NASDAQ / Last 4 quarters

CEVA earnings call analysis

CEVA, Inc.. AI-assisted transcript summaries focused on management tone, evasions, goalpost moving, catalysts, risks, and data-center exposure.

4 storedJun 10, 2026

Research summary and source transcript

readyJun 10, 2026

CEVA reported a strong start to FY2026 Q1 with 11% year-over-year revenue growth to $27 million, driven by an 18% increase in licensing and related revenue to $17.8 million, its strongest licensing quarter in three years. Management highlighted strategic wins in Bluetooth HDT, 5G/NTN, UWB, and AI, emphasizing a shift toward integrated, full-stack solutions that increase value per design and long-term royalty potential. While royalties were flat year-over-year due to seasonal mobile softness, non-mobile royalties grew 8%, reflecting strength in IoT, industrial, and AI-driven applications. The company upgraded its full-year revenue growth outlook to the top end of its 8%-12% range and raised non-GAAP operating income and net income expectations by 40%-50% year-over-year.

Management knows today that the Bluetooth HDT win with a leading US-based semiconductor company represents a foundational capability for the upcoming Bluetooth 7 standard and is part of a broader pattern of expanding integrated system-level engagements with existing customers, which increases value per design and enables multi-generation royalty streams. This shift from component licensing to full-stack solutions (including internally developed RF, modem, and software) is not yet reflected in market expectations, as the full royalty ramp from these engagements will take 12-24 months to materialize in production volumes, particularly in automotive and edge AI applications like the Toyota RAV4 deployment and NXP collaboration. The market likely underestimates the long-term margin expansion potential from combo chips and higher ASP product mixes in industrial and automotive IoT, where associated royalty revenues grew 19% despite unit declines.

Licensing execution, value-per-design expansion through integrated solutions, and royalty ramp from prior engagements.

  • Shift toward integrated, full-stack solutions (connectivity, AI, audio)
  • Expansion of engagements with existing customers across multiple technologies
  • Growth in non-mobile royalties and smart edge markets (IoT, industrial, automotive)
  • Momentum in Wi-Fi 6/7 and Bluetooth 6/7 adoption and combo chip traction
  • AI at the edge as a growing opportunity, with AI representing >20% of licensing revenue
  • Disciplined capital allocation and selective M&A focus on complementary IP
  • Bluetooth HDT win as a 'foundational capability for Bluetooth 7' and 'strongest licensing quarter in three years'
  • Renaissance R-car V4H platform in production in 2026 Toyota RAV4 as 'first mass volume automotive AI deployment'
  • Newport Nano NPU winning a leading AI award at Embedded World 2026
  • Wi-Fi shipments reaching an 'all-time high' of 91 million units, up 158% year-over-year
  • Bluetooth Wi-Fi combo chip volumes doubling year-over-year

Management exhibited a confident, direct, and credible tone throughout the call, providing specific details on wins, technical differentiators (e.g., internal RF development, TSMC 12nm Lynx 200), and measurable outcomes (e.g., shipment volumes, revenue splits). Executives avoided vague optimism, instead grounding excitement in concrete engagements, customer expansions, and production milestones (e.g., Toyota RAV4, NXP collaboration). When questioned about risks like memory pricing or M&A, they acknowledged challenges but offered logical, evidence-based reasoning (e.g., IoT diversification, historical second-half volume trends) without evasion or overpromising. The tone reflected operational discipline and strategic clarity, particularly in explaining non-GAAP adjustments and guidance upgrades.

  • No clear dodged analyst question was detected by the local fallback; manual review should still check whether Q&A answers quantified conversion, margins, and guidance.
  • There may be a benchmark or metric-framing issue worth manual review, especially around adjusted metrics, timelines, or changed expectations.

CEVA appears to be strengthening its competitive position through successful execution of its integrated solution strategy, winning expanded engagements with existing customers, and gaining traction in high-value edge AI and automotive applications. The company is differentiating itself by offering full-stack platforms (including internally developed RF) that reduce customer development risk and increase time-to-market advantages, which supports its value-per-design and long-term royalty thesis. While facing competition in wireless connectivity and edge AI, the breadth of its portfolio and depth of customer relationships suggest it is holding or improving its position in key smart edge markets.

  • Total revenue: $27 million, up 11% year-over-year
  • Licensing and related revenue: $17.8 million, up 18% year-over-year (66% of total revenue)
  • Royalty revenue: $9.2 million, flat year-over-year (34% of total revenue)
  • Non-mobile royalty growth: 8% year-over-year
  • Wi-Fi shipments: 91 million units, up 158% year-over-year (all-time high)
  • Cellular IoT shipments: 66 million units, up 38% year-over-year
  • Industrial IoT royalty revenue: up 19% year-over-year despite 18M units down from 34M
  • Cash and equivalents: approximately $216 million
  • Ramp of automotive AI royalties from Toyota RAV4 and NXP collaborations
  • Expansion of Bluetooth HDT and 5G/NTN engagements into higher-volume production
  • Growth in combo chip adoption driving higher ASPs and value per design
  • Continued penetration of Wi-Fi 6/7 and Bluetooth 7 in consumer and industrial IoT
  • AI licensing pipeline progressing toward production in wearables, surveillance, and smart home
  • Inventory normalization and seasonality supporting stronger high-end smartphone royalties in H2
  • Near-term royalty softness in mobile due to seasonal factors and memory/inventory constraints
  • Dependence on customer ramp of integrated solutions for future royalty recognition
  • Execution risk in expanding RF and full-stack capabilities across multiple process nodes
  • Competitive pressure in wireless connectivity and edge AI IP markets
  • Foreign exchange headwinds from Euro and shekel strengthening against USD
  • Reliance on successful transition to Wi-Fi 7, Bluetooth 7, and UWB for sustained growth

CEVA's technology is focused exclusively on the edge—enabling connect, sense, and infer capabilities in devices such as smartphones, IoT, automotive, and wearables. There is no mention of data center exposure, AI training, or server-centric applications in the transcript. The company's AI strategy centers on efficient, ultra-low-power inference at the edge, with explicit references to hybrid models where complex processing remains in the cloud, but CEVA's IP is positioned on the device side. Any data center impact is indirect and speculative, limited to potential increased demand for edge AI devices that complement cloud infrastructure, but no direct linkage or revenue contribution from data centers is discussed.

  • What is the expected timeline and volume ramp for royalties from the Bluetooth HDT and 5G/NTN engagements?
  • How much of the 158% YoY Wi-Fi shipment growth is attributable to new customers versus migration from legacy Wi-Fi?
  • What is the anticipated royalty rate uplift from integrated full-stack solutions compared to legacy component licensing?
  • What percentage of the AI licensing pipeline is expected to convert to production within 12-18 months?
  • How will the company address foreign exchange impacts given the strengthening Euro and shekel?
  • What are the specific criteria and valuation parameters for potential M&A targets in complementary IP?
  • What is the breakdown of royalty revenue by end market (auto, industrial, smart home, mobile) and their respective growth trends?
  • How sustainable is the 8% non-mobile royalty growth given the decline in industrial IoT units?

FY2026 Q1 earnings call transcript

45,684 chars
NASDAQ:CEVA Q1 2026 Earnings Call Transcript Generated on 6/6/2026 Betsy | Conference Operator: Good day, and welcome to the SEVA, Inc. First Quarter 2026 Earnings Conference Call. All participants will be in a 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 1 on a touch-tone phone. To withdraw your question, please press star, then 2. Please note this event is being recorded. I would now like to turn the conference over to Richard Kingston, Vice President of Market Intelligence and Investor Relations. Please go ahead. Richard Kingston | Vice President of Market Intelligence and Investor Relations: Thank you, Betsy. Good morning, everyone, and welcome to SEVA's first quarter 2026 earnings conference call. Joining me today are Amir Panoush, Chief Executive Officer, and Yaniv Ariyeli, Chief Financial Officer of SEVA. Before handing 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 the results of SEVA to differ materially from those expressed or implied by such forward-looking statements and assumptions. We will also be discussing certain non-GAAP financial measures, which we believe provide a meaningful analysis of our core operating results and comparison of quarterly results. Please see the earnings release we issued this morning for our reconciliations of our non-GAAP financial measures. Our earnings release can be found in the SEC filing section of our investor relations website. With that said, I'd like to turn the call over to Amir, who will review our business performance for the quarter and provide some insight into our ongoing business. Amir. Amir Panoush | Chief Executive Officer: Thank you, Richard, and good morning, everyone. We are pleased to report a strong start to 2026. Building on our momentum from 2025, we exceeded our expectations on both revenues and non-GAAP EPS, including licensing and related revenues of $17.8 million, our strongest licensing quarter in three years, reflecting the strength of our pipeline, customer momentum, and future earnings power. This performance reflects strong executions and alignments with key market trends. including the convergence of edge AI and wireless connectivity, rising system complexity, and growing demand for integrated solutions that accelerate time to market. As the industry faces increasing constraints in scaling centralized AI compute, the reality of shifting towards running inference at the edge and leveraging local resources is becoming more critical. Against this backdrop, intelligence-connected device shipments are expected to exceed 40 billion units annually by 2030, reinforcing the value of our connect, sense, and infer strategy. In the quarter, we signed several multi-technology engagements and three strategically important deals that demonstrate our strategy is translating into results. Starting with connectivity. In early 2025, we introduced our SIVA Waves LINX 200 platform to deliver fully integrated, system-level wireless solutions across RF, basebands, and software, helping customers accelerate time to market. This quarter, we secured a major licensing win for a complete Bluetooth High Data Throughput, or HDT, solution, a foundational capability for the upcoming Bluetooth 7 standard. We license this full solution, including modern software and RF, to a leading US-based semiconductor company. Bluetooth 7 is expected to enable higher throughput and more advanced use cases, including multichannel audio, wireless video, XR and gaming peripherals, and AI-enabled edge devices. Our HDT solution is a key building block enabling this next generation of high-performance wireless and AI-enabled edge devices. This builds on our prior Bluetooth engagement with the same customer, which is now approaching high-volume production, and further expands our footprint through a more integrated RF, modem, and software platform engagement. This also reflects a border shift in the industry from internally developed connectivity to licensing-proven platforms. We believe that moving to a full-stack solution increases value per design for SIVA through higher licensing fees and greater royalty content, while also deepening integration and enabling multi-generation engagement. For the quarter, we expect it to deliver faster time to market and lower development risk, allowing them to focus on their core differentiation while leveraging our proven IP, ultimately driving a stronger return on investment for both parties. Turning now to 5G and satellite communication. During the our Pentagy NTN 5G advanced modern platform, sending our cellular portfolio into satellite communication. Non-terrestrial networks, or NTN, an emerging market expected to scale to billions of devices over the coming decade, as satellite connectivity becomes an integral part of global communications infrastructure, complementing and, in some cases, extending beyond traditional terrestrial 5G networks. This is being driven by a wide range of use cases, including direct-to-direct remote and undeserved area coverage, asset tracking, and industrial IoT, where ubiquitous, always-on connectivity is critical. It is also increasingly important for enabling more resilient and independent communications infrastructure. Customer response has been highly encouraging, with clear momentum building across our pipeline. Building on this, we expanded an existing customer relationship with a satellite OEM from DSP cores to a more integrated baseband processing solution. As with our Bluetooth HDT engagement, this reflects a deepening relationship with an existing customer and an expansion in the scope and value of our IP within their platform. In Ultra Wideband, during the first quarter, we introduced our next generation UWB platform and secured a new customer win with a major US-based MCU provider, augmenting its internal UWB capabilities. With our IPN combining its system expertise with our proven connectivity solution to accelerate development and reduce risk. This engagement also builds on a broader relationship with the customer, who has licensed multiple SIVA technologies over the past two years. We are seeing a transition in UWB towards higher-value industrial, automotive, and enterprise applications, driven by demand for precise, secure location awareness in use cases such as access, asset tracking, and indoor navigation. As the market expands, customers are increasingly choosing to license proven IP to accelerate time to market and reduce development risk. Because of these wins, a clear pattern is emerging. The Bluetooth NTN and UWB engagements we highlighted this quarter are all within existing customers who have expanded their use of SIVA IP over the past two years. More broadly, customers are increasingly adopting more integrated system-level solutions from SIVA, expanding our value-per-design while strengthening long-term royalty and margin potential. Incensing We continue to see growing traction for our special audio solutions as demand for immersive audio experience expands. During the quarter, Lenovo launched its latest ThinkPad headset, powered by our RealSpace special audio with head tracking, building on recent wins with consumer brands like Nothing and Bolt. Finally, in AI, we continue to execute on our strategy to enable efficient, scalable inference at the edge, with AI representing more than 20% of our licensing and related revenues, and the signing of two new licensing agreements in the quarter. We are seeing a structural shift towards hybrid AI, where inference is increasingly moving to the device, while more complex processing remains in the cloud or across connected systems. This right AI model, right place, right time approach enables real-time on-device decision-making while maintaining the flexibility to scale compute as needed. As a result, demand for highly efficient, ultra-low-power solutions is growing across wearables, automotive, industrial, and smart home applications. And IP and AI content per device is increasing as more products require local connect, send, and infer capabilities. We believe the rise of hybrid and agent-based AI will further accelerate the shift towards distributed intelligence at the edge, where devices need to locally sense, infer, communicate, coordinate, and act in real time while selectively leveraging cloud AI resources. This trend is expected to drive growing demand for efficient AGI processing alongside advanced wireless connectivity across increasingly complex connected systems. This is now translating into production. Renaissance R-car V4H platform, which integrates our AI DSP and accelerator, is now in production in the 2026 Toyota RAV4, one of the highest volume passengers vehicle globally, marking our first mass volume automotive AI deployment. We believe this represents the beginning of a meaningful, long-term royalty stream with going AI content per device. We also announced a collaboration with NXP during the quarter, integrating our AI DSPN accelerator into their S32E2 and S32Z2 software-defined vehicle processors, further validating our position in automotive AI. In addition, our Newport Nano NPU won a leading artificial intelligence award at Embedded World 2026, further emphasizing our leadership position. Our AI licensing pipeline remains strong, with multiple evaluation and investment negotiations underway across a broad range of end markets. Stepping back, overall, we signed 14 licensing agreements in the quarter, including two with OEMs. In addition to the deals I highlighted earlier, we secured a Wi-Fi 7 design targeting consumer IoT, a Wi-Fi 6 Bluetooth combo engagement with a leading edge AI SOC platform company, and multiple additional Bluetooth and Wi-Fi winds across our connectivity portfolio. Turning now to royalties. We continue to see encouraging momentum across our diversified smart edge market. with growth in IoT, industrial, and AI-driven applications. While total royalties were flat year-over-year, non-mobile royalties grew 8%, reflecting strengths across our smart edge markets, partially offset by softness in smartphones. Wi-Fi shipments reached an all-time high in the quarter, driven by record Wi-Fi 6 volumes, highlighting the continuing expansion of this market as customers ramp deployments across a broad range of devices. More broadly, Wi-Fi and Bluetooth continue to be durable, multi-year growth drivers. As customers scale current generation technologies, such as Wi-Fi 6 and Bluetooth 6, they are also developing next-generation platforms, including Wi-Fi 7 and Bluetooth 7. These overlapping cycles are expected to support sustained unit growth increase IP content per design, and long-term margin expansion. We expect the continued shift towards combo chips to further reinforce our strategy, as customers integrated multiple SIVA technologies into a single design, increasing value per device and driving stronger overall economics. AI-driven royalties also continue to grow, highlighted by our automotive AI deployment at Toyota and a ramping AI SOC for surveillance, representing early signs of the long-term contribution we expect from edge AI across multiples and markets. Against these tailwinds, first-quarter royalties were impacted by typical seasonal softliness in mobile, combined with near-term effects for memory availability constraints and challenge inventory in the lower tier segments. We view this mobile dynamics as largely timing-related and expect improvements as the year progresses, supported by inventory normalization and typical seasonality, along with what we anticipate will be stronger high-end smartphone royalties in the second half. Overall, this quarter reinforces our ability to execute on our strategy and increase value per design as we move towards more integrated, higher-value engagements. I will now turn the call over to Yaniv for the financials. Yaniv Ariyeli | Chief Financial Officer: Thank you. I'll now review the financial results for the first quarter, which reflect the strong licensing performance and continued execution Amir just outlined. Revenues for the first quarter increased 11% year-over-year to $27 million. The revenue breakdown is as follows. Licensing and related revenue increased 18% year-over-year to $17.8 million, reflecting 66% of our total revenues. Royalty revenues were $9.2 million, in line with last year, reflecting 34% of total revenues. Gross margins were 86% on GAAP bases and 87% on non-GAAP bases. Our total gap operating expenses for the first quarter were $28.4 million, just over the mid-range of our guidance. Total non-gap operating expenses for the first quarter, excluding equity-based compensation expenses, amortization of intangibles and deal costs, were $23 million, just over the mid-range of our guidance. Gap operating loss for the first quarter was $5.1 million as compared to gap operating loss of $4.4 million in the same quarter last year. Non-gap operating margins and income were 2% of revenues and half a million dollars. Income was $1.9 million compared to $2.1 million for the first quarter of 2025. Taxes were approximately $1.3 million Gap net loss for the first quarter was $4.5 million and diluted loss per share was $0.16 as compared to net loss of $3.3 million and diluted loss per share of $0.14 for the first quarter of 2025. Non-gap net income and non-gap diluted earnings per share for the first quarter of 26 were $1.1 million and 4 cents respectively as compared to non-GAAP net income of $1.4 million and non-GAAP diluted earnings per share of 6 cents for the first quarter of 25. With respect to other related data, we shipped 458 million units of SIVA power devices. up 9% for the first quarter of 2025. Of the 458 million reported, 46 million units, or 10%, were for mobile handset modems, down from 49 million units in the first quarter last year. 394 million units were consumer IoT devices. up from 337 million units for the first quarter last year. 18 million units were for industrial IoT products, down from 34 million units in the first quarter last year. However, associated industrial IoT royalty revenues were up 19% year over year. reflecting a better mix of higher ASP product shipments, including 5G wireless infrastructure and automotive AI. Local shipments were 206 million units in the quarter, down from 233 million units in the first quarter of last year. Cellular IoT shipments were 66 million units, up 38% year-over-year, And Wi-Fi shipments were a record 91 million units, up 158% year-over-year. As for the balance sheet items, our cash equivalent balances, marketable securities, and bank deposits were approximately $216 million, providing strong financial flexibility. We remain focused on disciplined capital allocation, including continued investments in our roadmap and a selective approach for strategic M&A opportunities that can accelerate our growth. Our DSOs for the first quarter of 26 was 59 days. During the first quarter, we used $4.9 million of cash in operating activities. Ongoing depreciation and amortization was $0.9 million. And purchase of fixed assets was $2.3 million, including approximately $1 million related to leasehold improvements. At the end of the first quarter, our headcount was 430 people, of whom 348 were engineers. Moderator: Now for the guidance. Yaniv Ariyeli | Chief Financial Officer: As Amir highlighted, we delivered a strong start for the year, supported by continuing enhancements to our IP portfolio, solid licensing execution, and growing fundamental for future royalty expansion. From a financial perspective, we continue to view 2026 as a year of growth across multiple dimensions. Reflecting our first quarter performance, we're upgrading our annual outlook towards the higher end of our previously communicated range. For the full year, we now expect total revenue growth to be at the top end of our 8% to 12% range over 2025. With a typical seasonality profile of lower growth in the first half, and stronger growth in the second half, subject to memory pricing dynamics and supply conditions. On the expense side, we maintained focus on cost discipline and operating leverage, while continuing to manage foreign exchange headwinds with the strengthening of the Euro and the Israeli shekel against the U.S. dollars. Overall expenses cost of revenues and OPEX combined are expected to increase approximately 8 percent over 2025. As we continue to invest to support growth, we expect a portion of the incremental revenue to be translated to the bottom line, driving continued improvement in non-GAAP operating income, net income, and EPS. Based on our performance to date and current business momentum, we now expect non-GAAP operating margins and non-GAAP net income to increase by 40 to 50 percent year over year, which is above our prior expectations. Guidelines for the second quarter of 2026. Revenues are expected to be in the range of $26 to $30 million. reflecting continued growth both sequentially and year-over-year. Gross margin is expected to be 87% on GAAP bases and 88% on non-GAAP bases, excluding an aggregate $0.2 million of equity-based compensation expenses and $0.1 million of amortizations of required intangibles. Gap OPEX for the second quarter of 26 is expected to be similar to the first quarter and in the range of $27.7 to $28.7 million. Of our anticipated total OPEX for the second quarter, $5.3 million is expected to be attributed to equity-based compensation expenses, $0.1 of amortizations of required intangibles, and $0.1 million of cost associated with business acquisitions. Non-GAAP OPEX is also expected to be similar to the first quarter and in the range of $22.2 to $23.2 million. Then interest income is expected to be approximately $1.7 million. Taxes for the second quarter is expected to be approximately $1.5 million. And the share count for the second quarter of 26 is expected to be approximately 28 million shares for GAAP and 29.7 million shares for non-GAAP. Moderator: Betsy, we could now take questions, please. Betsy | Conference Operator: We will now begin the question and answer session. To ask a question, you may press star, then 1 on your touchtone phone. 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 2. At this time, we will pause momentarily to assemble our roster. The first question today comes from Ruben Roy with Stifel. Please go ahead. Operator | Conference Specialist: Ruben, your line is open. You may ask your question now. Sorry, guys. Ruben Roy | Analyst, Stifel: I was on mute somehow. Hi, Amir. Hi, Yaniv. Congrats on the nice start to the year. I guess to start, Amir, on the Bluetooth HDT win, I'm not sure if you guys had RF wins previous, but it seems to me like that would be a nice step up in your value per design strategy that you've been talking about. So can you maybe just talk a little bit more about what you're doing for the RF? And also, I guess, as part of that, is that sort of an architecture that you can replicate across other areas of the business, eventually Wi-Fi, ultra-wideband, et cetera? and anything you talk about in terms of the royalty rate relative to your traditional Bluetooth licenses. Thank you. Amir Panoush | Chief Executive Officer: Yeah, Roy, first, thanks a lot for the congratulations. Yeah, definitely, this is a very important win for us. As you pointed out, this is a win of a full system solution, all the way so-called from the antenna up to the full stack and the software, including our own internal developed RF, which is an investment that we've put in the last year or two to really build those system up. The key value here is really that our customers, they can get the full solution. They don't need to do more of the pre-testing validation of those things. And we provide them that as a full solution, then time to market and ability to be successful in the market is much higher. And even more so with this customer and overall other customers, what we see, that really helps them to drive more and more at so-called the mix versus by decision and move away from so-called internal development to a complete solution based on our technology. So we are very happy with that, with the RF, and we expect more of those means to come through the year and then, of course, in the next few years. The other piece that you pointed out, this is definitely a technology that we are planning to expand beyond the Bluetooth HDT. We have multiple other wireless technology with digital IP and the same strategy we are going to basically deploy and apply in the marketplace. More and more integrated solution, complete system around our leadership in wireless connectivity. So we are super, super excited about this momentum and that's what can build for the future. Last piece that you point on the royalty. As I mentioned in the previous course, at the end of the day royalty comes back to what value we bring to our customers. In this case, because it's not just the whole different components of the system, it's the fact that it's fully integrated, our customers definitely appreciate it, and we see meaningfully higher royalty than so-called 1 plus 1 is more than 2, and that will help us to drive much more royalty growth in the future with overall very strong flywheel across our wireless connectivity technologies. Ruben Roy | Analyst, Stifel: That's great. Thank you, Amir, for all the detail. I guess if I could ask a quick follow-up just on sort of the way the year is playing out. You continue to expect a stronger second half, and I think you gave us a lot of, you know, sort of data points and, you know, kind of visibility into how you're thinking about that. But you do have, you know, some, you know, factors coming into play. You mentioned memory pricing and, you know, you know, overall, you know, sort of macro, you know, sort of dynamics going on. So either Amir or Yaniv, can you maybe just give us a little bit of detail on what you're hearing from customers relative to some of those, you know, impacts that we might see as we kind of go through the year? I think, you know, memory pricing has started to impact some of the end markets. We're hearing from PC guys, et cetera, you know, talk about potential impacts there. Any additional detail on how you're thinking about the second half versus the first half and what you're hearing from customers would be great. And that's all I have. Thank you. Yeah, definitely. Amir Panoush | Chief Executive Officer: One thing first I would say, just if we look at this quarter, as we started the year, I'm extremely encouraged by the fact that even though so-called mobile hasn't been that strong, considering the challenge with memory allocation and so-called inventory utilization, we still deliver really great results. driven by one, very good execution across the licensing and solution-based offering. And second, we see a very good momentum overall in the border IoT. And going back to what you asked about the memory, if we look at the IoT, it's a market that is less impacted by that. We have a great access across a very diversified set of customers, use cases, and products and technologies. So I think overall we can do so-called better than others in terms of potential impact from memory allocation. And specifically on mobile, with the inventory drawdown that happened this quarter and maybe to some degree through the first half, it probably will put us in a good spot as we go to the second half, which on top of that, of course, what we expect is increased market share in the premium tier. So I think overall we are well-positioned. and going through so-called debt challenges overall in the marketplace. And it goes back to how we execute basically driving our licensing and ensuring that our customers are happy with the ramp-up of our technologies. Yaniv Ariyeli | Chief Financial Officer: Ruben, maybe we'll add one more thing. Historically, if you look at the volumes of shipments of our royalties, our customer shipments in the second half of every given year in the last three years, you'll see about a 40% increase. And then every year there is some issues, whether it's pricing or inventory or now memory. So with that said, the trend was mainly around 40% sequential growth, second half versus first half. And we are building that in also in our prospects for 2026. Ruben Roy | Analyst, Stifel: Got it. Thank you, Yannick. Thank you. Thanks, Amir. Betsy | Conference Operator: Thank you. Ruben Roy | Analyst, Stifel: Thank you, Roy. Okay. Betsy | Conference Operator: The next question comes from Suji De Silva with Roth Capital. Please go ahead. Suji De Silva | Analyst, Roth Capital: Hi, Amir. How are you? Congratulations on the progress here. Thank you. Amir, maybe you can talk about your – as you came in, you talked about sense, connect, and infer. And maybe today you could give us an update on that in terms of the example of traction at the same customer, two of those or three of those versus just one. That would be helpful to understand. Amir Panoush | Chief Executive Officer: I definitely suggest. I think several names that we mentioned in the past, including this time, we see them basically licensing multiple technologies from us. It can be multiple technologies across Connect, but also we have more and more across multiple technologies of Connect and Infer. And in some cases, the whole thing connects Sense and Infer. And so we see that progression going very well. And we expect more as we keep driving those technologies into the marketplace. But definitely what drives the baseline flywheel or success with our customers is very high appreciation of our wireless connectivity portfolio. And on top of that, our investment and expansion in the AI or Infer, overall portfolio. Other things, as we pointed out, Lenovo with their headset this quarter, we announced that basically They've been using or start ramping with our real space or 3D spatial audio technologies. And they are also a wireless connectivity basically customer to the semi guys that are delivering those solutions to them. Suji De Silva | Analyst, Roth Capital: Okay. I appreciate that, Amir. Great. And then in the connectivity specifically, Bluetooth is already well penetrated. Can you update us on where Wi-Fi is in the attached curve going up in terms of attachment? And then, will UWB follow a similar path, or is that more of a niche technology? Thanks. Amir Panoush | Chief Executive Officer: Yes, on the Wi-Fi, and if you can point more into the specific numbers, but we're extremely encouraged with the ramp that we've seen first through all 2025, and now continuing and even more in Q126, where we reach an all-record high volume this quarter, and we expect that to continue with a very nice ramp moving, so-called, from the more legacy Wi-Fi into Wi-Fi 6, And then within a year or two, we'll start seeing the transition into Wi-Fi 7 plus lots of the combos of the Wi-Fi and Bluetooth. So overall from a pattern and penetration in the marketplace, we expect, as we mentioned on other calls, right, that Wi-Fi shipments will reach a very high volume above the half a billion and more as we keep progressing and then basically augment very nicely our penetration with Bluetooth plus the combos. In terms of UWB, this is, I would call it overall, a newer technology. There are lots of very good indication in the marketplace from the use cases. And with that, the potential demand for the technology, we've seen more penetration right now in smartphone from there into different type of edge devices for location-based, for access and control. So we are very encouraged with that. Now we just got a major license deals with a US customer. And there will be more to follow. But overall, from a volume penetration, I'd say we are highly penetrated with Bluetooth. We are getting to the same level with Wi-Fi. And the next to follow will be UWB. Okay, great. Yaniv Ariyeli | Chief Financial Officer: Thank you. The only comment that I can add for that, Suji, is we talked about, Amir mentioned the combo chips. If you look at the Bluetooth Wi-Fi combo chip year over year, the volume has doubled. We haven't opened that number up yet. We'll do it in due time. But some of the reason also that we mentioned that the Bluetooth is going down because we are counting those combo chips is combo and not Bluetooth necessarily. There is no issue in the market. It's just our count and ASPs for those combo chips are higher than the individual Wi-Fi or Bluetooth solutions in the past. Suji De Silva | Analyst, Roth Capital: Can you be counting those in Wi-Fi units? Is that what you think? Yaniv Ariyeli | Chief Financial Officer: The combo of Bluetooth and Wi-Fi units, yeah, double year-over-year for Q1. Moderator: Okay, thank you, thank you. Betsy | Conference Operator: Sure. The next question comes from Sameek Chatterjee with JP Morgan. Please go ahead. Sameek Chatterjee | Analyst, JP Morgan: Great, thanks for taking my questions, and congrats on the strong results here. Maybe just another follow-up on Wi-Fi, the 91 million number that you had there, it's pretty strong considering a seasonal sort of, you typically see a seasonal downtick into 1Q. Can you just outline if there was anything in terms of a new customer, volume, et cetera, ramping into 1Q that drove that seasonality? And from this sort of 1Q base, should we expect to see a similar pickup into the second half that you've historically seen from first half to second half perspective in Wi-Fi? Thank you. And I'm a follow-up. Amir Panoush | Chief Executive Officer: Yes, Amik, this is a great question. And actually, the ramp or the volume in Q1 of our Wi-Fi shipments is not related to seasonality, as you pointed out. It's really the migration of multiple customers adopting our technology. So either migration from Wi-Fi 4 to Wi-Fi 6, or many of them actually new customers that start ramping with the Wi-Fi 6. And I will remind everyone that we talked about more than 30 licenses agreements that we have made in the last two, three years of Wi-Fi technology. And those basically customers are now coming more and more into production. So that momentum, we expect to continue. And actually we should expect second half to be stronger than the first half, both based on the seasonality, plus basically more and more new customers and new program basically ramping in volume for Wi-Fi. So Wi-Fi, we are really still in the ramp up in terms of market penetration. and our customers basically ramping their portfolio and their product line. Correct, correct, correct. And just maybe... And it's true, by the way, Samik, both to industrial and consumer. So we are really doing well on both fronts with our Wi-Fi technology. Sameek Chatterjee | Analyst, JP Morgan: Okay. Just for a quick follow-up, any updates on how you're thinking about capital allocation, particularly in relation to M&A? given that it's a pretty strong year, you'll generate more cash. How are you thinking about sort of the alternatives in front of you, including if you do go CM&A, what would be the more sort of targeted technology areas that you would look for? Thank you. Amir Panoush | Chief Executive Officer: Yeah, definitely. This is a key important item within what we're looking to execute and our overall strategy to scale up the company and looking into an M&A option for us. The focus there will be around so-called technologies that complement our success in the smart edge era. We have more focus on IP overall in order to build the scale. So, you know, we talk about connections and infer within those technologies and augmented technologies. I think that's what we are really targeting, and hopefully we'll be able to talk about it as we progress through the year. Operator | Conference Specialist: Thank you. Sameek Chatterjee | Analyst, JP Morgan: Thanks for taking my questions. Operator | Conference Specialist: Thank you. Yeah, thanks for me. Betsy | Conference Operator: The next question comes from Gary Mobley with Loop Capital. Please go ahead. Gary Mobley | Analyst, Loop Capital: Hi, guys. Thanks for taking my question. Looking specifically at the SEVA wavelengths, the RF subsystem there, I know the highlight that you put in front of us today is more of a system-level license agreement, you know, including the RF. But if I'm not mistaken, that RF subsystem might be unique to a specific manufacturing process node, TSMC 12 nanometers specifically. Can you speak to how you might move forward in broadening that, I guess, the scope of the RF subsystem across different process nodes and different boundaries and how that might affect the overall licensing for wavelengths? Amir Panoush | Chief Executive Officer: Yeah, Gary, great question. So yeah, the Lynx 200 that we announced previously was around 12 nanometers TSMC. And overall, what we are executing our strategy is actually to go beyond one process node or one fund B. And also, I think we are well positioned with the access that we have in the market from the number of customers that have licensed our digital IP technology. to have very good sense of where the roadmap is heading in terms of the process node needs, as well as the type of funders that they are looking to partner with. And yeah, we are not going to support all different options out there and permutation, and definitely some customers will build with their own RF, but I'm very confident that we can so-called go and support the majority or the significant portion of where the market is heading in terms of the process need and the funders. So we'll have so-called multiple options there, but we are not going to cover the whole spectrum. Gary Mobley | Analyst, Loop Capital: For a follow-up, I want to ask in general about the license pipeline. You know, how does it look, you know, compared to maybe a year ago? And if you can give us an update as to what might be recurring in license revenue and what percent still remains, you know, one time in nature. Amir Panoush | Chief Executive Officer: There are several so-called fundamental trends that encourage us and we feel good with the perspective of our licensing business. One, we see more and more customers, repeating customers coming again, going from one generation to the next. The other one is more customers are coming to license multiple technologies, either by adding additional technology or just from the start go looking for multiple technology. And the last piece is what we are highlighting this quarter, is really coming in licensing solutions, which at the end of the day brings more value to our customers and help us, so-called, to have better economics of the deals, including the licensing portion. When we take all those three into account, overall, we feel good. We feel confident with where we are in terms of the pipeline, our ability to execute our licensing business. And I think the last few quarters have shown that, including this quarter. So I would say overall, we look at the year as a good growth year in licensing. the pipeline really supports as well. Moderator: Thank you. Thank you, Gary. Betsy | Conference Operator: The next question comes from Josh Buckalter with TD Cowan. Please go ahead. Josh Buckhalter | Analyst, TD Cowen: Hey, guys. Thanks for taking my question, and congrats on the results. Maybe I wanted to start big picture. You know, we're seeing sort of a lot of positivity in the CPU space as compute resources are moving, you know, increasingly away from or in addition to being complemented by outside of the AI server rack. Could you maybe reflect on where we are on the embedded side in that adoption curve and specifically any updates or major momentum on the MPU side from the quarter you wanted to highlight? Thank you. Amir Panoush | Chief Executive Officer: Yeah, great question, Joe. So first from the so-called momentum of CPU, And this is what we have been talking about for the last few quarters about so-called the hybrid AI model and things are more moving into the edge. So this is very encouraging to see that that's really happening in the market. And also other customers are able to, other players in the market are able to basically execute to that and show that progress. And we need to keep in mind that when we look at our Connect, Sense and Infer IP portfolio, It actually complements extremely well CPU, whether that CPU based on that architecture or the other RISC-V architecture. So we are really indifferent to that and we can support both. So that puts us in a good position. On the NPU specifically, that's where we are building, again, a portfolio of NPUs that goes along any kind of CPU architecture. And I think that's where we're also uniquely positioned. focusing on NPU technology itself as accelerator to the CPUs that are out there. The more CPU drives more adoption of AI at the edge, the more opportunities we will see with our NPUs. All those things are encouraging so-called activities and potential tailwinds for us as we progress through the year and next year. Josh Buckhalter | Analyst, TD Cowen: Thank you for the color there. And then maybe I can follow up on the second half outlook. A lot of companies have flagged potential cuts in the second half from the memory headwinds. Have you guys seen anything yet that's impacted your customers? And then I was also hoping you could maybe walk through what are the expectations that you have in your second half outlook for the large North American smartphone customer that has some of your IP on their modem. Thank you. Yaniv Ariyeli | Chief Financial Officer: Sure. So, you know, we built some of our expectations top down with knowing that the markets in the second half with the seasonality of Christmas and the ramp up for introduction of new products around that timeframe is strong. We'll need to see how the market deals with the memory pricing and shortages. but right now we haven't heard anything specific from our customers other than what we have seen in the mobile space and the low tier phones that we have seen in other companies have talked about Qualcomm ARM in the first quarter of the year with mentioning the recovery going forward. So I don't think we have seen anything yet. I think the market has its way to overcome some of the Hurdles when we get to the high season, and we have built all that in, including the North American OEM that doesn't share its internal plan, that doesn't share exactly the timing of introduction of new products, whether they're based on their own motive or not. And we have our own estimates that we have built in this model. The rest will look and get the royalty reports on a quarterly basis. And based on that, they'd be able to report. Again, historically, and the more we have added the combo chips, like we talked about today with higher ASPs, the more that we have the automotive AI, NXP and Renaissance helping us this year, which weren't around last year with royalty contribution. the more 5G networks that started the year very strong, and then the OEM opportunities in the U.S., it looks like a stronger and promising second half. And this is the reason we took our guidance to the top range of the previous annual guidance of the 8% to 12%. Operator | Conference Specialist: Thank you. Moderator: Thank you, Joshua. Betsy | Conference Operator: The next question comes from Madison DiPaolo with Rosenblatt Securities. Madison DiPaolo | Analyst, Rosenblatt Securities: Please go ahead. Hi, this is Maddie calling on behalf of Kevin Cassidy. I was just wondering which end markets here are expressing the most interest for in the NeuroPro? Moderator: Say that again, Maddie? Yaniv Ariyeli | Chief Financial Officer: Sorry. Madison DiPaolo | Analyst, Rosenblatt Securities: Which end markets are expressing the most interest in the NeuroPro? Amir Panoush | Chief Executive Officer: It's board-based, I would say. We see it in automotive, we see it in some industrial applications, we see it also in smart home and consumer applications. If we look at the 10 plus more deals that we so-called licensed last year, it's really across all those four markets that I mentioned. So I can't point to one that is much more than the others, very significantly. It's nicely distributed and wide-based. So we mentioned also last quarter a PC OEM, so we are in the PC market consumer. Again, smart home surveillance and automotive industry. Madison DiPaolo | Analyst, Rosenblatt Securities: Okay, thank you. Operator | Conference Specialist: You're welcome. Thank you. Betsy | Conference Operator: The last question today comes from Martin Gang with Oppenheimer. Please go ahead. Martin Gang | Analyst, Oppenheimer: Hi, thank you for taking my question. My first question is on the Bluetooth radio. Is there any plan or intention to extend the IP to other connectivity products, notably Wi-Fi? Amir Panoush | Chief Executive Officer: Yeah, Martin, good question. Yeah, definitely. So we started and announced this product first. At the end of the day, we have very strong capabilities across the spectrum of wireless connectivity technology. And the intention and the plan is definitely to expand this to so-called a full solution offering across our wireless connectivity portfolio. So starting with Bluetooth, as you mentioned, the next natural thing will be Wi-Fi, and then also UWB and our other technologies. Definitely that's the plan. And overall, also this quarter we announced on the satellite side that we're also moving more into complete so-called basement solutions. not just so-called offering the components like DSP accelerators, but really the whole basement subsystem. And that resonates very, very nicely with customers, especially customers that want to make a decision moving from make to buy, because they need to rely on more of a so-called ready-to-go solution to help them with time to market and success overall. Martin Gang | Analyst, Oppenheimer: Thanks, Amir. A follow-up on your answer, you mentioned that satellite communication, is that primarily still on the market deployment regarding smartphone with satellite-based messaging capabilities, or are you seeing more emerging applications of that? Amir Panoush | Chief Executive Officer: No, we're actually seeing much more potential on the emerging applications as well. So if we look at the different types of OEM out there, they're basically moving to provide more and more as a service, and part of the service, they need a complete solution end-to-end, and we are offering the wireless communication both from the terminal side as well as from the satellite side, and then they will build a so-called complete end-to-end offering with the service, and that service is really to be able to have ubiquitous type of connectivity, whether it's for industrial use cases, logistical use cases, and so on, or even places where there is very little coverage of wireless infrastructure, and they want to provide that augmentation. So those are all about system well beyond just mobile. Operator | Conference Specialist: Thank you. Moderator: Thank you, Morten. Yeah, thanks a lot, Morten. Betsy | Conference Operator: This concludes our question and answer session. I would like to turn the conference back over to Amir Panoush for any closing remarks. Amir Panoush | Chief Executive Officer: Thank you. In closing, we believe SIVA is well positioned as the industry continues to evolve towards physical AI. where connectivity, sensing, and inference converge at the edge. Our expanding portfolio, combined with our strategy to deliver more integrated, system-level solutions, is enabling us to increase our value per customer and strengthen our long-term royalty model. We remain focused on executing our strategy, deepening customer relationships, and driving sustainable growth. Thank you for your continued support. Richard, I will hand over to you to wrap it up. Richard Kingston | Vice President of Market Intelligence and Investor Relations: Thank you, Amir. As a reminder, the prepared remarks for this conference call are accessible through the investor section of our website. With regards to upcoming events, we will be participating in the following conferences. Oppenheimer 27th Annual Israeli Conference on May 18th in Tel Aviv. The JP Morgan 2026 Global Technology, Media and Communications Conference, May 20th in Boston, Massachusetts. TD Cowen's 54th Annual Technology, Media, and Communications Conference, May 27th in New York. DEEPL's Boston Cross-Sector One-on-One Conference, June 2nd in Boston. The 6th Annual Rosenblatt Technology Summit, The Age of AI, June 10th, being held virtually. And the 16th Annual Roth London Conference, June 16th to 18th in London, England. 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. Betsy | Conference Operator: The conference has now concluded. Thank you for attending today's presentation. You may now disconnect. jsPDF 3.0.3 D:20260606090041-00'00'

Research summary and source transcript

readyJun 10, 2026

CEVA reported a record Q4 2025 revenue of $31.1 million, up 7% YoY, driven by strong licensing growth (11% YoY) and a breakthrough in AI processor licensing, including a significant NPU deal with a leading PC OEM. Full-year 2025 revenue increased 2% to $109.6 million, with non-GAAP net income up 20% YoY. The company is building a licensing-to-royalty flywheel, with $125 million in estimated lifetime royalty potential from 2025 deals, though realization is multi-year and dependent on customer deployment. Management remains focused on expanding its AI and connectivity IP portfolio to capture long-term value in the physical AI era.

Management knows that six of the NPU customers signed over the last one to two years are expected to enter production by end of 2026, with potential royalty contributions beginning in early 2027—a timeline not yet reflected in market expectations, which may not anticipate the ramp of AI-related royalties until later. This insight, derived from Yaniv Ariely’s comment about monitoring customer progress and expecting royalty kick-in in 2027, represents a 6-24 month information gradient, as the market may not yet price in the near-term revenue contribution from these AI licensing wins.

Licensing revenue growth, royalty revenue resilience through diversification, and expansion of AI processor (NPU) licensing as a higher-margin, long-term royalty driver.

  • AI and NPU licensing momentum and pipeline
  • Connectivity strength in Wi-Fi 7, Bluetooth, and combo solutions
  • Diversification across smart edge markets (consumer, automotive, industrial, infrastructure)
  • Physical AI as the next growth frontier
  • Licensing-to-royalty flywheel and lifetime value of deals
  • Balance sheet strength and use of capital for M&A
  • Amir Panoush’s emphasis on the PC OEM NPU deal as a 'breakthrough' and 'strong validation' of on-device AI adoption
  • Yaniv Ariely’s specific reference to six NPU customers expected in production by end of 2026 with royalty potential in 2027
  • Amir’s description of the PC OEM win as strategically important on two fronts: customer trust and ecosystem tipping point
  • Repeated references to 'physical AI' as a transformative opportunity beyond traditional markets
  • Highlight of 20 billion cumulative devices shipped and exceeding 21 billion by Q4 2025 as a credibility milestone

Management exhibited a confident, detailed, and credible tone, particularly when discussing specific wins like the PC OEM NPU deal and providing granular data on shipments, licensing counts, and financials. Executives avoided vague optimism, instead grounding excitement in concrete achievements (e.g., 10 Newport NPU agreements in 2025, 21 billion cumulative units shipped). Their willingness to discuss timelines for royalty realization (2027) and acknowledge macro challenges like FX and memory pricing enhanced credibility. There was no defensiveness or overpromising; tone was measured and evidence-based.

  • No clear dodged analyst question was detected by the local fallback; manual review should still check whether Q&A answers quantified conversion, margins, and guidance.
  • There may be a benchmark or metric-framing issue worth manual review, especially around adjusted metrics, timelines, or changed expectations.

CEVA appears to be winning competitively in the smart edge AI inference space, particularly in licensing NPUs to leading PC and embedded OEMs, as evidenced by the strategic PC OEM win and broader pipeline momentum. The company is differentiating through its integrated Connect-Sense-Infer portfolio and gaining share in Wi-Fi 7, Bluetooth HDT, and combo solutions. While competition exists (e.g., ARM in NPUs), CEVA’s focus on performance-per-watt and co-architecture flexibility appears to be resonating with customers seeking differentiated AI edge solutions.

  • Q4 2025 revenue: $31.1 million, up 7% YoY and 10% sequentially
  • Licensing revenue: $17.5 million in Q4 2025, up 11% YoY and 9% sequentially
  • Royalty revenue: $13.8 million in Q4 2025, up 2% YoY and 12% sequentially
  • Non-GAAP operating margin: 18% in Q4 2025, up from 15% in Q4 2024
  • Non-GAAP net income: $4.9 million in Q4 2025, up 86% YoY
  • Full-year 2025 SIVA power device shipments: 2.1 billion units, up 6% YoY
  • Estimated lifetime royalty potential from 2025 licensing agreements: $125 million
  • Royalty ramp from 2025 AI licensing deals expected to begin in early 2027
  • Continued market share gains in Wi-Fi 7 and Bluetooth HDT across IoT and consumer markets
  • Potential for additional PC OEMs to follow the lead of the disclosed NPU design win
  • Expansion of physical AI applications in robotics and edge sensing/inference
  • Use of strengthened balance sheet ($222M cash) for strategic M&A in IP domain over next 12 months
  • Seasonal strength in H2 2026 revenue, consistent with historical patterns
  • Royalty growth remains exposed to memory pricing and supply constraints affecting smartphone shipments
  • AI processor licensing revenue realization is multi-year and dependent on customer deployment and market adoption
  • Foreign exchange headwinds from Euro and Israeli shekel strength could add ~$5M in annual non-US dollar-based expenses
  • Dependence on a limited number of high-end PC OEMs for AI upside, despite broader pipeline claims
  • No explicit mention of new customer concentration or diversification beyond existing smart edge markets
  • Guidance assumes 8%-12% revenue growth in 2026, which may not materialize if AI licensing does not convert to royalties faster than expected

CEVA’s discussion of AI is focused exclusively on on-device inference at the smart edge, particularly in PCs, IoT, automotive, and robotics. There is no mention of data center AI, cloud inference, or any direct or indirect exposure to data center semiconductor markets. The company positions itself as enabling local AI processing to reduce reliance on the cloud, suggesting a strategic orientation away from, rather than toward, data center-centric AI workloads. Any AI/data-center impact is absent from the transcript.

  • What is the expected quarterly royalty ramp trajectory from the six NPU customers anticipated in production by end of 2026?
  • What percentage of the $125 million lifetime royalty potential is attributable to the PC OEM deal versus other NPU wins?
  • How is CEVA addressing foreign exchange exposure from non-US dollar-based R&D costs in Europe and Israel?
  • What specific criteria is CEVA using to evaluate M&A targets in the IP domain, and what is the expected timeline for deal execution?
  • Beyond robotics, what are the top three physical AI application areas CEVA expects to drive incremental royalty growth?
  • What is the attach rate of CEVA’s NPU IP in the PC OEM’s SoC, and what royalty rate or ASP assumptions underlie the $125 million estimate?

FY2025 Q4 earnings call transcript

42,356 chars
NASDAQ:CEVA Q4 2025 Earnings Call Transcript Generated on 6/6/2026 Operator | Conference Operator: Good day, and welcome to the SEVA Inc. Fourth Quarter and Year-End 2025 Earnings Conference Call. All participants will be in a 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 a touch-tone phone. To withdraw your question, please press star then two. Please note this event is being recorded. I would now like to turn the conference over to Richard Kingston, Vice President of Market Intelligence and Investor Relations. Please go ahead. Richard Kingston | Vice President of Market Intelligence and Investor Relations: Thank you, Betsy. Good morning, everyone, and welcome to SEVA's fourth quarter and full year 2025 earnings conference call. Joining me today on the call are Amir Panoush, Chief Executive Officer, and Yaniv Ariely, Chief Financial Officer of SEVA. Before handing 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 the results of SEVA to differ materially from those expressed or implied by such forward-looking statements and assumptions. We will also be discussing certain non-GAAP financial measures, which we believe provide a meaningful analysis of our core operating results and comparisons of quarterly results. Please see the earnings release we issued this morning for our reconciliations of our non-GAAP financial measures. Our earnings release can be found in the SEC filing section of our investor relations website. And with that said, I'd like to turn the call over to Amir, who will review our business performance for the quarter and provide some insight into our ongoing business. Amir? Amir Panoush | Chief Executive Officer: Thank you, Richard. Welcome, everyone, and thank you for joining us today. 2025 was a landmark year for SIVA. We strengthened our foundation, reinforced our leadership position in wireless connectivity, and accelerated our expansion into AI for the smart edge. Throughout the year, we continued executing on our long-term strategy, partnering closely with customers to solve their most critical technology challenges through a comprehensive best-in-class portfolio of IP platforms that enable smart edge devices to connect, sense, and infer data locally. This strategy matters now more than ever. The shift of AI inference from the cloud to the edge and toward hybrid AI continues to accelerate. And the next wave of innovation is increasingly about physical AI, where devices must connect to and sense their environment, process data locally, and infer in real time to make decisions. Siva is uniquely positioned for the physical AI era. By offering a comprehensive portfolio of IP building blocks spanning connect, sense, and infer use cases, we provide the flexibility our customers need Whether licensed individually or in multi-IP configurations, these technologies drive superior customers' outcomes and strengthen our long-term economic model. Before reviewing the year and our key achievements, I'll first provide an overview of our fourth quarter performance. For the fourth quarter, we delivered the highest quarterly revenue in SIVA's history, which was 7% higher year-over-year excluding the intrinsic design services business, which we divested in 2023. Licensing revenue increased 11%, exceeding our expectation through strong execution across all three of our technology's pillars and reflecting broad demand across multiple end markets. In the quarter, we signed 18 licensing agreements, including three NPU licensing deals, multiple Wi-Fi 7 and combo connectivity wins, and a meaningful software engagement, reinforcing the breadth of our portfolio. Of the 18 deals signed, five were with OEMs. Turning to licensing highlights regarding AI, we reached one of the most significant AI milestones for SIVA to date during the fourth quarter, signing an NPU licensing agreement with one of the world's leading PC OEMs developing its next-generation AI personal compute architecture. Their selection of Civa's Newport NPU portfolio is a strong validation of our technology and represents a breakthrough for on-device AI adoption in the PC category. This win underscores our ability to set the standards for high-performance AI integration into next-generation computing. This partnership is strategically important on two fronts. First, it demonstrates top-tier customers' trust in SIVA's leading and optimized IP foundations to their AI roadmaps, allowing them to focus their engineering talent on software, model optimization, and user experience differentiation. Second, it confirms that the PC ecosystem has reached a tipping point where dedicated NPUs are a baseline requirement for competitive AI performance. As AI features proliferate across operating systems, creative workflows, productivity applications, and local LLM acceleration, the ability to deliver superior performance per watt is the new strategic differentiator, and SIVA is a key player in this transition. Importantly, Our AI momentum is also increasingly reflected in our financial mix as well as deal activity. AI processor licensing represented a meaningful portion of our licensing revenue in 2025. While AI design cycles can be longer than traditional connectivity deployments, disagreements typically carry higher pre-unit and longer-term royalty potential. expanding content per device, and strengthening the durability of our royalty model over time. As for licensing highlights in connectivity, our connectivity business delivered another strong performance in the fourth quarter, highlighting the depth and durability of our wireless franchise. Bluetooth and Wi-Fi IPs continue to see strong demand as customers upgrade to Wi-Fi 7 and Bluetooth high data throughput. This quarter's deals include Wi-Fi 7 for IoT, a multi-use Bluetooth HDT agreement, and three Bluetooth Wi-Fi combo wins. One notable win was with the semiconductor division of one of the world's largest white goods manufacturers, which licensed our Wi-Fi 6 and Bluetooth IP for a combo connectivity chipset, supporting smart home applications. This illustrates a broader trend. Consumer, industrial, and automotive OEMs are increasingly designing their own connectivity silicones to deliver tightly integrated, app-centric experiences, and selecting SIVA as a trusted partner for roadmap critical platforms. As for SendSync, another standout deal in the fourth quarter was a software licensing agreement with a leading TV platform planning to integrate our motion engine technology into its smart TV operating system used by multiple global TV brands. As TVs evolved into interactive experience hubs, motion-based inputs and enhanced user interactions are becoming increasingly important. SIVA's longstanding presence in this market provides deep domain expertise and platform credibility. Now turning to royalties, This was our strongest royalty quarter in more than four years. We also caused our diversified smart edge royalty customers more than offset mobile softness, underscoring the strengths and resilience of our business model. In the fourth quarter, Wi-Fi shipments reached a record high, up 31% year over year, reflecting increased deployment, often as part of combo connectivity chips. Cellular IoT shipments were up 30% year-over-year, driven by smart edge applications, and Bluetooth shipments continue to be our largest volume category. We also saw a recovery from China-based handset customers during the quarter. However, memory pricing and supply constraints continue to impact smartphone shipments. Now turning for the full year 2025 review. For the full year, total revenue increased 2% year-over-year. Licensing and related revenue grew 6%, reflecting strong demand across AI and advanced connectivity. Royalty revenue was down 2%, primarily due to smartphone softness and memory supply shortage impacting overall unit achievement. Importantly, royalties grew sequentially each quarter and we exited the year with our strongest royalty quarter in more than four years. SIVA power devices shipped in 2025 reached a record 2.1 billion units, up 6% year-over-year, with record Wi-Fi shipments, which grew 48% year-over-year, and record cellular IoT shipments, up 42% year-over-year. Overall, we signed 54 licensing agreements in 2025 across our extensive IP portfolio, including 10 OEMs agreements. Importantly, 12 customers licensed multiple SIVA technologies, a clear indication that our strategy to offer a broad portfolio across Connect, Sense, and Infer is resonating and enabling customers to address multiple requirements within a single engagement. Taking a step back, 2025 features several important milestones that reinforce our long-term opportunities. The strength of our connectivity franchise is defined by deep customer integration and scale. During the year, we signed nearly 30 new engagements for our Bluetooth and Wi-Fi IPs, underscoring continuous relevance across smart edge markets. We also secured Wi-Fi 7 agreements with two of our largest connectivity customers, who together have shipped more than 3 billion civil power devices, effectively establishing long-lived royalty engines that we expect to drive billions of units and tens of millions of dollars in royalties over the life of these programs. In addition, our ability to deliver integrated combo solutions continues to differentiate us and improve data economics over time. 2025 was a breakthrough year for SILVA in AI and NPU licensing. During the year, we signed 10 Newport NPU agreements, headlines by comprehensive Newport portfolio license with Microchip and a strategic engagement with a leading global PCOEM, underscoring our attraction across embedded consumer automotive, industrial, and compute markets. This momentum is increasingly reflected not only in deal activity, but also in our financial mix, with AI processor licensing representing a meaningful portion of licensing revenue in 2025. Strategically, the licensing agreements we signed during 2025 are building long-term royalty trajectory and visibility. Based on these science agreements, and our insights into customers' roadmaps, we estimate that they represent an aggregated lifetime royalty potential of $125 million over their expected product life. While this value will be realized over multiple years and is dependent on customers' deployment and market adoption, the magnitude of this opportunity relative to our current royalty base underscores the strength, durability, and accelerating momentum of the licensing and royalty flywheel we are building. In terms of scale and credibility, we celebrated reaching 20 billion cumulative SIVA power devices shipped to date during the year, and in fact, exceeding 21 billion cumulative units by the end of the fourth quarter. These milestones reflect the trust we have built with the industry over decades and positioned SIVA strongly for the physical AI era now underway. A key strength of our business that is often underappreciated is our diversification across smart edge and markets. In 2025, smart edge applications generated 86% of total revenue, driven by market share gains by SIVA-powered customers across consumer, automotive, industrial, and infrastructure markets. As intelligence continued to move into physical devices, this diversified and expanding customer footprint positioned SIVA to evolve naturally from enabling the smart edge to enabling physical AI, where connectivity, sensing, and inference converge to drive the next phase of growth. Entering 2026, we are focused on extending our leadership in established categories and deepening our integration with our customers' roadmaps. By providing a more complete IP stack, we are becoming an even more essential partner to our customers, effectively increasing the value per device. Now, I will turn the call over to Yaniv to review the financials. Yaniv Ariely | Chief Financial Officer: Thank you, Amir. Good morning. I'll now start reviewing the results of the operations for the fourth quarter of 2025. Revenue for the fourth quarter increased 7% year over year and 10% sequentially to an all-time record high of $31.1 million. The revenue breakdown is as follows. Licensing and related revenue increased 11% year over year and 9% sequentially to $17.5 million, reflecting 56% of our total revenue. Royalty revenue increased 2% year-over-year and 12% sequentially to $13.8 million, reflecting 44% of our total revenue. Quarterly gross margin were 88% on GAAP basis and 89% on non-GAAP basis. Total gap operating expenses for the fourth quarter were $28 million, and total non-gap operating expenses for the fourth quarter, excluding equity-based compensation expenses, amortizations of intangibles, and deal costs, were $22.2 million. Gap operating loss for the fourth quarter was $0.4 million, as compared to gap operating income of $0.1 million for the same period last Non-GAAP operating margins and income were 18% of revenue and $5.7 million and grew 20% and 26% year-over-year, respectively, as compared to non-GAAP operating margins of 15% and non-GAAP operating income of $4.5 million recorded for the fourth quarter of 2024, respectively. Financial income was $1.4 million compared to a net loss of $0.1 million for the fourth quarter of last year. Gap and non-gap taxes were approximately $2.2 million, higher than our guidance of $1.8 million, and affected by the first tax asset write-off associated with the utilization, limitation of withholding taxes, and from a regular geography relocation of revenue recognized from deals and royalty revenues in the quarter. Gap net loss for the fourth quarter was $1.1 million and diluted loss per share 4 cents as compared to a net loss of $1.7 million and diluted loss per share of 7 cents for the fourth quarter of 2024. Non-gap net income and non-gap diluted income per share for the fourth quarter of 25, increased 86 and 71% to $4.9 million and 18 cents year-over-year, respectively. Compared to non-GAAP man income of $2.7 million and non-GAAP diluted income per share of 11 cents for the fourth quarter of 24. With respect to other related data, shipped 606 million units of SIVA power devices, down 3% from the fourth quarter of last year. Of the $606 million reported, 108 million units, or 18%, were for mobile handset modems. 479 million units were for consumer IoT products, up from 459 million for the fourth quarter of last 19 million units were for industrial IoT products, down from 35 million units in the fourth quarter of last year. Bluetooth shipments were 303 million units for the quarter, down from 343 million units in the fourth quarter of 24. IoT shipments were a quarterly record 60 million units, up 30% year-over-year. and our Wi-Fi shipments were a record 86 million units, up 30% year-over-year. As for the year, total unit shipments were a record 20.1 billion devices in 2025, up 6% year-over-year, which is equivalent to approximately 66 SEMA-powered devices sold every second in 2025. Annual modem shipments were down 18% year-over-year to 280 million units, reflecting softness in smartphones. Bluetooth shipments were 1.1 billion units, similar to last year. Annual consumer IoT related shipments were 1.7 billion units, up 14% year-over-year. Annual industrial IoT-related shipments were 87 million units, down 31% year-over-year. Wi-Fi, cellular IoT, and audio AI shipments all showed strong year-over-year growth of north of 40% each. In terms of royalty contributions, Wi-Fi royalties were up 70% year-over-year, reflecting higher volumes and ASPs from our Wi-Fi 6 customers and cellular IoT royalties were up 20% year-over-year. On an annual financial metrics, revenue increased 2% to $109.6 million, in line with our updated outlook we shared in May last year. Non-GAAP gross profit remained strong at 88%. Our non-GAAP net income increased 20% year-over-year, and diluted EPS increased 17% year-over-year, all contributing to sustainable and gradual growth and profitability. As for the balance sheet items, at the end of the year, cash, cash equivalent, balances, marketable securities, and bank deposits were approximately $222 million. In the fourth quarter, we successfully executed a $3.5 million share follow-on offering for approximately $63 million net to strengthen our balance sheet. Our DSAs for the fourth quarter were 57 days. And during the fourth quarter, we generated $8.7 million of cash from operating activities. Our ongoing depreciation and amortization was $1.1 million. And purchase of fixed assets was $1.5 million. At the end of the fourth quarter, our headcount was 424 people, of whom 343 were engineers. Now for the guidance. Amir highlighted our achievements in 2025 and the strong fundamentals we have in place to build long-term growth and profitability. From a financial perspective, this execution translates into solid progress across key metrics with annual non-GAAP net income increasing 20% year-over-year and non-GAAP fully diluted EPS growing 17%. These results were supported by record high revenues in the fourth quarter of 25 and non-GAAP operating margin of 18%, reflecting both operating discipline and improving mix. Building on the consistent progress we have made over the last two years gives us the confidence as we enter into 2026. which we view as another year of growth across multiple financial and business dimensions. In licensing and related revenues, we expect growth to be driven by continued expansion of AI adoption across multiple industries, an increasing mix of higher value, more integrated engagements, and our leadership in wireless connectivity supported by diversified product portfolio of connectivity, AI, and sensing IPs. On the royalty side, we are encouraging momentum across our connectivity product lines, including 5G handset modems, Bluetooth, Wi-Fi, and cellular IoT as deployment broaden and program license in recent years continue to ramp. While we do not have the control and the precise timing of royalty growth and continue to monitor factors such as memory pricing and the broader market condition, the underlining trajectory for our business and our diversified end market exposure positions us well moving into 2022. On an annual basis, our total revenue is expected to grow 8% to 12% over 2025, with lower growth in the first half of the year and higher in the second half, similar to prior years and seasonal trends, and subject to the memory pricing fluctuation and supply challenges. On the expense side, we continue to demonstrate strong cost disciplines and operating leverage. Excluding currency impacts, our overall 2026 non-GAAP expense base, including both cost of goods and operating expenses, is expected to increase in the range of 1% to 3%, significantly below our expected top line growth, reflecting the scalability of our business model, but excluding any FX During the second half of 2025, and so far this year, the strengthening of the Euro and the Israeli shekel against the US dollar has created foreign exchange headwinds across the industry, particularly for companies with global distributed engineering teams. As a result, our non-US dollar-based expenses, which are mainly the research and development teams, in Europe and in Israel are expected to increase by approximately 10 percent year-over-year, representing an incremental impact of around $5 million. Taking both factors into account, modest organic expense growth with FX impact, we expect total non-GAAP expenses in 2026 to be in the range of 104.4 to $108.4 million, with non-GAAP cost of goods sold increasing by approximately half a million dollars, and non-GAAP operating expenses increasing by approximately 6.1 million. Importantly, this outlook reflects our continued focus on disciplined investments, efficiency, and maintaining flexibility as we support growth across our diversified smart edge markets. From the guidance and activities we have just discussed, we anticipate non-GAAP operating income and non-GAAP net income to increase significantly by approximately 35% to 40% year over year. Annual 2026 equity-based compensation expenses is forecasted to be between $22 and $23.5 million, and the amortization of acquired intangibles and costs associated with business acquisition, approximately $0.4 to $0.5 million each. Gross margin is expected to be approximately 88% on GAAP basis for the year. Specifically for the first quarter of 26, with traditional seasonality in shipments of consumer IoT and mobile products post the holiday season, revenues forecasted to be between $24 to $28 million, sequentially lower than the record fourth quarter we just reported, but still significantly higher than the first quarter of 2025 at the middle. Gross margin is expected to be approximately 86% on GAAP bases and 87% on non-GAAP bases due to lower seasonal royalties, excluding an aggregate of $0.2 million of equity-based compensation expenses and $0.1 million of amortization of acquired intangibles. GAAP OPEX for the first quarter is expected to be between the range of 27.6 to $28.6 million, higher than the level we just reported for the fourth quarter of 25, at the midpoint of our guidance range, mainly due to the FX effect that I just walked through. Of our anticipated operating expenses for the first quarter, $5.2 million is expected to be attributed to equity-based compensation expense, $0.1 million for demortization, of acquired intangibles and another $0.1 million of costs associated with business acquisitions. Non-GAAP OPEX is expected to be in the range of $22.2 to $23.2 million. Net income is expected to be approximately $1.7 million. Taxes for the first quarter are expected to be approximately $1.3 million. And the share count for the first quarter of 26 is expected to be approximately 27.7 million shares on GAAP and 29.4 million shares for non-GAAP basis. Betsy, you could now open the Q&A session, please. Operator | Conference Operator: We will now begin the question and answer session. To ask a question, you may press star then 1 on your touchtone phone. 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. At this time, we will pause momentarily to assemble our roster. The first question today comes from Kevin Cassidy with Rosenblatt Securities. Please go ahead. Kevin Cassidy | Analyst, Rosenblatt Securities: Yes, thanks for taking my question and congratulations on the great results. For your NPU pipeline, can you just give an idea of the scale? How many more engagements do you have right now compared to, say, this time last year, and maybe even what the end market exposures are? Amir Panoush | Chief Executive Officer: Yeah, Kevin, thanks a lot for congratulating us and for the question. First of all, we started to I'm very, very encouraged by how we executed in 2025 our penetration into the AI. And that was a year of very significant market share gain, as well as more than 10 deals that we have been able basically to capture. With that, we have built a complete portfolio of NPUs for all the different type of smart edge markets and as that transition to the physical AI. So overall, we are well, well positioned right now going to 2026. The pipeline overall keeps growing across pretty much all the different type of sub-market segments that we see across the smart. This is true for consumer, different type of computing devices, different type of embedded MCU type of applications, as well as in the industrial, as well as in automotive. Really, we see a very healthy pipeline across all these sub-markets. Very encouraged with how we have executed and how we see the future going to 2026 on that. Kevin Cassidy | Analyst, Rosenblatt Securities: OK, great. And just as a follow up, a little clarification on the PCOEM. And congratulations on that. But I just wanted to make it clear. I think you said a dedicated NPU. So is this a separate chip or is it integrated in a CPU package too, like in the same silicon with the CPU? Amir Panoush | Chief Executive Officer: Yeah, so first, it's definitely a design win or a deal that we're extremely, extremely excited about. This is one of the top PC OEMs out there. And this is for an OEM that decided to build so-called their own internal AI and NPU functionality within so-called the SoC platform that they are integrating into. So basically what we are delivering them is the whole core NPU functionality. And then they integrated into the SOC that they are building. spk09: So a separate chip. Yeah. A separate chip for a... For NPO. spk07: For NPO. Okay, great. Thank you. Thank you, Kevin. Operator | Conference Operator: The next question comes from Ruben Roy with Stifel. Please go ahead. Ruben Roy | Analyst, Stifel: Thanks, and echo the congrats on a nice end to 25. Amir, maybe I could follow up on Kevin's question and just... talk a little bit more about the NPU win. Can you talk a little bit about the competitive dynamics for that? Because you have others like ARM sort of integrating NPUs. So how should we think about the functionality? Are there going to be multiple NPUs, do you think, in PCs going forward as the AI workloads evolve? Or is this something where you know, from a competitive basis, you guys were able to displace, you know, sort of the, you know, existing solutions maybe that are available to the OEM. Thank you. Amir Panoush | Chief Executive Officer: Yeah, definitely, Ruben. So first I would say that the way that we see right now the landscape and definitely for the high-end compute devices is that there is stronger and stronger need to really best in class performance. And by that I mean the power per watts that you can generate, the so-called latency or the performance of throughputs per token that you generate. This really requires so-called a co-architecture and flexibility of the co-architecture to deliver best in class, what we call PPA, Power Performance Area, that deliver basically a very competitive landscape for our customers. With this specific OEM, they looked at what is available out there and they want to make sure that they have complete internal integration between the hardware and the software to drive the so-called the high performance that they need. But what they need is the underlying core silicon IP technology with the software's come on top of that, that deliver for them the best in class performance. And I think we are well positioned competitively and that's why they picked us in this specific basically design phase. Ruben Roy | Analyst, Stifel: Right. Okay. Very helpful. And then as a follow-up, just to go through the guidance again a bit here, you guys talked a little bit about recovery in China from a handset customer, and obviously there's some moving parts with memory pricing, et cetera. So in thinking through sort of the first half or the second half commentary, can you just give us a little bit of a little more detail on how you're thinking about sort of end demand relative to dynamics out of your control, like memory pricing, et cetera, on first half? Is it much different, would you say, from typical seasonality? I mean, if you look at, as Janice said, you're up year over year at the midpoint, and seasonally it looks pretty similar to what you saw last year. So I'm just wondering what some of the assumptions on things out of your control might be in the first half, if that's much different from typical seasonality. Thank you. Amir Panoush | Chief Executive Officer: Yeah, I would start first that our business, a significant portion of our business is really not so-called dependent on mobile. It's well-ware diversified across the different sub-markets of the smart edge. And in that market, we keep gaining market share. Our customers keep ramping with our different type of technologies. And overall, we expect similar seasonality as we have seen in previous years. But with that seasonality, we keep increasing our market share. Now more specifically on mobile where potentially there is so-called more dependency or can be some impacts related to the memory supply. First again, we are going to see increase in market share thanks to the mobile OEM that is going to integrate more and more the internal modem, at least that's our expectation moving forward. But on a so-called integrated basis, With the other smartphone OEM that we have, definitely, again, there is potential impact coming from the memory shortage. And even there, we do expect meaningful seasonality between the first half and the second half. So on an aggregated basis, we're still expecting quite strong seasonality in 2026 as well, while driven by a market share gain across all the different markets for us. Yaniv Ariely | Chief Financial Officer: Ruben, I'll maybe add to that that our customer in China that you referred to, most of his sales are export to the rest of the world. India is a big market. Latin America, Africa, Eastern Europe type. So it's not necessarily domestic use. And therefore, the demand, the end demand is good. The question is how they will perform with the memory shortages and prices. That's just a little bit of another anecdote with regards to demand and demand at least for the products. And back to your first question, another reference is to the MPU. We came up with another press release of highlighting the entire, not just Q4, but the entire activity and results and achievements we had with AI. And in that press release this morning, we are saying that six of the NPU customers that have signed with us over the last year to two years should be ready in production by the end of the year and then probably or hopefully a royalty contribution in the beginning of 2027 for us from this relative new product line. So that's quite encouraging and we'll wait and continue to monitor their progress. Ruben Roy | Analyst, Stifel: That's really helpful, Yaniv. And I guess you just made me think of another question. So apologies, but I just love to follow up on that last point that you made, which is Amir talked about the $125 million in lifetime royalty potential, and you've got a PC and PU deal here. PC design cycles maybe are a little bit quicker than some of the stuff that you might expect from, let's say, a microchip that's much more broad-based into a lot of different markets. So, you know, if we think about the waterfall of the $125 million, it sounds like you're going to start to see some of that in 27. Any way to think about that pipeline relative to, you know, how it'll flow into the model outside of what I just said? You know, PCs may be a little bit faster than some of the broader markets or anything else you can add on the pipeline. That'd be great. Thank you. Yaniv Ariely | Chief Financial Officer: I think that over time, and not necessarily these first six, part of them, yes, we're going to see on one hand, the high royalty contribution, because as Amir explained, our offering today is both the high end and low end, very sophisticated automotive, PC, type of application as well as the IOT and wearables and the low power type of devices so the most important thing is higher volume for these new royalties but on top of that also higher ASPs on at least the higher end stuff. It's all a mix and this is a little bit more difficult to predict exactly how 2027 will look like and when it's going to hit, whether the first half or the second half. But when we monitor these customers of ours and when we support them in their design cycle, these are the dates and the opportunities we see in front of us. Overall, an increase in dollar revenue content from a new market for us. This is on top of the connectivity. This is on top of the IoT and mobile. It's essentially the third leg of AI. We did very well in licensing. Just over 20% of our licensing revenue for the first time ever in 2025 came out from that market. And potentially in 27, we could see also those royalties start to kick in. Indeed, exciting times. Amir Panoush | Chief Executive Officer: Yeah, I would just add to that, Ruben, that definitely we are extremely excited and encouraged by the fact that those design wins are going to generate, per our estimation, $125 million in terms of royalty potential. And you pointed out very correctly on that in consumer, PC, and so on, the time to royalty is shorter, and definitely we expect with that market type of design wins that it will also start generating in 2027. spk10: Perfect. Thank you. Thank you. Operator | Conference Operator: The next question comes from Sujit Silva with Roth Capital. Please go ahead. Sujit Silva | Analyst, Roth Capital: Hi, I'm here. Congratulations on the strong year and the progress here. The PCOEM win, just keeping up on that, is it more likely that it was a one-off special case for this OEM, or would you think, on the other hand, there's pipeline potential for additional OEMs to follow suit considering SIVA-based solutions as well? Amir Panoush | Chief Executive Officer: First, the PC landscape is such that the number of customers, of course, is not super large versus, let's say, the other more diversified IoT market segments that we're addressing as well. But within that landscape, having the ability to internalize the AI capabilities, and with that, the software hardware integration and the specific optimization to the use cases they want to drive, it's a big value add. So definitely there is potential that other will follow suit with the same type of configuration. And regardless of that, of course, we are extremely excited by the fact that after very significant lengthy type of evaluation, we came at the top based on very, very strong performance metrics that we can provide to, in this case, to the PC OEM, but for potentially other PC customers as well as in other high-end compute devices that need the high-performance type of metrics. Sujit Silva | Analyst, Roth Capital: Thanks, Amir. Very interesting. And then separately, you highlighted in your prepared remarks, Amir, physical AI. I was curious, you know, what pipeline opportunities there are there or current opportunities there are in physical AI that you would call out in terms of apps? And which physical AI app categories are the largest incremental royalty opportunity for you as that ramps up? Amir Panoush | Chief Executive Officer: I think what is emerging more, and this is so-called the growth area beyond so-called our traditional market segments that you're after, is everything related to robotics. And we're already addressing, we will keep gaining market share in the type of like automotive and under-industrial application and the border IoT. But what is really exciting right now, so-called specifically related to physical AI is is the expansion of those capabilities all across wireless connectivity. They need, of course, to sense and understand the environment and then make an inference or decision based on all that information that really is going to happen across robotics. And now robotics moving so-called from a small volume in, let's say, warehouses to potentially be everywhere and supporting all human beings worldwide. So there is very big potential there. Of course, as the year progresses, we will see the real impact of that. spk10: Thanks, Amir. Thanks, Julie. Operator | Conference Operator: As a reminder, if you would like to ask a question, please press star, then 1 to join the question queue. The next question comes from Alec Valero with Loop Capital. Please go ahead. Alec Valero | Analyst, Loop Capital: Hey, guys. Thank you for taking my questions. This is Alec on for Gary. My first question is on your fiscal 2026 guidance. What specifically would need to improve in fiscal 26 to turn toward the high end of your guidance range or even above the high end of the guide? Yaniv Ariely | Chief Financial Officer: Yeah, obviously in guidance, you know, you have the two aspects, revenue and expenses. On the revenue front, 8% to 12% was our long-term growth trajectory back from 2010. analyst day that we did back in december of two or so three years ago so that's that's still intact maybe we've been behind in 25 but we're back to back to that stronger licensing obviously it could help us loyalty ramp up for many of these markets that we talked about this year no less or more effect from memory that those are the normal typical events that could influence the royalty level, obviously the timing of different product ramp-ups and things like that. On the expense side, the biggest element for us this year is less associated to the organic plans and running the company. It's more of a macro thing, which I talked about earlier, the currency exchange rate differences between this year and last year. dollar compared to many other currencies around the world. And while most of our R&D is outside the US, this is hurting us. If there will be some type of future change throughout the next six months or so, one way or the other, that could shorten or increase the gap. But on the other hand, we are fully in control to still offset that or enjoy that if it's on the positive side. So I think these are the more or less moving pieces in our business from a cost and management. We came out with a pretty low expense increase and are managing our investment very, very tight and efficient to try to maximize shareholder value. Amir Panoush | Chief Executive Officer: But maybe just to add on that, Alec, in terms of unpacking Circle, what are the drivers for the top-line growth as we look at 2026? First, definitely our very strong leadership in wireless communication. We see us keep gaining more both on licensing and the royalty keeps increasing very, very nicely across all those different types of sub-markets. And the second, of course, is our momentum in AI. Extremely encouraged about what we have seen in 2025, and we have all the so-called capabilities from a product portfolio and engineering capabilities to drive that momentum even further in 2026. And then last but not least is overall our expectation we'll keep gaining market share both in mobile and Wi-Fi from a royalty basis. Mobile coming from the U.S. mobile OEM and on the Wi-Fi coming from just the continued penetration of our technology and the transition into Wi-Fi 6 and 7 and Bluetooth 7, the driving high royalty per unit. spk09: Got it. I really appreciate all that color. Just a quick follow-up. Alec Valero | Analyst, Loop Capital: So with your recent equity capital raise, I believe you are about at $200 million in the balance sheet. spk09: How do you think about M&A today, and what do you think about the current valuations? Yaniv Ariely | Chief Financial Officer: I think you guys are the expert for that, right? We wanted to strengthen our balance sheet. We're looking for non-organic growth to grow faster and gap that licensing to royalty 18 to 24 months timeframe. That's the merit in raising that cash. And that's our goal. That's our goal for the next 12 months to find the right fit technology-wise, market-wise, business-wise to increase that. Hopefully when the market, if we do well and continue to execute and the market understands that Siva is a very interesting AI play which I'm not sure we're yet being recognized for that. I see a lot of value for shareholders, but that's your forte, not ours. We'll manage the business. Amir Panoush | Chief Executive Officer: Yeah, one thing to add, thanks, Yanni. One thing to add, Alec, in terms of the balance sheets or the cash position, I strongly believe we really have built... excellent, excellent IP enterprise in terms of being able to deliver so-called IP licensing across many different markets. And the goal, of course, is to utilize that balance sheet to find additional assets out there in the IP domain that we can take on and expand even further our potential for growth and profitability. And so this really helps us to have the financial strength to go and be able to expand it further. spk07: Got it. Super helpful. Thank you very much. Well, congrats. This concludes our question and answer session. I would like to turn the conference. Richard Kingston | Vice President of Market Intelligence and Investor Relations: Yeah, we're back to, I think Amir has some closing remarks. Amir Panoush | Chief Executive Officer: In closing, I want to thank our employees worldwide for the dedication and execution through 2025. We enter 2026 from a position of strength with a diversified business model and deep customer integration across the market, driving the emergence of physical AI. With leadership in connectivity, accelerating traction in AI, and a portfolio designed to scale across, connect, sense, and infer, we believe SIVA is well positioned to continue building long-term value for our customers and shareholders. Richard, I will hand over to you to wrap it up. Richard Kingston | Vice President of Market Intelligence and Investor Relations: Thank you, Amir. Thank you, Amir. As a reminder, the prepared remarks for this conference call are accessible through the investor section of our website. And with regards to upcoming conferences, we will be participating in the following events. Mobile World Congress, March 2nd through 5th in Barcelona, Spain. Loop Capital Markets, 7th Annual Investor Conference, March 10th in New York. The Stifel 2026 New York City Technology 101 Conference, March 11th in New York. and the 38th Annual Roth Conference, March 22nd in California. 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. Operator | Conference Operator: The conference is now concluded. Thank you for attending today's presentation. You may now disconnect. jsPDF 3.0.3 D:20260606090042-00'00'

Research summary and source transcript

readyJun 10, 2026

The transcript is a webinar presentation by SIVA employees discussing NPU architecture and AI trends, not an earnings call for CEVA, Inc. No financial results, guidance, or company-specific performance metrics are disclosed. Therefore, there is no basis to assess changes in CEVA's business, financial health, or strategic position.

The transcript contains no information about CEVA, Inc.'s financial performance, product roadmap, customer wins, or competitive positioning that would constitute an information gradient. As this is a SIVA-hosted educational webinar on AI architecture with no reference to CEVA's earnings, guidance, or material business developments, the market already possesses all information available in the transcript — there is no asymmetric knowledge.

Not assessable; the transcript does not discuss CEVA's business model, revenue drivers, or operational variables.

  • NPU scalability across edge-to-cloud
  • Architectural extendability for evolving AI models
  • Power and energy efficiency via sparsity and quantization
  • Software toolchain support for configuration and accuracy estimation
  • Future-proofing AI hardware against model obsolescence
  • Detailed explanation of dynamic group quantization and its accuracy-preserving benefits
  • Emphasis on VPU integration for flexible operator support without CPU fallback
  • Highlight of NewPro-M architecture's configurability for performance, power, and area trade-offs
  • Pride in SDK tools enabling customers to simulate and validate NPU configurations pre-silicon
  • Confidence in backward compatibility and migration path for legacy IP customers

The presenters (Roni Wattelmacher and Asaf Ganor) speak with technical confidence and detail, using specific architectural examples and tool references to support claims. Their tone is educational and promotional, focused on demonstrating expertise in NPU design rather than financial performance. There is no evasiveness or overstatement detectable in the Q&A; responses are direct and grounded in the described technology. However, as this is not a CEVA earnings call, the tone reflects SIVA's marketing and engineering messaging, not CEVA management's credibility or forthrightness regarding financial results.

  • No clear dodged analyst question was detected by the local fallback; manual review should still check whether Q&A answers quantified conversion, margins, and guidance.
  • There may be a benchmark or metric-framing issue worth manual review, especially around adjusted metrics, timelines, or changed expectations.

Not assessable; the transcript does not contain any information about CEVA, its products, market position, or competitors.

  • Key figure to verify: Our customers ship around 2 billion chips and devices annually using our technologies.
  • The quarter appears to be moving from story to evidence: operating momentum is showing up in revenue, royalties, or backlog rather than only in management narrative.
  • Customer activity looks healthier than a one-quarter spike because the transcript points to both retention/renewal work and new-account activity.
  • AI and data-center exposure look strategically relevant rather than cosmetic, because management ties demand to compute-heavy end markets instead of treating it as a generic buzzword.
  • The transcript gives limited margin evidence, so the quality of revenue still needs corroboration from gross margin, operating leverage, and cash conversion.
  • Transcript misidentified as CEVA earnings call; actual content is SIVA webinar with no CEVA financial or operational data
  • No revenue, margin, backlog, bookings, or guidance figures disclosed
  • No discussion of CEVA's customer base, design wins, or market share in licensing segments
  • Absence of any mention of CEVA's financial condition, cash flow, or capital allocation
  • Inability to evaluate competitive position or business performance due to wrong source material

The transcript discusses NPUs for edge and cloud AI inference in general terms but contains no specific information about CEVA's exposure to data center AI spending, cloud provider engagements, or AI accelerator licensing in data center contexts. Any inference about data center impact would be speculative and unsupported by the transcript.

  • Confirm whether this transcript is actually from a CEVA, Inc. earnings call
  • If not, obtain the correct CEVA FY2025 Q3 earnings call transcript
  • Identify CEVA's actual revenue, licensing income, and royalty trends for the quarter
  • Assess CEVA's backlog or bookings growth in key segments (Mobile, Consumer, Automotive, Industrial)
  • Evaluate any updates to CEVA's guidance or capital return policy
  • Determine CEVA's progress in AI-related IP licensing and design wins

FY2025 Q3 earnings call transcript

45,451 chars
NASDAQ:CEVA Q3 2025 Earnings Call Transcript Generated on 6/6/2026 Roni Wattelmacher | Director of Product Marketing, Vision and AI Business Unit, SIVA: Hello everyone and welcome to our webinar, what it really takes to build a future-proof AI architecture. I'm Roni Wattelmacher, Director of Product Marketing for the Vision and AI Business Unit at SIVA. Joining me today is Asaf Ganor, our Director of AI Architecture. We're excited to discuss, show, and teach you how one should build his AI chip architecture for it to be future-proof and why it's important. Before we start, a few logistics. First of all, please fill out the survey that you see at the top of your screen. It's important to us. Secondly, we'll be doing a live Q&A session at the end. If you have questions, please write them in the Q&A box to your right. You can enter them at any time, and we will answer them later. On the left, you can see a resource list with links that you can download, links to interesting web pages, and contact information for Asaf and me if you wish to send us a message. Let's start. Let me quickly walk you through what we'll cover in today's session. We'll start with a quick overview of the AI market landscape and how it's evolving. Then we'll dive into key technology trends driving AI workloads, both at the edge and in the Cloud, and how the industry is shifting. Next, we'll talk about why future-proofing is no longer optional in AI hardware and software design, especially as workloads grow more complex and distributed. Then Asaf will take over and go deeper into how NPU solutions are designed to address these challenges, ensuring scalability across multiple generations of AI workloads. And finally, we'll open it up for your questions. Feel free to drop them in the chat during the session. For those of you not familiar with SIVA, we license silicon and software IP to semiconductor and OEM companies. Our customers ship around 2 billion chips and devices annually using our technologies. Our IPs address three core use cases across multiple end markets, Connect, Sense, and Infilm. Connect will provide IPs for Bluetooth, Wi-Fi, UWB, cellular IoT, 5G modems, from base station to satellites. In the sense, we offer IPs and software for vision, audio, radar, motion, and microphone sensing powered by DSPs and sensing algorithms. In fair, today's webinar focuses on this domain, our AI solutions. We license a scalable family of NPUs and the unified AI SDK to enable efficient inference at the edge. This chart shows how different use cases require different performance, which correlates with the power consumption and area. Milliwatts to many watts, less than one square millimeter to few cores of few square millimeters. As we move right on the chart, power consumption increases, and as we move up, inference capability increases. AI inference happens across a wide range of devices, from tiny wearables consuming milliwatts of power to large multi-core systems in the cloud. Multi-core architectures are needed as we move towards complex AI workloads like automotive, co-pilots, and cloud inference. Computing has evolved in distinct eras, each marked by a dominant platform and a step change in the number of deployed units. We started with mainframes in the 1960s, which had about a million units deployed and moved through many computers, PCs, desktop internet, and the mobile internet. Each era saw 10x to 1,000x growth in the number of connected devices. Today, we're entering the AI era, which is expected to doff all previous computing cycles. Forecasts suggest there will be tens of billions of AI-enabled devices, making this the largest compute cycle ever in terms of unit volume. Just like CPUs powered the PC era and GPUs scaled cloud compute, NPUs are the compute engines of the AI era. NPUs are optimized for AI inference and are designed to handle computational needs for AI workloads efficiently. What's unique about the AI era is that compute is not just in the cloud anymore. It's moving to the edge, powering devices like wearables, autonomous vehicles, smart cameras, and IoT nodes. Cloud still plays a role for training and large-scale inference, but real-time AI happens at the edge, where NPUs deliver the required performance and energy efficiency. This growth isn't incremental, it's exponential. Let's talk briefly about how cloud AI infrastructure is evolving to support the demand of the AI era. Cloud providers are the home for training massive foundation models and running large-scale inference that serves billions of users worldwide. Developers today don't have to manage the complexity of this hardware. All of this compute power is abstracted into AI services, fully managed, scalable, and easy to deploy. We're seeing a shift where models and hardware are being designed together. Transformers, for example, are being optimized to run efficiently on new silicon, driving better throughput and lower latency. Energy consumption has become a critical concern for AI farms. Cloud providers are focusing heavily on performance per watt, creating chips that do more while consuming less energy. It's not just NVIDIA anymore. Cloud providers are building purpose-built AI accelerators like Google's TPU, AWS, Inferentia, and Tranium, and Microsoft Maya chips, each optimized for the specific AI workloads. Finally, it's important to understand that the cloud and the edge complement each other. Heavy compute workloads like model training and large batch inference happen in the cloud, but real-time latency-sensitive tasks run directly on the edge where fast response is critical. Industry leaders are now expanding from cloud-first approaches to edge-native intelligence. recognizing the need for low latency on-device AI across robotics, IoT, industrial systems, and consumer products. These examples you see here on the slide show how the entire industry is preparing for an edge-first AI future, where devices themselves are intelligent, not just connected to a cloud that is. More optimized and efficient models are accelerating edge deployment, smaller size, and lower memory use. Generative AI in edge devices offers a wide range of benefits. Lower energy consumption, reduced latency, enhanced privacy, and cost efficiency. Use cases include chatbots using Edge LLMs and SLMs, for example, for appliances, medical analysis based on text, text generation by IoT devices. Emergent Edge devices will support complex GenAI tasks with on-device learning, enabling personalized private and cloud-independent AI. Dedicated edge NPU hardware is required for boosting real-time performance and energy efficiency using advancements like compression, distillation, transformer support, and low-bit networks. We've all heard of LLMs, Large Language Models, which brought human-like text generation to mainstream AI. Then came LVMs, large vision models, which applied transformer architectures to images and video. Now we're entering the era of LMMs, large multimodal models, which process and understand text, vision, voice, and sensor data together, creating much more capable AI systems. Originally, these multimodal models were designed for the cloud, but now we've seen models optimized to run efficiently on mobile devices, wearables, and industrial edge nodes. This is possible thanks to the breakthroughs like low-bit quantization in, for example, precision and optimized transformer architectures. As shown here, the multimodal AI model sits at the center, integrating and processing text, images, audio, and video, enabling devices to better understand the world and act as it should in the world around it. Now, let's talk about agentic AI, a key evolution in edge intelligence. These are AI systems that don't just recognize patterns. They autonomously plan and execute tasks without human intervention. Agentic AI is built on three core capabilities. Autonomy, the ability to operate independently and make its own decisions. Adaptability, adjust its action based on changing environments or goals. And decision-making, selects the best action to achieve a given objective. Let's see some use cases. Smart personal assistants. Think about wearables or home hubs that don't just respond to voice commands, but proactively assist based on your context, adjusting the lightning, suggesting action, or providing health insights in real time. Predictive maintenance. In industrial settings, machines use agentic AI to monitor their own performance, detect anomalies, and even schedule maintenance before a breakdown happens, reducing downtime and saving costs. These use cases run directly on edge devices where real-time decision-making is critical and cloud latency isn't acceptable. So far, we've talked a lot about AI models and compute platforms, but it's also important to recognize how AI is being integrated into the physical world. Physical AI refers to a system where AI doesn't just process data, it controls real-world devices. robots, drones, autonomous vehicles, and smart appliances. One of the most advanced examples today is Waymo's fully autonomous vehicles operating on public roads with no human driver in the front seat. On the right, you can see market data from San Francisco showing how Waymo's market share in ride-hailing has been growing steadily. In fact, Waymo's fully autonomous fleet has now suppressed lift in gross booking share and continues to close the gap with Uber. This is real-world proof that AI-powered physical systems are commercially viable and scaling up in urban environments. Physical AI is not a future concept. It's happening now. From autonomous delivery robots to self-driving cars, AI is making decisions and interacting with the real world in real time. Okay, let's talk about future-proofing. When we talk about future-proofing AI architectures, we're really talking about preparing for change. Most of what I just showed you did not exist just a while back. The AI and machine learning landscape evolves extremely fast. Models that are state of the art today could be obsolete in 12 to 18 months. New hardware accelerators, better algorithms, and entirely new use cases emerge constantly. AI workloads are evolving fast. Models today are bigger, more complex, and optimized differently than they were just a few years ago. If your NPU only supports a narrow set of operations or model types, you'll be stuck when the next generation of models comes out. Future proofing is critical. Now let's see how a typical system flow looks like. This common block diagram represents how different hardware components work together to run vision and AI workloads efficiently. It starts with the sensors capturing raw data, sensors like cameras, lidars, and so on. Some systems will include an image signal processor to pre-process that data. Then the data flows into the NPU and vision processor, where the main AI inferencing happens. The NPU engine accelerates the AI workloads. The NPU controller manages the processing flow. L1 and L2 memories are tightly integrated to provide fast access to data and instructions. Outside the NPU, the CPU, usually ARM or RISC-V are used, runs the system software decision-making logic, but also can run simpler processing tasks. The GPU, if present, can be used for graphics or some parallel compute workloads. The SRAM provides additional fast memory where needed, The system also interacts with peripherals, and for larger data, accesses external DDR memory. This is the high-level view. Now I'll hand it over to Asaf, who will take you deeper into how the AI and vision flows work and how to prepare for the future. Asaf Ganor | Director of AI Architecture, SIVA: Thanks, Roni. Hello, everyone. Great to be with you today. Let's take a step further into what it really takes to design AI architectures that stand the test of time. Building a future-proof AI architecture comes down to three foundational pillars. Scalability, which is the ability to dynamically scale compute resources to match a wide range of workloads. This isn't just within a single product, but across your entire product portfolio, from edge devices to the cloud. Extendability, the ability to seamlessly add support for new workloads, operators, and system flows, while keeping both software and hardware overhead minimal. And finally, sustained efficiency, ensuring maximum computerization, even as workloads diversify and evolve over time. These three pillars, scalability, extendability, and efficiency, are what enable resilient, long-living AI systems. Let's take a deeper look into the first pillar. Scalability is the ability to efficiently scale computes, memory, and system resources across workloads from ultra-low power wearables to cloud-scale data centers without compromising performance, power, or area. But this is easier said than done. Here are some of the key scalability challenges architects face. Resource imbalance occurs when an architecture optimized for high compute workloads is first forced to run lightweight tasks. In these scenarios, large computer arrays remain underutilized, and the overhead from logic and memory leads to unnecessary power consumption, making the solution inefficient for these tasks. Efficient workload distribution is another core challenge. The workload is initially optimized for a specific set of system resources. But as new workloads are introduced, this must share resources with legacy tasks. Over time, this puts pressure on both hardware and software tool chains to maintain efficiency and performance across use cases. Memory systems add their own layer of complexity. A truly scalable AR architecture must be able to handle a wide range of memory configurations, from remote high latency memory to ultra bandwidth systems like HBM or 3D DDR. It must dynamically balance memory access and processing cycles, maximizing data reuse and minimizing memory latency exposure. Next, physical design complexity. As we scale up compute capabilities, physical challenges intensify. Efficient floor planning, architecture wear implementation, and advanced physical design tools are critical to meet frequency and power targets while maintaining scalability. And finally, customization. One size does not fit all. Different applications may prioritize different KPIs, performance, power, area, or a balance of the three. A scalable architecture allows you to tune the resolution of processing elements, enable efficient workload-specific optimization across your product portfolio. Let's walk through a high-level block diagram of SIVA's Nucor-M AI architecture as an example. This diagram shows the main units of the IP and how each one addresses key scalability challenges we discussed earlier. At the top, we have the NPM engine, which serves as the main computer. Let's break down its components. The tensor processing units. This is the core MAC array optimized for metrics-heavy operations like conclusions and metrics multiplication across various data types. The NPM core's job is to keep the MAC read as busy as possible to maintain high utilization throughout the inference process. Vector processing unit, or VPU, is a flexible DSP core tuned for programmable workloads and integrated into the hardware pipeline. It's ideal for supporting new parameter types and specialized control logic without stalling performance. Sparsity acceleration. This unit takes advantage of the inherent sparsity in many neural networks. By skipping zeros, it reduces memory bandwidth demands, saves power, and boosts performance. Streaming nonlinear and quantization acceleration. Responsible of nonlinear operations like activations, such as railing, pooling, normalization, quantization, and even softmax. It processes data in a streaming pipeline, avoiding back pressure to the MAC grid and sustaining throughput. L1 memory fabric is a dedicated ultra-high bandwidth local memory that feeds data into each engine with minimum latency and maximum efficiency. Engine scheduler, a hardware level controller that orchestrates data flow between units within the engine. It ensures smooth data transition and prevents stalls across the pipeline. The number of NPM engines in the core is configurable. You can tailor the compute density to your application's needs, from low power to high performance use cases. Now let's look at the L2 subsystem, which manages data movement between external memory and the compute engines. The L2 fabric acts as a central buffer between the memory, like DDR, and the engines. The system DMAs handle the transfer of data from the remote memory to L2. These are optimized to offload complexity from the software stack, making 2-chain linear and more efficient. Weight decompression. Compressed weights are stored in memory and decompressed on the fly to reduce bandwidth and memory footprint. Data reshaping converts data layouts from DDR to L2, from L2 to L1 in each engine, and vice versa, to match the expected format and dimensions of each compute unit. This minimizes overhead and maximizes throughput. System scheduler manages coordination between engines. It enables dynamic workload sharing, task splitting, and buffer synchronization so engines can collaborate efficiently. Lastly, the controller core plays a crucial role. It interfaces with the host CPU, orchestrates application-level tasks, and handles pre- and post-processing when needed. Let's take a closer look at the configurable building blocks of the new Pro-M core. One of the key strengths of this architecture is its scalability through configuration, enabling precise optimization for performance, power, and area. First, the number of NPM engines is configurable to match your workload scale. Whether you're designing for edge AI or high-performance compute, the architecture scales to fit. but flexibility does not stop at the top level many of the compute resources within each engine are also highly customizable you can adjust the number of mugs per cycle the supported data types are selectable from a wide arrayed range both in the integer and floating points and realms in every product variant only the needed data types are implemented helping maintain maximum area and power efficiency and you can tailor the mac array topology itself adjusting its structure to match specific goals for example In an area-optimized configuration, int for max can be reused in a laddered setup to support int8. Meanwhile, a power-optimized implementation might dedicate separate mark hardware for each data type to minimize switching and improve efficiency. The same scalability applies to the nonlinear quantization accelerator itself, which may be provisioned to match your expected network workload. And finally, the software toolchain enables accurate PPA modeling, help you analyze trade-offs early and make informed design decisions based on your application KPIs. In uProM, flow control is managed through a dedicated hardware sequencing mechanism. This means that once tasks are set up, the hardware handles much of the execution autonomously, reducing the need for constant software oversight. The benefit of this approach is that it helps maintain consistent utilization across the compute engines, even when system configurations vary, whether you're working with a minimal setup or a highly parallel one. On the software side, the design supports VMA virtualization and abstraction. Rather than having to manually program buffer sizes or adapt drivers for every product variant, the architecture supports a consistent driver interface decoupled from hardware-specific parameters. This approach can significantly reduce integration efforts and code duplication when building multiple products on the same architecture, especially when working with different memory layouts, bandwidth constraints, or compute profiles. Scalability goes beyond just compute. It extends into memory architecture and system-level scaling. First, memory resources are tunable to match the system class. Whether you're targeting a microcontroller with PS RAM or a high-end platform with HBM or 3D DDR, the architecture supports the right bandwidth and size trade-offs to fit your performance and power envelope. Second, at the system level, NewProM is built to scale across multiple cores and even across multiple clusters of cores. That means you can grow the architecture to handle larger models or higher throughput without re-architecting your design from scratch. Together, these features ensure that scalability isn't just about adding compute. It's about growing all system components in harmony. Let's now look at how efficiencies are addressed across different layers of NewProM architecture. from hardware to software and from block level design to system behavior. First, at the hardware level, the design incorporates several well-established techniques for power efficiency. Fine-grained clock gating helps eliminate unnecessary switching in inactive models. Power domain shutdowns allows different blocks to be completely powered down when not in use. And memory retention supports fast wake up without data loss during sleep transitions. These techniques together support dynamic power management, which is essential for workloads that fluctuate in activity level. At runtime, the system provides an efficient execution environment, enabling smart allocation of compute and memory resources depending on task needs. This is especially important when running multiple models or switching between workloads with very different compute profiles. From a physical design perspective, the macro level layout of NewPro-M is engineered to be scalable and portable. That means it can adapt across different silicon variants such as power optimized or area constraint design with minimal layout effort. And finally, the compiler plays a critical role in maintaining efficiency. It adapts the tiling strategy, how data and operations are split based on available compute and memory resources. This allows for better data reuse, lower transfer overhead, and ultimately higher throughput per watt. Together, this element supports a design that can scale effectively, both across SQL configurations and across real-world workloads. When evaluating AI architectures, it's easy to get caught up in the wrong metrics, numbers that look impressive but don't necessarily reflect real-world efficiency or scalability. Let's start with a few commonly used metrics that can be misleading. Power in watts. A low power number might seem desirable, but it doesn't mean the design is efficient. It could simply indicate underutilization. In other words, the system isn't doing much work to begin with. Area in millimeter square. Less area often seems desirable. It suggests lower costs and higher integration potential. But if that smaller area isn't used efficiently, you may end up needing more silicon overall to meet performance targets. In the end, this can actually increase cost and power, defeating the original intent. And inference per second, IPS. This seems like a solid performance metric, but it lacks normalization. IPS can artificially be increased by scaling up hardware, so it doesn't reflect architectural quality or efficiency on its own. Instead, we should focus on normalized efficiency metrics, ones that tell us how much useful work the system is doing relative to cost or resource usage. Two key metrics are IPS per millimeter square. This is your area efficiency. It tells you how much inference throughput you get per unit of silicon, a critical metric when optimizing for cost integration or DICON strain, and IPS per watt. This is your energy efficiency. It tells you how much useful work you're doing for every unit of power consumed, which is key in both battery power devices and thermal limited applications. On the right, we see how these two axes combine into a performance map. The top right corner, high RPS per watt and high RPS per millimeter square represent the scalable sweet spot. That's where you want to be. Other quadrants reveal trade-offs. For instance, dense but slow designs might be efficient in area but drain too much power. And energy hobs might perform well but consume more power than they're worth in dive area. Evaluating AR architecture based on these normalized metrics give a far more realistic picture of how well a system will scale, especially when moving from proof of concept to real deployment. So when someone claims high RPS or low power, the next question should be, for what? Efficiency only becomes meaningful when we anchor it to area and energy context. Let's now shift to the second pillar of future-proof AI architecture, extendability. In simple terms, extendability means the ability to add new operators and system flows to your existing AR hardware and software without having to redesign the architecture every time something changes. This has become increasingly important as the AI landscape evolves so quickly, and it brings a few challenges worth highlighting. First, we have rapid model evolution. New neural networks with new layer ops or patterns are emerging all the time. Architectures that can't adapt will fall behind quickly. Second, many accelerators rely on fixed operator sets. They work well for today's model, but struggle when something new appears. This limits flexibility and makes it harder to stay up to date. Third, the fallback strategy is offering to the CPU, but that's not sustainable. Offloading unsupported ops to the CPU consumes more power and slows down inference, especially at scale. Fourth, it's not just about operators anymore. We're seeing full system flow changes with new quantization strategies, memory layouts, and normalization schemes that affect the entire system pipeline. A rigid system can't handle this gracefully. And lastly, the toolchain becomes critical. Hardware extendability is only useful if your SDK and software tools let you actually use it. Supporting new flows in software is what makes real-world deployment flexible and maintainable. So, while performance and efficiency often steal the spotlight, extendability is what keeps an AI architecture relevant over time, especially in a fast-moving field like this. Let's come back to our new Pro-M example. One of the key enablers for architectural extendability is having a flexible non-linear pipeline. Instead of hard coding activations, NewPro-M uses software-controlled non-linear functions, making it future-proof as new activation type emerges in models. This flexibility goes beyond just math. It enables custom system flows where operations like softmax, normalization, or dynamic quantization can run in tailored data streams, independent of the main path. In some use cases, this pipeline can operate in a standalone mode, bypassing the L1 memory completely, saving latency and energy for ops like pooling. Finally, dynamic tensor shaping lets the system adjust data on the fly, so even if the model changes layout or shape mid-inference, the hardware can handle it without stalling. A flexible pipeline needs tight orchestration, and that's where hardware-controlled sequencing steps in. Instead of relying on software to micromanage timing, NewProM Engine's scheduler enforces timing in hardware while still following software's high-level policies. This approach also allows for modular integration of external compute engines. Each new unit connects through a standard data control handshake, keeping the system clean and scalable. And with centralized flow control, you avoid data mismatches or deadlocks, since each stage in the data path is synchronized. This level of control keeps performance stable, even as workloads get more dynamic or complex. To support unknown or emerging operators, NewPro-M integrates the vector processing units, or VPU, directly into the pipeline. You can think of the VPU as a configurable DSP, designed to handle a wide range of operator types. It's not locked to specific logic. Instead, it supports everything from basic element-wise operations to complex model-specific custom kernels. Importantly, these kernel extensions are not isolated. They are supported end-to-end by the software toolchain, and they plug into the existing system flow without requiring structural changes or manual workarounds. This makes the VPU a powerful tool for efficient kernel mapping. Instead of falling back to a CPU or relying on its external cores, future operators can run directly on the VPU with minimal overhead and high throughput. And because the VPU is tightly coupled with the main pipeline scheduler, data can move fluently between the VPU and tensor data path, eliminating idle time and keeping the pipeline efficient even as new workloads are introduced. Let's now look at the third pillar of the future-proof architecture, efficiency. Efficiency is becoming more complex to maintain as models evolve, especially with a shift towards low-precision computing. The industry is clearly moving towards smaller data types, like INT4 and also FP4 and FP8, because they offer compelling benefits. They reduce power consumption, lower memory footprint, and enable smaller MAC units, leading to more compact circle designs. But with those gains come non-trivial trade-offs. As we shrink data precision, we also increase the risk of degraded accuracy. The math becomes more sensitive, and models can lose fidelity if precision is cut too aggressively. To counter that, systems must support advanced quantization techniques and sometimes even mixed precision execution, balancing small formats with higher precision fallbacks where needed. Another efficiency strategy is sparsity, skipping over zero values or redundant computations. While this can reduce bandwidth and improve performance, it's not always a drop-in solution. Many models need retraining to regain accuracy after a sparsity is introduced. And since sparsity techniques are often model or platform specific, they can reduce the portability and reuse of pre-trained models across different systems. In short, efficiency today isn't just about smaller formats. It's about managing the trade-offs they introduce and designing systems that can adapt intelligently to them. As precision shrinks in modern AI workloads, Maintaining accuracy becomes a key concern, especially with formats like INT4 or FP4. Traditional quantization typically assigns one scale factor per tensor or channel. That's simple, but it often results in degraded accuracy, especially for low precision formats where fine-grained variations get lost. Dynamic group quantization takes a more nuanced approach. Instead of one scale for the entire tensor, it applies different scale to smaller element groups, say a few rows or columns at a time. This boosts local precision without needing more bits. The challenge, of course, is to implement this without sacrificing performance on the weights and the dynamic data generated within the network itself, which is exactly what we've set out to solve. Let's illustrate the problem with traditional quantization. In this simplified metrics example, both weights and activations are quantized using a single scale parametrics. That means the entire range of values is compressed using a common scale, regardless of local variation. This course granularity may work fine for some data, but in real world scenarios, it often leads to poor value representation, especially at low bit width. Several values differences get lost and the model may suffer from accuracy degradation as a result. Dynamic group quantization solves this by applying multiple scales across smaller element groups. For example, rows or columns can be grouped and scaled individually. You can see that in the coloring. Each group has its own scaling and zero point. This makes the quantization more adaptive to local variation, preserving accuracy where it matters most. What's especially important is that this more detailed scaling doesn't compromise efficiency. The compute pipelines remain streamlines, even when activations are quantized dynamically at runtime from one layer to the next. The equation at the bottom just expresses the composite scaling and zero-point logic needed to reconstruct accurate results from quantized values. To show the impact of dynamic quantization in practice, here's a quick benchmark using LAMA-27B. The baseline model is FP16, and we compare it to dynamic quantization configuration using 4x8 bit quantization, 4 bits for the weight, 8 for activations. The results, a 4x performance gain and 4x smaller memory footprint for weights, while keeping the accuracy degradation to just 2%. This is a great example of how smarter quantization strategies like dynamic quantization can unlock meaningful efficiency gain without paying a heavy price in model quality. One of the most effective ways to boost efficiency in AI workloads is by leveraging sparsity, specifically the ability to skip over zeros in data and weights during inference. New Pro-M enables this through support of unstructured sparsity, which means it can dynamically detect and skip randomly scattered zeros without requiring any special pruning patterns. This allows the architecture to take advantage of the native sparsity present in many neural networks, achieving real performance gains without needing to retrain the model. Of course, structured and semi-structured sparsity is also supported, and retraining can be used when there is a need to guarantee a specific acceleration factor, even if that comes with a small accuracy trade-off. For example, many transformer models can be proven to run 50% sparsity without degrading accuracy. That unlocks several key advantages to performance, DDR bandwidth reduction, and power consumption. Let's wrap things up by revisiting what it really takes to build a future-proof AI architecture. AI is evolving fast, so the system we build needs to evolve just as quickly. And that comes down to three fundamental principles. First, scalability. Your architecture needs to stretch across a full performance spectrum, from the smallest edge devices to the largest cloud deployment. That means modular building blocks, flexible configurations, and the ability to serve a wide range of applications without redesign. Second, extendability. AI workloads don't stand still. Operators change. Models evolve. New techniques emerge. A robust architecture must support that evolution, allowing new features and flows to plug in over time without ripping out what already works. And finally, sustained efficiency. It's not just about having high peak performance on paper. It's about keeping the hardware busy even when the workloads are messy. That includes handling low precision data, exporting sparsity, and minimizing memory bottlenecks. Together, these three pillars, scalability, extendability, and efficiency, are what makes an architecture resilient, adaptable, and ready for what's next. Roni Wattelmacher | Director of Product Marketing, Vision and AI Business Unit, SIVA: Okay, thanks Asaf for the deep dive into the NPU design challenges and solutions. I'd like to emphasize that SIVA's award-winning NewPRO architecture brings together the key pillars of a future-proof AI solution as was just explained. Scalability across edge-to-cloud deployments, extensibility to adapt to evolving AI models and workloads, sustained efficiency through advanced quantization, sparsity, and power management. New Pro-Am is available today, powering a wide range of future-ready AI products. We hope you've enjoyed this. Let's move on to the live Q&A section and take your questions. Moderator | Webinar Moderator: Thank you, Roni and Asaf. Moderator | Webinar Moderator: We'll now move to the live Q&A session. You will find an icon with a question mark at the bottom of your screen where you can type in questions. Also note the survey at the bottom of your screen. If you can answer that, that would be great. And we'll start to answer some of the questions which you have fed here. Roni Wattelmacher | Director of Product Marketing, Vision and AI Business Unit, SIVA: so the first question is for you ronnie sure how can i pick the right mpu a member which is right for the models i plan to run well thank you and good question so siva provides um sdk a few tools that will allow you also with that and we have the architecture planner It helps you to analyze the model's compute, memory, and bandwidth, and according to the requirements, gives you the results. It allows you to simulate different NPU configuration and find the optimal match to those requirements, the performance, power, and area. After you do that, and you select the suitable architecture, you can use the AET tool, the accuracy estimation tool, to verify and to evaluate the impact of quantization and other hardware-specific optimization on the model accuracy. Together these tools help you to confidently choose the right SIVA MPU for your application, but you don't do it alone. SIVA's support personnel are by your side helping you during the entire process, so they guide you and take you all the way until you find the optimal solution. Moderator | Webinar Moderator: Thanks, Roni. Another question here is about Probably for you, Asaf, what is the estimated power saving when using the dynamic sparsity and quantization you spoke about in the session today? Sure, I'll take this one. Asaf Ganor | Director of AI Architecture, SIVA: So the power saving is very model-specific and also data-specific. Features like the dynamic sparsity give you more the more sparseness you have in the network itself, or the more aggressive you want to be with the tuning versus the accuracy trade-offs. Regarding the data types and quantization, again, advanced techniques allow you to go lower in data type without sacrificing the accuracy. The more aggressive you are and the more the network is comfortable with quantization, you can get more benefits. You don't have to guess. The SIVAS toolchain allows you for each network to input the number, the types of data types and quantization strengths you want to apply. You have accuracy estimation tools that you see results of your pruning, so you can know exactly what you get before you get running. Maybe adding a few sentences, what are the dominant power factors from these features, right? So if we talk about sparsity and also quantization, and the things that you say for me are DDR power, for one. DDR bandwidth is one of the more significant factors in the power of the entire inference. When you get less data stored and transferred from the DDR to the IP itself, you pay less in power and you have a lower memory footprint and also power. And of course, when you get acceleration, when you get a cycle reduction, you can benefit from the dynamic power. Also allows you to go with aggressive power management techniques like race to halt, meaning you perform your inference and you shut down your IP until you need it again. Or you can go with a DVFS approach, meaning if you can go with a lower frequency, you can sometimes lower the voltage and get the benefit of the power reduction as a result of that. Moderator | Webinar Moderator: Maybe there is a continuation question here from the audience about how does dynamic grouping work? Maybe you can continue with that one. Sure. Asaf Ganor | Director of AI Architecture, SIVA: So I tried to touch a little bit about the dynamic group quantization. I can talk a little bit about the algorithm. So what you want to achieve, as I said, you don't want to use the same scale for the entire population of elements. Let's say that you have data and you multiply it by the weights. If you want to do quantization, you need to represent the entire population with a small number of bits, four bits, eight bits, and this is inaccurate. You cannot fit all of the numbers to this small number of bits. So what you do, while you run during inference, you can group up the data while you generate it. Every set number of elements, you attach a scale, a more accurate scale to them. And then while you do the computation of the next layer, you do p-quantization, And you gain back the accuracy so that you get effective data type, which is higher than what you pay for in memory bandwidth and power consumption. Of course, there is a big challenge of doing so. How do you do it without compromising your performance? And how do you do it without compromising your accuracy? And this is exactly the expertise of SIVA, and this is what we know how to do well. So maybe it's a few more seconds. You can do it separate. The data and the weights don't need to be the same group size. If you want to keep, let's say, your weights in 4 bits and the data, you can entertain 8 bits. Then you can have a bigger group of them that go back to a scale or a channel of a tensor in the activations. And you can be more aggressive, let's say, 32, a group size of 32 elements in the weights. So these are separate decisions that you're And of course, you can see what accuracy you get as a result using SIVA tools. Moderator | Webinar Moderator: Thank you, Asaf. There was a general question about whether the slides will be available or the video itself. The slides and the video from this webinar will be available on demand right after this session. So you can view it or share it with colleagues at any time. Moving on to the next question, probably for you, Roni. How does SIVAD SDK ensure extensibility without requiring hardware redesigns as new operators emerge, as was discussed here about future Wolfram? Roni Wattelmacher | Director of Product Marketing, Vision and AI Business Unit, SIVA: So the SDK is built with the operator level abstraction and support software control, nonlinear pipelines. This enables new operators and flows to be integrated directly through the software updates. And we do it with what we call our tool SIVA Invite. And so this is how we do that. Moderator | Webinar Moderator: Thanks, Roloni. Another question for you, Asaf. How does Siva help customers balance performance, power, cost when optimizing for either edge or cloud deployment? Asaf Ganor | Director of AI Architecture, SIVA: Either edge or cloud. So I take this question, tell me if this is what you meant, guys, but I take this question as a question about how to select the right topology, hardware topology, versus what I talked before, which is how to choose the right configuration in the IP that I have in my product, right? So I will answer that. So when a customer comes to decide which configurations he wants, which KPIs he wants to get in his application, Then because the architecture is super scalable and very customizable, we start by getting to know the KPIs, the main networks, the main use cases the customer wants to get. And we can choose many different nodes. We can analyze it and show trade-offs to the customer, depending on what he cares about most, about the area, about power consumption, about the performance, raw throughput. And there are a lot of different scalability options and extendibility options I discussed before regarding the MAC accounts, even the MAC topology, the memory size, bandwidth, all according to the use case. So we can fit everything according to the specific use case. I'll give you an example. If the customer has a very aggressive power KPI, right, and he doesn't care about area, just a playground example to emphasize, we can even offer a higher MAC account, which costs more KPI area, But it allows you to go lower in frequency, and then you can run at a lower voltage and frequency point. So it all can be analyzed according to the use cases, and we can offer a very specific configuration to the customer. Moderator | Webinar Moderator: Thanks, Asaf. Another question. One is probably for you. What does the support lifecycle look like for Ziva NPU customers with AI workloads, models change every few years? Roni Wattelmacher | Director of Product Marketing, Vision and AI Business Unit, SIVA: So SIVA provides multi-year support with a roadmap that tracks AI model evolution, SDK updates, introduce new operator libraries, silicon partners receive long-term architectural enhancement, everything to ensure customers can evolve their products over time. So, yes. Moderator | Webinar Moderator: Okay. Maybe another question related a little bit for you, Roni. So the customers that already use legacy SIVA AI IPs, what is the migration path to the new program discussed here? Roni Wattelmacher | Director of Product Marketing, Vision and AI Business Unit, SIVA: Yes, so C20 ProM was designed with backward compatibility and software usability in mind. The SDK allows graph recompilation with minimal changes, and existing NPUs and also DSP integrations can be upgraded through standard IP replacement flows. So it is supported. Moderator | Webinar Moderator: Thank you. I see there are a few more questions here on the window. We will not answer all of them now, but we will make sure to get back to you directly, to anyone who is typing the questions here. We'll follow up. As said, the slides can be downloaded after this session on demand. The video can be viewed again. Thank you, everyone, for joining. and see you at one of our next webinars thank you jsPDF 3.0.3 D:20260606090043-00'00'

Research summary and source transcript

readyJun 10, 2026

CEVA reported Q2 2025 revenue of $25.7 million, down 10% year-over-year, with licensing down 13% and royalty down 5% YoY but up 16% sequentially. The company highlighted progress in AI licensing, securing four strategic NPU deals and noting expansion into infrastructure and data center markets. While licensing execution remains strong with five sequential quarters of $15M+ licensing revenue, operating results remain weak with GAAP operating loss of $4.5M and non-GAAP operating margin of only 3%. The 20B device shipment milestone was emphasized as a foundational achievement, but near-term profitability remains elusive despite sequential improvements in royalties and shipments.

Management knows today that the four strategic NPU customer agreements signed this quarter—including two Newport Nano deals for audio in embedded applications, two Newport M deals for diverse use cases, an agreement with Ali Corporation for set-top boxes, and a deal with a photonic computing company for cloud AI inference acceleration—will likely translate into royalty streams beginning in 18-24 months, based on their stated typical licensing-to-royalty timeline. The market does not yet know the volume, pricing, or adoption speed of these specific NPU designs in end products, nor whether the infrastructure/data center opportunities mentioned (e.g., PoEM architecture for cloud inference workloads) will materialize beyond early evaluations. This creates a clear information gradient: management has visibility into signed deals and customer engagement timelines, while the market must wait for shipment ramps and royalty recognition to validate the AI-driven growth narrative.

Licensing revenue from IP portfolio (Connect, Sense, Infer), royalty revenue from device shipments by licensees, and expansion into high-value AI-enabled NPU licensing as a driver of future royalty economics.

  • AI licensing progress and NPU deal wins
  • Sequential growth in royalty and shipments
  • 20 billion device shipment milestone
  • Strength in consumer IoT, Wi-Fi 6, and cellular IoT
  • Automotive momentum via DSP and sensor fusion wins
  • Expectation of second-half revenue strength due to seasonality
  • Four strategic, high-impact NPU customer agreements signed this quarter
  • Entry into broad adoption phase for edge AI NPUs
  • Newport Nano and Newport M deals validating market readiness
  • POEM architecture enabling intelligence workloads and low-latency inference for infrastructure
  • Over 20 billion SIVA power devices shipped as a foundational milestone

Management presented with a mix of confidence and caution—highlighting strong execution in licensing, sequential growth in royalties and shipments, and milestone achievements like the 20B device shipment, while acknowledging year-over-year declines in revenue and royalty due to smartphone weakness. The CEO used detailed, enthusiastic language when discussing AI deals and technical applicability (e.g., PoEM, NPU scalability), suggesting genuine excitement about long-term prospects. The CFO was more measured, focusing on historical patterns (e.g., Q4 strength) to justify optimism without overpromising. There was no evident defensiveness or evasion in tone; instead, management balanced optimism about AI and IoT with transparency about near-term headwinds, resulting in a credible and direct communication style.

  • No clear dodged analyst question was detected by the local fallback; manual review should still check whether Q&A answers quantified conversion, margins, and guidance.
  • There may be a benchmark or metric-framing issue worth manual review, especially around adjusted metrics, timelines, or changed expectations.

The company appears to be maintaining or strengthening its competitive position in licensing, particularly in AI-enabled NPUs where it secured multiple strategic deals this quarter and claims market leadership. The 20B device shipment milestone and growth in Wi-Fi 6, cellular IoT, and automotive DSPs suggest foundational relevance in connectivity and edge compute. However, YoY declines in total revenue and royalty, coupled with softness in legacy segments like low-end smartphones and industrial IoT, indicate that while the company is winning in high-growth areas (AI IoT, automotive, Wi-Fi 6), it is not yet offsetting weakness elsewhere, resulting in a mixed but cautiously optimistic competitive outlook.

  • Q2 2025 revenue: $25.7 million, down 10% YoY
  • Licensing revenue: $15 million (59% of total), down 13% YoY but up 5% in H1 2025
  • Royalty revenue: $10.7 million (41% of total), down 5% YoY but up 16% sequentially
  • GAAP gross margin: 86%, non-GAAP: 87% (down from 90%/91% YoY)
  • Non-GAAP operating margin: 3% ($0.8 million income)
  • GAAP operating loss: $4.5 million (vs. $35,000 loss YoY)
  • Shipped units: 488 million, up 16% sequentially and 6% YoY
  • Cellular IoT shipments: 66 million units, up 66% YoY (record high)
  • Royalty revenue up 16% sequentially, indicating near-term shipment momentum
  • Consumer IoT shipments up 21% sequentially and 60% YoY
  • Cellular IoT shipments at all-time record high of 66 million units, up 66% YoY
  • Wi-Fi 6 shipments up 113% YoY, reaching new record high
  • Five sequential quarters of $15M+ licensing revenue demonstrating execution stability
  • Expectation of strong Q4 royalty growth based on historical seasonality and smartphone share gains
  • GAAP operating loss of $4.5 million reflects ongoing profitability challenges despite sequential revenue improvements
  • Royalty revenue down 5% YoY due to weak low-end smartphone sales, indicating exposure to cyclical markets
  • Industrial IoT shipments down 16% YoY (24M vs. 28M), signaling softness in certain end markets
  • Bluetooth shipments down 5% YoY (254M vs. 266M), though management attributes to quarterly mix
  • Operating expenses remain elevated, with GAAP OPEX at $26.6M above guidance due to employee-related provisions
  • Reliance on seasonal H2 strength for annual goals introduces execution risk if smartphone or IoT demand falters
  • AI licensing progress may not translate to meaningful royalty contribution within expected 18–24 month window
  • No specific financial guidance provided for FY 2026, limiting long-term visibility

Management discussed direct opportunities to expand the NPU business into infrastructure and data center markets, citing a deal with a photonic computing company developing a cloud AI inference acceleration platform where CEVA’s PoEM architecture paired with its AI software stack was selected for multi-core performance under tight silicon and power constraints. They noted that as AI workloads grow more complex, traditional infrastructure faces pressure to improve performance and efficiency, and that their NPUs can complement GPUs by supporting high-bandwidth 3D DDR and running additional arithmetic for inference. This indicates a nascent but intentional strategy to address edge-of-cloud and enterprise AI inference workloads, though no current revenue, customer volume, or timeline was provided for this opportunity, making it speculative at this stage.

  • What is the expected timeline and ramp rate for royalty recognition from the four strategic NPU deals signed this quarter?
  • Can management provide any early data on shipment volumes or customer production timelines for the Ali Corporation set-top box or photonic computing company infrastructure deals?
  • What specific design wins or production ramps in automotive (beyond the Qualcomm/Autotox and 4D radar deals) are expected to contribute to royalty growth in H2 2025 or 2026?
  • How sustainable is the 16% sequential royalty growth, and what portion is attributable to holiday seasonality versus underlying market share gains?
  • What are the gross and operating margin implications as AI licensing (currently contributing meaningfully to licensing revenue) scales, given its higher value but potentially longer royalty tail?
  • Beyond the PoEM architecture discussion, what is the current stage of engagement with data center or infrastructure customers, and what revenue contribution is plausible by 2026?
  • Given the decline in industrial IoT and Bluetooth shipments YoY, what specific end-market or customer shifts are driving this, and is there evidence of stabilization or recovery?
  • How does the company plan to achieve double-digit non-GAAP net income growth YoY when Q2 non-GAAP net income was down significantly from $4.2M to $1.8M?

FY2025 Q2 earnings call transcript

32,799 chars
NASDAQ:CEVA Q2 2025 Earnings Call Transcript Generated on 6/6/2026 Rocco | Conference Operator: Good morning and welcome to the SEVA Inc. Second Quarter 2025 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 if you withdraw your question, please press star then two. Please note today's event is being recorded. Now I'd like to turn the conference over to Richard Kirschman, Vice President, Market Intelligence, Investor and Public Relations. Please go ahead. Richard Kirschman | Vice President, Market Intelligence, Investor and Public Relations: Thank you, Rocco. Good morning, everyone, and welcome to SEVA's second quarter 2025 earnings conference call. Joining me today on the call are Amir Panoush, Chief Executive Officer, and Yaniv Ariely, Chief Financial Officer of SEVA. Before handing 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 the results of SEVA to differ materially from those expressed or implied by such forward-looking statements and assumptions. Forward-looking statements include statements regarding our strategy and growth opportunities, including with respect to expanding our NPU business into infrastructure and data center markets, market positioning, trends and dynamics, including with respect to increasing importance of AI and integration of our AI into consumers' products, into our customers' products, expectations regarding demand for and benefits of our technologies and revenues, and our financial goals and guidance regarding future performance. SEVA assumes no obligation to update any forward-looking statements or information which speak as of their respective dates. We will also be discussing certain non-GAAP financial measures, which we believe provide a more meaningful analysis of our core operating results and comparison of quarterly results. A reconciliation of non-GAAP financial measures is included in the earnings release we issued this morning and in the SEC filing section of our Investors Relations website. With that said, I'd like to turn the call over to Amir, who will review our business performance for the quarter and provide some insights into our ongoing business. Amir? Amir Panoush | Chief Executive Officer: Thank you, Richard, and good morning, everyone. This quarter was marked by strong licensing execution across all our core offering pillars, Connect, Sense, and Infer. We secured 13 license agreements, including five first-time customers and four OEM customers. highlighting the breadth and strength of our IP portfolio. We saw a healthy sequential rebound in our royalty business, driven by increased shipments from our consumers and smartphones customers. In licensing, this quarter marked by a pivotal moment for our AI business as we entered the broad adoption phase for our edge AI NPUs. Following extensive evaluations with leading customers, we secured four strategic, high-impact NPU customer agreements, validating the market's readiness in our innovative, market-leading NPU portfolio. This includes two Newport Nano deals related to audio in embedded applications and two Newport M deals targeting two diverse use cases. AI is increasingly central to the next-generation audio experiences. In earbuds and hearing aids, it enables adaptive noise cancellation and personalized sound profiles. In smart speakers, it powers far-field voice recognition and context-aware processing. And in smartwatches, it expands voice commands and health diagnosis capabilities. These are just a few of the powerful capabilities that on-device AI can enable in the smallest, most power-constrained devices, which is why a broad base of our customers are integrating AI into their products. One of the Newport Nano agreements was signed with an existing high-volume connectivity customer expanding into AI-powered audio, reflecting the growing trends of customers integrating multiple SIVA IPs into a single chip. This approach boosts product capabilities, enhances deal economics, and increases royalty per device. It also marks the second major connectivity customer to adopt our Agile AI NPUs in recent quarters, reinforcing our strategy of deepening relationships through multiple IP agreements. We also signed a new NPU agreement with Ali Corporation, a leader in set-to-boxes chipsets to integrate Nupo Nano and Nupo M into their next-generation video platforms. As AI becomes essential in set-to-boxes and smart displays, our NPUs offer scalable, energy-efficient performance for advanced edge workloads. Another key deal was with a photonic computing company developing a next-generation communication acceleration platform for cloud AI inference. Their high throughput, low latency systems require scalable NPUs, and our new PoEM paired with our AI software stack was selected for its ability to deliver multi-core performance within tight silicon and power constraints. As AI workloads grow more complex, traditional infrastructures faces pressure to improve performance and efficiency. Our new POEM architecture is designed to address these challenges, enabling intelligence workloads, orchestration, adaptive data routing, and low latency inference. We see significant opportunities to expand our NPU business into infrastructure and data centers markets. In automotive, we secured two strategic agreements this quarter. One was a licensing deal with Qualcomm following the requisition of Autotox. a long-time SIVA customer. Our DSPs are integral to Autotox V2X solutions, now part of Qualcomm's Snapdragon digital chassis, supporting global V2X rollouts. With Autotox already in volume production, this collaboration is poised to accelerate global V2X rollouts while reinforcing SIVA leadership in next-generation automotive connectivity. The second deal involves our sensor fusion DSP for a US customer developing next generation 4D radar platform, which is gaining traction in ADAS and autonomous vehicles. Our automotive momentum continues to build. In Q2, a leading semiconductor began production of level 2, 3 SOCs using our vision DSPs and AI accelerators. And another top tier customer, is set to begin production on a SIVA-powered platform. These wins, along with the new ProAM design win at Nextchip and several others, position us for meaningful long-term royalty streams in automotive. Now turning to royalties, we saw good sequential growth across most of our markets, with royalties up 16% sequentially. On a year-over-year basis, royalty declined by 5%, mainly attribute to the lackluster smartphone sales at the lower end of the market, where widespread softness has been reported by our peers and which we also experience. With regards to the higher end of the smartphone market, our share is expected to grow at a leading US OEM using our technology in their in-house 5G modem. Outside of mobile, our consumer IoT customers showed strong sequential and year-over-year growth in shipments, driven by record high cellular IoT and Wi-Fi 6 shipments. Overall, consumer IoT shipments were up 21% sequentially and 60% year-over-year. We expect that the sequential growth in royalty will continue throughout the rest of the year as our customers build towards the holiday season and our share grows at our US OEM smartphone customer. Last week, we also announced a major milestone, over 20 billion SIVA power devices shipped. This achievement places us among a very small and select group of elite IP companies alongside the likes of Arm Holdings to reach this scale. It reflects our position as a foundational technology leader in the mobile and IoT eras and positions us strongly for the smart edge era now underway. Our board IP portfolio across Connect, Sense and Infer is increasingly sought after as reflected in both our licensing and royalty performance. With AI now contributing meaningfully to licensing revenue, we are well positioned to become the NPU IP of choice across the semiconductor industry. The trust we have built over the past two decades give us a strong platform to scale our AI business and deepen our role as a strategic partner to the world's leading chip makers. We view the 20 billion shipments milestone not as a finish line, but as a launchpad for CIVA's next chapter, becoming the trusted IP powerhouse of smart edge era and delivering long-term value for our shareholders. I will now hand the call over to Yaniv for the financials. Yaniv Ariely | Chief Financial Officer: Thank you, Amir. I'll now start reviewing the results of our operations for the second quarter of 2025. Revenue for the second quarter was $25.7 million, down 10% compared to $28.4 million for the same quarter last year. The revenue breakdown is as follows. Licensing and related revenue totaled $15 million, representing 59 percent of our total revenue for the quarter. This reflects a 13 percent year-over-year decline primarily due to the catch-up in licensing revenue recognized in the second quarter of 24 following a slip in the first quarter of last year. Licensing revenue for the first half of 2025 reached $30.1 million, a 5 percent increase compared to $28.7 million for the same period in 2024. This growth reflects the strength and stability of our expanded IP portfolio, the growth opportunity in AI licensing, and the solid execution of our global sales organization. Royalty revenue for the quarter was $10.7 million, reflecting 41% of total revenue. 16% sequential increase, but a 5% decrease year-over-year. The first half of 2025 royalty revenue totaled $19.9 million compared to $21.8 million in 2024. The year-over-year decrease reflects a slower start in the handset market during the first half of 2025. However, we anticipate sequential growth in the second half of the year with particularly strong momentum in the fourth quarter. Gross margins came in in line with guidance, 86% on GAAP and 87% on non-GAAP basis compared to 90% and 91% on GAAP and non-GAAP respectively a year ago. Total GAAP operating expenses for the second quarter or $26.6 million, above the high end of our guidance, due mainly to higher employee-related benefit provision after a slower first quarter result and associated adjustments. We're also continuing to build on our strategic investments in AI, strengthening our leadership position and fueling future growth. Total non-GAAP operating expenses for the second quarter, excluding equity-based compensation expenses, amortization of intangibles and related acquisition costs, were $21.6 million, also just above the high end of our guidance for the same reasons I just mentioned. Non-GAAP operating margins and income were 3% of revenue and $0.8 million. Operating margins of 15% and operating income of $4.4 million were recorded in the second quarter of last year, respectively. Gap operating loss for the second quarter was $4.5 million as compared to gap operating loss of $35,000 for the same period last year. Gap and non-gap taxes were $1.1 million just below our guidance and affected by the geographies of deals signed. Gap that lost for the second quarter was $3.7 million and diluted loss per share was 15 cents as compared to a net loss of $0.3 million and diluted loss per share of 1 cent for the same period last year. 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. In the same period last year, net income was $4.2 million and diluted net income per share was $0.17. With respect to other related data, shipped units by SEVA's licensees during the second quarter of 25 were 488 million units, up 16% sequentially and up 6% from the second quarter 2024 reported shipments. Of the 488 million units reported, 55 million units or 11% were for mobile handset modems. 409 million units were for consumer IoT markets up 16% from 353 million units in the second quarter of 24. 24 million units were for industrial IoT markets, down 16% from 28 million units in the second quarter of last year. Bluetooth shipments were 254 million units in the quarter, down 5% from 266 million in the second quarter of 24. Cellular IoT shipments were all-time record high at 66 million units, up 66% year over year. Wi-Fi shipments were 62 million units, up 80% from 35 million units a year ago. Wi-Fi 6 shipments reached a new record high and were up 113% year over year, as we continue our Wi-Fi 6 customer ramp-up in the consumer and industrial markets. Overall, good sequential growth in royalties and shipments in many of our customer and markets, while softness was evident in the smartphone and some areas of industrial. As for the balance sheet items, As of the end of June this year, SEVA's cash, cash equivalent balances, marketable securities, and bank deposits were approximately $157 million. In the second quarter this year, we were more active on our buyback program and repurchased 300,000 shares for approximately $6.2 million. As of today, around 725,000 shares are still available for repurchase under the repurchase program as expanded in November of last year. Our DSOs for the second quarter were 42 days, lower than the norm and lower than prior quarters. During the second quarter, we generated $1.2 million of cash from operation activities Ongoing depreciation and amortization was $1.1 million, and the purchase of fixed assets were $0.7 million. At the end of the second quarter, our headcount was 435 people, of whom 354 are engineers. Now for the guidance. Our licensing pipeline and potential deal flow especially around edge AI prospects, look healthy entering into the third quarter and second half of the year. We have demonstrated strong licensing execution in 2025, notably achieving five sequential quarters with about $15 million or above in licensing revenue. Roti revenue historically are stronger in the second half of the year due to seasonality, a new product deployment, a shipment ramps ahead of the holiday season. We're encouraged by the strength of many of our customers and in market demand, particularly around our seller IoT and Wi-Fi 6 product lines. We also anticipate growth in smartphone royalties in the second half of the year, driven by share gains at a US OEM smartphone customer using our technology for their in-house 5G motor. Such we're maintaining our overall revenue guidance growth as discussed in the prior earnings call. We continue with our long-term investment in AI and other new technologies to enrich our IP portfolio along with continued focus on expenses. We reiterate our belief that we will reach a double-digit percentage increase of non-GAAP net income and fully diluted non-GAAP EPS relatively to 2024. As for the third quarter, Total revenue is expected to be between $26 to $30 million. Gross margin is expected to be 1% higher than the second quarter, approximately 87% on gap basis and 88% on non-gap basis, excluding an aggregate of $0.2 million of equity-based compensation expenses and $0.1 million amortization of acquired intangibles. Gap OPEX is expected to be at a similar level to the second quarter in a range of $26 to $27 million. Of the anticipated total OPEX for the third quarter, $4.7 million is expected to be attributed to equity-based compensation expense, $0.2 million for amortization of acquired intangibles, and $0.1 for expenses related to business acquisitions. Non-GAAP OPEX is also expected to be quite similar to the second quarter in the range of $21 to $22 million. Net interest income is expected to be approximately $1.3 million. Taxes for the third quarter is expected to be approximately $1.8 million. And the share count for the third quarter is expected to be 25.8 million shares. Rocco, you could now open our Q&A session, please. Rocco | Conference Operator: Yes, sir. If you'd like to ask a question, please press star and 1. If your question has already been addressed and you'd like to remove yourself from queue, please press star and 2. Today's first question comes from Kevin Cassidy, Rosenblatt Securities. Please go ahead. Kevin Cassidy | Analyst, Rosenblatt Securities: Yes, thanks for taking my question and congratulations on the great results. You know, as you get an increasing of your licensing in NPUs, would this adjust, you know, it's a higher valued IC, so would we expect in the future royalty revenues would have higher leverage or, you know, see an acceleration? Amir Panoush | Chief Executive Officer: Hi, good morning, Kevin. Thanks for the question. Yeah, definitely, as we discussed also previously, right now we see Great momentum with overall licensing, our NPUs. I personally, with the management team, are very encouraged about how we see the prospect of winning those deals. But more looking into the long term, as you pointed out on the royalty side, those deals definitely have better economics. The complexity of the technology and the needs of the technology is higher than our so-called average typical royalty that we have today. And with that, we're expecting the royalty to have a meaningful increase, definitely per unit, as those devices will deploy in the marketplace. Kevin Cassidy | Analyst, Rosenblatt Securities: Okay, great. And maybe just as the timing for that is, I guess, is the time from licensing to royalty longer with this more complex design? And also, because the AI market is moving so fast, would we expect that the tail for the royalty, meaning the product life cycle, is that going to shorten compared to your past, especially wireless customers? Amir Panoush | Chief Executive Officer: Yes. So typically, Kevin, what we see is that the time between so-called when we license the technology until we start getting royalty reports or the royalty is between 18 and 24 months. And in this case, I would say the valuation sometimes takes longer, but actually the time for our customer to take it into production is similar. In consumer, it can be even shorter than 24 months. It can be 18 months. In a little bit more complex systems, in infrastructure and so on, it can be 24 months or so. Overall, I would say our customers, as you pointed out, AI is moving quickly, and customers really need to deploy it as soon as possible from their perspective. So we do expect the royalty to take the same as typically in consumers to some degree, maybe even slightly faster. In terms of the tail of that, again, depends to what system it goes to. If it goes to the typical consumer devices, so we should expect the same cycle. If it goes to more the infrastructure side, then the tail is much longer, whether it's automotive, whether it's wireless infrastructure, whether it's more on cloud enterprise support and so on. Kevin Cassidy | Analyst, Rosenblatt Securities: I see. So end market is more important. Great. Okay. Thank you. Rocco | Conference Operator: Thank you, Kevin. Kevin Cassidy | Analyst, Rosenblatt Securities: Yeah, thanks a lot, Kevin. Rocco | Conference Operator: Thank you. And our next question today comes from Suzy DeSilvo at Roth Capital. Please go ahead. Suzy DeSilvo | Analyst, Roth Capital: Hi, I'm Mary. Hi, Yaniv. Congratulations on reaching 20 billion units. I'd be curious what the number was when Yaniv joined the company. The royalty stream being stronger in the second half, fourth quarter, I presume flagship smartphone customers are a key part of that. I'm wondering if that number or that contribution matters. would be going up in 26 as the flagship customer continues to mix in its in-house modem, or do you have any visibility there? Amir Panoush | Chief Executive Officer: Yeah, I would say first, Suji, we haven't guided anything for 26 yet, so it's not that we are going to comment on the specifics on that. But generally speaking, our expectation that we see the success of our technology penetrating as the customer technology keeps wrapping up, Definitely, we're assuming that for the second half of this year. For 26 and beyond, of course, as we get closer, we will be able to share more. Yaniv Ariely | Chief Financial Officer: But, Suzy, I would add, even if you exclude that U.S. customer, historically, Q4 has been the strongest on the high-volume, low-cost smartphones for many, many years. It's not something new. Q1 is the slowest quarter and the slow start for the year, and then it ramps up. with Q4 always being the strongest. So even historically, we have pretty good data to back that up unless something that we don't anticipate will happen for this year. That's where the confidence is coming from on the rest of it. Suzy DeSilvo | Analyst, Roth Capital: That sounds like a good tailwind there. And then on AI, congrats on the progress there. You talk about sort of scaling AI. I mean, you're hitting kind of a scale point. Is that software tools, ecosystem? Can you give us the elements of what gives you sort of a scaled opportunity in the Edge AI ecosystem? And, you know, maybe you can talk specifically about what data center might be for you guys. It's an interesting avenue that you guys might be charting out. Thanks. Amir Panoush | Chief Executive Officer: Yeah, definitely. Thanks. So first from a, I would say the applicability of our NPOs and why we are very encouraged with the interest and the momentum that we see in the market from a scalability point of view that you asked. There are two aspects to that. One from the hardware IP, the Silicon IP capabilities. We are really providing a very scalable platforms going from hundreds of jobs all the way to hundreds of tops and type of product line. And actually our customers can tune and fine tune that to their own requirements. And we have with that a very good fit to what they need. On top of that, from a software perspective, we really give them the complete SDK or software stack to support it. and one that is quite easy to integrate into their own system and to support their own customers. So on both forms, the Silicon IP and the software IP that comes on top of that, the scalability that we provide resonates extremely well with our customer base. And that's why this quarter, we really got four deals, two more, I would say, on the lower end of the spectrum, so called Nupo Nano, and two on the higher end of the spectrum, the Nupo M. Within that, actually, as we pointed out this quarter, We are starting to see an interesting fit of our so-called Edge NPU solution, where everything is about low power, very efficient utilization, high performance, and also small size, becoming applicable also for inference on the cloud. This is not really to replace every socket like I hand GPUs, but it's where you need to complement with NPUs to run the additional arithmetics and to support the very high bandwidth 3D DDR that typically is used in those systems. So it's great to see the applicability happens there that adds for us additional markets to support and to go after. But of course, the core of our solution is how to make it very optimized for edge, which become also applicable in some cases for cloud inference AI as well. Okay, appreciate the clarification. Amir, thanks. Rocco | Conference Operator: Thank you, Suzy. And our next question today comes from Chris Reamer of Barclays. Chris Reamer | Analyst, Barclays: Please go ahead. Yeah, thanks for taking my questions and congratulations on strong results. I was wondering if you could talk a bit about the pipeline. You mentioned last quarter that you had several new products coming to the market. Are they already working and do you have anything else coming new looking toward the end of the year? Yaniv Ariely | Chief Financial Officer: Sure. So on AI, I think Amir highlighted that this is a pivotal point in our business. For a long time, we've been talking about AI. For the last year, we came out with new products, mid last year, the end of last year, with new products around AI for the higher end and lower end markets and use cases. So that's something that the traction It was record high for us in licensing in the second quarter, but we have seen a deal per quarter over the last three quarters before that. So obviously, we are trying and will try to get that in the market and in different customers. And we have multiple evaluations on these technologies as we speak. So that is a very strong add-on. to our portfolio on top of the typical connectivity, different Wi-Fi, Bluetooth, audio solution and the rest of the portfolio. I think we're continuing to invest and come up with new features, new technologies all the time. The AI is an interesting add-on and as we've talked about in the past, different cellular IoT, segment of the market. Finally, we are seeing the benefits in royalties with a record high in volume shipments or Wi-Fi 6, which is not necessarily new in licensing. We're seeing its great results with a record high also in the second quarter in volume. So it's the same mix of enhancing our portfolio of licensable technologies on one hand, and over time, the royalties kick in and help us from different markets and new customers that are starting to ramp Chris Reamer | Analyst, Barclays: Got it. Yeah. And just looking at shipments, nice uptick this quarter. And you did mention the addition of the customer in the US, the smartphone expected shipments to go up. But how should we be looking at some of the other segments? Do you have any color on other segments of the market that may be shipping higher? Amir Panoush | Chief Executive Officer: Yeah, maybe I'll give a good color about each of the end market that we are looking at. So I'll start actually with the smartphone industrial that have been slower than what we expected in Q1, Q2, so-called the first half. These two we expect to have a very strong, so-called sequential growth in the second half of the year. But overall, smartphone, generally speaking, I would say it's flattish to a little bit soft overall. But we are gaining market share and with that we're expecting a very strong sequential growth in the second half of the year. The rest of the market, the consumer IoT and the infrastructure and all the other markets that we are supporting actually has been doing very well and we expect additional sequential growth in the second half. That's all also what we see is our Wi-Fi 6 is keep ramping very, very strongly. So that's a strong tell for us in the second half, as well as the solo IoT that we pointed out in these earnings as well. Chris Reamer | Analyst, Barclays: Okay. Yeah, thanks. That's good, Kelly. That's it for me. Yaniv Ariely | Chief Financial Officer: Thank you, Chris. Rocco | Conference Operator: And as a reminder, if you'd like to ask a question, please press star then 1. Our next question comes from Martin Yang at Oppenheimer. Please go ahead. Martin Yang | Analyst, Oppenheimer: Thank you for taking my question. I want to ask about Bluetooth. That product has been growing pretty consistently year-over-year in the past few quarters. What contributed to this quarter's decline on a year-over-year basis? Anything on customer or market dynamics worth pointing out? Amir Panoush | Chief Executive Officer: There wasn't something specific this quarter that I will point to on the Bluetooth 4.0. I would say, generally speaking, we expect the second half good sequential growth for our Bluetooth technology as well. There is the shift that is coming right now to adopt more the Bluetooth 6.0 in production. Of course, we are already Bluetooth 7.0, which will drive significant growth for us in the 26, 27. Overall, it's very healthy for us. It can be a little bit the mix of our customers for this quarter, but nothing more than that. Yaniv Ariely | Chief Financial Officer: Yeah, in general, Martin, it's about a quarter of a million devices, which is not that bad because annually, last year, we powered 1.1 billion devices. And again, Q1 and Q2 tend to be slower than the second half of the year, so I think that number will pick up in the next two quarters. Amir Panoush | Chief Executive Officer: And on that one, also generally speaking for our top customers, we see they've been doing well this quarter and we expect good sequential growth in the second half as well. Thank you. Martin Yang | Analyst, Oppenheimer: And in the context of overall annual guidance, do you think overall Bluetooth will give you year-over-year growth for the year on units? Yaniv Ariely | Chief Financial Officer: Well, it's hard to guess. This is why we don't give annual guidance on specifically licensing on royalties. There's so many moving parts. There's so many different markets. There's so many different use cases from hearing aids to watches and to a lot of IoT devices. Very difficult to know all that. In general, if you look at the last couple of years, the answer is yes. We've grown Bluetooth year over year. We've grown Wi-Fi significantly year over year. Cellular IoT took a long time, many, many years to pick up. But in the last two years, we're seeing tremendous growth in that market. So I think the answer should be yes without having a bottom-up type of analysis. But from a top-down and the customer use case and the customers that come back, and license newer generation of each of these technologies, we are seeing a lot of good momentum there. Amir Panoush | Chief Executive Officer: Yeah, but just to add on that, if you look at actually at the first half of the year for 24, that was, for 25, sorry, that has been a growth over 24. So first half, the Bluetooth volume growth is already there and we expect for the full year to keep seeing an increase year over year for our Bluetooth shipment. Yaniv Ariely | Chief Financial Officer: which is also true for Wi-Fi and cellular IoT. First half of this year was higher than the first half of last year for all of these three different markets and technologies. So good sign from the company. Martin Yang | Analyst, Oppenheimer: Thank you, Adib. Yaniv Ariely | Chief Financial Officer: Yeah, sure. Martin Yang | Analyst, Oppenheimer: Thank you, Amir. Amir Panoush | Chief Executive Officer: Thank you. Rocco | Conference Operator: Thank you. This concludes today's question and answer session. I'd like to turn the conference back over to Amir Paloush for any closing remarks. Amir Panoush | Chief Executive Officer: Yeah, thank you. On behalf of the SIVA team, thank you for joining us today. We continue to execute on our strategy to democratize Edge AI to our portfolio of technologies that enable connectivity, sensing, and inference. Our strong licensing performance, expanding royalty base, and milestones of over 20 billion devices shipped underscore the trust our customers place in us as a foundational technology provider. With AI adoption accelerating across consumer, industrials, and automotive markets, and our IP portfolio more relevant than ever, we are well positioned to drive long-term growth and shareholder value. We look forward to meeting many of you during the third quarter at investor conferences. Richard, I will hand over to you to wrap it up. Richard Kirschman | Vice President, Market Intelligence, Investor and Public Relations: Thank you, Amir. As a reminder, the prepared remarks for this conference call are filed as an exhibit to the current report on Form 8K and accessible through the investor section of our website. With regards to upcoming events, we will be participating in the following conferences. The Oppenheimer 28th Annual Technology Internet of Communications Conference, August 13th, being held virtually. The Rosenblatt Virtual Tech Summit, August 19th, being held virtually. Sixth Annual Needham Virtual Semiconductor and Semicap One-on-One Conference, August 20th. Jefferies Semiconductor IT Hardware and Communications Technology Conference, August 26th in Chicago. The Evercore Semiconductor IT Hardware and Networking Conference, August 27th in Chicago. TD Securities Technology Growth Cap Summit, September 4th in New York. and Jefferies Tech Trek 2025, September 11th in Tel Aviv, Israel. For 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. Rocco | Conference Operator: Thank you. This 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. jsPDF 3.0.3 D:20260606090044-00'00'