Astera Labs, Inc. Common Stock (NASDAQ:ALAB) Q3 2025 Earnings Call Transcript

Astera Labs, Inc. Common Stock (NASDAQ:ALAB) Q3 2025 Earnings Call Transcript November 4, 2025

Astera Labs, Inc. Common Stock beats earnings expectations. Reported EPS is $0.49, expectations were $0.39.

Operator: Good afternoon. My name is Eric, and I will be your conference operator today. At this time, I would like to welcome everyone to the Astera Labs Q3 ’25 Earnings Conference Call. [Operator Instructions] I’ll now turn the call over to Leslie Green, Investor Relations for Astera Labs. Leslie, you may begin.

Leslie Green: Good afternoon, everyone, and thank you, Eric. Welcome to Astera Labs Third Quarter 2025 Earnings Conference 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 Mike Tate, 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 filed and filings we filed from time to time with the SEC, including the risks set forth in our most recent annual report on Form 10-K and our upcoming filing on Form 10-Q.

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 any forward-looking statement. In light of these risks, uncertainties and assumptions, the results, events or circumstances reflected in the forward-looking statements discussed during this call 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 conference call, except as required by law.

Also during this call, we will refer to certain non-GAAP financial measures, which we consider to be an important measure of the company’s performance. These non-GAAP financial measures are provided in addition to and not as a substitute for financial results prepared in accordance with U.S. GAAP. The discussion of why we use non-GAAP financial measures and reconciliations between our GAAP and non-GAAP financial measures is 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?

Jitendra Mohan: Thank you, Leslie. Good afternoon, everyone, and thanks for joining our third quarter conference call for fiscal year 2025. Today, I’ll provide an overview of our Q3 results, followed by a discussion around the current trends within AI Infrastructure 2.0. I will then turn the call over to Sanjay to walk through Astera Labs near- and long-term growth profile. Finally, Mike will give an overview of our Q3 2025 financial results and provide details regarding our financial guidance for Q4. Astera Labs delivered strong results in Q3 with our revenue and profitability metrics coming in above our outlook. Quarterly revenue of $230.6 million was up 20% from the prior quarter and up 104% versus Q3 of last year. Growth within the quarter was broad-based across our signal conditioning, smart cable module, and switch fabric products.

Scorpio P-Series continued its initial volume ramp at our lead customer, and we are excited that our P-Series revenue will further broaden with recent new design wins across a variety of AI platforms at multiple hyperscaler customers. Scorpio X-Series is shipping in preproduction quantities with a volume ramp expected throughout 2026. Our Aries portfolio continues to perform well with PCIe 6 solutions contributing robust growth during the quarter. Our Aries 6 products are the industry’s first and only PCIe 6 retimer solutions ramping in high volume today. Taurus drove strong growth during the quarter as incremental opportunities began shipping in volume across both AI and general purpose systems, and we expect further growth in 2026 as we expand to 800-gig switching platforms.

For Leo CXL memory expansion products, customers are exploring AI inference use cases, especially to offload memory from expensive on-package HPM to large pools of DDR5 memory. This will broaden Leo opportunities beyond the current general purpose compute applications that we continue to support. On the organizational front, we have grown aggressively and plan to exit 2025 with a global team of more than 700 employees, up 60% compared with the beginning of the year. Lastly, we are happy to report that our non-GAAP operating margin of 41.7% marked a new record level for the company. In addition to our strong financial performance and new design wins, we continue to lay the foundation for future growth with the advancement of our technology roadmap and scaling of our team and capabilities.

In October, we announced that Astera Labs had entered into a definitive agreement to acquire Xscale Photonics, a provider of leading-edge fiber chip coupling technologies. This acquisition will help enable us to develop photonic scale-up solutions by combining Xscale’s fiber chip coupling capabilities with Astera Labs connectivity and signal conditioning expertise. We envision future Scorpio scale-up switches to be enabled with photonic solutions to optically expand rack scale cluster sizes containing hundreds of connected AI accelerators. This acquisition represents an important step within our long-term optical journey to intercept a large additive market opportunity associated with scale-up photonics. The industry continues to see strong momentum with major announcements pointing to ongoing rapid growth in large-scale AI infrastructure deployments.

Increasing AI use cases are driving higher monetization and surging demand for compute as evidenced by token generation doubling every 2 months and significant year-over-year increases in LLM user activity. To meet this demand, the industry is rapidly adopting rack-scale infrastructure, which analysts forecasting CapEx at the top 4 U.S. hyperscalers to surpass $500 billion in 2026. This shift to AI Infrastructure 2.0 will require ultra-low latency all-to-all connectivity for large workloads, and Astera Labs is advancing its intelligent connectivity platform to deliver high-performance, energy-efficient fabric switching solutions that maximize AI platform efficiency and productivity. To achieve the goal of providing our customers with a wide choice of innovative, flexible and efficient connectivity solutions, we are building our portfolio based on open standards.

Our full portfolio of standard-based solutions was in display at the 2025 Open Compute Project Global Summit with the support of 15 industry partners, all highlighting the importance of an open ecosystem for AI rack-scale infrastructure. We believe the proliferation of open standards-based AI rack-scale platforms will allow the industry to leverage broad innovation and enable interoperability while providing a diverse multi-vendor supply chain. We are particularly enthusiastic about the continued momentum behind the UALink scale-up connectivity standard, which exemplifies the open ecosystem approach by combining the low latency of PCIe and the fast data rates of Ethernet to deliver best-in-class end-to-end latency and bandwidth. UALink was built from the ground up with broad contributions from market-leading AI infrastructure participants to specifically solve the mounting challenges of scale-up networking.

We believe UALink delivers the bandwidth efficiency and the ultra-low latency needed to unlock full accelerator performance and enable effective scaling as AI clusters expand. Customer activity around UALink continues to be strong. We are engaged with several leading hyperscalers and AI platform providers in the RFP and RFQ stages to align on the designs and applications that fit best with their technology and business requirements. We continue to expect a portfolio of UALink solutions to be available to customers in the second half of 2026 with early revenues generated in 2027. With that, let me turn the call over to our President and COO, Sanjay Gajendra, to outline our vision for growth over the next several years.

Sanjay Gajendra: Thanks, Jitendra, and good afternoon, everyone. Today, I want to provide an update on our recent execution, followed by an overview of the meaningful market opportunities that will fuel Astera Labs growth over the next several years. Astera Labs has a singular goal to deliver a purpose-built intelligent connectivity platform, including silicon, hardware and software solutions to customers for rack-scale AI deployments. The forthcoming evolution to AI Infrastructure 2.0 will not only be defined by faster silicon and larger AI clusters, but also by open connectivity standards and software that promote innovation at scale. In short, standardized high-speed interconnect technologies will be essential to deliver AI open racks that are highly performant while operating as one cohesive unit.

During Q3 of 2025, Astera Labs continued its high-growth trajectory and further diversified our overall business to deliver another record quarter. We are excited to report several new design wins at multiple hyperscalers during the quarter for Scorpio P-Series fabric switches across a variety of AI platforms supported by both merchant GPUs, including NVIDIA’s GB300 and B300 as well as designs based on custom AI accelerators. Additionally, our Aries PCIe 6 smart retimer business and customer opportunities are now expanding as AI racks built around custom AI accelerators and new merchant accelerators begin to adopt PCIe 6. This dynamic is poised to further accelerate the broader adoption of PCIe 6 across the ecosystem and further drive our dollar content opportunity.

Overall, our PCIe 6 solutions contributed in excess of 20% of our Q3 revenues, illustrating our market-leading position. We see a similar dynamic taking shape within the Ethernet market with the transition to 800-gig links putting additional strain on signal integrity. Given faster speeds and larger AI cluster sizes, system architects are turning to Ethernet AEC applications to solve the reach challenges of passive cabling. This transition is expected to drive market growth with increasing overall volumes and a generation over generation ASP lift. While we expect strong continued demand for our 400-gig solutions throughout 2026, we also believe our customer base will diversify with 800-gig solutions, driving a new layer of growth for our Ethernet smart cable modules.

We believe our approach to enable multiple cable partners with our smart cable modules, supports the scale and flexibility that is preferred by hyperscalers. Looking ahead, we are gearing up for Scorpio X-Series to shift to high-volume production over the course of 2026. With this ramp of Scorpio X-Series for scale-up connectivity topologies next year, we expect our overall dollar content opportunity per AI accelerator to significantly increase, representing another step-up from a baseline revenue standpoint. Overall, given the extreme importance of scale-up connectivity to AI infrastructure performance and productivity, we see Scorpio X-Series solutions as the anchor socket within next-generation AI racks. Our early engagements are providing us valuable insights in terms of both hardware and software requirements to deploy scale-up switching networks for a diverse set of GPUs and AI accelerators.

Beyond the connectivity protocol like PCIe, UALink or Ethernet, there are additional functions, both in the data part and management of scale-up networks that can make or break the performance and deployment of scale-up networks. We are learning this every day, building a competitive moat and ensuring our solutions are ready for real-world deployments at scale. From an implementation perspective, the architecture of our Scorpio X family was built to support multiple platform-specific scale-up protocols and customizations. We are actively expanding our PCIe-based scale-up fabric solutions. And in parallel, we are working on future UALink products for applications that need higher bandwidth. For PCIe, we are engaged with over 10 AI platform providers with opportunities that are expected to drive revenue growth across multiple generations of AI platforms over the next several years.

We view UALink opportunities to be meaningfully additive to our PCIe scale-up revenues. Our flexible fabric architecture, hands-on experience with scale-up networks, support for diverse workloads that run on training and inference clusters of various scale and complexity and open approach puts us in an excellent position to win next-generation designs. As we look to 2026 and beyond, our playbook remains the same: one, stay closely aligned with the multigenerational technology roadmaps of our customers and partners; two, innovate exponentially in everything we do; and three, separate the noise from reality and continue to be laser-focused on execution needed for a thriving, durable business. In conclusion, we are motivated by the meaningful opportunity that lies before us, and we will continue to passionately support our customers by strengthening our technology capabilities and investing in the future.

With that, I will turn the call over to our CFO, Mike Tate, who will discuss our Q3 financial results and our Q4 outlook.

Michael Tate: Thanks, Sanjay, and thanks to everyone for joining the call. This overview of our Q3 financial results and Q4 guidance will be on a non-GAAP basis. The primary difference in Astera Labs non-GAAP metrics is stock-based compensation and its related income tax effects. Please refer to today’s press release available on the Investor Relations section of our website for more details on both our GAAP and non-GAAP Q4 financial outlook as well as a reconciliation of our GAAP to non-GAAP financial measures presented on this call. For Q3 of 2025, Astera Labs delivered quarterly revenue of $230.6 million, which was up 20% versus the previous quarter and 104% higher than the revenue in Q3 of 2024. During the quarter, we enjoyed revenue growth from our Scorpio, Aries and Taurus product lines supporting both scale-up and scale-out PCIe and Ethernet connectivity for AI rack-level configurations.

Scorpio P-Series demand for PCIe Gen 6 scale-out applications was robust during the quarter. Aries demonstrated solid growth during the quarter for both Gen 5 and Gen 6 solutions. With the transition to PCIe Gen 6, we gained increased dollar opportunities with both our Scorpio and Aries Gen 6 products as demonstrated with our Gen 6 revenues exceeding 20% of our Q3 revenues. Taurus growth during the quarter was driven by increasing shipments for 400-gig scale-out connectivity and AI systems. Q3 non-GAAP gross margin was 76.4% and was up 40 basis points from the June quarter levels with product mix remaining largely constant across higher volumes. Non-GAAP operating expenses for Q3 of $80 million were up $9.4 million from the previous quarters due to higher payroll taxes and the continued expansion of our R&D organization.

Within Q3 non-GAAP operating expenses, R&D expenses were $57.2 million. Sales and marketing expenses were $10 million, and general and administrative expenses were $12.8 million. Non-GAAP operating margins for Q3 reached a new record level of 41.7%, up 250 basis points from the previous quarter. Interest income in Q3 was $11.5 million. Our non-GAAP tax rate for Q3 was 18%. Non-GAAP fully diluted share count for Q3 was 180.6 million shares, and our non-GAAP diluted earnings per share for the quarter was $0.49. Cash flow from operating activities for Q3 was $78.2 million, and we ended the quarter with cash, cash equivalents and marketable securities of $1.13 billion. Now turning to our guidance for Q4 of fiscal 2025. We expect Q4 revenues to increase within a range of $245 million and $253 million, up roughly 6% to 10% from third quarter levels.

For Q4, we expect growth across our Aries, Taurus and Scorpio product families with particular strength from our Taurus smart cable modules. We expect Aries growth to be driven by a number of end customer platforms where we support scale-up and scale-out connectivity. Strong Taurus growth is expected to be driven by increased volumes on 400-gig designs for AI scale-out connectivity. Scorpio growth will be primarily driven by the continued deployment of our P-Series solutions for scale-out applications on third-party GPU platforms, while we expect Scorpio X-Series to ship initial volumes. We expect Q4 non-GAAP gross margins to be approximately 75% with the increased mix of our Taurus hardware modules in the quarter. We expect fourth quarter non-GAAP operating expenses to be in the range of approximately $85 million to $90 million.

Anticipated operating expense growth in Q4 is driven by the expectation of continued investment in research and development functions and also the incremental operating expenses from the Xscale acquisition anticipated to close during the quarter. Interest income is expected to be approximately $11 million. Our non-GAAP tax rate should be approximately 15%. Our non-GAAP fully diluted share count is expected to be approximately 183 million shares. Adding this all up, we are expecting non-GAAP fully diluted earnings per share to be approximately $0.51. This concludes our prepared remarks. And once again, we appreciate everyone joining the call. And now we will open the line for questions. Operator?

Q&A Session

Operator: [Operator Instructions] Your first question comes from the line of Harlan Sur with JPMorgan.

Harlan Sur: Again, great execution by the team. Post the announcement of UALink 1.0 specification in April, this is the industry’s first standard scale-up networking architecture. There have been a plethora of new scale-up announcements, mostly Ethernet-based scale-up architectures. I think the market has been concerned about these competitive architectures, but we know that GPU and XPU chip design cycle times are anywhere from 18 to 24 months and the scale-up architecture associated with these designs have been spec out like way in advance. So in other words, I assume that your design win pipeline and engagements, XPUs or GPUs that either have decided to use Scorpio X or UALink has not changed at all since the last earnings, maybe even expanded, but wanted to get the team’s view.

Sanjay Gajendra: Yes, absolutely, Harlan. We continue to see our market opportunity grow for our scale-up products, particularly Scorpio X product like you noted. Scale up, as you can imagine, it’s a very large market. We estimate it to be in tens of billions of dollars. And like you correctly noted, some of these design wins take last over multiple generations simply because of the investment that goes into developing the software and the hardware required for scale-up topologies. For us, if you think about our business, today, we are getting ready to ramp into production with our PCIe-based scale-up solutions. It’s been extremely popular. There are several customers that are using PCIe like protocols for scale-up. A new entrant was Qualcomm that publicly announced their new AI 200 inference racks that feature PCIe-based scale-up.

For Astera, we have engaged with over 10 AI platform providers, and we expect that these design wins and engagements that we have will continue to ramp. In fact, we expect this to go 2029 just based on some of the multi-generation nature of these design wins. For us, UALink is also a very meaningfully additive opportunity as customers start adopting it just based on the higher data rate support. The spec has been around, like you noted, for over a year now in terms of the consortium being formed. The spec is stable. The ecosystem is forming. Silicon development is in full gear. And many of these customers, we have currently engaged with RFPs and RFQs. So the momentum is really built up very nicely and continues to grow. So we do expect meaningful revenue from UALink to start coming in, in 2027.

There are, of course, other standards being defined, and that is to be expected. This is a market that will have multiple standards that will coexist. But for us, the bottom line is that we are in a good position to address all of the emerging scale-up market opportunities with the engagements we have, the learnings that we have had by being in the trenches over the last, let’s say, 9 to 12 months, developing scale-up solutions, understanding what is needed and what is not needed, and with production ramps happening in 2026. So overall, we feel very confident that this is going to present multiple opportunities for us, resulting in a multi business — multibillion-dollar business on the scale-up side based on all the opportunities that we see in the market.

Harlan Sur: No, I appreciate that. And the team has talked about the Scorpio X as an anchor product, right? In other words, customers design your switch fabric solution. This creates the opportunity for additional content pull in, right, whether that’s your signal conditioning products, your AEC, optical cable modules. Is this strategy playing out? In other words, if you look at, let’s say, all of your Scorpio X engagements, what percentage of these engagements are also using your retimers, your AEC or optical cable module solutions? And do you have a sense of the average content uplift per XPU on these incremental attach?

Sanjay Gajendra: Yes. So like you correctly noted, if you are a system designer at a hyperscaler, on day 1, when you decide on building a new platform, you generally think of 2 things. One is the accelerator and other one is the scale-up switch and the topology for it. So fortunately, we get invited to the conversation very early. And some of these conversations are multigenerational. So it gives us a good outlook for not just on requirements that we have in the near term, but also on the long term. We announced the acquisition that we are working towards for Xscale. And that was driven based on similar insights that we’ve been able to gather in terms of what is needed for us. In terms of content itself, once that we are in the sockets for our scale-up solution, it naturally opens up conversations around other products that we have, whether it’s retimers, gearbox devices, controllers and things like that, which we have been able to maximize in terms of how we can service.

In terms of dollar content, what I would say is that overall, if you look at some of these future design wins that will ramp up, they scale up to multiple thousands of dollars if you look at it from an accelerator and a rack level. So in general, we do see that having a strong presence in the scale-up network allows us to pull in several other products and technology that we currently have and also working on in terms of future product lines that we intend to offer to our customers.

Operator: Your next question comes from the line of Ross Seymore with Deutsche Bank.

Ross Seymore: The first one I wanted to follow on to the switch fabric side of things with Scorpio. You talked about more design wins across several platforms and more customers. I guess where I really want to get some more color is on the diversification theme. How are you seeing that business diversify? And when Scorpio X launches, does that naturally come back to some concentration? Or does that further diversify the business? And what I appreciate is a naturally concentrated market. But within that framework, how do you see that business diversifying over time?

Sanjay Gajendra: Yes. So in general, the theme that we have been working towards is to ensure that there is a good diversity, both with our product lines and customer base that we have. Like you correctly noted, the hyperscaler market is fairly concentrated. I mean that’s the occupational hazard that we all have to deal with. But to your point, today, like we have noted for things like PCIe-based scale-up and in the future, UALink and other protocols. Today, we have over 10 customer platforms that we are engaged with. We have made tremendous progress in the last quarter, making progress in terms of not just design wins, but also for some of the opportunities moving them forward from a technical POC software development and other aspects that are needed to deploy this technology at scale.

So at this point, given our presence with the fabric devices, that’s truly allowing us to be very broad-based. And this not only includes third-party GPU-based platforms, but also custom accelerator-based products. And that’s been an exciting momentum for us right now as we seek to add many more design wins to the customers that we have on Scorpio Series.

Ross Seymore: And I guess one for Mike on the gross margin side of things. It’s very, very impressive. I understand the mix dynamic and why it might be going down a bit in the fourth quarter, but it’s still well above your 70% long-term target. So I think investors are just wondering what would be the puts and takes that would drive it down from kind of the mid-70s to 70% over time, especially if the scale-up architectures and different products are going to become so important to you and as Sanjay said, are relatively accretive on the dollar amount, and I assume even on the gross margin side.

Michael Tate: Yes. So on the first order effect, generally, when we sell hardware products and modules versus silicon, that’s margin dilutive. So we do see an uptick in tours in Q4. So that’s the guidance to 75%. As we look longer term, we are going to greatly broaden our product portfolio and the design cycles are moving very fast. So in doing that, we will have a wider range of margins for our products generally because the market is moving so fast, we can’t have very pointed products for every opportunity. So some will have a cost structure which is a little more overburdened for the opportunity set, and that will be part of the mix. So we still encourage people to think about us going to our long-term model. But with that, we do see operating leverage as we grow our revenue dollars at a very good pace.

Operator: Your next question comes from the line of Blayne Curtis with Jefferies.

Blayne Curtis: Great results. Maybe I just want to start off on just level set. Obviously, you beat by a wide amount. I think you mentioned kind of the first 2 you mentioned was single conditioning and these SCM modules. I’m just trying to figure out if you can kind of — I know you don’t want to break out certain segments, but can you give us a little bit more color as to what drove the beat and what changed during the quarter?

Michael Tate: Yes. We saw breadth through all the 3 product lines. We generally want to be conservative because a lot of the revenue growth that we have are from new programs, and these programs are very complex. So we just want to give a little cushion in case there’s any delays in the product launches by our customers, but it was a very successful quarter for our customers in their deployments. So that enables us to deliver the upside.

Blayne Curtis: And I want to ask you, I mean, obviously, the — what NVIDIA has done with retimers was a lot of the talking points throughout the year, but you’re starting to see these ASIC platforms going to be more material next year. Is there a way to think about that Aries family as these ASICs ramp on a relative basis versus kind of the retimer content you’re seeing today?

Sanjay Gajendra: Yes. So just to kind of level set, right? So we do expect a significant growth in Aries revenue this year, and we do expect the revenue growth for Aries family to continue to next year as well. So in general, obviously, the ASP of the retimer business is different compared to the Scorpio or the switch fabric business that we have. And we do expect that Scorpio to be our largest product line from a revenue standpoint. And there are obviously several different design wins that we have that — that are expected to ramp to production volume in 2026. So in general, what I want to say is that the business has transitioned to some of these larger sockets and the higher ASP business that we have, and that trend will continue with the inflection point happening sometime in 2026, when Scorpio will overtake Aries and other product lines from a revenue standpoint.

Operator: Your next question comes from the line of Tore Svanberg with Stifel.

Tore Svanberg: Congrats on the strong results. So I had a question on the acquisition and you now penetration into the optical scale-up market. Just curious, material revenue time lines. I assume this is potentially the beginning of more to come. And maybe you could also discuss a little bit why you decided to intersect optical now versus perhaps prior or later.

Jitendra Mohan: Yes. Thanks, Tore. Maybe I’ll begin and then Mike can add on. Look, our vision has always been to deliver complete connectivity infrastructure at the rack scale. We have stated this many times, we call it AI Infrastructure 2.0, and we are laser-focused on building solutions for that. Today, we are focused on copper-based solutions, mainly because this is what our customers ask us to do. However, as data rates increase and scale-up domains go beyond 1 rack, clearly, at some point, you will need optical interconnects for scale-up. And there is already a big market for optical interconnects at a data center scale. So we view our kind of entry into optical as a big additive opportunity, and we will intercept the market with unique solutions that are aligned with our customers’ roadmaps in the sense of when they want to transition from copper to optical.

So as far as the timing is concerned, why now and why not earlier or why not later? It is part of the plan that we have with our customers on when we want to intercept with an optical solution. It is also important to note that with this acquisition of Xscale, we are adding capabilities to the company that we did not have before. Xscale allows us to get the glass components that are required to deliver a successful optical product, whether it is a CPO or an LPO or an NPO. But this is a technology, that’s very complementary to the signal conditioning and switching expertise that we have. So our vision would be to deliver a product line where our Scorpio family is optically enabled with photonic solutions to allow for higher data rates and longer reach in scale-up domains.

With the Xscale acquisition, we get a phenomenal team that we can kickstart this development and this acquisition is just again our commitment to enter this market and intercept at the right time.

Michael Tate: I’m sorry, Tore, you had a question on the timing of…

Tore Svanberg: Yes. As part of my first question, when can we start to expect material revenue coming from optical products from Astera? Is that ’27, ’28, ’29?

Michael Tate: More likely, the earliest for scale-up optical connections will be in the ’28, 2029 time frame.

Tore Svanberg: Very good. And just as my follow-up, you talked about Taurus driving strong growth here in Q4. I think you mentioned 400 gig. Is that also diversified growth? Is this with more than one customer? And when do you see the inflection happening for Taurus for 800-gig in 2026?

Sanjay Gajendra: Yes. So the 800-gig deployments, I want to say, are just starting in terms of the market need. So for us, we are engaged with several customers. Our business model for AEC is to offer the smart cable modules that then gets enabled through multiple cable vendors. And generally speaking, there is a little bit of lag between when customers start their initial POC or initial deployment to when they start scaling. So overall, we believe that from an 800-gig standpoint, our business — our revenue impact would start in 2026. I want to say, early part of ’26 as the qualifications complete and start ramping to production.

Operator: Your next question comes from the line of Mehdi Hosseini with SFG.

Mehdi Hosseini: A couple of follow-ups. I just want to go back to the target that Scorpio would be about 10% of your revenue. And if that’s the case, then does it imply that Scorpio would be like closer to 20% of the revenue in the December quarter?

Michael Tate: Yes. The 10% was for the full year. It started to launch materially in Q2. So the exit rate would be closer to the 20%. That’s correct.

Mehdi Hosseini: Okay. And then with the X ramping, let’s say, spring of next year, that’s when the contribution is going to actually accelerate. Am I thinking about this right?

Michael Tate: Yes. P will continue to grow given that we have new designs that will be ramping throughout the year. So P and itself is a nice growing piece of revenue for Scorpio. The X-Series is in kind of low initial volumes right now, but then it starts to ramp materially next year. What we said before is the X is ultimately a bigger opportunity, the scale-up opportunity. So at some point, and we’re not saying exactly when, it will be bigger than P, and we’re very excited about that potential.

Mehdi Hosseini: Sure. And then I have a rather clarification question. I’m new to the name, maybe just me, but when you say you have 10 AI platforms involved with your Scorpio product, what does that mean? Does that mean 10 different CSPs, 10 different customers? The AI platform, if you could just elaborate on it, it would be great.

Sanjay Gajendra: Yes. So we refer to the customer base that we have that includes the folks that are developing their own accelerators. It also includes the hyperscalers that are buying some of the third-party accelerators and integrating it into their AI servers. So those are the 2 broad categories to think of in terms of the 10 customers that we noted.

Mehdi Hosseini: Okay. So that basically implies the diverse set of customers that are adopting the UALink, the open-source, right? Is that — would that be fair?

Sanjay Gajendra: Yes. So that comment itself is correct. We do believe that there’s quite a bit of momentum around UALink based on the fact that it’s developed grounds up. But the 10-plus customer comment we made was in reference to folks that are using PCIe-like protocols for scale-up. However, we do believe that the folks that are using PCIe-like protocols would also be looking at UALink as an option to service platforms that require higher data rate, meaning from a physical layer standpoint. So to that standpoint, UALink would be additive to our PCIe customer base. At the same time, it will also provide an upgrade path for folks that want higher speed on specific AI platforms.

Operator: Your next question comes from the line of Quinn Bolton with Needham & Company.

Quinn Bolton: Let me offer my congratulations as well. I just wanted to come back to the Xscale Photonics acquisition as it’s your first entry into the optical side of things. I think this technology looks like it’s fiber chip coupling. It seems like you probably need some kind of silicon photonics capability to complete a scale-up CPO type solution. So just wondering, is that something you look to develop in-house? Would that be sort of an acquisition that you would look to pursue in the future? Just how do you complete that full scale-up CPO solution?

Jitendra Mohan: Quinn, good question. Yes. So as you have correctly pointed out, in order to build a full optical solution, you need 3 pieces. You need an electrical IC that takes the signals from, let’s say, a switch chip or an XPU chip, converts it into a format that’s applicable for a photonic chip. So that’s the second component that you need, a photonic chip that will now convert these electrical signals into light. And then you need a packaging technology that will couple this light into fibers and so on. And there are very specific requirements for each one of them. With the acquisition of Xscale, we solved 2 of the 3. So they are working on some very cutting-edge technology on package development. Once the acquisition closes, we will be able to reveal more about what that means to us and how we will intend to use it.

But they are, as you correctly pointed out, working on packaging technologies that is a very critical part of the equation. We also get a lot of photonics expertise as part of this acquisition as well. So we will look to put a team together internally to work on photonics. But at the same time, photonics is a very complex equation wherein customers also have a lot of say into what photonics to use. So we are open to not only work on our solution, but also use third-party photonics solutions to enable an overall optical solution that is suitable for our customers’ requirements. And then when it comes to electrical, we’ve been doing electrical chips for many years as part of Astera Labs and a long time before that. So we feel pretty confident in building the electrical components.

But all 3 of them put together is what makes a compelling optical solution. And we have some great ideas on how to build a unique solution as we enter — as we contemplate entering this space.

Quinn Bolton: And then I guess just wanted to come back to the comment about initial Scorpio X shipments in the fourth quarter. Is that kind of preproduction more sample units? Or are you starting to see the initial Scorpio X design win going to production? Is this the initial build of a production system?

Sanjay Gajendra: Yes. So this is the initial build. So we’ve gone through the qualification stage and all the intermediate stages. So we start shipping into production volumes — production systems end of the year, but the big ramp will happen in ’26.

Operator: Your next question comes from the line of Sean O’Loughlin with TD Cowen.

Sean O’Loughlin: Congrats again on the results. I wanted to ask maybe a bit of a technical one on the PCIe switch transition to a UALink native switch as we look towards that product’s launch next year. How much of a step change is that in terms of silicon complexity and design? Or is it much more on the, call it, firmware or SDK side and the silicon is largely similar since they’re both based on memory semantics rather than networking semantics? And then sort of related to that, you talked about your customers taking a look at the UALink protocol, even though they’re on PCIe today, how much of a lift is on their side when they’re looking at the scale-up communication kernel and what can you do to sort of derisk that transition for them?

Jitendra Mohan: Yes. It’s a great question. And you are very correct in pointing out that there are similarities between both PCI Express and UALink from a protocol standpoint and also that customers have made good investments into their software stack that is tuned to a particular type of protocol. So let me answer them one by one. So to begin with our Scorpio X family today supports PCI Express and PCI Express like protocol. And when we transition from PCI Express to UALink, it will indeed be a new chip that addresses the future generation of these AI systems. However, when we designed Scorpio X, we took into account future generations of this product where the line rates will go up. So the switching architecture and many of the features that go into the switching product, which are beyond the protocol are already ready for the next generation.

So while it is going to be a new development for us to go to a UALink switch, we will certainly leverage the development that we have done for the current Scorpio X generation very heavily, including all of the software features that are part of our COSMOS software stack that are responsible for optimizing, customizing and delivering a lot of diagnostics and telemetry to our end customers. In terms of the similarities between PCI Express and UALink, they are both load store-based protocols. PCI Express has been around for many, many years. It is a memory semantic-based protocol. So from an XPU perspective, the ASIC can simply say, I want to access this memory location, and it doesn’t matter whether that memory location is in the same GPU or the same XPU or a remote XPU.

This is the beauty of memory semantic-based protocol. And UALink carries forward the same thing. It carries forward the memory semantic-based protocol. It carries forward the lossless nature of the network and the software lift for an end customer is much easier. So we do see UALink as an evolutionary step for our PCI Express customers, as Sanjay mentioned before. At the same time, UALink does a few things that are very much customized for AI scale-up. The data rates are much faster. Obviously, we go from 64 and 128 gig to 200 and then beyond in the future. But more importantly, the protocol was built ground up for AI scale-up. It takes into account the AI workloads, the AI traffic patterns and simultaneously delivers low latency as well as increased throughput.

And most importantly, UALink is also an open standard. So it’s been around for 1 year now, 1 year officially, which in AI terms is probably a decade. And during this time, the IP ecosystem has become mature. The spec is very solid. And a lot of vendors are working on new silicon to deploy UAL-based switches in the 2026 time frame with revenues coming in, in 2027.

Sean O’Loughlin: I really appreciate the color there. And I’ll follow up with a quick clarification on the 20% PCIe Gen 6. I believe that was inclusive of both Scorpio P and Aries or was that an Aries-specific comment?

Jitendra Mohan: That’s inclusive of Scorpio, which is all Gen 6 product — in our Aries Gen 6 products.

Operator: Your next question comes from the line of Suji Desilva with ROTH Capital.

Sujeeva De Silva: Hi Jitendra, Sanjay Mike, congrats on the progress here. I know the optical revenue is down the line here. But just wondering in comparing pain points of bandwidth versus XPU density, which one kind of pushes customers faster to scale up using optical? Or is there a way to kind of handicap one versus the other?

Jitendra Mohan: Yes. I think what our customers have told us, and you can see this in the product announcements that various AI platform providers and hyperscalers have made is they prefer to stick with copper for as long as possible. And the reason for that is multifold. Clearly, copper is so far proven to be more reliable. It’s lower power. It offers better TCO. And so as part of the focus that we have on copper, we’ll continue to push copper for as long as possible. And that is copper is not going away anytime soon. However, as the topologies of scale-up networks evolve, you will end up with a practical limitation of trying to provide megawatts of power into one rack. And so as a result, at some point in time, we will have to disaggregate the rack into multiple racks, which will then be beyond the reach of copper.

So that is what we are planning for. And in the outer years, as Mike mentioned, in the 2028, 2029 time frame, we expect to see these optical deployments from POC and eventually turning into revenues.

Sujeeva De Silva: Okay. Great. So rack to rack. And then just a clarification on the 10 POC customers for PCIe, UALink for scale up. Are any of the customers pursuing anything Ethernet related with you? And are you working on any Ethernet stack efforts in-house yet? Or is it all PCIe to UALink roadmap today?

Sanjay Gajendra: Yes. So again, we can’t comment on what customers are looking at. But let me talk about what we are doing. Like we have highlighted many times, we are heavily engaged right now on scale up. Today, most of the deployments are PCIe like. And these are engagements that obviously will have — will live for multiple generations, and that’s probably something that perhaps is a little underappreciated. We do expect the revenues to go into 2029. And in terms of like other protocols, what I would say is that think about it this way, we believe in open standards. We believe in doing what’s right for the customers. Our Scorpio X-Series is developed today to support PCIe and it can easily upgrade to UALink, especially on the non-protocol-related functions.

So overall, what I would say is that if a time comes when customers require alternate implementations, we are well setup for it because one key thing to highlight is that although there is so much a focus on like the physical layer protocol, PCIe or Ethernet or other things, what we are learning is that the most important or some of the most important functionality is required in the data part, in the management side because these clusters are giant and having a link that is nonperforming or a subsystem on the data part not delivering the right performance could significantly impact the overall performance of the cluster. So to that standpoint, what we are seeing is that there are several things that needs to be done at the upper levels, and those are things that will remain constant for us irrespective of the physical layer that we end up supporting based on market and customer requirements.

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

Sebastien Cyrus Naji: I wanted to ask about the opportunity for Astera in China and in particular, the willingness for Chinese hyperscalers to maybe use more open technologies like PCIe or UALink.

Jitendra Mohan: Yes. So there is a difference in the hyperscaler opportunities in the U.S. relative to the hyperscaler opportunities in China. Because of the constraints that are placed on the availability of IP and technology, we actually see a lot of demand in China for PCI Express-based scale-up. And the reason that has to do with that is the IP availability there is limited in terms of the data rate, 200 gig is not readily available. PCI Express Gen 5 and add-in card formats are most common in China. And in order to build a larger scale-up network so that they can address the same problems that you might be able to solve with an 8 GPU cluster here in the U.S. might require 16 or 24 clusters of GPU to address the same problem.

So when you have more GPUs, our revenues are typically indexed by the number of GPUs. So when you have more GPUs and more accelerators, it is a bigger opportunity for us to sell both our switching solutions as well as solve our retiming solutions from both chip-down opportunities as well as active cable opportunities.

Sebastien Cyrus Naji: Got it. Okay. That’s really helpful. And maybe if I could do just one follow-up. Just I’d love to get your thoughts on NVIDIA’s shift to more of a cable-less design with their Rubin servers or Rubin rack. Does that design shift change Astera’s opportunity with Aries or Taurus at all?

Jitendra Mohan: So as we have said before, the opportunity for us for NVIDIA-based designs is when hyperscaler customers customize their design to deploy in their own infrastructure. That has been true of the Blackwell platform, and we believe that something like this will happen for the Vera Rubin platform as well. The choice of using a cable backplane versus a PC board-based backplane has to do with the number of GPUs that are present in the design. And certainly, we should let NVIDIA explain the rationale from going from one to the other. But the opportunity for Astera comes when hyperscaler customers take the very performant high-performance GPU platform and customize it for their use cases.

Operator: There are no further questions at this time. I would like to turn the call back over to Leslie Green for closing remarks.

Leslie Green: Thank you, everyone, for your participation and questions. And please refer to our Investor Relations website for ongoing information regarding upcoming financial conferences and events. Talk to you soon.

Operator: This concludes today’s conference call. You may now disconnect.