Arm’s Q2 beat says AI licensing is being underpriced, not just smartphone royalties

The print changes the debate because the upside was not a one-line accounting beat that can be dismissed as timing. Street expectations embedded revenue of $1,060.6 million and EPS of $0.33, so the actual $1,135.0 million revenue and $0.39 EPS delivered surprise of +7.0% and +18.6%. What was priced in was a continuation of Arm’s billion-dollar quarterly run, with investors already aware that Q1 FY2026 revenue was $1,053.0 million and that Q4 FY2025 had reached $1,241.0 million. What was not priced in was the mix of Q2 FY2026 revenue up +34.5% YoY, gross margin of 97.4%, and a management guide for Q3 that starts at $1.225 billion, plus or minus $50 million. The variant perception is that Arm is being valued too narrowly as an end-device royalty compounder, when this quarter shows licensing demand for AI compute products is already large enough to change the medium-term earnings setup before cloud and networking royalties become a larger percentage of the base.

That distinction matters because the company’s own commentary separated the two engines cleanly, and both exceeded the burden of proof in different ways. Rene Haas framed the reported quarter as “our third consecutive billion-dollar quarter,” with revenue of $1.14 billion, up 34% year-on-year, which matters less as a victory lap than as confirmation that the post-IPO step-up is no longer a single-quarter event. Jason Child gave the more investable detail: royalty revenue grew 21% year-on-year to a record $620 million versus guidance of mid-teens, while license and other revenue was $515 million, up 56% year-on-year. The surprise, therefore, was not just that royalty revenue was better than the guide. It was that licensing was nearly the same scale as royalties in the quarter, with $515 million against $620 million, and management linked that licensing growth to companies building next-generation AI products rather than to an ordinary smartphone refresh. That is the core mispricing: royalty growth of 21% year-on-year is valuable, but licensing growth of 56% year-on-year suggests customers are pulling Arm deeper into custom silicon road maps where revenue recognition can arrive well ahead of downstream unit royalties.

The financial trajectory supports that interpretation because Q2 FY2026 did not merely bounce from a weak sequential base. Revenue of $1,135.0 million followed $1,053.0 million in Q1 FY2026, producing revenue QoQ of +7.8%, after Q1 FY2026 had declined -15.1% QoQ from Q4 FY2025. The more important comparison is the YoY acceleration: Q2 FY2026 revenue YoY of +34.5% compares with Q2 FY2025 revenue YoY of +4.7%, while gross margin rose to 97.4% from 93.6% in Q2 FY2025. Investors who expected Arm to keep growing but normalize margins missed the operating leverage embedded in a quarter where revenue moved to $1,135.0 million and gross margin printed at 97.4%. The immediate caution is that Q3 FY2026 in the quarterly history shows gross margin of 94.2%, and Q4 FY2026 shows 93.1%, so the 97.4% Q2 margin should not be capitalized as the new steady state. But the thesis does not require 97.4% to persist. It requires that licensing and royalty expansion keep revenue above the level contemplated by consensus while OpEx growth funds AI platform work rather than simply consuming the upside.

That margin point connects directly to the spending debate, because the EPS beat was delivered before the guided OpEx step-up becomes more visible. Jason Child said non-GAAP operating expenses were $648 million, up 31% year-on-year and slightly below guidance, while non-GAAP operating income was $467 million, up 43% year-on-year, with non-GAAP operating margin of 41.1% compared with 38.6% a year ago. The market was right to expect investment intensity, but the quarter shows the company can spend into the AI opportunity and still deliver EPS upside when licensing converts. The Q3 guide makes the trade-off explicit: revenue of $1.225 billion, plus or minus $50 million, non-GAAP operating expense of approximately $720 million, and non-GAAP EPS of $0.41, plus or minus $0.04. This is not a low-cost beat-and-raise story. It is a willingness to take quarterly non-GAAP operating expense from $648 million to approximately $720 million while guiding EPS above the Q2 reported $0.39 on the company’s own basis. If the market penalizes the higher spend without giving credit for the ACV and license signal, it is misreading the investment cycle.

The durability of that investment cycle is visible in ACV and customer adoption, not only in this quarter’s P&L. Child said ACV grew 28% year-on-year, “maintaining strong momentum following the 28% year-on-year growth we reported in Q1,” and that wording matters because it commits to two consecutive quarters of the same 28% year-on-year ACV growth rather than a one-quarter surge. Haas added a concrete customer proof point: Google’s Arm-based Axion chip delivers up to 65% better price performance while using 60% less energy. That is not just a marketing datapoint for Alphabet (Google); it is a second-order read-through for custom cloud silicon as an Arm royalty and licensing vector. If cloud operators can point to up to 65% better price performance and 60% less energy, then the incentive to shift more internal workloads to Arm-based custom cores is economic rather than ideological. For NVIDIA, the implication is not that Arm displaces accelerators, but that Arm CPU and subsystem IP attaches to AI infrastructure designs where power efficiency and software portability shape platform decisions. For Qualcomm, MediaTek, Samsung, and Apple, the same quarter signals that Arm’s bargaining power is expanding beyond mobile, even as smartphones remain one of the “major markets” management cited for royalty growth.

The supply-chain read-through is therefore uneven in a useful way. Arm has no listed suppliers in the data pack, so the observable signal runs through its IP customers. Alphabet is the clearest beneficiary because the call explicitly tied Google’s Axion to up to 65% better price performance and 60% less energy, and because Arm architecture for custom cores is listed as the relationship. NVIDIA gets a platform read-through through IP, particularly if cloud and networking royalties trend higher, but the data pack does not quantify NVIDIA-specific revenue, so the magnitude we can anchor is Arm’s royalty revenue of $620 million and licensing revenue of $515 million. Qualcomm, MediaTek, Samsung, and Apple remain IP customers exposed to the same royalty pool, and Haas said royalty growth was driven by all major markets including data center, smartphones, automotive and IoT. The magnitude is not vague: royalty revenue reached $620 million, up 21% year-on-year, while license and other revenue rose to $515 million, up 56% year-on-year. The second-order point for those customers is that Arm’s monetization base is becoming less dependent on any single end market, which strengthens Arm’s hand as customers renew access to architectures used across phones, autos, IoT, and AI compute.

The competitive comparison sharpens the same conclusion, because Arm’s economics look more like scarce IP than conventional semiconductor product revenue. In the EDA_IP peer set, SNPS reported $2,276.0 million revenue, gross margin of 72.3%, and revenue YoY of +41.9%; CDNS reported $1,474.2 million revenue, gross margin of 95.8%, and revenue YoY of +18.7%; Arm’s latest peer-table quarter shows $1,490.0 million revenue, gross margin of 93.1%, and revenue YoY of +20.1%. Against CDNS, Arm is not the gross-margin outlier, since CDNS gross margin is 95.8% versus Arm at 93.1%, but Arm’s growth of +20.1% is above CDNS at +18.7% on a similar revenue scale of $1,490.0 million versus $1,474.2 million. Against SNPS, Arm grows slower than +41.9%, but carries a higher gross margin at 93.1% versus 72.3%. The market’s mistake would be to compare Arm only with device-exposed semiconductor royalty names and ignore that its gross margin profile and licensing cadence increasingly overlap with design infrastructure economics.

The call delivery also supports the thesis, but with one important caveat that PMs should not ignore. In the tone history, Q2 FY2026 sentiment was 0.39, the highest value in the displayed sequence that includes Q3 FY2025 at 0.31, Q4 FY2025 at 0.28, and Q1 FY2026 at 0.35. Guidance_tone was 0.47 in Q2 FY2026 versus 0.29 in Q1 FY2026, which aligns with a management team willing to guide Q3 revenue to $1.225 billion, plus or minus $50 million. Prepared_sentiment was 0.71 in Q2 FY2026, above Q1 FY2026 at 0.62, while qa_sentiment was 0.17, below Q1 FY2026 at 0.22. The caveat is that the call was more confident in prepared remarks than in Q&A, and qa_evasiveness rose to 11.2 from 3.6. That does not break the thesis, but it tells us where the debate sits: management was emphatic on reported momentum and guidance, less expansive in Q&A where investors probed sustainability, China exposure, and cloud royalty trajectory.

The China and cloud details are the places where the numbers create tension rather than a clean bull case. Lee Simpson said, “I see China is maybe 22% of sales this Q,” which is material enough to matter if export controls, local substitution, or customer mix shift against Arm. Separately, Harlan Sur said cloud and networking accounted for about 10% of royalty revenues last fiscal year, and Child said a trajectory “somewhere in the 15% to 20% range” is a reasonable expectation for where management expects to trend throughout the year. The tension is that cloud and networking are still not the majority of royalties, even though the AI narrative drives much of the multiple. The bull case must therefore be framed correctly: Q2 does not prove that AI cloud royalties have already transformed the company. It proves that licensing tied to AI products is already large at $515 million, up 56% year-on-year, while the royalty base of $620 million, up 21% year-on-year, has a credible cloud and networking mix-up story from about 10% last fiscal year toward 15% to 20% throughout the year. If that mix does not move, the AI royalty thesis gets pushed out, but the licensing thesis still has current-quarter evidence.

This is why the Q3 guide is the fulcrum rather than a footnote. Child’s guidance for royalties to be up just over 20% year-on-year and licensing to be up 25% to 30% year-on-year implies that management expects both legs to keep expanding even after a Q2 where licensing grew 56% year-on-year. The Street may focus on deceleration in licensing growth from 56% to 25% to 30%, but that would miss the base effect after Q2’s $515 million license and other revenue. It also misses that the company is raising the operating expense run rate into demand, not retreating after a beat. Child quantified an increase from last quarter as about $50 million, then clarified it went up $52 million to about $178 million and called that a good run rate to assume going forward. The wording is messy because the transcript includes a correction from “hundreds of billions” to “hundreds of millions” when discussing investment, but the substance is consistent: Arm is spending at a scale consistent with building platform IP, not harvesting a static royalty stream.

The stock reaction should therefore be judged against a narrow checklist, not against whether every AI claim becomes revenue immediately. The confirming evidence next quarter is concrete: Q3 revenue needs to land within the guided $1.225 billion, plus or minus $50 million; non-GAAP EPS needs to hold within $0.41, plus or minus $0.04; royalties need to be up just over 20% year-on-year; licensing needs to be up 25% to 30% year-on-year; and non-GAAP operating expense should be approximately $720 million without erasing operating leverage. The thesis breaks if Q3 shows revenue below the lower end of the $1.225 billion, plus or minus $50 million range while OpEx is still approximately $720 million, or if royalty growth falls short of “just over 20% year-on-year” at the same time licensing misses the 25% to 30% year-on-year guide. It is further pressured if cloud and networking fail to move from about 10% of royalty revenues last fiscal year toward the 15% to 20% range management described, or if China at maybe 22% of sales becomes the explanation for volatility rather than a manageable exposure. Until those numbers crack, the right read is that Q2 FY2026 was not merely an $0.39 EPS beat versus $0.33. It was evidence that Arm’s AI licensing cycle is already in the reported numbers, while the royalty mix shift is still ahead.