CEVA’s Miss Is the Head Fake; the Royalty Mix Is Turning Before Licensing Re-accelerates

CEVA reported the kind of quarter that usually gets dismissed by screens: revenue of $25.7 million missed the Street’s $28.0 million by -8.3%, while EPS of $0.07 beat the $0.05 estimate by +40.0%. The market had effectively priced in a straight-line licensing rebound after Q2 FY2024’s $28.4 million comparison and Q4 FY2024’s $29.2 million exit rate, so the revenue miss matters. What it missed, though, is that this was not a demand collapse quarter. The company’s own reported basis shows royalty revenue of $10.7 million, a 16% sequential increase, while licensing and related revenue still represented 59% of total revenue. That is the key tension in the print: the near-term licensing cadence disappointed, but the royalty base is improving in the exact product categories that should matter more for CEVA’s operating leverage over the next several quarters.

The priced-in expectation was a revenue beat or at least a clean match to the $28.0 million estimate, and the stock deserved to be penalized for the -8.3% miss if investors were underwriting licensing as already normalized. The actual surprise was elsewhere: non-GAAP EPS came in at $0.07 versus $0.05, and Yaniv Arieli explicitly framed the quarter as better on profitability than the internal bar, saying, “Non-GAAP net income and diluted income per share for the second quarter of '25 was $1.8 million and $0.07, respectively, better than forecasted.” That wording matters because management is not claiming broad demand upside; it is claiming that the model can still produce positive non-GAAP earnings when revenue is below consensus. The distinction is important for a semiconductor IP asset, where a single licensing pushout can obscure whether the installed base is still throwing off royalty leverage.

That distinction shows up in the revenue composition more clearly than in the headline. Licensing and related revenue totaled $15 million, or 59% of total revenue, while royalty revenue was $10.7 million, or 41% of total revenues. The first figure explains the miss, because licensing remains the larger and lumpier piece; the second explains the EPS beat, because royalties rose 16% sequentially even though the total company was still below the Street’s revenue number. Arieli’s first-half framing is also useful, but only if kept in the right box: licensing revenue for the first half of 2025 reached $30.1 million, a 5% increase compared to $28.7 million for the same period in 2024. That does not erase the Q2 shortfall, but it argues against treating the licensing business as broken. The first-half royalty comparison is less forgiving, with $19.9 million versus $21.8 million, so the right view is not “everything is accelerating”; it is that Q2 contains an early sequential royalty turn while the year-over-year base still has to be repaired.

The financial trajectory therefore supports a narrower thesis than a generic recovery call: CEVA has been pinned in a mid-$20 million quarterly revenue band while gross margin has stayed in the high-80s, and the debate is whether Q2’s mix is the bridge out of that range or another false start. The Q2 reported gross margin of 86.2% is below the 89.7% level in the year-ago quarter, so there is no case for margin expansion yet. But the guide calls for gross margin to be “1% higher than the second quarter,” with approximately 87% on GAAP and 88% on non-GAAP. If that happens while revenue moves into the $26 million to $30 range, the model starts to look less like a stalled IP licensor and more like a company absorbing uneven licensing with better royalty follow-through. If it does not, the EPS beat will look like a one-quarter cost and tax artifact rather than a signal.

The cost structure is the main reason not to overstate the quarter. GAAP operating expenses were $26.6 million, above the high end of guidance, and non-GAAP operating expenses were $21.6 million, also just above the high end of guidance. Management attributed both to employee-related benefit provision after slower first-quarter results and associated adjustments, but the numbers still mean CEVA had little operating cushion on a $25.7 million revenue base. Non-GAAP operating margins and income were 3% of revenue and $0.8 million, versus operating margins of 15% and operating income of $4.4 million in the same quarter last year. That is the clearest rebuttal to the bull case: the EPS beat does not mean the income statement is healthy yet. It means the company can defend positive non-GAAP earnings at depressed revenue, which is useful, but not enough unless Q3 revenue actually steps up.

The unit data is where the print becomes more interesting for semiconductor investors, because the royalty mix points away from handset dependence and toward a broader connectivity recovery. CEVA reported 488 million units, with 55 million units tied to mobile handset modems and 409 million units in consumer IoT. Consumer IoT was up 16% from 353 million units in the second quarter of 2024, while industrial IoT was down 16% from 28 million units. The market may be underweighting that split because total royalty dollars were still down year-over-year, but the unit mix says the demand problem is not universal. The embedded connectivity categories that feed CEVA’s long-duration royalty stream are showing different speeds, and the largest bucket is already growing.

The product-level breakdown sharpens the read-through. Bluetooth shipments were 254 million units, down 5% from 266 million, which explains why the royalty recovery is not yet visible on a year-over-year dollar basis. But cellular IoT shipments reached 66 million units, up 66% year-over-year, and Wi-Fi shipments were 62 million units, up 80% from 35 million units. For customers, that is a differentiated message. MediaTek, which uses CEVA-X DSP core and subsystem IP, gets a read-through that modem-adjacent and connectivity attach remain active even as CEVA’s total revenue missed. NXP, tied to Bluetooth LE IP and SensPro AI DSP, should care that Bluetooth units are still down 5%, but the broader IoT unit base is not collapsing. Renesas, tied to Wi-Fi 6 and Bluetooth IP plus DSP for ITS, has the clearest positive signal from Wi-Fi units up 80%. Broadcom and Samsung are less clean reads from this pack, but CEVA’s 55 million mobile handset modem units show the handset modem exposure is not the dominant royalty engine this quarter.

That customer read-through also frames the peer comparison. CEVA’s 86.2% gross margin is structurally above the broader fabless comparables in the pack, including MediaTek at 46.3% and NVIDIA at 74.9%, which is exactly what an IP licensing model should deliver. The problem is not gross margin architecture; it is revenue scale and timing. On growth, CEVA’s reported Q2 revenue was down -9.7% year-over-year in the quarterly history, while MediaTek’s latest reported revenue YoY was -2.7%. That gap matters because CEVA should not be granted a premium IP-model narrative if its licensing cadence cannot keep pace with customers’ own semiconductor cycles. The variant call is that CEVA does not need a broad semiconductor upcycle to work from here; it needs licensing to stop missing consensus and royalties to convert the cellular IoT and Wi-Fi unit strength into dollars.

Management’s capital allocation slightly supports that view, but it is not a substitute for revenue delivery. CEVA repurchased 300,000 shares for approximately $6.2 million in the quarter, which is meaningful relative to a company producing $1.8 million of non-GAAP net income. The buyback says management sees value at current levels, but the better interpretation is defensive confidence rather than offensive acceleration. If the licensing pipeline were already visibly converting, the revenue guide would likely carry a cleaner statement than $26 million to $30. The wide range leaves room for another timing miss, and the market is right to demand evidence before paying for the royalty recovery.

The call delivery reinforces that mixed but improving setup. The tone history shows Q2 FY2025 sentiment at 0.42, up from 0.31 in Q1 FY2025, while guidance_tone jumped to 0.49 from -0.03. That is not just a mood shift; it aligns with a quarter where EPS beat despite revenue missing and where management guided revenue back into the $26 million to $30 range. But tone_confidence fell to 0.44 from 0.58, which is why the improvement should not be treated as all-clear. The prepared message is more constructive, while the lower confidence score says the transcript still carries caution around execution and timing.

The language beneath that tone score matters because management committed to near-term operating inputs rather than vague optimism. Arieli guided Q3 total revenue “between $26 million to $30,” and paired that with non-GAAP OpEx “in the range of $21 million to $22 million.” Those are the two constraints that define the next quarter’s operating leverage. If revenue lands near the low end, OpEx discipline alone will not create much upside; if revenue lands closer to the upper end, the same cost base can show whether the Q2 EPS beat was an early signal. The most important hedge in the call is not demand commentary but the width of the guide, because a $26 million low end is only modestly above Q2’s $25.7 million company-reported revenue, while the $30 high end would mark a clear licensing recovery.

The market’s likely mistake is treating the -8.3% revenue miss as the only clean signal and the +40.0% EPS surprise as lower quality. Some of that skepticism is fair because GAAP net loss was $3.7 million and diluted loss per share was $0.15, worse than the net loss of $0.3 million and diluted loss per share of $0.01 in the same period last year. But dismissing the entire EPS beat ignores the royalty unit data and the Q3 margin guide. This is a stock where the sign of the next revision depends less on whether investors like Q2 and more on whether they believe the 66% cellular IoT shipment growth and 80% Wi-Fi shipment growth can outrun Bluetooth softness and licensing lumpiness. The print did not prove that thesis, but it made it more plausible than the revenue miss alone suggests.

The risk to the thesis is that gross margin and OpEx are leaving little room for another licensing stumble. Q2 GAAP gross margin was 86%, and management said it was 90% a year ago on the same basis. GAAP OpEx is expected to remain in a $26 million to $27 million range for Q3, so a revenue print near $26 million would keep the company operating close to breakeven pressure on a GAAP basis. That is why the EPS beat should not be annualized. The correct action is to separate the revenue timing issue from the royalty signal, then demand confirmation on the Q3 bridge. If CEVA misses the low end of revenue guidance or fails to deliver the promised 87% GAAP gross margin, the argument that Q2 was a licensing-timing head fake breaks quickly.

What to watch next is therefore concrete. For Q3, the thesis requires revenue inside the $26 million to $30 guide, with more weight on whether it clears the midpoint implied by management’s range than on another penny of EPS. Gross margin needs to move to approximately 87% on GAAP and 88% on non-GAAP, because management explicitly guided a 1% lift from Q2. Non-GAAP OpEx must stay in the $21 million to $22 million band, since Q2’s $21.6 million already sat just above the high end of prior guidance. On royalties, the confirmation would be continued sequential progress from the $10.7 million base and no reversal in the categories that carried Q2, especially cellular IoT at 66 million units and Wi-Fi at 62 million units. Break the thesis if Q3 revenue is at or below $26 million, if gross margin does not rise from Q2, or if royalty units show that the 16% sequential revenue increase was a one-quarter restock rather than a base recovery.