Good day, and welcome to the CEVA Inc. third quarter 2026 earnings conference call. All participants will be in listen-only mode. Should you need assistance, please signal a conference specialist by pressing the star key followed by zero. After today's presentation, there will be an opportunity to ask questions.
To ask a question, you may press star then one on your telephone keypad, and to withdraw your question, please press star then two. Please note, today's event is being recorded. I would now like to hand the conference over to Richard Kingston, Vice President, Market Intelligence and Investor and Public Relations. Please go ahead.
Thank you, Rocco. Good morning, everyone, and welcome to CEVA's third quarter 2026 earnings conference call. Joining me today are Amir Panush, Chief Executive Officer, and Yaniv Arieli, Chief Financial Officer.
Before handing the call over to Amir, I would like to remind everyone that today's discussion contains forward-looking statements that involve risks and uncertainties, as well as assumptions, that if they materialize or prove incorrect, could cause our results to differ materially from those expressed or implied by such statements.
Important factors that can cause actual results to differ include, but are not limited to, those stated in our earnings release issued earlier today and in our filings with the SEC. We will also discuss certain non-GAAP financial measures, which we believe provide investors with additional insight into our core operating performance.
Reconciliations between our GAAP and non-GAAP results are included in the earnings release issued this morning and available on the Investors section of our website. With that said, I will turn the call over to Amir. Amir?
Thank you, Richard, and good morning, everyone. During the first quarter, we made good progress across our business. AI licensing revenues reached a record, accounting for more than 35% of total licensing and related revenues. Our customers shipped 623 million CEVA-powered devices, matching our all-time quarterly record, and non-GAAP earnings per share improved.
Overall licensing revenues was lower than we anticipated, as a significant multi-year connectivity agreement with a key strategic customer that we expected to sign in the third quarter was delayed. It has since been signed early in the fourth quarter, and I continue to be very encouraged by our diversified licensing pipeline across Connect, Sense, and Infer. In royalties, we saw strong growth in mobile and broad-based IoT, with royalties up 18% sequentially, although memory market headwinds tempered volume across compute-intensive consumer devices such as smartphone, PCs, and TVs.
These developments reflect the continuous execution of our strategy to broaden our addressable markets, deepen our relationships with leading semiconductor companies, OEMs, and system players, and help our customers bring successful CEVA-powered products to market, which drives our future royalties. Turning now first to licensing, we signed 10 agreements during the quarter, including three with first-time customers.
AI-related deals were particularly strong, accounting for six of these agreements. We believe our record quarterly AI licensing revenues demonstrate that our investment in AI technologies is translating into meaningful business today and will drive long-term royalty growth. Building on this momentum, we recently introduced our expanded NeuPro family of AI processors, with new market-leading features and significant improvements in performance and energy efficiency. With the newest addition to the family, NeuPro Ultra, we are extending CEVA reach beyond the edge into AI infrastructure.
We expect this will significantly expand our addressable market as data center solutions providers seek greater efficiency across inference systems. The expanded portfolio builds on our existing base of more than 50 NeuPro customers and opens new opportunities in areas such as memory and connectivity infrastructure, where efficient AI processing is becoming critical.
We are already seeing customers' interest across these new applications. Including in the quarter, a new customer selected NeuPro-M for an AI infrastructure application focused on improving the efficiency of data centers' inference systems. Other edge AI agreements targeted consumer, automotive, and industrial markets. We also made further progress towards NPU royalty generation.
As highlighted in our November 2nd announcement, our customer, Mobilint, recently launched its REGULUS AI SoC, which incorporates both our NeuPro-Nano NPU and SensPro AI DSP technologies in the same SoC, illustrating how adoptions of multiple CEVA technology can expand our royalty opportunity per device. Several other NeuPro customers completed chip tape-outs during the third quarter, and we expect initial NeuPro royalties around the end of this year or early 2027.
Turning to Connect, our wireless connectivity business continues to benefit from the growing need for multiple connectivity technologies in increasingly intelligent devices. Our leadership in this market was recently reaffirmed by IPnest, which ranked CEVA number one in wireless interface IP for the third successive year, with approximately 68% market share by revenue. During the quarter, three of our highest volume connectivity customers extended their longstanding agreements with CEVA for multiple years and added more CEVA IP.
One licensed our Wi-Fi 6 and Wi-Fi 7 technologies, complementing our Bluetooth and DSP IP it already ships in high volume. Combining connectivity technologies within the same customer is a core part of our strategy, and more than 30 customers now license both Bluetooth and Wi-Fi from us, increasing our footprint within their products and our royalty content per device. In Sense, we secured a new agreement for our MotionEngine Hex technology to enable advanced motion-based interaction for smart TVs. Our customer is targeting one of the world's largest television brands.
Together with last year's agreements with a leading smart TV platform provider to integrate MotionEngine into its operating system, this extends our presence across both the platform and device levels of the smart TV ecosystem, enabling more intuitive and precise user experiences as televisions evolve into increasingly interactive entertainment and gaming platforms.
Turning now to royalties, revenue increased 3% year over year on shipments of 623 million CEVA-powered devices, matching our all-time quarterly record. Our IoT units volume reflect the scale and penetration of our connectivity business, with strong Bluetooth volumes, record cellular IoT shipments, and record Wi-Fi 6 unit shipments, which carry higher royalty value.
This strength was partly offset by weakness in the compute-intensive consumer devices, such as smartphones, PCs, and TVs, where memory market headwinds weighed on demand. A key development was continued ramp of internally developed modem at a major U.S. customer, contributing to growth in mobile royalties. This is the first time our customers has moved its internally developed modems into high-volume production across its mainstream flagship smartphone models, an important step for CEVA in this market.
We expect this volume ramp to accelerate significantly in the fourth quarter, supported by a full quarter of device shipments across multiple models, and to continue into 2027 and beyond. We believe this, together with continued share gains in broad-based IoT, will support future royalty growth. Yaniv will provide more detail on the royalty results in his financials review. In summary, the first quarter showed progress across each part of our strategy.
AI is becoming a meaningful licensing business. Our connectivity franchise continues to scale across IoT, and a new smartphone royalty cycle is underway. At the same time, three of our highest volume connectivity customers added more CEVA technologies and extended their relationships with us well into the future, supporting a continued increase in our average selling price per design. With that, I will turn the call over to Yaniv to review our financials results.
Thank you, Amir. Good morning, everyone. I will now review our financial results for the third quarter. Revenue for the third quarter increased 2% year over year to $29.1 million, supported by 18% sequential royalty growth of shipments of 623 million CEVA-powered devices, matching our all-time quarterly record. The revenue and other financial breakdowns are as follows. Unless otherwise stated, all comparisons are to the third quarter of 2025.
Licensing and related revenues increased 2% to $16.3 million, representing approximately 56% of our total revenues. AI licensing reached a quarterly record and accounted for more than 35% of our total licensing and related revenues, reflecting the growing commercial adoption of our AI portfolio and expanding our future royalty opportunity. Royalty revenues increased 3% year over year and 18% sequentially to $12.7 million, representing approximately 44% of total revenues.
Growth was led by the continued ramp of internally developed modems at a major U.S. customer and record royalty revenues related to our wireless IoT business lines. GAAP gross margin was 85% compared to 88%. Non-GAAP gross margin was 86% compared to 89%. GAAP operating expenses were $28.3 million, at the low end of our guidance, reflecting expense control.
Non-GAAP operating expenses, excluding equity-based compensation expenses, amortization of acquired intangibles, and acquisition-related costs, were $22.5 million, also at the low end of our guidance. GAAP operating loss was $3.5 million compared to GAAP operating loss of $2.1 million. Non-GAAP operating income was $2.5 million compared to $3.1 million. Non-GAAP operating margin was 9%. Net financial income was $2.3 million compared with $1.2 million, attributed to better hedging.
Income taxes expense was approximately $1.1 million compared to $1.7 million, with a decrease mainly due to lower withholding taxes. GAAP net loss was $2.3 million, or $0.08 per diluted share, compared to GAAP net loss of $2.5 million, or $0.10 per diluted share. Non-GAAP net income increased 39% to $3.8 million, and non-GAAP diluted earnings per share increased 18% to $0.13. Non-GAAP net income increased 67% sequentially from $2.3 million, and non-GAAP diluted earnings per share increased by 63% from $0.08.
With respect to other related data, during the quarter, customers shipped 623 million CEVA-powered devices, an increase of 8% compared to the third quarter of 2025 and 10% sequentially. Of those shipments, mobile shipments reached 82 million units, or approximately 13% of the total, an increase of 19% year-over-year and 34% sequentially.
Mobile royalty revenues increased 20% year-over-year and 30% sequentially. Consumer IoT shipments increased 4% to 518 million units. Industrial IoT shipments increased 120% to 23 million units. Looking at our connectivity technologies, these shipment metrics demonstrate the breadth of our royalty base across multiple end markets.
Bluetooth shipments increased 9% to 329 million units. Cellular IoT shipments reached a new quarterly record of 69 million units, up 2%. Wi-Fi shipments were 78 million units, down 6%. We saw a significant increase in royalty contribution from higher-value Wi-Fi 6 products, reflecting the improving technology mix of our customers' deployments, with Wi-Fi 6 representing 57% of Wi-Fi unit shipments, up from 24%.
As for the balance sheet items, at the end of the quarter, with approximately $221 million in cash equivalents, marketable securities, and bank deposits, providing significant financial flexibility to support continued investments in our technology roadmap while maintaining a disciplined approach to capital allocation, including selective strategic M&A opportunities that can accelerate our growth. Days sales outstanding were 56 days, lower than the 70 days in the prior quarter. During the quarter, we generated $1.5 million of cash from operating activities.
Depreciation and amortization expenses were $0.8 million, while capital expenditures totaled $1.1 million. At the end of the quarter, we employed 394 people, including 315 engineers, reflecting the continued investments in innovation while maintaining disciplined expense management. Turning to our outlook. In the first nine months of 2026, licensing and related revenues increased 14% year-over-year, led by strong demand for AI technologies.
Our pipeline is robust, with healthy deal flow across Connect, Sense, and Infer heading into 2027. In royalties, customers shipped CEVA-powered devices at a level matching our all-time quarterly record, reflecting the scale of our IoT connectivity business despite memory market headwinds in compute-sensitive consumer markets. Smartphone modems add a multi-year growth driver for us through 2027 and beyond. We expect a strong fourth quarter, with approximately 24% sequential revenue growth at the midpoint of our guidance.
This would represent the highest quarterly revenues in CEVA's history, and we expect non-GAAP diluted EPS to more than double sequentially. Our fourth quarter outlook is based upon expected step-up in smartphone modem royalties at a major U.S. customer, typical fourth quarter seasonality, and our robust licensing pipeline. It also reflects a disciplined assessment of the opportunities we expect to convert in this quarter.
At the midpoint of our fourth quarter guidance implies full year 2026 revenue growth of approximately 11%. We will continue to invest in our roadmap while carefully managing costs and mitigating foreign exchange headwinds, and now expect full year non-GAAP expense growth, which is the cost of revenues and the OpEx, of approximately 7% or $7 million, which is below our previous expectations. Fourth quarter guidance.
Revenues are expected to be in the range of $34 million-$38 million. Gross margin is expected to be approximately 88% on GAAP basis and 89% on non-GAAP basis, excluding approximately $0.2 million of equity-based compensation expense and $0.1 million of amortization of acquired intangibles.
GAAP operating expenses are expected to be between $27.6 million and $28.6 million, including approximately $4.3 million of equity-based compensation expenses and $0.1 million of amortization of acquired intangibles, and $0.1 million for acquisition-related costs. Non-GAAP operating expenses are expected to be between $23.1 million and $24.1 million.
Net financial income expected to be approximately $2 million. Income tax expense is expected to be approximately $1.8 million, and our weighted average diluted share count is expected to be approximately 30.1 million shares on GAAP and non-GAAP basis. Rocco, we are ready to take questions now.
Thank you. We will now begin the question-and-answer session. To ask a question, you may press star then one on your telephone keypad. If you are using a speakerphone, please pick up your handset before pressing the keys. If at any time your question has been addressed and you would like to withdraw your question, please press star then two. Today's first question comes from Kevin Cassidy at Rosenblatt Securities. Please go ahead.
Yes, thanks for taking my question, and congratulations on the results. But maybe just to look at the mobile market, I know you have the U.S. customer that has started shipping, but how about the rest of the market? Can you give us a feel for how the mobile market is, and what are your customers saying about unit shipments this quarter and even into next year?
Yeah, thanks for the question, Kevin. This is Amir. There are several things that we need to keep in mind as we look at the mobile market. I will start first, as you mentioned, with the U.S. customers. We expect a very significant ramp in Q4 and as we go into 2027. Definitely in this quarter, Q3, it was only just a very short period of time within the quarter, within the end of the quarter, so the contribution in this quarter was quite minimal comparing to what we expect in Q4 and next year.
As we look at the overall smartphone market, as we also commented in the script, overall, this market, I would say also the so-called what we call compute-intensive consumer market, whether it's so-called the TV, PCs, and mobile, that market is under pressure because of the price increase in the market.
If we compare this year versus last year, definitely we have seen decline in the overall so-called addressable market, the size of the market. Having said that, and we also mentioned that in the previous quarter, we are seeing our low-tier customer that supplies our technology, while they started the year very, very soft in Q1, we really see a very strong sequential growth as we go through the year. So Q2 was much stronger, now Q3 is stronger.
While the overall market is soft, we are seeing them actually also gaining market share. So net year-over-year, it's still lower than last year, but for us at least, the impact is much less because the low-tier customer is actually gaining share, going quarter-over-quarter. Specifically with the U.S. customers, we are looking for a very significant ramp as we go to Q4 and beyond.
Kevin, maybe I would add one more thing, that if you look at the mobile royalty revenues, with all the headwinds and the difficulties, they grew 20% dollar-wise for us year-over-year, and 30% sequential. Still we're seeing higher, more four and 5G-type phones versus two and 3G phones a year ago, which bears higher ASPs, and 20% year-over-year. It doesn't yet, before the big U.S. guy starts to kick in with much higher volume and dollars, it's still a very solid contribution for mobile this quarter.
Okay, great. Thank you. Maybe as you're entering into the AI infrastructure market with the NeuPro, is there going to be a difference in time to market or time to royalty in the AI infrastructure versus AI edge?
Yeah, thanks for the question, Kevin. First, we are super excited how we are doing in the AI market overall with our NeuPro product line. As we mentioned, now we have more than 15 customers that already licensed the technology. Based on that success on the edge side, in terms of the performance and what we can support the customer base, what we've seen is actually a very good capability to expand it into the infrastructure.
Within the infrastructure market, we see actually two types of customers adopting our technology. One is more the customers that want so-called to internalize those capabilities and develop their own NPU ASIC, and they don't want to start from scratch. It can be OEMs or semiconductor customers.
We also see, whether it's the memory supplier or the so-called the high-speed wire connectivity suppliers, looking to add NPU so-called on their own edge within the infrastructure to accelerate the performance. Our technology that really can customize and scale up and down, a great fit in providing extremely good performance to run inference on the infrastructure.
In terms of time to market, to your question, actually it's quite similar. Overall, AI is moving very, very rapidly, so there is very strong push and demands to get those products to market as quickly as possible. This is not longer than what we have seen on the edge. It's our typical 18 to 24 months to get to royalty.
The other thing that I will mention on the infrastructure, actually we already started licensing this new technology last quarter, and this quarter we got another customer that's licensing the technology. While we just came and announced it more publicly this quarter, this technology we have been and it's already starting to get integrated by several customers.
Okay, great. Thank you.
Thanks, Kevin.
Thank you. Our next question today comes from Suji Desilva at Roth Capital. Please go ahead.
Hi, Amir, hi, Yaniv. On the Q4 licensing versus Q3, I know there's a large deal signing moving into Q4. Can you give us a sense of that magnitude and the impact there? Then the run rate into 2027, what's the normalized rate through the movement of that deal?
Yeah, it's an important deal for us. It's an existing customer, a multi-use type of technology, a significant deal. It's a bit of a timing issue of not being able to close it in the third quarter, but it was already closed in the beginning of the fourth quarter. We saw a pretty nice run rate in the first two quarters of the year, about $18 million level. We're confident with these types of levels.
We're adding more technology, more offerings. We have been able to license new technologies to existing customers this quarter. So, for the time being, we are very comfortable at these higher levels than last year, and we are looking at a nice growth year-over-year with the licensing, which will give us strong opportunities to increase royalties in years to come.
Yeah, Suji, I may add a few other data points here. So if we look at the first three quarters of this year, licensing is 14% higher than the previous year. So overall, licensing has been strong and good. Unfortunately, a meaningful deal here, as Yaniv mentioned, with existing customers that they pushed to Q4, which we already signed. This deal, by the way, is a great deal. It is with existing customer that will expand and adding additional connectivity technology to their portfolio.
So we are very excited about the deal that got signed. As well as overall, if we look at backlog of Q4 and the deals that we have in the pipeline, we are very confident with our ability to go and deliver in Q4 and actually finish the year very nicely on the licensing overall.
Okay, great. Thanks for that, Amir and Yaniv. My second question, maybe, following up on Kevin's question, you answered 18, typical time to royalties of 18 to 24 months, even for the AI NeuPro infrastructure even. Based on the wins and the pipeline for that AI infrastructure and NeuPro in general, when may be the elbow or the kind of the inflection point in the royalties, like from here looking forward? Just to understand what the timeframe to that might be based on your visibility here.
Yeah, so we mentioned also on the prepared remarks that we are expecting some of the royalties that probably will start in the coming Q4, but the more ramp will start through 2027 and then even further in 2028. Another data point that I can share is the large PC customer that we mentioned previously, actually, we have taped out the device with them.
So we really are very, very encouraged with the execution and how that is moving forward. We have now multiple customers that so-called achieved the tape-out, and they are targeting to ramp within 2027. So overall, very, very encouraging signs on that front.
Great. Appreciate that color, man. Thanks, guys.
Thank you. Our next question today comes from Joseph Cardoso at JPMorgan. Please go ahead.
Hi, good morning. Thanks for the question. Maybe for the first one, Wi-Fi units softening again this quarter on a sequential basis, and we're just curious if you could help us think through the ramp here going forward, particularly as you highlighted Wi-Fi 6 record unit or record quarter here. Maybe just help us think about how do we think about the crossover of Wi-Fi 6 and Wi-Fi 5 or the legacy Wi-Fi products rolling off as we think about going forward. Then I do have a quick follow-up. Thank you.
Sure. Happy to add some color here. Wi-Fi shipments this quarter were 78 million units, not the highest. It was down 6% from a volume perspective year-over-year. But we mentioned the very interesting data point, which we have been talking about for a while now, and that's replacing the existing Wi-Fi 4 with Wi-Fi 6 technology.
We have said for a while that the ASPs and the technology is more expensive and much higher R&D for us, and we have doubled the volume of Wi-Fi 6 for north of 50% of that volume, up from a year ago, 24%. What that means is that the royalty contribution on those 78 million devices more than doubled year-over-year while the units went down.
That is exactly why the way to achieve that is to replace with the newer Wi-Fi technologies, and we are seeing it very strongly this quarter. As soon as that continues and Wi-Fi units continue to increase, that will have a double impact on the overall revenue and the dollars we collect.
Maybe I will add a few more data points. First, I would say that we have really established extremely strong leadership in the marketplace with our wireless connectivity, both Bluetooth and Wi-Fi technology. If we look at Wi-Fi, now we have more than 30 customers that already now have licensed both the Wi-Fi and Bluetooth technology from us, so-called combo technology. And within Wi-Fi 6, we now have more than 40 customers that already licensed our Wi-Fi 6.
So really, quarter by quarter, we keep seeing more customers ramping now their products into volume shipments and then keep growing that volume shipment. So we expect the Wi-Fi volume shipments to grow significantly, while, as Yaniv pointed out, also the transition from Wi-Fi 4 to Wi-Fi 6 will come in a very, very strong phase, as we already now have seen in this quarter. We are very encouraged about that and the trajectory as it is going to move forward.
Thank you for the color, both there. As my follow-up, it sounds like the scope of opportunities are expanding for you guys. I think now this quarter you also talked about television program that is in the pipeline, but simultaneously, engineering headcount continues to trend lower. Can you maybe just double-click on this dynamic? What is driving it, and how should we think about the sustainability of that going forward and implications to the cost structure here? Thank you.
Let me start with the cost structure maybe. The biggest hit for us in 2026 was unfortunately the dollar weakening against the euro and the shekel, which are the two main currencies, which probably about 300 or 400 engineers are located at. And that was about a 20% hit overall year-over-year. So that is something that is the major expense item on our books this year.
And even that, we have managed to decrease a bit and state that if in the prior guidance, the annual guidance, we talked about $8 million or 8%, now we have tuned that down to about 7% overall expense increase, which is about $7 million. So from a number perspective, hopefully that headwind is behind us, and the comparison for next year will not be as hard as it was of 2025 versus 2026 from just the FX perspective.
Headcount fluctuates up or down every quarter, but we are still investing a lot in R&D, especially the AI, and also in new markets like the RF add-on to Bluetooth and then to Wi-Fi technology. In some places, we are trying to be a bit more lean, and on others, we are continuing to invest heavily. Amir, maybe you want to add there?
Yeah. What I would say is that first, from an investment profile and how we look at the business, there are three things we want to make sure that we are going to execute well, and that is our very strong growth driver, and we will keep investing in R&D. One, of course, is the AI and now the expansion into infrastructure. The second is overall our leadership in wireless connectivity, and then driving solutions with the RF and overall complete software and hardware.
All those things we have been able pretty much to execute within the baseline of the headcount that we have had. As Yaniv mentioned, specifically in 2026 versus 2025, we had the headwinds of the exchange rate. For the most part, this is probably behind us as a comparison to next year. We are well set basically to keep investing in R&D while, at the end of the day, driving improved profitability as a top-line growth.
Thank you. Our next question today comes from Josh Buchalter at TD Cowen. Please go ahead.
Hey, guys. Thank you for taking my question. Maybe to start, it sounds like there's a good deal of renewed optimism in the smartphone market, specifically given what's going on with the modem share. Could you maybe speak to your philosophy on this business? Is this kind of in cash cow mode where you're monetizing previous investments, or are you guys investing to continue to expand opportunities specifically in the smartphone modem market? Thank you.
Yeah, great question, Josh. I would say first, strategically for the company, as also we have indicated previously, our strategy is diversification. Diversification of our product portfolio as well as market. We are not targeting mobile as a specific market to so-called tailor our IP and customize just for that. We are building a broad-based IP in order to get it into the different markets.
Specifically for mobile, there are two trends that definitely we would like to see our technology so-called supporting that, and we see very good opportunities moving forward. One is all the things related to the convergence between satellite communication and the cellular communication, so-called 5G-enhanced, the 3GPP, plus satellites.
We mentioned already in the past design wins that we have in that perspective, and we are offering our PentaG solution as a complete software and hardware offering that our customers can take into the satellite communication as well as the cellular communication as a single solution. The next is, of course, 6G that is coming in the coming few years. Our technology is developed such that it is upgradable for that, and we are supporting the next generation of this technology.
This is another opportunity for us as we move forward. But overall, we'll continue developing our IP, and not just for mobile, not just mobile as a target market, but as a broad-based IP offering. And we are very happy to see that inflection coming right now, just this quarter, and really ramping in Q4 and beyond. This is a great ramp that we will see as we go through 2027 and beyond.
Makes sense. Thank you for the color there. I wanted to ask about the 2026 guidance. I realize there was a timing issue related to the licensing deal that flipped into the fourth quarter, but the magnitude of that change, I guess I was wondering why the annual guidance came down from your previous outlook, or was that just mainly memory pricing impacting overall royalty volumes in your key end markets? Thank you.
Yeah, great question. I will give a little bit colors here. As we look at Q4, how we guided Q4, right? First, the tailwinds or the thing that we are very encouraged by, of course, is the mobile ramp of the customer. This is seasonality overall that typically Q4 is very strong, as well as our backlog and the pipeline that is, we are very encouraged to see that it is a very strong one for Q4. Having said that, as you pointed out, overall, if we look at the market, all the compute-intensive markets are under stress right now. There is some pressure there, whether it is smartphone overall, TVs, or PCs.
As we mix all those things together, a disciplined assessment of how we look at the quarter, also considering that this quarter, at the end of the day, we have not executed to our expectation, we came with that guidance. Overall, we are very encouraged about how the quarter should look like, but this is a disciplined assessment based on all those inputs.
Obviously, with that said, we are still guiding to the best licensing revenue and royalty revenue ever for CEVA, which is the $34 million-$38 million, which is a very nice new place to be at, if all goes well.
Yeah, just to clarify, the total revenue, the Q4 guided at the midpoint is the highest quarter ever for CEVA.
Thank you both.
Thank you.
Our next question today comes from Gary Mobley at StoneX. Please go ahead.
Hi, everybody. Thanks for taking my question. I wanted to ask about NeuPro Ultra, in some of the early days traction into the AI data center infrastructure market. Essentially, what I want to know is, who is the customer there, or the target customer, and how does it compare to Arm or x86 CPUs for agentic or inference workloads, or even GPUs for that matter?
Yeah. Thanks, Gary, for the question. First, just to clarify, this is not so-called to replace the very high-performance GPU that runs, let us say, training on the cloud. This is really to support acceleration and power efficiency optimization of doing inference on the cloud. There are two different type of customers that are looking to license such technology from us.
They are the OEM or the Tier 1s, so-called the semiconductor in that space, that wants to drive their own accelerators development internally, and they do not want to start from scratch, and they are coming to us to license our NPU architecture. On top of that, they build so-called a complete hardware-software stack to optimize their performance. That can go alongside GPUs, or CPUs, in the overall system. Then within those racks or systems, there are definitely the memory suppliers and the high-speed connectivity suppliers.
What everyone sees is that those are the bottlenecks in the system, whether you access to the memory or the interconnectivity. Having accelerators on that specific interface or memory, you can significantly improve the efficiency on the system, reduce the latency, and increase the overall power efficiency. Those type of customers also looking to have NPU or that type of capabilities, accelerators, and of course, they do not have internally, and they are coming to us to add it into their systems.
Helpful, Amir. My follow-up was to ask about the progress on the automotive front. I know you've got a number of automotive processor licenses there, and I believe it's very centered on NeuPro, so maybe if you can give us an update or some sort of outlook on when the initial automotive royalties might come to fruition.
Yeah. I think in a quarter or two quarters ago, we mentioned that our AI DSP technology is now ramping in volume productions in automotive markets with both Renesas and NXP. For example, the RV What's the name of the Toyota car?
The RAV4.
Yeah, the RAV4 Toyota car, which is in very high volume right now in the U.S. That car, in their platform, basically using our AI DSP technology. This is right now already so-called in volume ramp production, and we see the royalty coming, and it's growing quarter after quarter. In addition, several of the NPUs designed that we announced with Mobileye and others are basically going to production within this year.
Then we will see more ramping in 2027 with the different design wins that we had. Interestingly, also, we see overall in the automotive a request for very high-performance type of multi-core configuration, and our NPU with the NeuPro Ultra capabilities is a great fit for that. Thanks, Gary.
Thanks, Gary.
Our next question today comes from Charles Shi at Needham & Company. Please go ahead.
Hi. Thanks for taking my question. Maybe a follow-up on NeuPro Ultra. You mentioned the two types of customers, and at least the second type, memory suppliers, connectivity suppliers. It sounds like a DPU or data processing unit type of application, or you may call it the SmartNIC, but wondering if that's the case, and is this more like NPU attached to DPU type of application, if you can clarify.
On the first type of customers, you just announced OEM, Tier 1. I'm still a little bit confused at what exactly this product is for. You said it's not the CPU, it's not GPU. It may work along with those, but what exactly is that application, if you can clarify? Thank you.
AI.
Yeah. Thanks, Charles. As for the first question, I will not go into the specific application, whether it is a DPU or something similar, but it is related to attaching to either the memory subsystem or the wire connectivity subsystem. The thinking there is basically to accelerate the decode part of that system, as basically the data moves between the different setups within the whole chipset. So that is basically where we are supporting there. In terms of what was the second question? I could not follow this.
No, no. The first type of customers. You mentioned that there are two types of customers, right? The first type is OEM, the Tier 1 semiconductor companies, driving their own accelerator internally, and they are looking at NeuPro Ultra. You already addressed the second type of customer, memory suppliers, connectivity suppliers. But I would like to have a little bit more color on the first type of customers. What exactly is the application of NeuPro Ultra? If you can provide some simple mental model for us to think about what this is for, and we have a better way to think about the future opportunity. That is it.
Yeah. So we see the Ultra actually fits in two places. Here we are talking about hundreds to thousands of TOPS, top of implementation, where you are running multi-core, multi-user, multi-batching configuration, that can either be cloud inference, where you want to be further to the GPU to have much more optimized, more efficient power NPU to accelerate some of the specific models that you want to run on the cloud inference.
As well as some of the edge applications like automotive, as you move to really higher level of autonomy, all the ways to later on to autonomous driving, and you want that level of capabilities to run those models with very low latency.
Thank you. That will be all from me. Thank you.
You are welcome, Charles.
Thank you.
Thank you.
Thank you. Our next question today comes from Martin Yang at Oppenheimer. Please go ahead.
Hi. Thank you for taking my questions. I want to ask about the third quarter licensing revenue growth. When you look at the breakdown of it, how much are from renewals or expansions among existing customers? Or maybe give the similar color regarding the first nine months of this year. Thank you.
Yeah. A big portion of our customers are recurring customers. They come back to license either newer generations of, especially in the Connect field, whether it is new Wi-Fi or Bluetooth standard that come out and are introduced every year to two years. Same applies for handsets and that type of standards. A newer type of existing customers have started in the last five years or so to come and license multiple types of technologies.
We have seen many customers license combo Bluetooth and Wi-Fi. We are seeing recently add-ons of RF technologies to those digital modems, and that is a newer type of customers. Now we are seeing that same type of existing customers come also and utilize different AI configurations and technologies for their development.
I would say that every quarter, if we sign anywhere between 10 to a dozen, 14 deals, you will see anywhere between one to three, four new customers that join us and are first-time customers. The rest of them are existing customers, but in many, many cases, to newer technologies.
That has been the sticking point and the strategy that we have taken over the last couple of years to be able to offer those different players, whether it is the Sense, Connect, or Infer, different technologies to help them out time-wise and the cost of development when we could offer it as a one-stop shop.
Yeah. Maybe I will add a little bit more color. What we see with For example, this quarter, we announced 10 deals out of that three new customers, but the majority of the existing customers, they are either coming to us to purchase new technology that they have not. Like, for example, one of our customers this quarter, they used to be a DSP customer, then they added last year Bluetooth, and this year now they added this quarter Wi-Fi into the basket of technology that they are using from us.
As well as some of the customers are coming to basically get a newer, upgraded, better technology within the same so-called product line. So very few are so-called pure renewal. Almost most cases are about getting better, newer technology or additional technology.
Thanks for that color. A follow-up on that comment is, when you look at the new licensees for NeuPro, do you see them as incremental or new customer base separate from your connectivity and other solutions, or do you already see a big overlap between NeuPro licensees and your existing licensee base?
Yeah. Very good question, Martin. If we talk about market that we are less present in, like for example, what we talk about the infrastructure, where we talk about new expansion in the high-end automotive market and so on, that we then use in many times our new customers licensing our NPU technology. But as we look more into the broad-based consumer market, we are so heavily penetrated with our connectivity technology that in many cases, having NPU is an addition on top of the wireless technology that we are already providing.
Yeah. If you look at the deal count, out of 10 deals this quarter, six were for AI. Out of those six, we had few that were new first-time customers, and all the rest are first-time AI customers, but based on existing customers that use completely different connectivity or other type of solutions from us. In a sense, maybe that same OEM or the same semi guy has the same name, but the application that they are using is completely new and could be perceived as a new customer for that add-on or new technology that we are offering.
Thank you very much. That is it for me.
Thank you, Martin.
Thank you. That concludes our question and answer session. I'd like to turn the conference back over to Amir Panush for any closing remarks.
Thank you. Before we close, I would like to put this quarter in the context of our long-term strategy. CEVA's business is built on turning today's licensing design wins into tomorrow's royalty streams. AI licensing revenue reached a record this quarter, with initial NPU royalties approaching, and our customers' modem transition is expanding our participation in smartphones.
Our strategy remains consistent, leveraging the breadth of our portfolio, selling multiple technologies into each customer, and moving towards more complete solutions offering that increase our value and royalty content per device. Together with our leadership in connectivity, this gives us multiple drivers of growth and confidence in the opportunity ahead. Richard, back to you.
Thanks, Amir. As a reminder, the prepared remarks for this conference call are accessible through the investors section of our website. With regards to upcoming conferences, we will be participating in the following upcoming conferences: the 15th Annual Roth Technology Conference, November 18th in New York; UBS Global Technology and AI Conference.
We'll be there December 3rd in Scottsdale, Arizona; and StoneX 15th Annual Discovery 1-on-1 Conference, December 10th in New York. For further information on these events and all events we will be participating in can be found on the investors section of our website. Thank you and goodbye.
Thank you. That concludes today's conference call. We thank you all for attending today's presentation. You may now disconnect your lines and have a wonderful day.