CEVA’s revenue miss masks the better signal: royalties are finally absorbing the licensing lumpiness

The print says CEVA is becoming less fragile than the revenue miss implies, and that is the variant perception. A software-IP company that misses revenue by -8.5% and still posts $0.11 of EPS against a $0.10 estimate is telling investors that mix, royalties, and expense control mattered more than the absolute quarterly license close. The Street was set up for $31.0 million of revenue, or $31.03 million on the estimate basis, and got $28.4 million, or $28.38 million on the actual basis; that miss was real and should not be dismissed. What actually surprised was the earnings line: $0.11 versus $0.10, a +10.0% surprise, with gross margin at 88.0% and sequential revenue growth of +10.5% after Q2 FY2025 revenue of $25.7 million. The stock should not be treated as if the quarter proved demand acceleration across all licensing buckets. It proved something narrower but more investable: the royalty base is starting to carry more of the model at the same time management is guiding Q4 revenue to a level that, at the midpoint language, does not require a heroic step-up from Q3.

That distinction matters because the setup going into the print was top-line cleanliness, not earnings resilience. Revenue had been uneven for eight quarters: $29.2 million in Q4 FY2024, $24.2 million in Q1 FY2025, $25.7 million in Q2 FY2025, and now $28.4 million in Q3 FY2025. The Street estimate at $31.0 million effectively priced a quarter closer to the prior local high, while the company delivered a recovery from the trough rather than a full reacceleration. Yet the progression is not the profile of a business losing relevance: revenue was +4.3% year over year in Q3 FY2025 after -9.7% in Q2 FY2025, and gross margin expanded to 88.0% after 86.2% in Q2 FY2025 and 85.6% in Q1 FY2025. The market may over-penalize the missed licensing timing because the reported financials show the model is absorbing that timing better than in earlier down quarters, when Q1 FY2025 revenue of $24.2 million came with diluted EPS of -$0.14 and Q2 FY2025 revenue of $25.7 million came with diluted EPS of -$0.15.

The financial trajectory therefore argues for a narrower discount rate on CEVA’s earnings power, not a free pass on bookings. Revenue of $28.4 million in Q3 FY2025 still sits below the $31.3 million printed in Q4 FY2025 in the quarterly history, and the history also shows Q1 FY2026 revenue at $27.0 million, so investors should not pretend the business has escaped quarterly lumpiness. But the gross-margin line is the counterweight: 88.0% in Q3 FY2025 compares with 85.4% in Q3 FY2024, 86.2% in Q2 FY2025, and 85.6% in Q1 FY2025. That is the margin evidence behind the EPS beat. When Amir Panush opened with “With revenue of $28.4 million and non-GAAP EPS of $0.11,” the important part was not the revenue number, which the Street already disliked; it was the pairing of a missed top line with positive non-GAAP EPS. A PM should take that as evidence that CEVA’s cost and mix sensitivity is improving, provided Q4 does not require a one-off license catch-up to sustain the margin.

The mix detail strengthens that view because royalties supplied the higher-quality evidence in the quarter. Yaniv Arieli gave the split that matters: “The revenue breakdown is as follows: licensing and related revenue totaled $16 million, representing 56% of our total revenue for the quarter.” That means the rest of the discussion should not anchor only on deal timing. Royalty revenue was $12.4 million, 44% of total revenue, with a 16% sequential increase and a 6% increase year over year. Unit activity was also constructive: licensee shipments were 559 million units, up 19% sequentially and 11% year over year, while IoT reached 510 million units, up 13% year over year. Those figures explain how CEVA can miss a quarterly revenue estimate yet still argue that the installed base is compounding. The bear case is that licensing revenue at $16 million remains the swing factor; the bull case, which this print supports more than the headline miss suggests, is that royalties are becoming large enough at $12.4 million to soften the quarterly license-close volatility.

The second-order read-through is most relevant for customers that consume CEVA connectivity and DSP IP rather than for external suppliers, since the data pack lists no suppliers to CEVA. For MediaTek, which is tied to CEVA-X DSP core and subsystem, the quarter’s 559 million licensee units and 11% year-over-year unit growth suggest CEVA-linked designs are still shipping through end devices even though MediaTek’s peer table revenue was 149,150.5 million with revenue YoY of -2.7%. That contrast is important: CEVA’s unit royalty signal is not simply mirroring MediaTek’s reported top-line direction. For NXP, attached to Bluetooth LE IP and SensPro AI DSP, the mixed connectivity message is precise rather than uniformly good: Bluetooth shipments were 303 million units, down 1% from 306 million in 2024, but IoT units hit 510 million and industrial IoT totaled 10 million units. For Renesas, linked to Wi-Fi 6 and Bluetooth IP and DSP for ITS, the Wi-Fi datapoint is the sharper read-through, with shipments at 82 million units, up 73% from 47 million units a year ago. For Broadcom, tied to DSP cores, the quarter confirms DSP royalty pull-through broadly through the 559 million unit base but does not isolate a Broadcom-specific magnitude beyond that base. Samsung, tied to DSP cores for LTE/5G modem, gets a similar read: the print supports continuing modem/DSP shipment relevance, but the supplied data quantifies the aggregate 559 million unit activity rather than a Samsung-specific split.

The peer comparison also argues against reading CEVA like a conventional fabless revenue miss. CEVA’s gross margin at 88.0% sits above the peer-table gross margins for MediaTek at 46.3%, NVIDIA at 74.9%, and META at 81.9%, reflecting the IP-licensing economics rather than wafer-cycle leverage. That does not make the $28.4 million revenue base comparable to NVIDIA’s $81,615.0 million or the peer-table hyperscaler figures, but it does mean a one-quarter license shortfall has a different earnings translation when gross margin is 88.0%. The better comp point inside the table is not scale; it is margin durability. CEVA’s model can turn a missed revenue quarter into a positive EPS surprise if royalties and operating expense land inside a narrow band, while many product peers need larger revenue beats to offset manufacturing and inventory costs. This is why the -8.5% revenue miss and +10.0% EPS surprise should not be averaged into a neutral read. They indicate that investors need to separate the volatility of license revenue from the economics of the royalty base.

The expense line is the constraint on pushing the thesis too far. Management said total operating expenses for Q3 were $27.1 million and non-GAAP operating expenses were $22.1 million, both at the higher end of guidance, with the non-GAAP variance tied to higher employee benefit provisions associated with better financial results. That is a specific reason, but it still means CEVA did not generate the EPS beat by underspending the business. The Q4 setup keeps that pressure in place: Yaniv Arieli guided total revenue by saying, “As for the fourth quarter, total revenue is expected to be in the range of $29 million to $33 million.” He also pointed to gross margin at approximately 88% on a GAAP basis and 89% on a non-GAAP basis, while non-GAAP OpEx is expected to be higher than Q3 in the range of $22 million to $23 million. The thesis depends on Q4 revenue being high enough within $29 million to $33 million to absorb that $22 million to $23 million non-GAAP OpEx envelope. If Q4 lands near $29 million and OpEx is at $23 million, investors will question whether Q3’s EPS beat was mix and timing rather than sustainable leverage.

The call delivery reinforces the same conclusion: management sounded more positive in prepared remarks, but uncertainty rose, so the right stance is constructive with a defined tripwire. The tone history shows Q3 FY2025 sentiment at 0.49, above Q2 FY2025 at 0.42, while prepared_sentiment rose to 0.72 from 0.45. That is a meaningful improvement in scripted confidence. But uncertainty also jumped to 64.4 from 34.7, and qa_sentiment slipped to 0.30 from 0.38. The prepared narrative was better than the Q&A digestion. That conflict matches the financials: the company can point to 559 million shipped units, $12.4 million of royalty revenue, 88.0% gross margin, and Q4 revenue guidance of $29 million to $33 million, but investors were still left with a top-line miss against $31.0 million and a license bucket that remains lumpy at $16 million. The call was not evasive on the numbers, but the elevated uncertainty index says the path from unit growth to predictable revenue is still not fully settled.

The tone history also prevents over-reading the Q3 script as a clean inflection. Q3 FY2025 guidance_tone was 0.37, unchanged from Q4 FY2024 at 0.37 and below Q2 FY2025 at 0.49, even as prepared_sentiment hit 0.72. That split matters because management’s prepared language around the installed base was more upbeat than the forward guidance tone. In later history, Q4 FY2025 guidance_tone rose to 0.53 and Q1 FY2026 guidance_tone to 0.57, but uncertainty stayed high at 52.0 and 60.2. The machine read is consistent with a company that has better royalty evidence but not yet a low-variance forecast. For portfolio construction, that means the print is a reason to own the royalty compounding setup if valuation embeds the revenue miss as structural weakness, not a reason to underwrite a straight-line licensing recovery. The difference is material: CEVA has shown Q3 FY2025 revenue of $28.4 million, Q4 FY2025 revenue of $31.3 million, and Q1 FY2026 revenue of $27.0 million in the history, so any model that assumes a monotonic quarterly ramp is fighting the company’s own cadence.

The cleanest debate after the print is whether royalty momentum can keep offsetting license timing. On that score, the best numbers are the royalty and shipment data, not the headline revenue miss. Royalty revenue of $12.4 million was up 16% sequentially and 6% year over year, licensee shipments of 559 million units were up 19% sequentially and 11% year over year, and Wi-Fi shipments of 82 million units were up 73% from 47 million units a year ago. Bluetooth, however, was not part of the same acceleration, with shipments of 303 million units down 1% from 306 million in 2024. That mix argues CEVA’s next leg is more Wi-Fi and IoT-led than broad-based across every connectivity standard. It also means the company’s customers will feel the read-through unevenly: Renesas has the clearest positive data point through Wi-Fi 6 and Bluetooth IP, NXP has mixed Bluetooth and IoT signals, and MediaTek has a royalty-unit signal that diverges from its peer-table revenue YoY of -2.7%. The surprise in CEVA’s print is not “everything is accelerating.” It is that enough of the royalty base is accelerating to protect EPS when licensing misses.

What to watch next is concrete. For Q4 FY2025, the confirmatory range is the company’s $29 million to $33 million revenue guide, gross margin of approximately 88% on a GAAP basis and 89% on a non-GAAP basis, and non-GAAP OpEx of $22 million to $23 million. The thesis is confirmed if revenue lands toward the upper part of $29 million to $33 million while gross margin stays near 88% GAAP and royalty revenue does not give back the Q3 improvement from $12.4 million. It is broken if Q4 revenue lands near $29 million, non-GAAP OpEx is near $23 million, and shipments fail to sustain the Q3 base of 559 million units. On the next call date after this 2025-11-10 event, listen for whether Wi-Fi can remain near the 82 million unit level and whether Bluetooth can recover from 303 million units, down 1% from 306 million in 2024. The stock’s debate should move from the -8.5% revenue miss to royalty durability only if those unit and margin markers hold; if they do not, the Street was right to price the quarter as a licensing miss rather than an earnings-quality inflection.