The shares were up 117.1% year to date against the S&P 500 up 7.6%, up 59.5% over the trailing 30 days and up 93.8% over the trailing twelve months.
• The six weeks before the print: ARM closed at $134.96 on March 24, the day of the Arm Everywhere event, and rose 16.4% to $157.07 the following session. From the event to the pre-print close the shares gained 75.8%.
• After-hours and premarket: shares fell roughly 5.6% in premarket trading on May 7.
• Reaction session: ARM opened May 7 at $222.40, a 6.3% gap down, traded a $210.23 to $232.23 range, and closed at $213.31, down 10.1% or $23.99. The S&P 500 fell 0.4% the same session.
• Volume: 22.4 million shares against a 30-day average of 9.5 million, 2.4 times normal.

A stock that beats on revenue, beats on earnings, beats its own operating expense guide and guides the next quarter above consensus does not usually fall 10%. The explanation is entirely in the setup. ARM entered this print having risen 75.8% in six weeks on a product announcement, at an all-time high, with the whole of that move built on a demand narrative for a chip that has not shipped. The print's job was not to be good. It was to validate a re-rating that had already happened.

It failed that job in two specific places. The first is the supply cap. Management doubled the demand figure and held the revenue outlook, which converts the AGI CPU from an uncapped option into a capacity-constrained product with a $1 billion ceiling and no date for lifting it. Anyone who bought the March move bought the second billion; this print told them it is not yet purchasable. The second is royalties. At 10.5% growth and 3.4% below consensus, the IP line that supports the entire valuation posted its weakest quarter of the year in the same release that quantified the new business as a roughly 30% gross margin operation. Sell-side commentary flagged the royalty shortfall against models directly, and separately noted that operating expense growth had slowed relative to trend, which is the offset that likely kept the decline to ten points rather than more.

The reaction is best read as position-clearing rather than thesis revision. At least 14 brokerages raised price targets after the print, which is not the behaviour of a Street that thinks the quarter was bad. It is the behaviour of a Street that thinks the business is fine and the entry point was not. Volume at 2.4 times normal is elevated but well below the 4.1 times we saw on the February print, and the close at $213.31 still leaves the shares up 95.2% year to date and 92.4% above where we initiated three months ago.

Street Perspective

Debate: Is the royalty deceleration a comparison effect or the start of a trend?

Bull view: The hard comparison was pre-announced in February, the full year still landed above 20%, data center royalties doubled for a third consecutive quarter and are guided to double again, and management has guided the line back to roughly 20% growth in the very next quarter. One soft print against a known comparison is arithmetic, not deterioration.

Bear view: Royalties decelerated 16 points, undershot management's own "low teens" language, and missed consensus, in a quarter when mobile units turned negative for the first time. The insulation argument has been asserted for three quarters and this is the first quarter it could be tested against real unit weakness. It did not obviously pass.

Our take: The bulls have the better of the specific quarter and the bears have the better of the burden of proof. The comparison explanation is checkable and was given in advance, which is worth a great deal. But the guide asks for an immediate recovery to 20% from 10.5% while the memory constraint persists and the comparison relief is only partial, and management chose not to re-quantify the sensitivity that would let anyone verify the claim. We would rather grade this on next quarter's royalty line than argue it now.

Debate: Does selling silicon expand the opportunity or compress the multiple?

Bull view: Arm is monetising engineering it already does. The expensive compute die is the compute subsystem it builds anyway, incremental headcount runs to dozens rather than hundreds, the business turns operating-profit positive next year, and demand doubled to $2 billion in six weeks with more than 50 ecosystem partners endorsing it. Management's FYE31 frame of $25 billion and more than $9 of EPS is 23.7 times a terminal year at the current price.

Bear view: A roughly 30% gross margin, 35% long-run operating margin business is being attached to one at 98.3% and 65%, and management's own five-year plan has the lower-margin half at $15 billion against $10 billion. That is a mix shift toward worse economics, higher capital intensity and supply-chain exposure Arm has never carried. Companies that move from royalties to products get product multiples.

Our take: Both are right and they resolve at different horizons. Over five years the revenue opportunity is genuinely larger and the incremental cost genuinely low, so the earnings case is stronger with the chip business than without it. Over the next twelve months the disclosure works against the multiple, because the market now has the margin structure and does not yet have the revenue. We flagged at initiation that we would rather own this after the March event clarified the SoC ambition. The event clarified it, and the clarification argues for a lower multiple on the incremental dollar, not a higher one.

Debate: Is the related-party concentration improving or just being diluted?

Bull view: The parent's contribution was flat at $200 million while third-party licensing grew, so concentration fell from 16% of revenue to 13.4% and the ex-agreement licence decline narrowed from 24.3% to 2.4%. Six new Total Access licences and two next-generation CSS agreements are unambiguously third-party demand. The problem is shrinking on its own as the rest of the business grows into it.

Bear view: Revenue from external customers, the reported line, grew 5.9% in the quarter and 7.4% for the year. Related parties are 30.5% of FY26 revenue against 20.5%. Three quarters in there is still no contract term, and $646 million of the associated revenue is sitting unbilled on the balance sheet. Dilution of a ratio is not disclosure.

Our take: The bull case wins on trajectory and the bear case wins on what is knowable. Both can hold, and the resolution is not analytical. Arm could close this debate with a single paragraph describing the agreement's term, scope and recognition pattern, and has declined to do so through three reporting cycles. Until it does, we apply a discount to the related-party revenue and we note that the reported external-customer line grew at a mid-single-digit rate for the year.

Model Update Needed

Our February estimates were set at initiation. This quarter's actuals and the FY27 framework require revisions in both directions: revenue and earnings up, cash conversion down.

Valuation: at the May 7 close of $213.31 and 1,068 million diluted shares, ARM carries a market capitalisation of roughly $227.8 billion. Net of $3,601 million in cash and short-term investments with no borrowings disclosed, enterprise value is approximately $224.2 billion, or 45.6 times FY26 revenue. Against reported FY26 non-GAAP EPS of $1.77 the trailing multiple is 120.5 times; against our FYE27 estimate of $2.05 it is roughly 104 times. Trailing free cash flow of $882 million is a 0.39% yield. Management's FYE31 frame of more than $9 in EPS puts the shares at 23.7 times a terminal year five years out.

Valuation impact: we are again not publishing a price target. At initiation we withheld one because the March event and the SoftBank terms were both unresolved and both looked resolvable within two quarters. One has resolved, in a direction that lowers the appropriate multiple on the incremental revenue rather than raising it. The other has not moved. What has changed is that the market has now supplied its own answer, re-rating the shares from 63 times forward to roughly 104 times on the strength of a five-year plan whose largest component is supply-capped at half its demand. We would rather be explicit that the outcome distribution here is wide than publish a target that implies otherwise.

Thesis Scorecard Post-Earnings

Scored against the pillars established at initiation, unchanged in wording.

Overall: the thesis is unchanged in structure and better in evidence. Of the four risks we set in February, one is now contained, one has separated into a resolved half and a worsening half, one is materialising exactly as described, and one improved in ratio while its disclosure gap stayed open. Of the three bull pillars, one strengthened, one held, and one was tested for the first time and came through with a weaker result than the preceding three quarters. That is a business performing well and a disclosure set that has not kept pace with what the market is being asked to pay.

Action: Hold. One of the two upgrade triggers we named at initiation was met on its stated wording, and we mark it met: the March 24 event clarified the chiplet and complete-SoC ambition, and capital intensity did not materially rise, which was the condition we attached to it. The trigger was aimed at the wrong variable. The risk in moving to silicon was never the capital, which this quarter shows is modest, but the margin structure, and the clarification put that at roughly 30% gross and 35% long-run operating or EBITDA against 98.3% and 65% for the IP business. A trigger that is met and resolves against the multiple is not a reason to upgrade. The second trigger, disclosure of the SoftBank agreement's terms, is unmet after three quarters. Meanwhile the forward multiple has gone from 63 times to roughly 104 times while forward earnings rose 18%, so the price has done substantially more work than the business. We would upgrade on disclosure of the SoftBank agreement's structure, on a reported data center revenue line, or on a de-rating toward roughly 75 times forward earnings. A move to Underperform would require the royalty recovery to 20% in Q1 to fail alongside a second quarter of sub-30% cash conversion.