Arm’s Q3 Beat Was Small; the Mispriced Signal Is the $200 Million Licensing Run Rate and Data-Center Royalty Mix

The right read on this quarter is that the market got the direction right but likely underpriced the composition. What was priced in was a good quarter: Street revenue was already at $1,226.9 million and EPS at $0.41, leaving limited room for a classic beat-and-raise reaction when reported revenue came in at $1,242.0 million and EPS at $0.43. What actually surprised was more specific: royalty revenue reached $737 million with 27% year-on-year growth, license revenue reached $505 million with 25% year-on-year growth, and Jason Child tied $200 million of that license line to the SoftBank Technology Licensing and Design Services agreement. The variant perception is that the $200 million figure is not merely a one-quarter related-party oddity to discount away. On the call, Child answered the run-rate question with language that matters because it converts a debated contribution into a planning baseline: “I would expect that $200 going forward is the right run rate going forward.” If that holds, the Street’s focus on the modest +1.2% revenue surprise misses a licensing floor that supports operating leverage even as Arm keeps spending through the product cycle.

That composition matters because Arm’s financial trajectory has now separated from the choppiness that characterized earlier post-IPO quarters. Revenue was $824.0 million in Q3 FY2024, $983.0 million in Q3 FY2025, and $1,242.0 million in Q3 FY2026, so the latest quarter’s +26.3% year-on-year growth is not a one-period snapback from a weak base. The sequential line also recovered from Q1 FY2026’s $1,053.0 million and Q2 FY2026’s $1,135.0 million to Q3 FY2026’s $1,242.0 million, while gross margin remained at 94.2% after 97.4% in Q2 FY2026 and 94.3% in Q1 FY2026. The margin giveback from Q2 FY2026 is visible, but it is not the main story because the business is still producing software-like gross margin while scaling both royalty and license revenue. The historical context is more important than the single-quarter margin downtick: Arm’s revenue has moved from $633.0 million in Q4 FY2023 to $1,242.0 million in Q3 FY2026, and the company has reported four consecutive quarters above the billion-dollar level on management’s own basis.

The forward guide reinforces the idea that the quarter was more about durability than surprise magnitude. Jason Child guided Q4 revenue to $1.47 billion, plus or minus $50 million, and non-GAAP EPS to $0.58, plus or minus $0.04, with non-GAAP operating expense of approximately $745 million. The quarterly history shows Q4 FY2026 revenue at $1,490.0 million, gross margin at 93.1%, revenue growth of +20.0% quarter on quarter, and revenue growth of +20.1% year on year. A portfolio manager should care less that Q3 revenue only beat by +1.2% and more that the model is being asked to absorb a step from $1,242.0 million in Q3 FY2026 to $1,490.0 million in Q4 FY2026 while gross margin stays above 93.1%. That forward step is the real test of whether licensing is becoming a more visible base rather than a volatile deal-timing swing factor.

The spending line is the cleanest objection to the bull case, but the numbers argue for intentional reinvestment rather than uncontrolled dilution of the model. Non-GAAP operating expenses were $716 million, up 37% year on year, while non-GAAP operating income was $505 million, up 14% year on year, yielding a non-GAAP operating margin of about 41%. That is the conflict in the print: OpEx growth of 37% is outrunning operating income growth of 14%, yet the company still delivered non-GAAP EPS of $0.43 against the Street’s $0.41 and guided non-GAAP EPS of $0.58, plus or minus $0.04. The market may penalize the 37% OpEx growth if it views Arm as a mature IP toll collector. It should be more willing to fund that expense base if the license contribution of $200 million is a recurring run rate and the royalty mix is shifting toward higher-core-count compute. The burden of proof for management has risen, however, because approximately $745 million of Q4 non-GAAP operating expense means the next quarter has to show revenue conversion rather than only R&D ambition.

The reason to tolerate that spending is that the royalty base is changing from units alone to compute density inside named hyperscale platforms. Rene Haas said, “Royalties increased 27% to a record $737 million, driven by record units with strength across AI and general-purpose data center.” The wording matters because management did not attribute the royalty record only to smartphone recovery, where unit sensitivity can cut both ways; it called out AI and general-purpose data center as explicit contributors. Haas also said the data-center royalty business has grown more than 100% year on year and that management expects data center in a few years to be the largest business, larger than mobile. Child later framed the time horizon more tightly, saying that over the next two to three years investors should expect data center to become similar to or maybe larger than the smartphone business, which he described as 40 to 45% of total business. That is the central variant perception: Arm is giving investors a time-bound mix shift, and the quarter already contains the numerical evidence in $737 million of royalties and more than 100% year-on-year data-center royalty growth.

The customer read-through is consequently more meaningful than the aggregate revenue beat. For Alphabet (Google), the print supports the economics of internal Arm-based server silicon: Google’s Axion-powered m4a instances were described as delivering up to 2x better price performance and 80% better performance per watt versus comparable x86 offerings, and Haas separately referenced Alphabet announcing a $180 billion CapEx spend. For NVIDIA, Arm’s data-center royalty growth of more than 100% year on year is a reminder that more AI infrastructure value can accrue to IP and CPU attach around accelerators, not only to the accelerator vendor itself. For Apple, Qualcomm, MediaTek, and Samsung, the smartphone exposure remains central because Child quantified the sensitivity: a minus 20 degree unit impact would have at most a four to 6% revenue impact specifically within smartphones. That number prevents a lazy “mobile risk” critique from overwhelming the print; the smartphone business is still large, but management is quantifying a bounded revenue impact while pointing to data center as a business that can become similar to or larger than 40 to 45% of total business over two to three years. Arm has no suppliers listed in the data pack, so the direct supply-chain implication here runs through customers rather than upstream procurement.

The competitive comparison also supports the view that Arm should not be analyzed only against traditional EDA revenue growth. In the latest EDA_IP peer set, Arm’s Q4 FY2026 line is $1,490.0 million with 93.1% gross margin and +20.1% revenue growth, while SNPS reports $2,276.0 million with 72.3% gross margin and +41.9% revenue growth, and CDNS reports $1,474.2 million with 95.8% gross margin and +18.7% revenue growth. The useful comparison is not that one stock is simply cheaper or better, since valuation is not in the data pack; it is that Arm is now roughly revenue-scale comparable to CDNS on this latest reported-quarter table, with $1,490.0 million versus $1,474.2 million, while carrying a gross margin of 93.1% versus CDNS at 95.8%. SNPS is growing faster at +41.9%, but its 72.3% gross margin is materially lower than Arm’s 93.1%. That places Arm in a different earnings-quality debate: the market is paying for IP scarcity and margin structure, so the decisive question is whether the data-center royalty and license run-rate claims keep revenue growth near +20.1% rather than fading back toward the lower-growth smartphone profile.

The call delivery was more mixed than the financial story, and that nuance matters because management’s words are carrying a larger part of the bull case. The tone history shows Q3 FY2026 sentiment at 0.29, down from 0.39 in Q2 FY2026, with guidance_tone at 0.24, down from 0.47, prepared_sentiment at 0.60, down from 0.71, and qa_sentiment at 0.06, down from 0.17. That is not a management team projecting unqualified ease in the quarter being reported. Yet the Q4 FY2026 tone data moves in the other direction: sentiment rises to 0.32, guidance_tone rises to 0.51, tone_confidence rises to 0.41, prepared_sentiment rises to 0.72, qa_sentiment rises to 0.15, and ai_optimism rises to 0.38. The conflict is that uncertainty also rises to 63.9 from 47.5 and qa_evasiveness jumps to 50.2 from 3.8, even as guidance_tone improves by +0.27. The interpretation is not that management was evasive about the near-term numbers, because the Q4 revenue guide of $1.47 billion, plus or minus $50 million, and EPS guide of $0.58, plus or minus $0.04, were explicit. The higher evasiveness more likely reflects investor pressure on sustainability of the $200 million SoftBank contribution and the timing of data center becoming larger than mobile.

That tone split should influence position sizing around the thesis. Prepared remarks give investors concrete anchors: Q3 revenue of $1.24 billion, royalty revenue of $737 million, license revenue of $505 million, non-GAAP OpEx of $716 million, non-GAAP operating income of $505 million, and Q4 revenue guidance of $1.47 billion, plus or minus $50 million. The Q&A, however, is where the sustainability debate lives, and the tone history’s qa_evasiveness of 50.2 in Q4 FY2026 versus 3.8 in Q3 FY2026 tells us not to overfit to a single declarative answer. A fair stock reaction should therefore separate near-term de-risking from medium-term proof. Near term, the company beat EPS by +4.9%, beat revenue by +1.2%, and guided a material Q4 revenue step. Medium term, investors need confirmation that the $200 million SoftBank contribution is a run rate and that data-center royalty growth above 100% year on year is not simply early hyperscaler ramp noise.

The most underappreciated second-order implication is for x86 displacement narratives in cloud infrastructure, because Arm supplied customer-specific performance metrics rather than generic adoption language. AWS’s fifth-generation Graviton processor was described as having 192 cores, doubling the core count from Graviton four, delivering 25% higher performance, and up to 33% lower latency versus Graviton four. Google’s Axion-powered m4a instances were described as delivering up to 2x better price performance and 80% better performance per watt versus comparable x86 offerings. Those figures matter for Arm’s royalty model because core count and deployment breadth can scale economics even when the end customer’s procurement budget is framed as capex rather than chip resale. Haas also said Arm has a developer ecosystem over 22 million developers, more than 80% of the global total, and that hyperscaler share is expected to reach 50% among the top hyperscalers. If those adoption claims convert into royalty revenue, the smartphone comparison changes from a risk factor to a benchmark that data center can overtake.

The bear case is still coherent, but it has to be stated numerically. Q3 gross margin fell to 94.2% from 97.4% in Q2 FY2026, diluted EPS in the quarterly history was $0.21 in Q3 FY2026 versus $0.22 in Q2 FY2026 despite revenue rising to $1,242.0 million from $1,135.0 million, and non-GAAP operating expenses grew 37% year on year while non-GAAP operating income grew only 14% year on year. If investors believe the $505 million license line is inflated by a related-party $200 million contribution that fails to recur, or that data-center royalties remain too small despite more than 100% year-on-year growth, then the +1.2% revenue beat is not enough to defend a premium multiple. But the data pack pushes against the harshest version of that bear case: management explicitly said $200 million is the right run rate going forward, ACV grew 28% year on year after 28% year-on-year growth in Q2 and Q1, and Q4 FY2026 revenue in the history is $1,490.0 million with +20.1% year-on-year growth. The debate has moved from “is Arm growing?” to “is the growth increasingly contractual and data-center weighted?”

What to watch next quarter is therefore tightly defined. Confirmation starts with Q4 revenue landing within the guided $1.47 billion, plus or minus $50 million, and closer to the $1,490.0 million shown in the quarterly history than to the low end of the range; a break would be revenue nearer the bottom of the guide without an offsetting license explanation. EPS needs to track the $0.58, plus or minus $0.04, guide while non-GAAP operating expense stays near approximately $745 million, because a miss with OpEx above that level would validate concerns about 37% year-on-year spending growth. Licensing quality should be judged against the $200 million SoftBank run-rate statement and total license and other revenue of $505 million in Q3; the thesis weakens if management stops treating $200 million as recurring. Royalty durability should be judged against $737 million in Q3 royalties, 27% year-on-year royalty growth, and the claim that data-center royalty revenue grew more than 100% year on year. Finally, on the next call dated after this 2026-02-04 event, listen for whether management repeats the two-to-three-year path for data center to become similar to or larger than the smartphone business at 40 to 45% of total business; if that timing slips or the smartphone unit sensitivity expands beyond the four to 6% revenue impact cited for a minus 20 degree unit impact, the variant perception breaks.