CEVA reported its second quarter on Monday, 10 August 2026, before the market opened. We are writing about it today.
Errors or inaccuracies are possible. Spotted something that looks wrong? Write to me and I will correct it.
What CEVA Does
CEVA does not make chips. It sells the blueprints other people build them from.
Its development centre is in Herzliya and its registered headquarters in Rockville, Maryland. The product is silicon intellectual property: digital signal processors, wireless connectivity blocks (Wi-Fi, Bluetooth, cellular), and AI accelerators for the edge
- chips that run a model on the device itself rather than in the cloud.
From that follows a revenue model with two very different legs:
- Licensing - a one-off payment when a customer buys the right to use a design. Large, lumpy, and non-recurring.
- Royalties - cents on every chip actually sold in the world, years after the signature. Small per unit, but this is the annuity.
That distinction is the whole report. A quarter carried by licensing is a quarter of signatures; a quarter carried by royalties is a quarter of shipments.
The Quarter
| The quarter | A year ago | |
|---|---|---|
| Total revenue | $29.0 million · up 13% | |
| Licensing and related | $18.2 million · up 21% | |
| Royalties | $10.8 million · up 1% | |
| GAAP gross margin | 87% | 86% |
| GAAP operating loss | $2.1 million | $4.5 million |
| GAAP net loss | $2.9 million | $3.7 million |
| GAAP diluted loss per share | $0.10 | $0.15 |
| Non-GAAP operating income | $3.1 million · 11% | $0.8 million · 3% |
| Non-GAAP net income | $2.3 million | $1.8 million |
| Non-GAAP diluted EPS | $0.08 | $0.07 |
Every line in that table improved.
Revenue rose 13%. Licensing rose 21% to its highest level in three years. The GAAP operating loss more than halved. And non-GAAP operating income nearly tripled, from $0.8 million to $3.1 million, with the margin rising from 3% to 11%.
That is operating leverage in its cleanest form: revenue rose 13%, and adjusted operating profit rose far more than that, because an IP company's cost base barely moves with sales.
The Signature That Is the Story
The central win of the quarter is the selection of the NeuPro-M NPU for next-generation custom AI silicon by what the company describes as a leading global AI and computing platform company.
The company does not name it, so we will not guess. But it does say something about the nature of the customer: a party that controls both the hardware and the operating system of its own platform.
Why that matters: a customer like that does not buy a single block. It embeds the component into its design for years, and the volume that follows is measured on an entirely different scale from an ordinary chip customer. This is precisely the case where licensing today is royalty pipeline tomorrow.
Also in the quarter: a high-volume US semiconductor company added to its portfolio a third-party chip based on CEVA's Wi-Fi 6 and Bluetooth Low Energy IP; an existing US customer expanded a licence from a single baseband component to a complete baseband processing subsystem; and CEVA launched RealSpace Elevate, extending its Microsoft-certified spatial audio technology into PC gaming.
The One Line That Should Bother You
Royalties rose 1%.
That is the easiest line to miss in a report that is otherwise all green. Licensing rose 21%; royalties rose one percent.
Why it matters: licensing is lumpy, one-off revenue. A company can sign ten agreements in one quarter and three in the next, and revenue jumps and falls without the underlying business changing at all. Royalties are the measure of how many chips carrying CEVA technology are actually being sold in the world.
And the defence of that line exists, and it belongs here: royalties rose 17% sequentially, which the company attributes to strong wireless connectivity shipments, the continued ramp of automotive AI programmes, and improving smartphone royalties. So the quarter-on-quarter direction is positive; it is the year-on-year comparison that stayed flat.
Trailing-twelve-month licensing revenue also rose 13% to $69.6 million - so this is not a single lucky quarter.
What the Release Does Not Contain
The second-quarter release contains no written numerical guidance for the coming quarter or the year.
In its forward-looking statements section the company refers to "our forecast of financial measures for the following quarter and 2026", and management held a conference call at 8:30 a.m. Eastern Time on the day of publication.
In other words, the outlook was delivered verbally on the call, not in the text of the release.
The Market Reaction, and a Gap the Release Does Not Explain
The stock fell 17.5% on Monday and closed at $31.92, against $38.67 on Friday.
This has to be said plainly, without dressing it up: the release published is better than its predecessor on almost every line, and the stock lost roughly a sixth of its value. The document I am relying on does not contain the explanation for that gap.
The plausible explanation is that something said on the conference call - guidance, run rate, or commentary on the mix between licensing and royalties - was weaker than the market had priced. But that is an assessment, not a data point, and I am not going to present it as fact. What is known with certainty: the written report is positive, and the price fell.
The Balance Sheet
Cash, short-term deposits and marketable securities: roughly $220.7 million ($44.3 million in cash and equivalents, plus $176.4 million in securities and deposits).
For a company of this size, that is a very substantial cushion relative to the scale of the accounting loss.
הזווית שלי
דעה אישית של אילן אברמוב - לא ייעוץ ולא המלצה
This is one of the most interesting reports of the week, precisely because it does not add up.
When a company reports 13% revenue growth, a 21% jump in licensing to a three-year high, a near tripling of adjusted operating profit and a narrower loss, and the stock falls 17.5% - something that mattered was not in the report.
And here I am being careful. I do not have the transcript of the conference call, and I am not going to invent guidance I did not hear. A number that has not been verified does not go in.
What can be said from the document itself is this: this quarter leaned on licensing, not royalties. And when the market prices an IP company, it treats royalties as the annuity and licensing as an order book - important, but lumpy and hard to extrapolate. A quarter where licensing rises 21% and royalties rise 1% is a quarter it is hard to draw a line through.
What does change the forward picture is the NeuPro-M signature. If a customer that controls both its hardware and its operating system has chosen to put CEVA's accelerator inside its own design, that is the kind of customer that turns licensing today into large royalties in two or three years. But that is exactly the gap: the engine of the thesis sits in the future, and the price trades in the present.
What I will watch is one line, and it is not revenue: royalties year over year. As long as it is flat, the story rests on signatures. The moment it starts compounding at a double-digit rate, this becomes an entirely different story - and today's licensing acquires meaning in retrospect.
