Good afternoon. My name is Connor Meek, and I will be your conference operator today. At this time, I would like to welcome everyone to the Astera Labs Q3 2026 earnings conference call. All lines have been placed on mute to prevent any background noise. After management remarks, there will be a question-and-answer session. If you would like to ask a question during this time, simply press star followed by the number one on your telephone keypad. If you would like to withdraw your question, press star one again. Thank you. I will now turn the call over to Leslie Green, Investor Relations for Astera Labs. Leslie, you may begin.
Thank you, Connor. Good afternoon, everyone, and welcome to the Astera Labs third quarter 2026 earnings call. Joining us on the call today are Jitendra Mohan, Chief Executive Officer and Co-founder, Sanjay Gajendra, President and Chief Operating Officer and Co-founder, and Desmond Lynch, Chief Financial Officer. Before we get started, I would like to remind everyone that certain comments made in this call today may include forward-looking statements regarding, among other things, expected future financial results, strategies and plans, future operations and the markets in which we operate. These forward-looking statements reflect management's current beliefs, expectations, and assumptions about future events, which are inherently subject to risks and uncertainties that are discussed in detail in today's earnings release and the periodic reports and filings we file from time to time with the SEC, including the risks set forth in our most recent annual report on Form 10-K.
It is not possible for the company's management to predict all risks and uncertainties that could have an impact on these forward-looking statements or the extent to which any factor or combination of factors may cause actual results to differ materially from those contained in our forward-looking statements. In light of these risks, uncertainties, and assumptions, the results, events, or circumstances reflected in the forward-looking statements discussed during this call may or may not occur, and actual results could differ materially from those anticipated or implied. All of our statements are made based on information available to management as of today, and the company undertakes no obligation to update such statements after the date of this call, except as required by law. Also during this call, we will refer to certain non-GAAP financial measures, which we consider to be important measures of the company's performance.
For example, the overview of our Q3 financial results and Q4 financial guidance refer to various non-GAAP financial measures. These non-GAAP financial measures are provided in addition to, and not as a substitute for, financial results prepared in accordance with US GAAP. A discussion of why we use non-GAAP financial measures and reconciliations between our GAAP and non-GAAP financial measures and outlook are available in the earnings release we issued today, which can be accessed through the investor relations portion of our website. With that, I would like to turn the call over to Jitendra Mohan, CEO of Astera Labs. Jitendra.
Thank you, Leslie. Good afternoon, everyone, and thanks for joining our third quarter 2026 conference call. Today, I will cover our Q3 results, market trends, and traction across our AI fabrics and signal conditioning portfolios. I will then turn the call over to Sanjay to discuss our long-term growth profile, and Des will cover our Q3 financial results and Q4 guidance. Astera Labs delivered outstanding results in Q3, with record revenue of $584.9 million, up 49% sequentially and up 154% year-over-year. Operational execution was strong during the quarter as we broadened our customer base with new design wins and expanded our intelligent connectivity platform with the launch of multiple new products. Q3 revenue growth was broad-based, led by the high-volume ramp of Scorpio X-Series for scale-up networks. Scorpio was our largest product family in Q3, with Scorpio X-Series revenue crossing over our P-Series.
Together, Scorpio represented over $1 billion in annualized revenues. This is an important milestone in our journey to be a critical rack-scale AI infrastructure solutions provider. Since our last earnings call, we have meaningfully expanded our portfolio with a record number of product launches. This includes the launch of Aries 7, the industry's first PCIe Gen 7 copper and optical connectivity solutions, Leo 2 and Leo X-Series smart memory controllers for AI and general-purpose use cases, and Orion, our new compute-agnostic rack-scale architecture targeting enterprise AI and inference applications. Together, these innovations broaden the markets and applications we serve, creating greater value for our hyperscaler and enterprise customers while supporting a diversified long-term revenue growth profile. Our design-in activities picked up significantly in Q3 at both new and existing customers.
Scorpio is increasingly establishing itself as a diversified AI fabric platform, with momentum extending across both X-Series and P-Series portfolios. In Q3, X-Series shipped to two customers across multiple platforms, and we expect to deliver pre-production volumes to additional new customers during the current quarter. The P-Series similarly broadened its footprint across multiple platforms, including two U.S. hyperscalers in Q3. Beyond AI fabrics, we see additional vectors for diversification. Taurus 800G solutions are qualified and shipping in volume to our lead customer and are currently sampling across several additional customers. Leo momentum continues with qualifications completed across several customers, including two hyperscalers, and we expect to ramp volume shipments in mid-2027. Our custom solutions business is also seeing a substantial increase in design activity, as the number of opportunities have broadened over the course of the year, with over 10 active engagements.
Finally, our platform business, anchored by our Orion reference architecture, is generating strong early customer interest, with initial shipments expected to begin in 2027. AI infrastructure demand continues to strengthen, with our hyperscaler customers reporting strong revenue growth from AI services. Yet the physical inputs, wafer capacity, grid power, and memory supply remain tightly constrained. That combination puts a premium on utilization. At the same time, the workload mix is shifting to inference, where continuous reasoning loops multiply the traffic crossing the fabric. Operators now optimize for two things: latency, to deliver faster time to first token, and accelerator utilization, to drive dollars per token. Our Scorpio AI fabric is purpose-built for exactly these metrics. As token economics become increasingly important to our customers, let me highlight the differentiated capabilities driving Astera's success in AI fabrics. First, Radix.
Our portfolio of Scorpio X devices, including the industry's largest 320-lane configuration, enables large cluster sizes for PCIe-like scale-up AI fabrics. Scorpio X's high lane count helps to maximize scale-up domain sizes while enabling an all-to-all flat topology to allow AI accelerators to communicate at full bandwidth with low latency. This is particularly important for AI inferencing, as fewer hops reduce collective communication latency, allowing resources to be spent on compute rather than moving data between switches. Increased utilization results in higher tokens per second, and ultimately, greater monetization from the same accelerator investment. Second, intelligence. Conventional switches just move packets. Scorpio X-Series is AI workload-aware and participates in collective communication and computation. Hardware-accelerated Hypercast and in-network compute offload AI operations, such as all-reduce and broadcast, from the accelerators directly into our fabric, delivering up to a 2X improvement in collective performance.
Similarly, fabric-attached Leo memory controllers for KV cache offload deliver significant improvement in time to first token and tokens per second. By utilizing advanced AI-aware capabilities enabled by Scorpio and Leo products, customers can improve token economics and increase token output with the same number of accelerators. Third, software. COSMOS is a software layer that transforms our AI fabric and smart connectivity solutions into infrastructure that our customers can deploy, manage, and operate efficiently at scale. COSMOS software spans our entire portfolio, is broadly adopted across the AI ecosystem, and extends from device-level diagnostics to fabric and rack management. COSMOS speeds rack deployment by resolving connectivity issues in software instead of costly hardware spins, then keeps rack online with fabric-wide telemetry and predictive fault detection.
It also configures and manages Hypercast and in-network compute, so our software directly participates in the AI workload running on XPUs, in addition to managing the connectivity infrastructure. Faster time to revenue and higher utilization of AI infrastructure are the reasons why COSMOS is driving new design wins across our product portfolio. These differentiated capabilities are optimizing performance and increasing XPU utilization, which is rapidly expanding our opportunity funnel and accelerating design win conversion. As a result, we expect to deliver strong diversified growth with our AI fabric portfolio through increasing value to our customers and continued expansion of our dollar content per AI accelerator. Today, our content per XPU across fabric switches and copper signal conditioning can exceed $1,000 per XPU. We expect our next-generation AI fabric products alone to contribute well beyond $1,000 per XPU.
Over time, we aim to address the entire scale-up switching TAM, expanding from PCI Express and NVLink Fusion-based platforms today to systems based on UALink and other protocols. With the addition of memory controller and optical connectivity products, increasing AI fabric attach rates driven by next-generation scale-up protocols, including UALink, larger cluster sizes, and higher ASPs associated with each data rate transition, we see a path to doubling our total silicon dollar content per XPU with each new generation of AI platforms. In summary, Q3 marked another important step in Astera Labs' evolution. Record results were led by the high-volume ramp of Scorpio X-Series and continued growth across our signal conditioning portfolio. We also saw strong design win momentum across new and existing customers and launched a record number of products during the quarter.
Looking ahead, we see significant opportunities across optical connectivity, custom silicon, memory solutions, and rack-scale infrastructure that build on our core franchises, expand our addressable market, and position us for long-term growth. With that, let me turn the call over to our President and COO, Sanjay Gajendra.
Thanks, Jitendra, and good afternoon, everyone. This quarter was an important inflection point for Astera Labs. We have evolved from a component-level supplier into a rack-scale infrastructure platform provider. We are now bringing together silicon, hardware and software to solve connectivity challenges across the AI rack and pods over both copper and optical interconnects to improve XPU utilization and higher ROI. As we broaden our intelligent connectivity portfolio, we are increasing our silicon dollar content to thousands of dollars per accelerator across AI and cloud infrastructure while expanding our addressable market opportunity. Our foundation today comprises of two core revenue-generating franchises, AI fabrics and signal conditioning, and is rapidly expanding across four additive market vectors, including optical connectivity, custom silicon, memory solutions and our rack-scale infrastructure platforms. As AI systems scale, the challenge is no longer simply connecting more accelerators inside a server.
As Jitendra mentioned, customers are building larger scale-up domains that need more bandwidth, lower latency and smarter ways to move data to improve token economics. Our real-world experience deploying scale-up networks in production volume is helping us define workload-specific, high-value feature sets. Based on these factors and increased attach ratios, we believe we can more than double the dollar content per XPU in each subsequent generation. Signal conditioning remains a critical problem to solve as data rates double and AI clusters scale. As a result, we expect ASPs and attach rates for our new PCIe Gen 7 portfolio across optical and copper to expand versus the prior generation. In addition, our Aries gearbox and redriver solutions are expected to address new sockets within AI platforms.
Taurus for 800G applications began shipping in volume during Q3, and we anticipate solid growth within scale-out Ethernet AECs heading into 2027, driven by higher ASPs and new program ramps. We are also sampling 3.2T retimers to multiple customers. Looking ahead, we are adding four growth drivers that materially expand our addressable market. Starting with optical, we continue to make great progress on our fiber connectivity business. At OCP last month, we introduced our first optical products, including smart optical drivers and TIAs optimized for scale-up PCIe Gen 7 topologies, addressing near-packaged and linear pluggable optics applications. We also demonstrated integration of our optical technology with our Scorpio X-Series fabric switches. Our smart TIAs and laser drivers will begin sampling this quarter to select hyperscaler and OEM partners for NPO applications, and we are actively working with multiple customers to extend PCIe fabrics beyond the rack to the pod level.
With near-package optics for scale-up application, we see our TIAs and drivers unlocking additional content opportunity per XPU, and we expect these optical products to begin initial revenue contribution in late 2027. Longer term, we are developing fully integrated optical engines with the goal of building an optically enabled Scorpio X-Series fabric switch. in 2030, we estimate the scale-up optics interconnect opportunity to be a $40 billion market, which we will service through our high-density fiber-attached solutions and optical engines for NPO and CPO applications. Moving to custom solutions, the AI accelerator landscape continues to broaden and diversify, driving demand for purpose-built intelligent connectivity. Interest and activity for our custom offerings have accelerated in 2026, and we are engaged with 10-plus opportunities now. We are developing custom fabric solutions, custom memory controllers for specialized inference applications, and IO chiplets for next-generation accelerator platforms.
We expect initial shipments of these solutions to commence in late 2027 and ramp to volume in 2028. We are being selective and focusing on high-value custom programs, looking to leverage our deep portfolio of connectivity IP, robust supply chain partnerships, and system-level expertise. Next is our memory solutions business, where we recently expanded our Leo Smart Memory Controller family to address new applications and use cases. Leo X-Series is our new fabric-attached memory controller that pairs with our Scorpio X-Series AI fabric to create a dedicated memory tier for KV caches and agent context. We also introduced Leo 2 E-Series and P-Series, which double memory bandwidth and capacity versus the prior generation and support both DDR4 and DDR5 DIMMs. As memory constraints intensify, Leo lets customers expand memory capacity, pool standard memory across a rack, and reliably reuse previously deployed DDR4 alongside new DDR5 DIMMs.
Engagement and design activity remain strong across a broadening set of hyperscaler and inference cloud customers. I am happy to report that Leo 2 products are now shipping in pre-production volume, with high-volume ramp projected in mid-2027. Turning to platform solutions, we recently announced Orion, our compute-agnostic rack-scale reference architecture built to enable turnkey AI infrastructure solutions for inference cloud and enterprise customers. Orion spans six production-ready reference designs, scales from eight to 160 accelerators per cluster, and integrates with COSMOS, our software management platform that is now enhanced to include rack and fabric manager functionality. These reference designs leverage our portfolio of intelligent connectivity solutions, including AI fabrics, smart signal conditioners, advanced memory solutions and COSMOS to provide a tightly integrated, highly performant platform.
We are already seeing substantial interest from GPU and XPU suppliers, and Orion is on track to enter the market through our ODM and OEM partners in 2027. Fully loaded Orion rack-scale designs contain greater than $100,000 in Astera Labs content, creating another large additive market opportunity. In conclusion, we remain focused on execution and delivering substantial value to our AI and cloud infrastructure customers across silicon, hardware, and software. Design activity accelerated with wins across a broad set of programs and applications at multiple customers. Finally, a record number of product launches is broadening our portfolio, creating deeper synergies across our solutions and expanding both our market opportunity and silicon dollar content per AI platform. With that, let me turn the call over to Des.
Thank you, Sanjay, and good afternoon, everyone. Today, I will review our Q3 financial results and then discuss our Q4 guidance, both presented on a non-GAAP basis. We delivered outstanding third quarter results with record revenue of $584.9 million, up 49% sequentially and 154% year-over-year. Growth was broad-based across our AI fabric and signal conditioning portfolios, led by the volume ramp of Scorpio X-Series. Revenue also diversified across a broader base of customers and platforms during the quarter. Scorpio became our largest product family by revenue in Q3, also with X-Series revenue surpassing P-Series. X-Series revenue increased significantly as shipments expanded across multiple platforms and radix configurations, while P-Series also grew sequentially as existing deployments expanded and new programs entered production.
Within signal conditioning, Aries growth was led by strong PCIe Gen 6 adoption across scale-up and scale-out applications, while Taurus also gained momentum as our 400G and 800G solutions ramped across scale-out applications. Non-GAAP gross margins for the third quarter was 72.4%, above our guidance of 72%. Non-GAAP operating expenses for the third quarter were $162 million, up 19% sequentially, reflecting continued R&D investment to expand our addressable market across our core franchises and emerging growth sectors. Non-GAAP operating margins for the third quarter was 44.7%, up from 39.1% in the prior quarter, an increase of 560 basis points, reflecting significant operating leverage as revenue scaled. Interest and other income was $13 million, and our non-GAAP tax rate was 11.4%.
Non-GAAP net income was $243.4 million, resulting in non-GAAP diluted earnings per share of $1.32, up $0.52 or 65% sequentially from Q2. We ended the quarter with $1.38 billion in cash equivalents, and marketable securities, an increase of $125.3 million from Q2. The increase was primarily driven by record cash from operations of $179.2 million. Now, turning to our outlook for the fourth quarter. We expect revenue to be in the range of $730 million-$750 million. At the midpoint, this represents 27% sequential growth. Growth is expected to be driven by the continued expansion of Scorpio X-Series and P-Series across both customers and platforms. Signal conditioning growth is expected to be led by PCIe Gen 6 retimer adoption in Aries and the ramp of Taurus 800G solutions. We expect fourth quarter non-GAAP gross margin to be approximately 72%.
We expect fourth quarter non-GAAP operating expenses to be between $194 million and $200 million, reflecting continued investment in our core AI fabric and signal conditioning portfolios, as well as additional investments in the growth opportunities outlined by Sanjay. We expect non-GAAP operating margin to be approximately 45%. Interest and other income is expected to be approximately $13 million, and we expect our non-GAAP tax rate to be approximately 12%. We expect our Q4 share count to be 187 million diluted shares outstanding. Overall, we expect non-GAAP fully diluted earnings per share to be between $1.62 and $1.66. In closing, Q3 was an exceptional quarter marked by record revenue, significant operating leverage, and the strategic milestone of Scorpio becoming our largest product family by revenue. Looking ahead, our growth drivers continue to broaden. Scorpio X-Series and P-Series are scaling across new and existing customers and platforms.
In signal conditioning, Aries and Taurus are benefiting from growing adoption of next-generation connectivity standards. Beyond our core franchise, opportunities are emerging across optical, custom memory and platform solutions, reflecting the strategic R&D investments we have made to expand our portfolio and address full market. We will continue to invest across our portfolio, execute with discipline, maintain strong profitability, and deliver long-term value for our stockholders. This concludes our prepared remarks. I will now turn the call back to our operator to begin Q&A. Operator?
We will now begin the question and answer session. Please limit yourself to one question. At this time, I would like to remind everyone, in order to ask a question, press star, then one on your telephone keypad. We will pause for just a moment to compile the Q&A roster. Your first question is from the line of Harlan Sur of JP Morgan. Your line is open. Please go ahead.
Good afternoon. Thanks for taking my question, and as always, great execution by the team. On the new Orion rack-scale reference architecture, as large corporations like ours start to evaluate open source, open-weight models on premise for lower token inferencing costs, control on things like model fine-tuning, and just more importantly, like internal data security, these customers are looking for compute platforms that can fit into their private data center architectures, right? Air-cooled, compact cluster sizes, flexibility on XPU, CPU supported. Orion seems to be that sort of perfect solution. So, what has been the interest level? Have you engaged with large enterprises? And what is the Astera dollar content capture across the different configurations you offer?
Great question, Harlan. Thank you so much. No, we are very excited with Orion. Like you correctly pointed out, enterprise AI is something that is a significant market. It is a very fragmented market with different workloads, and all of them need compute in a form factor that is readily available, but more importantly, it is workload-specific. So what we have developed with Orion is that it is a GPU-agnostic rack, meaning you can bring any GPU that you want that suits your workload, or based on availability and things like that. The rest of the infrastructure is all defined, meaning the connectivity is all defined as part of the reference design, the power, the cooling, and other components that are essential. What Astera is providing are three main things. We are providing the entire reference architecture, whether it is at a box level or rack level.
We're providing baseboards that utilize our silicon that combines the fabric solutions, the memory solutions, the signal conditioning and other components that are required in a well-optimized, fully tuned, and fully validated way. Then we're providing the software, which we're excited about, because COSMOS has now gone from being a device-level software to being a rack-level and fabric-level. So it manages all the fabric, it manages the rack-level diagnostic, and other aspects that are involved in orchestrating the GPU performance. So overall, Orion is squarely addressing use cases like you outlined, and the Orion family will have multiple generations of it, which will obviously have higher and higher content. But the initial platform that we call Orion 1.0, that's based on PCIe scale-up, we estimate about $100,000 of Astera content that will be monetized through components, through boards, and software.
Thanks, Sanjay.
The next question is from the line of Joe Moore of Morgan Stanley. Your line is open. Please go ahead.
Great. Thank you, and congratulations. Can you talk about Scorpio X? You hit the billion-dollar threshold. It's been such a steep ramp. Do you anticipate that there could be a plateau as the products that Scorpio X is in sort of matures before you get to UALink? Or just help us out a little bit with what that progression may look like. It seems like you're anticipating a sort of long tail for PCI Express, but also this transition to other protocols. Can you just talk about that dynamic?
Hi, Joe. It's Des here. Thanks for the question. We've been really pleased with our sort of execution throughout 2026, which really has shown broad-based strength across the portfolio. If you look at our growth in the second half of the year, this has really been led by scale-up deployments and the production volume ramp of Scorpio X-Series, which started in Q3 and will continue into Q4. This led to us being able to deliver almost 50% revenue growth in Q3. Using the midpoint of our guidance for Q4, we do expect to see a further 27% sequential growth. If I look further ahead, what I would say is that the initial Scorpio X platforms have completed much of their initial deployment ramps, and they'll be contributing at scale going forward.
We expect Scorpio X to continue to grow, driven by new platforms and also additional customers layering into the model throughout 2027. If I look further ahead, given our experience with high-scale deployments, advanced feature sets, and our strong relationships with our customers, we are very well-positioned for the next-generation platform growth. Certainly beyond Scorpio, we're excited about the growth opportunities as we outlined in the call earlier. We continue to see growth in the new product areas of Aries 7, Leo, and optical products, which will all start layering into the model in 2027 and offer future growth opportunities. Overall, we're very pleased with our performance and overall momentum that we have across our portfolio.
Great. Thank you.
Thanks.
Your next question is from the line of Blayne Curtis with Jefferies. Your line is open. Please go ahead.
Hey, guys. Thanks for taking my question, and nice results here. I wanted to ask on NPO. You mentioned some design progress there. You had some new parts at OCP on the analog side. Can you just maybe walk us through how we think about all these different components layering in? Is there designs that you're working on for the full NPO solution now, or is that a bit later?
Absolutely now, Blayne. As we noted in the prepared remarks, we are planning to start sampling our NPO products to lead customers this quarter. We do expect roughly about six months of qualification and system-level integration and all that. So we do expect that revenue will start contributing towards the second half of 2027 from NPO deployments. These are, again, components that we're servicing for going inside the NPO module, the TIAs and laser drivers and so on. So we do expect hundreds of dollars to come in per accelerator with the NPO type of products and solutions that we're building. Going forward, NPO, we do expect to continue to drive the growth beyond 2027, with CPO volumes coming in probably mostly in the 2028, 2029 timeframe.
Thanks, Sanjay.
The next question is from the line of Papa Sylla of Citi. My apologies. The next question is from the line of Tore Svanberg of Stifel. Your line is open. Please go ahead.
Yes. Thank you, and congrats on another record quarter. You've got these four new big growth engines, and I agree, these are huge TAMs. You also stepped up the, or you're going to step up the OpEx quite a bit here in the December quarter. I'm just curious, should we think of a lot of the OpEx for these four new business opportunities, is that still in the come, or have you already made a lot of heavy lifting to target those four opportunities? Thank you.
Hi, Tore. Des here. We continue to invest in the business at the right level to support the high-growth opportunities ahead of us. You're correct, we did see the OpEx step up to about $162 million in Q3. We expect it to continue to grow into Q4. What I would point out is that our OpEx is growing slower than our revenue growth, so you're seeing a significant leverage within the model. I would say that our OpEx growth is really driven by three factors in the second half of the year. We continue to grow our headcount to support our customer programs. Our headcount has doubled year-over-year from here. Secondly, I would say that we've seen additional variable R&D expenses in the form of wafers, masks and IP expenses.
Then lastly, as we continue to outperform our P&L model, we are seeing some higher variable comp from there. Again, what I would really reiterate is that we are seeing the significant leverage within the model. I think going forward, the right way to think about OpEx in the longer term is around 30% of revenue, and we're certainly underperforming that today given the high revenue growth that we're seeing.
All right. Thank you.
Thanks.
The next question is from the line of Ananda Baruah of Loop Capital. Your line is open. Please go ahead.
Yeah. Good afternoon, guys. Congrats on the progress, and thanks for taking the question. Yeah, I guess question clarification. You guys mentioned a couple times of doubling the content for XPU with each new generation of platforms. I guess the question clarification is it each new generation of XPU, or is it each new generation of Astera Labs, like Gen 7, Gen 8? Or is it something else? Would just love to get context around that, how to think about that. Thanks.
Great question. This is Jitendra. If you look at the content that we have in some of the highly loaded systems today, it already exceeds $1,000 of content on a per XPU basis across our fabric switches and signal conditioning products. As we go to the next generation of our products, we aim to add $1,000 plus content just from switching opportunity on a per XPU basis. That kind of gets you to the first doubling as we speak.
Beyond that, as we go to the next generation of AI platforms or the next generation of XPU, we bring in additional content to bear on these platforms, which includes, for example, optical networking components that Sanjay just talked about, Leo memory controllers for KV cache offload and so on, as well as growing cluster sizes that drive higher ASP on a per product basis, as well as higher data rates that also drive higher ASP on a per product basis. If we model all of this out, we expect to get double the content on a per XPU basis on each successive generation of XPU.
Oh, boy. That sounds like a lot of content, Jitendra. I appreciate it, guys. Thanks a lot.
Yeah. Our customers continue to build larger and larger cluster sizes, and as these cluster sizes go bigger, we get a higher attach rate. Thank you.
The next question is from the line of Papa Sylla of Citi. Your line is now open. Please go ahead.
Thank you for taking my question, and congrats on the very strong Q3 results. Maybe Jitendra or Sanjay, just wanted to double-click a little bit on the prior 10 plus of Scorpio X engagement you previously talked about. Any color on how many of those you expect to be meaningful ramps in the first half of 2027? Tied to that, how are you still thinking about UALink timeline? Do you still kind of see more of a second half 2027 story? If that's the case, is it reasonable to assume that some of your current Scorpio X engagements will kind of turn into UALink engagement or opportunities?
Yeah, thanks for the question. Yeah, what we had guided was that there are more than 10 customers we are engaged on Scorpio X, so those engagements continue to make progress in a very positive way. What we noted was that we, of course, have our lead hyperscaler platform that is now in production.
That is reflected in the Q3 numbers and also on the guide that we provided for Q4. What we also noted is in Q3, we started shipping the X-Series Scorpio to an additional hyperscaler. So that is now in pre-production, so that should help us generate revenue in 2027. We do have backlog from multiple additional customers in Q4 for Scorpio X. So these are parts that will get shipped out this quarter, and then it will go through its development and qualification phases in the first half of 2027 and contributing revenue in the second half.
Now, what is also happening is that because of all the results and performance that we are able to deliver to our customers with Scorpio X, with many of the unique features that we highlighted in our prepared remarks, the customers that have designed in are now adding new platforms or essentially using Scorpio X on new platforms. So not only are we growing the number of customers that are adopting Scorpio X, we are also seeing customers rapidly expanding the use of Scorpio X in multiple platforms that they are developing. So overall, I want to say the Scorpio X has taken off very well, reflected both in the pipeline opportunities as well as the revenue that we highlighted with Scorpio family exceeding $1 billion of annualized run rate in Q3.
Maybe let me take the UALink question. We continue to see the UALink ecosystem strengthen. There is more IP available. There are more vendors that are coming in with products for UALink. So overall, there is not a lot of change in the UALink from a customer standpoint. The customers who had focused on memory semantic-based protocols, such as PCI Express, continue to stay in that swim lane and will, as you said, over time convert to UALink. Our development of UALink products also continues. We introduced our Taurus family earlier this year, and we continue to work on the anchor sockets on switching and so on in order to intercept them with the release of the UALink XPUs from our customers. Not much has changed. Those who are in the UALink swim lane continue there.
Those who are in the Ethernet swim lane will continue in that swim lane. Overall, our strategy is to address the entire scale-up switching TAM. We have started with PCI Express, gone on to the NVLink Fusion, and actively working on UALink. As we start to see more customer interest, we will address other protocols for scale-up as well.
Got it. That's very helpful. Thank you.
The next question is from the line of Sebastien Naji of William Blair. Your line is open. Please go ahead.
Thanks, and good afternoon. Thanks for taking the question. I wanted to stay on the theme of Scorpio in particular, but maybe ask a little bit about the P-Series. Scorpio X-Series revenue crossed P-Series in the quarter. Can you maybe just comment on how we should think about the growth in P-Series in Q4 and in 2027? It seems like X-Series has maybe more significant content, but is more concentrated, while P-Series is perhaps lower content, but has a broader set of customers. Just how should we think about the mix between those two products going forward?
Hey, Sebastien, and it's Des here. We've been really pleased with our performance in Scorpio P-Series. In the third quarter, it grew double-digit sort of growth, and I expect it to continue in Q4 at a similar sort of rate, sort of going forward. As we continue to look further ahead into sort of next year, what we continue to see is the hyperscalers continuing to grow. We talked about having two hyperscalers growing in revenue in the second half of this year, so that'll continue to grow into next year. We continue to see a variety of engagements across different customers that continue to ramp. So we've seen this being a continue to be a meaningful sort of part of the overall sort of Scorpio ramp. Obviously, Scorpio X-Series, given the higher ASPs, that's going to be the sort of dominant sort of growth driver going into next year.
P-Series continues to perform really well.
Thank you.
Thanks.
The next question is from the line of Quinn Bolton of Needham. Your line is open. Please go ahead.
Hey, guys. Let me add my congratulations on the outlook. I just wanted to ask a longer-term question. You gave us four new product categories on the call, optical, custom, the memory controller expansion, and then the Orion rack-level platforms. I will not hold you to it, but of those four, would you say one is more likely than the others to hit 10% of revenue at some point over the next couple of years? Or how should we think about the relative growth rates of those four new opportunities?
Yeah, what I would start off by saying is that all the four are significant TAMs, as you can imagine, with optical or the memory inferencing use cases. Custom is a huge opportunity out there, both for XPU attach and some of the other XPU-like opportunities. For us, and of course, the rack is going into several enterprise and inference application. The value proposition and form factor there is very different. I would say if you start off, these are all big markets for us. Now it comes down to how are we able to develop the right product, make sure that we execute and deliver, and it's a matter of timing is what I would say. In other words, the most near-term opportunity that we're expecting in second half is all the optical ramp that we're expecting, specifically on the NPO side.
That is something that will have the most immediate effect from a revenue contribution standpoint. Custom silicon, that one is a muscle that we have been building and making a lot of progress. Like we noted, we have over 10 engagements right now for doing custom fabrics, custom memory controllers, and so on. This one, I would say, is going to be a significant opportunity for us, but I would say probably more in the 2028 timeframe. The last one, the Orion, to be candid, it's been fantastic reception since we announced it at OCP. Lot of folks are coming to us, especially enterprise folks and folks that are building XPUs and other accelerators. That one, we will start launching the product, of course, through ODM and OEM partners that are building the system for us. That will start contributing revenue in 2027.
We're yet to sort of get a full handle on exactly what that numbers would look like for the Orion platform, so we expect to provide more concrete data on that in our next quarterly call. But in general, the key message here is that all of these are substantial market opportunities. They all require innovation and execution that we have proven ourselves to be able to deliver as a company. We are looking forward to playing in all this market and contributing additional growth engines to go along with the signal conditioning franchise we have and the fabric franchise that we have. Overall, if you can have six engines that are all pumping out great growth, I think we will have achieved the goal of being a great iconic company.
Thank you, Sanjay.
The next question is from the line of Karl Ackerman of BNP Paribas. Your line is open. Please go ahead.
Hi, this is Sam Feldman on for Karl Ackerman. Thanks for taking the question. You are seeing strong growth with Aries. Can you delineate this growth between content uplift from the transition to PCIe Gen 6 versus unit growth? More generally, how do you think about your retimer content for XPU as optics moves closer to compute?
Yeah, that is a great question. We had a record quarter for Aries, and Aries family continues to grow. We started the company actually with the Aries PCI Express Gen 4, and now we are going to Gen 5, Gen 6 and at OCP, we launched our Gen 7 portfolio. As you correctly pointed out, there are two factors that are causing the ASP to go up. One is just the increase in data rate. Every time we double the data rate, we can expect to get a 20%, 25%, 30% increase in ASP from generation to generation. The other factor is that we also ship these devices in multiple form factors. You have the chip-down opportunity where the chip is directly soldered on the board, but we also have the smart cable modules.
where the chip is sold in the form of a PC board that gets then built into an active electrical cable, and there we get a higher ASP. Just as importantly, we have these active electrical cables attaching into the scale-up part of the network, and therefore we get a higher attach rate and overall growth in revenue. On the high level, I would say that we have been very happy with the growth of the Aries family. It continues to grow. It has grown from 2024 to 2025 to 2026, and we expect it to grow in 2027 as well.
The last point that you made is, some of these links that were previously served with copper will get served with optics as we go into the next generation, and there is an opportunity for us to have even higher content because the dollars that we can get on a per link basis on optical links is much more than what we get on copper. In addition to that, there were several links that previously did not require any kind of retimers at all, which will now require both copper as well as optical content, and that's a great opportunity for us to continue growing the Aries portfolio.
The next question is from the line of of Simon Leopold Raymond James. Your line is now open. Please go ahead.
Yeah, thanks. Jeff Koche for Simon. Just wanted to ask you a quick question on the Marvell-Lumentum merger. Obviously, trying to bring silicon and photonics closer together, potentially enabling tighter co-design and improved power. I just want to get your take on this strategy. Do you think that this can provide a competitive advantage, and does it change how you are viewing how you're going to vertically integrate? Thanks.
Marvell and Lumentum are both great companies. We have good respect for both of them, so I'm sure they've thought through the strategy before getting into this combination discussion. I'll let them comment on what they think about it. But I'm happy to talk about our own strategy for optical. Our strategy for optical is more grounded in scale-up, right? Unlike the other companies, we are not yet focused on scale-up, scale-out, or scale-across. Our focus is on scale-up, which is not a big TAM today, but as we talked about during the prepared remarks, we expect that to be a $40 billion TAM going forward. And all of the components that we need to address that TAM, we are building ourselves, starting with the electrical ICs, laser drivers, and TIAs for the NPO generation, and more complex EICs for the CPO generation.
At OCP, we also showed our silicon photonics that goes into some of these NPO modules, as well as the connectors that allow us to build full optical engines for NPO as well as CPO. These optical engines will then get deployed with our hyperscalers, and hyperscalers will continue to source laser components just like they do today with their existing vendors.
Great. Thanks a lot, guys.
The next question is from the line of Tom O'Malley of Barclays. Your line is open. Please go ahead.
Hi, team. This is Trip Smith on for Tom O'Malley. Nice results, and thank you for the question. I was just curious if you could talk about the cadence of revenue growth into calendar year 2027. I think there's some expectations that your lead customers' current generation builds are kind of peaking out in the 1Q timeframe, and then the next gen is not really expected to ramp until kind of closer to calendar year 2028. So any color there would be super helpful. Thank you.
It's Des here. Thanks for the question. Obviously, we will not comment upon our end customers' volume ramp. I think I addressed some of this earlier in the Q&A, where I talked about our the large scale that we've been able to generate on the Scorpio X platforms. The initial deployment has taken place, and they'll continue contributing in scale going forward. We'll continue to see new platforms plus additional customers leaning into the model on a go-forward basis. I think it's important that we also remember the scale that we've been able to generate in terms of our Scorpio X family.
I think Jitendra talked about a $1 billion run rate being achieved in Q3, and we're optimistic about our growth prospects going forward as we go into next-generation platforms as well.
The next question is from the line of Bastien Faucon of Susquehanna. Your line is open. Please go ahead.
Hi. Thank you for taking my question. I'm filling in for Mehdi Hosseini. I wanted to come back to your comment about the content that you're capturing by Orion rack scale designs. You mentioned that you capture about $100,000 of content per rack. In your press release, you highlighted that those pods have from eight accelerators to 160. I wanted to know a little bit more about how to think about that $100,000. How many GPUs or accelerators are you taking into consideration for that estimate? My follow-up would be, how should we think about the gross margin profile of the rack scale solutions relative to your long-term gross margin targets of 70%? Thank you.
Yeah. Let me try to break it down for you. The Orion rack is designed to support up to 10 compute trays. Each compute tray can have up to eight GPUs or accelerators that are plugged in.
Our content, which includes the fabric, the switches, and so on, they reside at the compute tray level to interconnect the GPUs. We do have content there. Now, some customers, rather than operating the 10 compute trays individually, they want to create a larger cluster, meaning operate the 80 GPUs as one unit. For that, they need a scale-up switch and a tray as well. In those situations, there is additional hardware that is required. And of course, when you think of 160 GPUs, we are talking about two racks that are talking to each other with connectivity going from the east to west type of topologies.
In general, if you think about that kind of system, then the components that we are able to integrate in that includes, of course, all of our fabric and switch products, both at the compute tray level and at the rack level. It includes all of our signal conditioning products, retimers, and such that would be required, including AECs, right? And then we are also, for the first time, monetizing our COSMOS software. COSMOS software, like I noted, has upgraded from being a component-level solution to a rack-level solution, a fabric manager, and so on. That provides additional monetization points that adds to the $100,000 that I noted. And then the last point I would make is that, again, we are providing this as production-ready reference design.
Obviously, there is a certain amount of royalty that we get from our OEM and ODM partners that are taking that and providing it to their end customers, the enterprise AI and inference cloud guys. All in all, there are three monetization points. It is the baseboards that we sell, it is the software that we sell and royalty that we get for the reference design. That is the way to think about how the monetization happens. When we said 100K, we are just talking about one rack, meaning 80 GPUs, even though the reference design itself can support up to 160. The $100,000 refers to a single rack. That is the other point I want to emphasize. And overall, like I noted, this is just the beginning of the Orion story, meaning we will have additional versions that brings in UALink and liquid cooling and all that stuff.
In general, we do expect that we will be able to add more value by doing a vertical integration of the components and software. In terms of the margin itself, we do expect this to be accretive, just given the business model that I noted, where we are not only providing silicon, but we are providing software, and there is additional revenue coming in through the royalty approach that we are taking. Overall, it is meant to bring in top line. It is meant to be accretive to the margin that we have. I will also like to call out and say this is early days for this platform. The interest has been great. We will, of course, keep updating you guys on how the business is growing, and I think in Q1, we will be able to provide a lot more details on the impact for our overall P&L.
Thank you very much.
There are no further questions at this time. I will now turn the call back over to Leslie Green for closing remarks.
Thank you, everyone, for your participation and questions, and please refer to our investor relations website for information regarding upcoming financial conferences and events. Thank you so much.
This concludes today's conference call. You may now disconnect.