Rambus (RMBS) 2026财年第三季度业绩电话会议：营收创历史新高，第四季度产品展望强劲

核心要点

• Rambus公布2026财年第三季度营收达到创纪录的2.18亿美元，同比增长22%，环比增长5%，超出了公司指引的上限。
• 产品营收达到创纪录的1.14亿美元，同比增长22%，环比增长15%，主要得益于DDR5 RCD需求以及新一代产品的采用。
• Non-GAAP净利润为9170万美元，Non-GAAP稀释后每股收益同比增长35%，环比增长10%至0.84美元。
• 创纪录的经营活动现金流1.02亿美元为库存投资和股东回报提供了资金支持。Rambus回购了1.35亿美元的股票，其中包括1亿美元的加速股票回购。
• 管理层预计第四季度营收为2.27亿至2.33亿美元，产品营收为1.27亿至1.33亿美元。按中点计算，产品营收将同比增长34%，环比增长14%。
• 管理层表示，迈向2027年需求正在增强，尽管供应商交货周期正在延长，且预计供应将保持紧张。

核心财务业绩

业务与运营表现

Rambus将创纪录的产品营收主要归因于其DDR5寄存时钟驱动芯片（RCD）产品组合。其支持DDR5-9600的第六代RCD目前已向客户广泛送样。配套芯片继续保持环比增长，而最初与MRDM相关的出货量仍较少，主要集中在认证和预量产阶段。

管理层表示，内存速度的提升和更具异构性的服务器架构正在创造对额外时序、电源和安全管理功能的需求。公司预计，随着时间的推移，这将提升单内存模块的芯片价值量。基于DDR的服务器从8内存通道扩展至12或16通道，也可能支撑更高的模块需求。

Rambus拓展了其电源管理路线图，推出了用于LPDDR6服务器内存的PMIC。管理层预计，由于电源管理复杂性更高，新一代产品的单机价值量将高于先前的LPDDR方案。预计最初的市场部署将在2028年左右，不过管理层将其描述为AI基础设施中的细分领域。

硅IP业务在下一代HBM、PCIe控制器与交换机IP以及硬件安全领域斩获设计胜出。管理层表示，随着客户寻求先进内存接口、高速互连和基于硬件的安全方案，公司与定制芯片开发商以及超大规模客户的合作日益紧密。

公司将硅IP业务描述为年规模约1.3亿美元、增长率在10%-15%区间的业务。管理层表示，2026年的业绩正趋向该区间的上限。

Rambus在第三季度还增加了1200万美元的库存，以支持未来产品的产能爬坡并为客户提供更高的供应保障。管理层预计第四季度库存增幅类似。

管理层业绩指引

按中点计算，第四季度产品营收预计同比增长34%，环比增长14%。Non-GAAP每股收益中点为0.88美元，意味着同比增长30%，环比增长5%。

管理层表示，第四季度前景指引的中点意味着2026财年相比2025财年，实现约18%的营收增长、24%的产品营收增长以及24%的Non-GAAP每股收益增长。

Rambus预计第四季度产品毛利率将保持在其近期60%-63%的区间内，且可能出现微幅环比改善。管理层预计2026财年全年的产品毛利率将收于该区间中点附近。

公司计划随着时间的推移继续将40%-50%的自由现金流回馈给股东，同时保持对其产品路线图和增长机会的投资。

风险与关注点

• 随着半导体供应链趋紧，供应商交货周期正在延长。管理层表示，这并未影响第三季度业绩，预计也不会影响第四季度指引，但2027年供应可能仍将保持紧张。
• 供应链投入成本上升正使产品毛利率承压，逼近近期60%-63%区间的下限。Rambus正通过内部降本抵消部分影响。
• 管理层预计，由于平台采用错峰且该技术仍处于采用周期的早期阶段，MRDM带来的市场机遇增长将比标准RCD代际更渐进。
• 客户正更早地提交储备订单，提高了Rambus的可见度。然而，管理层表示很难确定下游客户是否在囤积库存。
• 基于LPDDR的服务器内存预计将补充而非替代传统的DDR配置，但管理层预计其在整体市场中的占比仍将较小。

分析师问答亮点

产品营收加速增长：管理层表示，第四季度的增长前景主要由新一代RCD和配套芯片的产能爬坡推动。与MRDM相关的营收仍较为有限。

2027年需求：进入2027年，Rambus看到需求增强且储备订单可见度提高。管理层仍预计上下半年会存在一定的季节性差异，且未提供2027年的定量指引。

硅IP扩展：公司正在更早地参与客户的架构规划，尤其是与超大规模客户和定制芯片开发商。管理层表示，这提供了有价值的洞察，可以为硅IP组合和未来芯片产品提供指导。

供应规划：Rambus正利用其资产负债表建立库存，并与供应商密切合作以支撑预期需求。尽管上游供应商交货周期有所延长，但客户交货周期并未增加。

内存架构组合：管理层预计，使用传统DDR内存的标准CPU服务器将占据绝大部分市场增长，而基于LPDDR的解决方案则应对能效尤为重要的补充使用场景。

业绩电话会议完整文字记录

---

完整财报电话会议逐字稿

管理层陈述

Operator

Welcome to the Rambus Third Quarter Fiscal Year 2026 Earnings Conference Call. [Operator Instructions] As a reminder, this conference call is being recorded.

I would now like to turn the conference over to Sumeet Gagneja, Chief Financial Officer. You may begin your conference.

Sumeet Gagneja

Thank you, operator, and welcome to Rambus Third Quarter 2026 Results Conference Call. I'm Sumeet Gagneja, Chief Financial Officer at Rambus. On the call with me today is Luc Seraphin, our CEO. The press release for the results that we will be discussing today has been filed with the SEC on Form 8-K. We are webcasting this call along with the slides that we will reference during portions of today's call. A replay of this call will be available on our website beginning today at 5:00 p.m. Pacific Time.

Our discussion today will contain forward-looking statements including our expectations for projected financial results and prospects, market growth, demand for our solutions and the potential effects of geopolitical and macroeconomic conditions. These statements are subject to risks and uncertainties that may be discussed during this call and are more fully described in our documents we filed with the SEC, including our 8-Ks, 10-Qs and 10-Ks.

These forward-looking statements may differ materially from our actual results, and we are under no obligation to update these statements. In our press release and on the call today, we present both GAAP and non-GAAP financial measures to provide greater clarity regarding our financial performance. A reconciliation of these non-GAAP financials to the most directly comparable GAAP measures has been included in our press release, in our slide presentation and on our website at rambus.com on the Investor Relations page under Financial Releases.

The order of the confident will be as follows: Lot will begin with an overview of the business, followed by my review of the financial results, and then we'll open the call to Q&A.

I will now turn the call over to Luc to provide an overview of the quarter. Luc?

Luc Seraphin

Thank you, Sumeet. Good afternoon, everyone, and thank you for joining us. In Q3, we delivered a truly exceptional quarter. Revenue and non-GAAP earnings both reached new record highs and exceeded the top end of our guidance ranges. Strong customer demand, fueled record product revenue and cash from operations and we expect an even stronger product performance in Q4. We introduced new chip and silicon IP products that extend our leadership in data center and AI infrastructure.

In addition, as complexity and pace accelerate in semiconductor development for AI, we are also investing in the tools and infrastructure that will keep us on the leading edge of technology. As such, we engaged with a fellow pioneer in the ecosystem and announced an exciting new partnership that propels us into the age of AI agent-driven chip and IP design.

As we continue to invest in our future, we are also pleased to have completed a $100 million accelerated share repurchase. This demonstrates the strength of our business model and our ability to return value to our stockholders while advancing our product road map to drive long-term growth. Taken together, the results and progress we delivered this quarter reflect the strength and long-term resilience of a strategy we have pursued over many years, investing in technology leadership, anticipating the key trends transforming the data center and positioning the company to capture the opportunities those trends create.

Our strategy is producing strong results today while at the same time, laying the foundation for future growth. As we outlined over previous quarters, AI continues to create strong tailwinds for Rambus across our businesses. As Frontier AI training continues to advance, and inference scales across a growing range of applications, customers require increasingly sophisticated memory and infrastructure architectures to move, manage and secure growing volumes of data. Agentic-AI is accelerating this trend as deployments evolve towards more distributed and heterogeneous architectures that require greater coordination across systems.

At the same time, orchestration of always-on workloads puts greater demands on the CPU, lifting CPU to GPU ratios and driving requirements for more main memory bandwidth and capacity across the data center. The latest agentic implementations amplify this need even further by assigning dedicated compute and memory resources to individual users or agents. All this creates accelerating demand for Rambus products and IP throughout the memory subsystem and across the broader infrastructure stack, while we continue to invest in the fundamental R&D and product road map required to address the ever-growing demand of memory management efficiency in the long run.

Now let me turn to the quarterly results, starting with products. We delivered a new all-time high in quarterly product revenue, reaching approximately $114 million, up 15% sequentially and 22% year-over-year. Our leadership in DDR5 RCDs and growing adoption of new generation products fueled our record results. We are seeing strong demand across multiple generations of our RCDs supporting our outlook for another record quarter of product revenue in Q4, which we expect to grow more than 30% year-over-year.

Looking ahead, we have a strong pipeline of products aligned with upcoming platform ramps positioning us well for next generation of server platforms. As part of that pipeline, we are pleased to share that our sixth generation RCD announced early in Q3 which supports DDR5 9600 is now sampling broadly with customers and is being very well received. We continue to execute successfully on our road map, recently expanding our product portfolio with an industry-leading PMIC for the upcoming LPDDR 6 so CAMI.

Earlier this year, we introduced our first SoCaM2-chipset that supports LPDDR5. This new PMIC solution extends our server memory product road map to LPDDR6 and reinforces our leadership in advanced server memory power management. It was developed in close collaboration with customers and ecosystem partners to enhance memory subsystem power efficiency and help scale tokens per watt for AI workloads.

We identified early on that power management would be critical to the success of AI deployments and have been investing in this area for many years. We are proud of our progress and continued leadership in this space. We have consistently said that as memory speeds increase and server architectures become more sophisticated and heterogeneous, the memory subsystem will require additional intelligence, power management and control functionality.

Our road map is designed to address this evolution across both DDR and LPDDR based memory with timing management, power management and security management creating opportunities for Rambus to participate in increasing module chip content over time.

In addition, DDR based server platforms are expanding from 8 memory channels to 12 to 16 channels, increasing the maximum available memory bandwidth and capacity. This raises the number of modules supported in the server and amplifies demand as platforms progress. Today, Rambus offers complete chipsets across all major industry standard DDR5 and LPDDR server memory module architectures. This broad portfolio positions us well for future server platforms and allows customers to leverage Rambus solutions across a growing range of memory architectures and deployment models.

Looking ahead, we believe we are exceptionally well positioned for future server generations and for the continued expansion of memory-intensive AI infrastructure.

Turning now to silicon IP. We delivered a very strong quarter with key design wins across our Silicon IP portfolio. The strength of those wins, combined with expanding strategic engagement across the industry, including custom silicon makers and hyperscalers reinforces our confidence in the long-term growth opportunity for the Silicon IP business and for Rambus. One of the defining characteristics of today's market is the rapid growth of custom silicon. The expansion of AI is fueling a new wave of custom silicon development, hyperscalers, AI infrastructure providers and emerging semiconductor companies are designing increasingly differentiated architectures and tailored silicon optimized for their own environments.

These systems depend on advanced memory technologies, high-speed interconnects and trusted security creating significant opportunities for Rambus Silicon IP. During the quarter, we drove increasing ecosystem engagement and new Tier 1 design wins for next-generation HBM supported by our deep architectural engagements ahead of standards being finalized. We continue to see increasing interest in PCIe controller and switch IP for scale up and scale out connectivity solutions. as customers seek higher levels of performance and efficiency across rapidly expanding AI infrastructure.

The importance of bleeding-edge IP continues to increase as customers pursue differentiated performance, power efficiency and time-to-market advantages. As a result, we are engaging earlier in the architectural planning cycle with both our customers as well as our customers' customers and participating in a broader set of strategic programs.

Leading edge memory and connectivity are fundamental to performance and as AI infrastructure scales and becomes more heterogeneous, hardware-based security foundation is becoming fundamental. Agentic-AI relies on distributed computing architectures that execute workloads across multiple systems, which increases the available attack surface for adversaries, a growing concern in the current environment.

As a pioneer and longtime leader in hardware-based security, Rambus IP solutions address that risk by providing a trusted foundation for the entire system and enhancing protection for valuable data and infrastructure. Building on our industry-leading security offerings, we recently added a new route of trust solution, designed specifically for AI and data center deployments that supports the Elektra specification.

Elektra was developed by key leading technology companies to establish a common hardware security foundation for data center class SoCs and reflects the growing recognition that hardware-based security is essential for next-generation infrastructure. Looking across both chips and silicon IP. We will continue to drive solutions that address the long-term foundation memory performance, high-speed connectivity and robust security demands of AI infrastructure.

At the same time, we are harnessing AI to advance the underlying tools and development infrastructure that enable us to move faster, extend our leadership and scale the business more effectively. As I mentioned at the beginning of the call, we recently announced a strategic investment and collaboration with chip agents, a fellow industry pioneer and leader in Agentic AI design platforms.

We see great opportunities to leverage the cheap agents AI-native design flows to accelerate our road map by speeding the pace of our chip and IP development, improving design quality and enhancing and generating productivity across our organization. We believe this collaboration further strengthens our competitive position, both for products and for IP and reinforces Rambus leadership role in next-generation of AI data center infrastructure.

In summary, the company delivered an exceptional third quarter, and we expect an even stronger Q4 as we continue to capitalize on the growing demand from AI. We achieved record revenue and earnings while expanding our chip and IP offerings. And at the same time, we are investing in our development infrastructure to address the growing opportunities ahead. These results and decisions reflect the strength of our strategy and the continued alignment between Rambus technology and the most important trends shaping AI and data center infrastructure.

We are executing well and more importantly, we are continuing to invest for the future and build on a strong foundation of technology leadership and customer partnerships. We are confident in our trajectory, our road map and our ability to drive continued profitable growth in 2027 and beyond.

As always, I want to thank our customers, partners and employees for their continued trust and support. And with that, I'll turn the call over to Sumeet. Sumeet?

Sumeet Gagneja

Thank you, Luc, and good afternoon, everyone. We delivered an outstanding third quarter, achieving record revenue and earnings and record cash flow from operations. These results reflect the strength of our business model, growing demand for our products and technologies and disciplined execution across the company. Consistent with our balanced capital allocation strategy, we completed a $100 million accelerated share repurchase program during the quarter.

The transaction demonstrates our confidence in the business and our commitment to returning capital to shareholders while preserving the flexibility to invest in the growth. Looking ahead, we believe Rambus is well positioned to capitalize on the secular growth in AI infrastructure, data centers and advanced computing through the strength of our product and technology portfolio.

With that, let me review our third quarter financial results. Third quarter revenue was $218 million, driven by strong product and contract revenue. This represents growth of 22% year-over-year and 5% sequentially. Product revenue was $114 million which is up 22% year-over-year and up 15% sequentially. Royalties revenue was $77 million. Contract and other revenue was $27 million, up 35% year-over-year and 13% sequentially. Contract and other revenue consists primarily of silicon IP, as a reminder, only a portion of our silicon IP revenues reflected in contracts and other revenue and the remaining portion is reported in royalties revenue.

Total non-GAAP operating costs including cost of goods sold for the quarter were $116 million. Operating expenses were $70 million, down sequentially due to slightly lower R&D and SG&A expenses. Interest and other income for the quarter was $7 million. Using non-GAAP tax rate at 16%, non-GAAP net income for the quarter was $91.7 million, resulting in Q3 non-GAAP earnings per share of $0.84, which is up 35% year-over-year and up 10% sequentially.

Now let me turn to the balance sheet details. We ended the third quarter with cash, cash equivalents and marketable securities of $777 million, down $48 million from Q2. Record cash flow from operations of $102 million were offset by $135 million of share repurchases, including the $100 million accelerated share repurchase program. Capital allocation remains a key priority for us. We intend to continue investing in the business while returning 40% to 50% of free cash flows to shareholders over time, supported by our strong balance sheet and continued cash generation.

Our capital allocation approach reflects our focus on driving shareholder value while investing in our growth and pursue long-term opportunities ahead. We also leveraged the strength of our balance sheet to increase inventory by $12 million during the quarter, positioning us to support future product ramps and provide customers with greater supply assurance.

Now let me turn to our non-GAAP outlook for the fourth quarter. As a reminder, the forward-looking guidance reflects our best estimates at this time. and our actual results could differ materially from what I'm about to review. We expect fourth quarter revenue to be between $227 million and $233 million. Product revenue is expected to be between $127 million and $133 million. At the midpoint of the guidance, it is an increase of 34% year-over-year, an increase of 14% sequentially. Royalties revenue is expected to be between $68 million and $74 million and contract and other revenue is expected to be between $26 million and $32 million.

We expect Q4 non-GAAP total operating costs which include cost of sales to be between $124 million and $120 million. We expect Q4 capital expenditures to be approximately $8 million. Non-GAAP operating profit for the fourth quarter is expected to be between a profit of $103 million and $113 million. We expect fourth quarter interest and other income to be approximately $6.5 million. Assuming a non-GAAP tax rate of 16% and $109 million diluted shares outstanding, we expect fourth quarter non-GAAP earnings per share to be in the range from $0.84 to $0.92.

At the midpoint of the guidance, this represents growth of 30% year-over-year and 5% sequentially. Overall, we are very pleased with our third quarter performance. Our record revenue, earnings and cash flow from operations demonstrate the breadth of our product and technology portfolio, the diversity of our revenue streams and the disciplined execution of our teams. The midpoint of our fourth quarter guidance implies that 2026 revenue will grow approximately 18% over 2025, including product revenue growth of approximately 24%.

With our continued focus on driving growth and operating efficiency, we expect non-GAAP earnings per share to grow approximately 24% over 2025. We remain focused on executing our strategy and capturing the opportunities created by AI infrastructure, data center and advanced computing markets. We are committed to translating these opportunities into sustainable growth, strong cash generation and long-term shareholder value.

With that, I'll turn the call back to the operator for Q&A. Operator, could we have our first question, please.

Operator

[Operator Instructions] Your first question comes from the line of Kevin Garrigan with Jefferies.

分析师问答

Kevin Garrigan

Luc, so Q3 product revenue growth was 22% year-over-year. the Q4 midpoint implies about 34%. I know you talked about strong growth across RCD generations. But can you give us a sense of how much growth is coming from other areas like companion chip ramps or any initial MRDM shipments?

Luc Seraphin

Thanks, Kevin. Yes, we're really pleased with the results for Q3 and the growth for Q4, 34% year-over-year. We always saw that the second half of the year would be stronger than the first half of the year. I would say it's a combination of market demand going up. and a lot of our products actually ramping into the market. The new generation of our RCDs make the bulk of that. At the beginning of the year, we said that our companionship were approximately low double-digit percentage of our product revenue and that it would grow sequentially quarter-over-quarter. We see this happening.

And margin remains very modest actually. We have shipped the margins for qual and preproduction this year. The AMD platform is starting to ramp. So it's still very modest. So we're really, really pleased with this 34% year-over-year growth based on the core, I would say, LCD and companion chip.

Kevin Garrigan

Okay. Great. And then as a follow-up, just on the MRDM opportunity. So the RCD playbook was always about being first to market and qualifying early. One of your competitors was designed for the first-generation MRDM platform. So what's different about this next generation that gives you confidence that you can take a leadership position?

Luc Seraphin

The difference with MRD is that is associated with the launch of the platform that is very different than what we have all seen for the past few years. There's only 1 platform that is supporting MRD now going into the market. We don't know what the -- what the portion of the margin is going to be compared to the standard be in that area. So it's going to be a ramp, and we're confident this ramp will happen. We still believe this is a great opportunity for us that we still size at $600 million in total.

But the rent will be shallower in our opinion, than the other generations of products, simply because all of the platforms are not ramping at the same time. This is a new technology, and our customers have to adopt this technology and go through the initial standards. There's also this thing that we see in the market of memory despecking. It's not that people decrease the number of modules, but they tend to decrease the capacity of each 1 of these modules. So we do see a growth of the number of modules in the market that explains our growth in Q4.

But we do see at the same time that each 1 of these modules tend to have lower capacity and it's due to the economics of the memory at this point in time. So between the memory economics, the stage ramp of the platforms and the introduction of new technology and supply associated with that, we see this as a great opportunity, but we see the ramp to be shallower than a standard ramp of an LCD product.

Operator

Your next question comes from the line of Gary Mobley with StoneX.

Gary Mobley

Congrats on strong execution as we navigate through the second half of the year. Clearly, the second half of the year is playing out with seasonal strength on the product side as you had been communicating throughout the year based on what seems to be a strong market environment, and your existing backlog, how would you call out the seasonality in the first half of the year, you see something a little atypical, maybe probably better than seasonal in the first half of the year? How do you see your visibility plan in the first half of the year on the product side?

Luc Seraphin

Yes. So you're correct, Gary, the second half of the year, we have this seasonal effect that is advantages to us. I would add to this, so this year that I think we're really pleased with the execution. We do see our companionship ramp in earnest, for example. There's a new generation being adopted at a higher rate than they used to be. When we look into 2027, I would say that we see demand strengthening. I think this is not going down. So we're quite optimistic for 2027. The demand is strengthening. We have a better visibility into our backlog is going into 2027.

I think we're going to continue to see some sort of seasonality between the first half and the second half in 2027. But in aggregate, we are confident with the health of the demand growth in 2027.

Gary Mobley

As my follow-up, I wanted to ask about the Silicon IP business. I think you've been communicating that this might be a $130 million annual business. I would assume that -- you're maybe trending above that. Can you confirm or give us any color on that? And then with respect to custom ASIC development, do you see an opportunity, whether it be on high-speed Seres interface IP, high-speed memory interface IP or security IP to forge a long-term development arrangement with some of the hyperscalers developing custom silicon like what we've seen with some of the other competitors in the IP space?

Luc Seraphin

Yes. So Gary, I'll start with the second question about the potential for silicon IP and then I'll ask Sumeet to answer on the revenue and the revenue growth. We are engaged more and more with the ecosystem on the silicon IP because people are building more and more complex chips that require more and more complex integration. And on the bleeding edge of technology, this is, for example, what we described in our last call with the engagement with a hyperscaler on next generation of memory.

So we see our confidence in the silicon IP keeps increasing. If you look at our product rollout, we talk about PCIE7 IPPCIE7 switch, HBM in ex generations of HBM, these dedicated security solutions for the data center. So we are engaged deeply with the ecosystem. that gives us the opportunity going forward to potentially continue to increase our portfolio, not only organically but possibly inorganically and also to partner with our customers to develop broader system.

So as part of our thinking for silicon IP, we have that notion of working more and more with our partners on the system level. By the way, the announcement we made about Agentic-AI helps to do that. the ability to have agents work with us on integrating IP in much more complex systems is actually enabled by this type of things. So we're building the infrastructure to do that. That's the direction we are going into. It's going to take us time, but that's definitely a great trend for us. I'll let Sumeet answer on the numbers, yes.

Sumeet Gagneja

Yes. Thanks, Luc. And just to add to that, Gary. So yes, so we've said our silicon IP business around $130 million, growing at 10% to 15%. But like Luc said, our confidence has been building throughout the year, and 2026 has been a great year for our silicon IP business, along with our strong momentum in the design win. So I would put it more like we're drawing more towards the higher end of that range. We have said 10% to 15%, and we continue to make progress in all aspects from memory interface and security aspects.

Operator

Your next question comes from the line of Sebastien Naji with William Blair.

Sebastien Cyrus Naji

Yes. Maybe first, I wanted to ask about the LPDDR 6 PMIC that you announced. Could you maybe give us a sense for how that increases your dollar content per module versus the SO CAM2 content that you had announced in April, I think it was 2 voltage regulators and an SPD Hub. And then when do you expect maybe some initial revenue for that product?

Luc Seraphin

Thanks, Sebastian. I would say that I'll be a bit discrete about the exact content on this so can 2 because it's not -- the final space has not been released yet. So you'd rather wait for the final spec to be released. But what I can say is that when you move from the first version of Slocan that we announced earlier this year to this new generation of SOCOM. The content is increasing. The power management is more complex. And as a result of that, the dollar content is going to increase from that first generation to that second generation.

The module performance itself is going to be much better as well, which will, I would say, trigger adoption at a higher level. But as every 1 of these Scan solutions, they provide a very good solution for these areas very high where power performance is really, really important. But this is a small portion of the market, and that's going to be continuing -- that's continued to be the case between the first version of SoCam and this will come to. SoCAM2 will hit the market probably in 2028. And the way to look at it is we see the development of the AI infrastructure as having multiple horizons for us, if you wish.

We want to make sure that we developed those solutions way ahead of the market. And when they deploy in the market, then we are present with solutions that work. This is what we've done in the past that explains, for example, our performance in Q4 this year that are based a lot on the products we developed in the past. they had little traction at the beginning, but they're starting to show revenues in Q4 of this year. This is going to be the same with SoCal.

SOCOM is going to address a niche portion of the AI stack, AI infrastructure. But the adoption of LP is going to happen. It's not going to happen fast. It's not going to be the same content as the current content of standard memory, but it's important for us to have that as part of our portfolio, while we help our customers and ecosystem partners to deploy their solutions in the market.

Sebastien Cyrus Naji

Great. Okay. That's very helpful. And then maybe my second question or my follow-up for Sumeet on gross margins. looks like non-GAAP product gross margins were about 60.4% in Q3. That's on the lower end of the, I think, 60% to 63% near-term range that you gave us. Could you maybe give us a sense for how much of the gap here is mix in terms of just more companion chip revenue, next-generation RCDs with early yields versus how much of it is wafer substrate or OSAT cost inflation. And then how should we think about Q4 product gross margins? Do they stay in that lower 60% area? Or can they move up a little bit?

Sumeet Gagneja

Yes. Thanks for the question, Sebastian. Yes. So our gross margin historically has been despite ban 60% to 63%. And you're right, in the Q3 was slightly above 60%. Our long-term target remains, which is 60% to 65%. In the near term, it's really the margin trending towards the lower end is primarily driven by the supply chain input cost. And we've been very successful in offsetting that cost input cost by driving internal cost savings and keeping our product gross margin in the range of 60% to 63% in the near term.

Q4, it's the same sort of setup going into Q4, but the investments and the cost reductions we have done especially in our operations side, there may be a slight uptick in the gross margin in Q4, but it's raw material, but it's still within the 60% to 63%. On a full year basis, it's more important to look at from the gross margin perspective. we expect 2026 to be right in the middle of that 60% to 63% range.

Operator

Your next question comes from the line of Mark Lipacis with Evercore.

Mark Lipacis

I had 2. I was hoping you could just talk about lead times from your own suppliers and any to extent that you're seeing tightening of supply on the front end of the back-end foundry business? And then Elie, can you talk about lead times to your customers? Are there any stretching of lead times to your customers from you? And any indication of your customers downstream or the supply chain downstream is increasing their own inventories given how tight everything seems to be becoming?

Luc Seraphin

Thanks, Mark. I'll start and let Sumeet comment. I would say that on the supply side, we do see lead times increasing. We signaled this 2 calls ago, we were expecting this. But because we were expecting this, we also put in place plans to mitigate for that. So we lever our strong balance sheet to actually build inventory on products that we believe are going to grow substantially with our customers, and we're in constant relationship with our suppliers to make sure that we use our balance sheet to make sure that we can supply to our customers.

So yes, definitely, longer lead times on the supply side, but it has not impacted our Q3 number, and it will not impact the Q4 number that we just talked about. So we've been able to manage that. But we have to continue to do that. Our lead times, the first approximation, our lead times to our customers have not increased at this point in time. But you understand that we are in constant discussions with them as well. the whole chain is tightening. The good thing is our business is very focused. We have very strong partnerships with our customers. and we work with them in making sure that we can supply what they need.

In summary, longer lead times on the supply side, we are using our balance sheet to deal with that. And at this point in time, we've not increased our lead times to our customers.

Mark Lipacis

And look, and just to be clear, do you have the ability to see if customers or customer customers downstream or increasing their inventories? Or is that like a black box to you, and it's really hard to determine.

Luc Seraphin

It's very hard to determine. We see -- what we see is that people place backlog earlier than what they used to a quarter ago or 2 quarters ago. I think everyone understands the current situation. So we have better visibility in the backlog. But at the same time, that gives us more opportunity to have a deeper, I would say, relationship with our customers once we understand the profile of their backlog then it's easier to work with them on how we can lead that backlog. So we don't see people building inventory, but we see people placing backlog.

Operator

Your next question comes from the line of Tristan Gerra with Baird.

Tristan Gerra

So if I heard well, you mentioned that your revenue in the quarter and the guidance is not constrained by supply. So it probably provides a little bit of a easy comp were just in the first half, where you did face some supply issues. What's your level of confidence that you can actually need to be accelerating demand that you described into next year? And I know you're not guiding beyond this quarter, but do you think that the acceleration that you're seeing in product revenue year-over-year is sustainable? Or do you see supply as an issue again next year? And is there any talk about longer-term agreements or anything that customers are trying to set up to ensure supply for you.

Luc Seraphin

So we see demand -- thank you, Tristan. We see demand strengthening into next year, and we started to work with our supply chain partners to meet what we believe is going to be required. In all honesty, this is something that we work on almost on a weekly basis. We try all avenues to support our customer base. So demand is strong going into 2027. Supply will remain tight, but as I said, as we did in 2026, we're working early with our supply base to look at every not we can turn to support our customer demand. that's how we look at this.

The good news is that the demand for the products we have in our portfolio is strong. The demand for CPUs, this CPU to GPU ratio is going in our favor. It's known from our customers all the way down to our suppliers. and we'll do our best to supply that growing demand into 2027.

Tristan Gerra

Great. And as a quick follow-up, I wanted to explore the difference in content opportunity medium term for x86 CPU and stand-alone Vera and cannot come up with maybe a for content in XED6 CD4o versus what you would get on the SOCEM-2side with -- but are you quantifying at all what the opportunity is and and the difference that you see in terms of opportunities on the XB6 and the on-base of CPUs next year and the years beyond.

Luc Seraphin

Yes. Thank you, Tristan. The way we look at it and the reason we are excited about this AI infrastructure deployment is that we look at this in an aggregate way. There are areas of AI like if you take the head nodes, for example, where -- SoCam and LTE memory can be used. It's not always used, but it can be used. And that's an opportunity for us from a technology standpoint. It allows us to develop our LP technology on all fronts.

But in aggregate, you do have a growth in these head nods that will create an opportunity for SoCAM, but you have higher growth in the traditional servers with a refresh necessary in order to continue to develop the standard range of applications, a range of standard applications. We have a higher growth in agentic standard CPUs as well for orchestration, tools management, memory management. So in aggregate, if you take the heads where you can have SOCAM.

And if you take the standard servers and the agentic server, the whole market is growing, but the majority of the growth is coming from CPUs with standard memory subsystems. So that's the way we look at it. We definitely don't look at this as SoCAM is cannibalizing the market for standard dam. It's actually complementary. It makes the system more efficient. The standard memory subsystems are going to increase at a higher pace going into 2027.

And the head nodes will drive new technologies like SoCAM to maximize this kind of application. So we see this as complementary. And that's why we want to have solutions for all of that. We really want to be agnostic to the memory subsystems while the infrastructure for AI is growing in the market? Because as that infrastructure is growing in the market, we're going to see hetrogenous system, and it's important that we have solutions for each 1 of those. But the vast majority is going to continue to be CPU with a standard GM configurations.

Operator

Your next question comes from the line of Kevin Cassidy with Rosenblatt Securities.

Kevin Cassidy

And I'll echo the congratulations on the great quarter. Just -- it's interesting, as you mentioned about the custom chips and that you're getting involved earlier into the design cycle. And I'm just wondering if -- is there any way does this help you to win more modules on those designs as they come into production?

Luc Seraphin

Thanks, Kevin. We're really excited with this silicon IP engagements with the ecosystem because they give us great insight into the architectural trends that these people are talking about. When we talk to hyperscalers with the opportunity for us to sell our silicon IP, we have insights into some of the architecture problems they're trying to resolve and where they see the world going. And as a consequence of that, that helps us develop -- continue to develop the right road map of products for this type of products.

But it also gives us insight as to the overall architecture of their systems and where they are going to be used, where LP is going to be used and why. What generation of DDR is going to be used and why? It just gives us that channel of information. It's almost through the discussion, and it gives us a good sense of the trends in terms of what type of technologies they are going to use not only on the silicon IP side, but also in terms of how they will configure their systems with different types of modules.

Kevin Cassidy

Right. And it seems like a big advantage over your 2 competitors that don't have IP story. And maybe just on the inventory 1 more time. Are you comfortable with the days of inventory now? Or should we expect to see that continue to increase even with the increasing demand next year?

Sumeet Gagneja

Yes, Kevin, good question. So inventory, as you mentioned, we've been growing inventory this year. In Q3, we grew it again $12 million, again, primarily leveraging the strength of our balance sheet to make sure we pared gridsurance to our customers. In Q4 as well, looking at, just like Luc mentioned throughout the call, like we have more visible look into the backlog and as demand is accelerating, we do plan to build increased inventory even in Q4 similar levels to like we build in Q3 and Q2.

Operator

At this time, there are no further questions. This concludes the question-and-answer session. I would like to turn the conference back over to the company for closing remarks.

Luc Seraphin

Thank you to everyone who has joined us today for your continued interest and support. We look forward to speaking with you again soon. Thank you.

Operator

Thank you. This concludes today's conference. You may now disconnect.