Credo Technology Group Holding LtdCRDONASDAQ
LOADING
|||
Operator: Ladies and gentlemen, thank you for standing by. At this time, all participants are in a listen-only mode. If you have a question, please press star then the number one on your telephone keypad. If you would like to withdraw your question at any time, simply press star 1 again. I would now like to turn the conference over to Dan O’Neil. Please go ahead, sir.
Daniel O'Neil: Good afternoon. Thank you for joining our earnings call for the second quarter of fiscal 2026. Today, I'm joined by Bill Brennan, Credo's Chief Executive Officer; and Dan Fleming, Credo's Chief Financial Officer. During this call, we will make certain forward-looking statements. The forward-looking statements are subject to risks and uncertainties discussed in detail in our documents filed with the SEC, which can be found in the Investor Relations section of the company's website. It's not possible for the company's management to predict all risks nor can the company assess the impact of all factors on its business 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. Given these risks, uncertainties and assumptions, the forward-looking events discussed during this call may not occur, and actual results could differ materially and adversely from those anticipated or implied. The company undertakes no obligation to publicly update forward-looking statements for any reason after the date of this call to conform these statements to actual results or to changes in the company's expectations, 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. These non-GAAP financial measures are provided in addition to and not as a substitute for or superior to financial performance prepared in accordance with U.S. GAAP. A 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 using the Investor Relations portion of our website. I will now turn the call over to our CEO. Bill?
William Brennan: Thanks, Dan, and thank you, everyone, for joining our fiscal '26 second quarter earnings call. I'll walk through our Q2 results, highlight the transformative developments we've announced since our last call and then share our forward outlook. After my remarks, our Chief Financial Officer, Dan Fleming, will provide a detailed financial review and our guidance for the third quarter. In the second quarter, we delivered record revenue of $268 million, representing 20% sequential growth from Q1 and an extraordinary 272% increase year-over-year. Non-GAAP gross margin came in at a robust 67.7%, and we generated approximately $128 million of non-GAAP net income. These are the strongest quarterly results in Credo's history, and they reflect the continued build-out of the world's largest AI training and inference clusters. AI clusters are no longer measured in tens of thousands of GPUs. They're now measured in hundreds of thousands and soon millions. The scale, density and complexity of these systems are pushing every aspect of interconnect. Reliability, power efficiency, signal integrity, latency, reach and total cost of ownership have all become mission-critical. This is the challenge our customers face, and it's where Credo is uniquely positioned to deliver the solutions they need to succeed. Our 3-tiered innovation framework, purpose-built SerDes technology, world-class IC design and a true system-level development approach, all wrapped with our industry-leading pilot debug and telemetry platform has allowed us to forge deep strategic partnerships.
William Brennan: Let me now walk through our business in detail. Starting with our active electrical cables. The AEC product line remains the fastest-growing segment in the company. AEC revenue again grew strongly, driven by rapidly increasing customer diversity. In Q2, 4 hyperscalers each contributed more than 10% of total revenue. The fourth hyperscaler is now in full volume ramp and a fifth started contributing initial revenue. Customer forecasts have strengthened across the board in the past months. AECs have become the de facto standard for inter-rack connectivity and are now displacing optical rack-to-rack connections up to 7 meters. At 100 gig per lane today and 200 gig per lane tomorrow, zero-flap AECs deliver up to 1,000x better reliability than traditional laser-based optical modules while consuming roughly half the power. When you're installing a 100,000 GPU cluster, Link Flaps can delay time to stability and time to revenue. And when you're training a model costing tens of millions of dollars, Link flaps can have a significant impact on overall uptime and productivity. It is this step function improvement in reliability and power efficiency that's driving the expansion of the AEC TAM in 100 gig and now 200 gig per lane generations. And we expect that trend to continue as customers densify racks and push cluster scale to new levels.
William Brennan: Next, our IC business, which includes retimers and optical DSPs, also continued the strong performance. We expect significant optical DSP growth in fiscal '26, driven by 50-gig and 100-gig per lane deployments with longer-term growth driven by our 200-gig per lane solutions. Live demonstrations last quarter of our 200-gig per lane Bluebird optical DSP drew significant interest and extremely positive feedback. Ethernet retimers remain important in both traditional switching fabrics and the fast-growing AI server segment, where features like MACsec encryption, gearbox functionality and rich software programmability are highly valued. Our PCIe retimer and AEC families are also progressing on plan. Customer silicon evaluations have consistently highlighted our best-in-class combination of reach, latency and power efficiency, a rare trifecta enabled by our unique purpose-built SerDes architecture. We remain on track for PCIe design wins in fiscal '26, followed by meaningful production revenue in fiscal '27. Our existing AEC and IC businesses both address multibillion-dollar market opportunities with excellent visibility for continued growth.
William Brennan: But the truly exciting part of this quarter is that we've added 3 entirely new growth pillars, each representing distinct multibillion-dollar market opportunities that significantly expand our total addressable market and extend the reach and depth of our connectivity leadership. The first new growth pillar is Zero Flap optics, the first laser-based optical connectivity family that delivers AEC class network reliability, enabled by a customized optical DSP that is tightly coupled with our pilot software and integrated with a switch level SDK, our Zero Flap optics integrate with our customers' network software. Link Health telemetry data on each optical link enables autonomous detection and mitigation of conditions that cause link flaps before they bring down the cluster. This enables a step function improvement in network reliability. We're currently in live data center trials with our lead partner, and we expect to begin sampling a second U.S. hyperscaler later this fiscal year. Our ZF optics solutions expand our addressable market to any length of connection within the data center. We anticipate initial revenue in fiscal '27 and long term, a market that will be a multibillion-dollar opportunity.
William Brennan: The second new pillar of growth was announced in September. Credo has combined forces with Ottawa-based Hyperlume, a team of experts specializing in high-performance microLED technology. Credo has been investing in micro LED innovation over the past 18 months with the intent of developing a new class of connectivity solutions. Uniting with the Hyperlume team will accelerate our time to market. As a first product, we'll develop and bring to market a pluggable optical solution that utilizes micro LEDs as the light source. Our same 3-tiered innovation playbook will be the catalyst to pioneering this entirely new connectivity category we call active LED cables or ALCs. ALCs will deliver the same reliability and power profile as an AEC, but in a thin gauge cable that can reach up to 30 meters and is ideal for row scale scale-up networks. Customer reaction has been very positive. We plan to sample the first ALC products to lead customers during our fiscal '27 with initial revenue ramping in fiscal '28. We believe the ALC TAM will ultimately be more than double the size of the AEC TAM.
William Brennan: Finally, we announced the third new pillar of long-term revenue growth, OmniConnect gearboxes, a family of products that will enable a disaggregated and optimized approach to XDU connectivity. In November, together with our lead customer, we unveiled the first gearbox that will address that memory wall by redefining memory to compute connectivity, a solution we call Weaver. Today's on-package high-bandwidth memory is capacity and throughput limited as well as expensive and supply chain constrained. Weaver allows designers to move to commodity DDR memory and achieve up to 30x more memory capacity and 8x the bandwidth. The key enabler for the OmniConnect family is Credo's purpose-built 112-gig VSR SerDes that enables a 10x improvement in beachfront I/O density and has a reach of up to 10 inches. The Weaver gearbox from 112-gig BSR to DDR effectively overcomes the physical and logical limitation of current memory to compute connectivity solutions. Our first customer for Weaver announced their plan to deliver an XPU targeted for inference with 2 terabytes of memory capacity, a complete game changer for workloads such as real-time AI video generation and full self-driving, where memory capacity and bandwidth are the primary gating factors. Industry forecasters project the memory to compute connectivity market to be a multibillion-dollar market by the end of the decade. We anticipate initial revenue in our fiscal '28 with significant scaling thereafter. The next OmniConnect gearboxes to be introduced will provide a future-enabled path to scale out, scale up and near package optics connectivity with XPUs.
William Brennan: In summary, we now have 5 distinct high-growth connectivity pillars: AECs, IC solutions, including retimers and optical DSPs, zero-flap optics, ALCs and OmniConnect Gearbox solutions. Together, they'll give Credo a combined total market opportunity that we believe will exceed $10 billion in the coming years, more than triple where we stood just 18 months ago. Looking forward, we couldn't be more excited about the combination of continued growth in our core AEC and IC businesses, plus the upcoming ramps of ZeroFlap optics, ALCs and OmniConnect gearboxes. We believe this combination gives us a strong outlook into continued revenue growth through fiscal '26 and well beyond. Team Credo continues to execute at an elite level, delivering record results quarter after quarter while simultaneously pioneering and launching new multibillion-dollar product categories. I'm proud of our world-class operational excellence and innovation. With that, I'll turn the call over to Dan Fleming for a detailed financial review and our Q3 guidance.
Daniel Fleming: Thank you, Bill, and good afternoon. I will first review our Q2 results and then discuss our outlook for Q3 of fiscal year '26. In Q2, we reported revenue of $268 million, up 20% sequentially and up 272% year-over-year and well above the high end of our guidance range. Our product business generated $261.3 million of revenue in Q2, up 20% sequentially and up 278% year-over-year. Notably, our AEC product line again grew healthy double digits sequentially to achieve new record revenue levels once again based on substantial year-over-year growth across 4 domestic hyperscale customers. Our top 4 end customers were each greater than 10% of revenue in Q2. As a reminder, customer mix will vary from quarter-to-quarter, and we continue to make progress in diversifying our customer base. We continue to expect that 3 to 4 customers will be greater than 10% of revenue in the coming quarters and fiscal year as hyperscale customers continue to ramp to more significant volumes and as we expect to begin to ramp an additional hyperscale customer in the coming quarters.
Daniel Fleming: Our team delivered Q2 non-GAAP gross margin of 67.7%, above the high end of our guidance range and up 11 basis points sequentially. Our product non-GAAP gross margin was 66.8% in the quarter, up 18 basis points sequentially and up 469 basis points year-over-year. Total non-GAAP operating expenses in the second quarter were $57.3 million, slightly above the midpoint of our guidance range and up 5% sequentially. Our non-GAAP operating income was $124.1 million in Q2 compared to non-GAAP operating income of $96.2 million in Q1, up demonstrably due to the leverage attained by achieving more than 20% sequential top line growth, while OpEx growth was in the mid-single digits. Our non-GAAP operating margin was 46.3% in the quarter compared to a non-GAAP operating margin of 43.1% in the prior quarter, a sequential increase of 319 basis points. Our bottom line once again demonstrated the substantial leverage we are delivering in the business. Our non-GAAP net income was $127.8 million in the quarter, a record high and a 30% sequential increase compared to non-GAAP net income of $98.3 million in Q1. And our non-GAAP net margin was 47.7% in the quarter as we drove significant leverage in the business.
Daniel Fleming: Cash flow from operations in the second quarter was $61.7 million, up $7.5 million sequentially. CapEx was $23.2 million in the quarter, driven largely by purchases of production MACsec. And free cash flow was $38.5 million, down from $51.3 million from the first quarter due to higher CapEx investments. We ended the quarter with cash and equivalents of $813.6 million, an increase of $333.9 million from the first quarter, up largely from the proceeds of our ATM offering, which began in October. We remain well capitalized to continue investing in our growth opportunities while maintaining a substantial cash buffer. Our Q2 ending inventory was $150.2 million, up $33.5 million sequentially.
Daniel Fleming: Now turning to our guidance. We currently expect revenue in Q3 of fiscal '26 to be between $335 million and $345 million, up 27% sequentially at the midpoint. We expect Q3 non-GAAP gross margin to be within a range of 64% to 66%. We expect Q3 non-GAAP operating expenses to be between $68 million and $72 million. We expect Q3 diluted weighted average share count to be approximately 194 million shares. These expectations are based on the current tariff regime, which remains fluid. As we look toward the end of fiscal year '26 and into fiscal '27, we expect sequential revenue growth in the mid-single digits, leading to more than 170% year-over-year growth in the current fiscal year. We expect each of our top 4 customers from Q2 to grow significantly year-over-year in fiscal year '26. We also expect revenue diversification to strengthen further with our fourth customer surpassing the 10% revenue threshold for this fiscal year. We expect non-GAAP operating expenses to increase year-over-year by approximately 50% in fiscal year '26. As a result, we expect our non-GAAP net margin to be approximately 45% for fiscal year '26. This should translate to net income more than quadrupling year-over-year. And with that, I will open it up for questions.
Operator: [Operator Instructions] And your first question comes from the line of Tom O'Malley with Barclays.
Thomas O'Malley: So you're seeing an expansion of the AEC market pretty rapidly here with the fourth and then now soon to be the fifth 10% customer kind of rolling on. I found it interesting amidst but was clearly like a really strong ramp in the AEC market, you're able to say that when you look at the ALC market, you could see that being double the AEC TAM. So talking big numbers, maybe you could spend a little time talking about whether that's unit-driven, whether that's ASP-driven, just surprised given the size -- given how well AECs have done.
William Brennan: So I think it's a combination of both the quantity as well as ASPs. Now when we think about ALCs, it's really an ideal product from the standpoint of delivering the same reliability as AECs, which is really the most critical factor right now with host to [indiscernible] or GPU to switch connections. Also, from a power efficiency standpoint, it's in the same class as AECs. I would say from a system cost perspective, again, in that same class. What it delivers is a thinner wire gauge and longer length. And as we see the scale-up networks really going from intra-rack to row scale, we're talking about a tremendous increase in the number of connections. We estimate that it could be up to 10x that of the number of scale-out connections. And so yes, we are really, really bullish and we're strong believers in microLED as a technology that's really going to be a game changer for us as well as our customers.
Thomas O'Malley: And then just on the timing and the scale of the other customers ramping on, maybe you could give us what the percentages of the top 4 were this quarter? And then any commentary on how large those other customers rolling on will be just to give us a feel for how one is kind of handing off the next over the next couple of quarters would be super helpful.
Daniel Fleming: Sure, Tom. So for Q2, as we mentioned in my prepared remarks, we had 4 10% customers. The largest was 42% of revenue, and that was the customer that we've, in the past, said we expect to be the largest customer this fiscal year. The second largest was 24%, which have to be our first hyperscaler to ramp a few years back. Third largest was 16%, which was our largest customer in Q1. And the fourth was 11%, which is our newest hyperscaler that we've discussed in the past. So when you kind of plot that all out, what you see and what we've been fairly consistent in saying is that the ramp at any given single hyperscaler is never really linear. So we're seeing that with our largest customer from last quarter, taking a bit of a pause this quarter. But our largest customer for this fiscal year, which had been down for the last few quarters is back up again. So there is definitely a give and take that we are managing through. And again, we have 12-month visibility, in some cases, even greater visibility with our customers in order to be able to manage that through the course of time. Hard to predict exactly how things are going to fall quarter-to-quarter longer term. But hopefully, that gives you kind of a flavor as to what we're seeing kind of on the ground right now today.
Operator: And your next question comes from the line of Tore Svanberg with Stifel.
Tore Svanberg: Yes. Congratulations on the solid results. Bill, I had a question on your sort of 3 new product lines here, obviously, all in the optical domain. First of all, could you maybe elaborate a little bit on how much you're focusing on the system-level products because obviously, in copper, yes, you have system-level products, but the 3 areas in optical, how much of those should we assume or systems? And then I have a follow-up.
William Brennan: Yes. So we've been very consistent in talking about our intent to continue to expand our portfolio at the system level. And I would say that, yes, ALCs as well as ZF optics, those are both optical solutions. But the OmniConnect family will be initially copper-based and then longer term, we'll offer near package optics options with that. And so I think that from the standpoint of delivering value to our customers, we are very much focused on delivering non-commodity [ non-IEEE ] standard well beyond what that standard calls for. And I think ZF optics is a great example of that in a sense that going back 18 to 20 months ago, we really heard strong feedback from multiple customers that reliability was the top priority as people became more familiar with the issues that they were seeing with building out AI clusters. And so with one of the customers that we were on stage at OCP, we've got to work thinking about how do we integrate higher level within the stock, how do we integrate within the network software for that customer within their AI cluster. And so the concept is how do we how do we provide more visibility, more telemetry, how do we make that information available and actionable at a network level. When we think about the opportunity that we're creating for our customer here is it's really creating almost like a check engine light being able to set a threshold on links, all of the links within a cluster enable it to sense real time when those -- any of those links are degrading, being able to set a threshold and once that signal integrity drops below that threshold, being able to action it by in an orderly fashion, taking that GPU out of the cluster before you see a link flap that could potentially take down -- take that cluster entirely. And so that type of value add is completely innovative, and it's defining a new class of optical connectivity. And it's very, very targeted for reliability. And that reliability is really with traditional laser-based optical transceivers. So that's been a big investment. It had -- we delivered a custom optical DSP. We have -- the whole pilot telemetry platform is something that we've been working on for many years, and there's a lot of special development work that was done to tightly couple that optical DSP. And then integrating within a switch level SDK, this is -- these are all things at a system level that needed to be done to be able to bring this product to market. When we think about ALCs, I've talked about that. That's a game changer. It's basically changing the light source to at a ground level, deliver better reliability. And when we think about our first gearbox in the Omnionnect family, it is a copper solution, but it's redefining how memory to compute or memory to XPU connectivity is done. And when we think about the alternative being HBM, all in package, all driving extremely challenging heat dissipation environments, that translates directly to reliability as well. So being able to move the memory up to 10 inches away, not only does it lift the logical and physical limitation that you've got by being in a package with a beachfront I/O density that's not as dense. It really eliminates the reliability issue that you've got with the heat that's generated by putting so much memory in a single package with the XPU. So it's really a combination of both optical and copper. But again, we're agnostic to the medium. We're agnostic to exactly how these solutions are put together, whether they're copper or fiber. It's ultimately delivering a solution to the customers that is just much better than it's in the market today.
Tore Svanberg: That's very helpful. And as my follow-up, and I'm very interested in the 112 gig VSR SerDes technology. I think you mentioned initially, it's going to be copper, but eventually, you're going to have an optical solution as well. I'm just curious, would you have to develop some silicon photonics technology there? Or would that still sort of be a solution in the pure silicon domain?
William Brennan: Yes. So regarding the SerDes that was developed specifically for this application. I love the arguments in the market about who's got the best SerDes and a lot of times in the market, people like to think the longest REIT SerDes is obviously the best. We look at it differently. We really look at it from an application-specific perspective. And so when we think about the idea of giving an XPU customer a piece of IP to integrate. We think about how do we achieve the smallest footprint, the lowest power. And ultimately, with this VSR SerDes that we developed on a max reticle die, we can fit 1,200 of these SerDes, creating 120 terabits per second of potential bandwidth. That's unprecedented. That has yet to be achieved, especially when you think about the reach being 10 inches. Other competitive solutions, the reach would be an inch or less. And so basically, forcing there to be a die-to-die connectivity versus moving a chip outside the package. And so it's a real breakthrough. I would argue that this SerDes is best-in-class, and it's enabling an entire family of solutions that range from I/O solution for memory, but also from a scale-out or scale-up perspective and long term with near package optics. And the near package optics will leverage the work that we're doing in microLED. So first step will be ALCs and then we'll roll that right into some additional game-changing applications like near package optics.
Operator: Your next question comes from the line of Vivek Arya with Bank of America.
Vivek Arya: Bill, which applications are your 4 customers using AECs for today? And which applications are they not yet using AECs for? And as they look into next year, there's a lot of talk of co-packaged optics coming into the mix. And I was hoping you could give us a sense for what impact that might have on the current or future AEC usage by your customers?
William Brennan: So the different applications that we've talked about in the past have been, number one, front-end network connections, which, of course, that was the first application to ramp with our first customer years ago. We've also talked about the scale-out opportunity as part of the back-end network of AI clusters. The third application we've talked about is switch racks. And that would be as opposed to buying a chassis filled with switches, you would stack those vertically in a rack. And that backplane connection within a chassis would become a short cable connection within the rack. Those are the first 3 applications we've talked about, and we are in production with all 3 of those with different customers. We're definitely not fully penetrated with all of our customers, but we are in production with all 3 of those applications. I would say the one remaining application that will be high volume is with the scale-up network as that network goes rack scale and then ultimately goes row scale, depending on the density and the number of racks that are being deployed. Now as it relates to co-packaged optics, this is a -- and this has been a long conversation really for the 12 years I've been involved in this industry, there's been a conversation about moving to a co-package. It's changed names or changed acronyms over time. But I think that before it takes off in a really big way, there's got to be answers to the pitfalls that are very, very well known, number one, related to reliability, serviceability, maintenance, cost, of course. And we think that there are solutions that are going to be more optimized from a power standpoint, more optimized from a reliability standpoint. And we're focused on bringing those solutions to market really maybe even within the same time frame as people talk about co-packaged optics. So there's been lots of demonstrations, but we don't see a lot of customers moving forward in a big way that would have an impact on us. I think it's safe to say that from that [indiscernible] connection perspective, reliability is top, absolutely top of the list on priorities. So I think we're -- that's obviously the high-volume part of the market, and we don't see that moving to CPO anytime soon.
Vivek Arya: And for my follow-up, I'm curious, how does your ASP lift from the 100 to 200 gig per lane compared to the lift that you have seen and are still seeing in the 50 to 100 gig transition? And does the competitive landscape change as you start moving to 200 gig, does that new application create a bigger opportunity for some of your competitors?
William Brennan: I think the question is a bit more complex than just thinking about moving from one lane speed to a faster lane speed. Our ASPs are highly dependent on the number of connectors in the SKU, the devices that we're using within those connectors and the length of the connections. And so it is safe to say that we have had some uplift going from 50 to 100 gig. And I believe there's going to be an uplift going from 100 to 200, but I can't kind of categorize it simply. But I do think that naturally, there will be an advantage as we move to faster lane speeds.
Operator: Your next question comes from the line of Quinn Bolton with Needham & Company.
Quinn Bolton: Congratulations on the nice results and outlook. I guess, Bill, I had a question just on the supply side of things. I think looking at the 10-K you guys put out, you list Bizlink as sort of your sole provider of AEC cable manufacturing. And obviously, the forecast here is getting pretty big pretty quickly. How are you feeling about AEC supply? Are there any constraints you're working through? Anything to call out on the supply front?
William Brennan: I think we're entering a period of time where we will be talking about supply constraints more frequently. Now as it relates to our AEC volumes, I don't see a concern related to the quantity that we can produce with our partners. I think we've proven that over the last year that we can ramp significantly in a very short period of time. So this is a completely different equation than thinking about a wafer foundry. And so I do -- over the last 2 to 3 months, we've really seen an increase in the conversations around what is the total potential wafer demand in the market for next year and the year beyond and the year beyond that. And I do think that if we look at it from a demand standpoint, the market in general is going to kind of quickly get to the point where we're kind of almost self-regulated by foundry capacity. And so I don't want to be too controversial talking about it. But I think from an advanced node perspective, I think it's going to be a more frequent conversation about capacity constraints at a wafer level. Now as it relates to Credo, I think that a couple of things give us an advantage. We've talked about -- if you remember, the [NYS-1] process strategy, which is we've always had a strategy to be in older geometry processes than our competition if we can deliver best-in-class power, best-in-class die sizes and best-in-class performance. And so we've done that successfully over time. So right now, 12-nanometer is our workhorse. And that's not as tight as, say, 3-nanometer or 5-nanometer. And so we feel pretty good about where we are just related to a situation where there's an increasing demand across the board for wafer volume. But I also will say that our foundry partners are very, very well in tune that the connectivity solutions, the relatively small die size connectivity solutions are critical for shipping GPUs. You can't really ship GPUs without connectivity solutions. And so I think for a couple of reasons, I think we're going to be able to manage our way through that. Now as far as our partners within the AEC product, yes, I mean, this is why being vertically integrated like we are gives us an advantage because I've got a system-level supply chain team that is making sure that all of the partners that supply components that go into the AECs that they've got as much visibility and much commitment as they need to go build out what we see needed in the future. And so I think -- yes, great question. I think it's going to be really topical as we go through calendar '26.
Quinn Bolton: And for my follow-up, Phil, just looking at your top 2 customers, I believe they are still at either 50 gig or 25 gig per lane. But just wondering if you could confirm that at least -- it still feels like the vast majority of your AEC business is at 50 gig per lane, and it sounds like there's an opportunity that as those customers ultimately transition to 100 gig per lane, there's probably some ASP lift you would see with that transition. I won't ask you to comment on specifically when they may transition. But just I don't think it's happened yet. I was hoping you could confirm that the top 2 guys are still either 25 or 50 gig per lane.
William Brennan: I'll confirm that we're in production with 25 gig per lane, 50 gig per lane as well as 100 gig per lane. And probably one of the fastest-growing parts of our business has been the shipments of 100-gig per lane solutions. And so the -- yes, you're right that it's not so simple for me to categorize it. I do think that over time, you'll see the entire customer base moving to 100 gig per lane and then making a move towards 200 gig per lane over the next 2 to 3 years. So I think generally, the trend that you're referring to, I think that we can say generally that it's accurate that we'll see uplift as the market migrates towards faster speeds.
Operator: Your next question comes from the line of Suji Desilva with ROTH Capital Partners.
Sujeeva De Silva: Congrats on the progress here. Maybe you could talk about your customers now when your first IPO, you had a handful of customers, now you're gaining with 5 customers here potentially. So the hyperscales that don't kind of use you at a scaled level 10%-plus type levels. I know you're in all of some level. What is the difference between the customers you haven't penetrated yet and the ones you're getting to 10-plus percent in various quarters?
William Brennan: We look at every one of the hyperscalers as almost a different market. They're all trying to maybe solve the same problem generally, but there's so many different ways to solve those problems. And so network architecture is very much specific to each one of our customers. And I can tell you that the first program that we engage with, typically, that's led to many other programs within the same customer. And I think that as we think about the customer base right now, we think of 6 hyperscalers in the U.S., and we think of a handful in Asia. But I think we're going to be talking more and more about that next tier of customer because I think it's going to be possible to drive pretty big numbers with that next tier of data centers. So generally, I think that as I think about our business, this has been a very significant quarter for us in a sense that we're showing a path to a much more diversified business, not just from a customer base, but from a protocol perspective with PCIe as well as from a just the overall TAM expanding by addressing a much broader market with our system-level solutions. And so if you think about it, you -- we talked about OmniConnect first, and we're really addressing die-to-die or chip-to-chip connections at a system level. And when you think about retimers, those connections can probably be up to 0.5 to 1 meter and typically on an appliance card or a switch card. Then we think about AECs up to 7 meters. We think about ALCs up to 30 meters, and we think about ZF optics up to a kilometer or beyond, whatever the maximum length is within the data center. And so in the past 90 days, this has been probably the most transformative from a news standpoint. We've been working on these things for 18 months or so. But now being able to talk about it, I think it shows a clear path to a much more diversified company long term as we think about moving the company from that $1 billion threshold in revenue annually to $5 billion and beyond over the next several years.
Sujeeva De Silva: Okay. Appreciate how you classify the products there for us. And then my other question is on manufacturing. I know with AECs, you're doing the cable manufacturing in-house selling the entire cable solution. Is it the same strategy for ALC? Will it be modules versus cables, outsourcing? Any color there would be helpful.
William Brennan: The strategy for ALCs will be very similar to the strategy for AECs. We intend to own the entire stack, take accountability for the entire system solution. And so we'll see as that develops that will be a very similar model to what we've got today with AECs.
Operator: Your next question comes from the line of Vijay Rakesh with Mizuho.
Vijay Rakesh: Bill and Dan, congratulations here. Just a quick question on the scale up. Do you see those revenues start to ramp in second half calendar '26? And I think you mentioned active AD cables ALC in calendar '27. Is that the way to look at it?
William Brennan: For scale-up specifically, we're going to enter that market with PCIe Gen 6 solutions. Now of course, we'll entertain Gen 5 in the interim, but really, the product is targeted for Gen 6. We'll bring both retimers and AECs to market simultaneously, and we see a big opportunity for both. If I think long term, that will migrate to faster lane speeds. And PCIe Gen 7 is absolutely a conversation. We'll deliver products to the market to meet that protocol, that Gen 7, 128 gigabits per second per lane protocol. But also, there's a huge conversation about 200 gig per lane. And over time, that -- the protocol war will settle down and ultimately 1 or 2 or even 3 will be in production. We feel good about that because all of the things being discussed right now at 200 gig per lane use the same IEEE SerDes. And so we've talked about being protocol agnostic for the 200 gig per lane generation. And I'll say that all of the products that we've talked about, we're going to be delivering solutions for scale-up with all of those products, including AECs and ALCs. And if the market moves towards longer connections, we will surely move to ZF optics as well, but that's yet to be determined.
Vijay Rakesh: Got it. And then longer term, as you -- you're obviously seeing a pretty strong AC ramp. How should we look at the gross margin profile as optical DSPs are starting to ramp as well? Just longer term, how to look at gross margins?
Daniel Fleming: Yes. We've been very consistent in saying our long-term expectation for gross margins is in the 63% to 65% range. So we are clearly at a point in time right now where we're a bit above that, but we don't expect that to be the case longer term. If you look at the more medium term, probably we guided to 65% at the midpoint. So we'll be kind of near that high end of that long-term expectation. But just longer term, I expect that to settle down into an area that historically, companies like us have been in.
Operator: Your next question comes from the line of Sean O'Loughlin with TD Cowen.
Sean O'Loughlin: And obviously, congrats on the great results. I had a question about the ALC technology and your relationship with Hyperlume, specifically, I think, Bill, you just alluded to having been working on this for the last 18 months or so. And I would assume that, that's kind of how the acquisition came together. And then, so maybe just talk about that. And then if we're sitting here in 2027, 2028 and ALCs haven't ramped, has it been -- is this because of a technology problem that you guys still need to figure out? Or is it because of a business problem that maybe the market didn't go the direction that you thought it would?
William Brennan: I think it goes back a couple of years ago that we first became very interested in microLED technology as an option to bring really differentiating products to market first with ALCs and then the second step is to answer some of the pitfalls of near package optics. The first couple of companies that we engaged with were independent companies. And ultimately, we made the decision earlier this year to basically bring the technology in-house, bring the team in-house so that we could have a very predicted outcome on time to market. And so at this point, what we've learned over the last 18 months to 2 years is significant. The team that we've combined forces with Hyperlume, we feel like right at this point, it's more of an execution play. It's not really a -- so the technology, I think, is well on the path to being developed. Now it's just an execution play, bringing it to market. And so I feel confident about bringing a product to market in our fiscal '27 and ramping -- first initial ramps in fiscal '28. From the standpoint of our confidence on this as far as this being a product that's going to resonate with our customers, I think that we've got several years under our belts bringing new products and categories of products to market. The initial conversations with customers, I don't see any major barriers to overcome from the standpoint of customers accepting the product. These are going to be bookended solutions. So they're going to be delivered in the form of a cable. And the bottom line is when we deliver on the promise of reliability, power efficiency and system cost that's equal to AECs, but you get a thinner wire and you get longer length. I mean it's really the perfect type of product. And so I really, again, view this as more of an execution challenge. And I feel very good about the team we've got in place, the additions that we've got planned over the next 12 months to make sure we get this right.
Sean O'Loughlin: Yes. And then a quick follow-up on ZF Optics. Understanding that they're a system-level solution, is there a -- is there a line rate or a lane rate that the ZF optics platform is targeting to intersect with? Or is it relatively agnostic, can be leveraged at 50 gig, 100 gig, 200 gig per lane type of applications?
William Brennan: It can be leveraged across the board. The first product that we'll go to production with is 100 gig per lane. It will be 800 gig ports that we're addressing. But there's definitely a path to 1.6T. But if we had a customer that wants to bring a product to market that was 50 gig per lane, there's no issue with that.
Operator: Your next question comes from the line of Christopher Rolland with Susquehanna.
Christopher Rolland: Congrats on the pretty amazing results here. And yes, these are probably going to be for Dan. So you guys, in a very good way, kind of blew up my old model. So I think we've talked -- you guys talked about more than 170% year-over-year growth. Perhaps we can put some brackets or talk about next year obviously, we'll have a deceleration. But do you think that we could grow at a rate similar to the Street where it was? Or do you think that needs to come down a little bit just given the better results for this fiscal year? Just any sort of commentary as to the growth rate for next year would be phenomenal.
Daniel Fleming: Yes. We -- so in my prepared remarks, I specifically said mid-single digits revenue growth sequentially through fiscal '27. So that's the expectation that we've set. We set that expectation probably 3 or 4 quarters ago as well if you go back and look at things, and we've outperformed that. So sure, things could change, but we're not setting any expectation that's different from...
Christopher Rolland: Okay. Okay. I guess I don't know, does that -- were we at the higher end of mid-single digits? And do we move lower just given the better results? And then my second question, Dan, is around OpEx. So a pretty big guided bump for next quarter. Maybe talk about that. Is that just a bunch of engineers coming on? Is that more onetime? And then do we take that bump and then just forecast it across every quarter moving forward? How should we think about that? And/or like just what do you think OpEx might grow at a percentage of revenue, if that's easier to discuss, however you would frame it?
Daniel Fleming: Yes. So -- so let me just first state that we're continuing to manage our operating expenses in order to support the revenue growth that we foresee. And as you know, all of these projects, all of these pillars of growth that Bill has talked about, there's a long gestation period for all of them, and you need engineers, engineering talent to be able to execute on that. So we feel like we're in a great position to do so. Our Q3 OpEx guide was up more than it has been in the last few quarters, up 22% at the midpoint. But as you dig into our Q, you'll see our R&D spend was sequentially down in Q2. So in one way, we're kind of making up for that small decline. So long story short, guiding to $70 million at the midpoint, that does set a new base. There's additional project-related spend within that plus additional hiring. We've brought the Hyperlume team on board. There's a lot of different factors that go into that number. But if you plug that number into your model and have a small sequential increase in Q4, you'll kind of end up at a 50% year-over-year OpEx growth from fiscal '25 to fiscal '26, which is -- was also in my prepared remarks. So hopefully, that's helpful.
Operator: Your next question comes from the line of Karl Ackerman with BNP Paribas.
Karl Ackerman: I have 2 questions, if I may. First, could you speak to why you are now licensing your active electrical cable IP to third parties and how this agreement reflects your perceived competitive moat in go-to-market for AECs?
William Brennan: To give background on the case that we filed with the [ITC] this is a market, the AEC market that Credo pioneered over the last 7 or 8 years. We spent tens of millions of dollars establishing the innovations required to build these products. And it's great to see that the product category is -- has realized what we felt would be a multibillion-dollar market potential. And so along the way, we've been pretty communicative with others that are in the market or showing intent to enter the market that we've got IP, and we're intending to make sure that there's respect for that IP. And so we got to the point where we felt it necessary to file with the ITC because we weren't getting great respect or great acknowledgment of the fact that the path was pretty well defined by Credo and our engineering team. And so I think we feel really good with where we are. We never thought about this market as being a market that would take off if there was a single supplier. And so our customers all want multiple suppliers. Ultimately, we've landed in a good spot with 3 of the conversations that we've had thus far. We've got a couple more in flight or maybe even more than a couple more in flight. This doesn't change anything competitively for us. We've always thought about competition as a challenge of moving more quickly and in a way that delivers what our customers want. And so it's a function of delivering what they want first having it be qualified first, ramping first, delivering flawlessly as they ramp and as they are in high-volume production. That's really our focus as a team, both at an engineering level as well as an operations level. And so I would say that we're satisfied with where we are competitively, and we're satisfied with where we are with the results thus far that we've seen as we've protected our IP.
Karl Ackerman: Got it. That's helpful, Bill. Maybe a question for Dan. You know that you've got 12 months of visibility with several hyperscale customers. I guess it begs the question, what was the largest delta in your upwardly revised outlook versus your prior outlook of 960 and change. I guess how much of it was simply the ramp of your -- the fifth hyperscale customer versus just higher order rates of existing programs or even new AEC applications cross scale out and disaggregated chassis at some of these other customers?
Daniel Fleming: What I'd say this is more of a general comment, and this is true of the last 3 to 4 quarters. We've seen continuing strengthening of our forecast throughout the quarters as they proceeded, which is how we've gotten into this particular rhythm that we're in right now. So it's not specific to a customer. It's perhaps more of an industry trend, and we've benefited from that. So that's what we've seen.
Operator: And your next question comes from the line of Joseph Cardoso with JPMorgan.
Joseph Cardoso: Maybe for my first one, I just wanted to touch again on the entry into the optical transceiver market beyond just DSPs. This is obviously a very large and expanding TAM, but perhaps one where current industry margins are somewhat below your long-term targets. So maybe can you just take a second and just talk about how you're thinking about the margin opportunity here? And also curious whether you're focused on just selling the full modules or if there's an opportunity to drive attach of the DSPs, pilot software, et cetera, and sell those building blocks of the zero flap solution to potential customers? And then I have a follow-up.
William Brennan: Yes. We're absolutely thinking about this product in a similar way that we think about the other system-level products. We're going to take full ownership and accountability for the entire system-level solution. So we're not necessarily competing in the current commodity market. If a customer is looking for that step function in reliability that you can get by going up the stack with the solution we're bringing to market. I mean, absolutely. But it's not a conversation about what is the price of a transceiver in the market. This is really a system-level solution. And so it's a combination of all things, DSP, it's a combination of the software that we're bringing. And it's really most importantly is the tight interaction that we've got with our customers. And so I think we've been really clear, and I think Dan specifically has been clear in stating that we don't see any change to our long-term gross margin model.
Joseph Cardoso: No. Got it. Very clear, Bill. And then maybe just a quick clarification on the fifth customer in the quarter. And maybe this is anecdotal, but it sounds at least like they're tracking ahead of expectations relative to last quarter with some initial revenue this quarter, but just wanted to clarify if that was the case, and as we think about that customer tracking into the back half of this year, is there any opportunity for this customer to be 10% at least on a quarterly basis now in the back half? Or have the denominator just kind of outpaced that opportunity there?
William Brennan: Things take time. And I think if you look at the base that we're talking about, that 10% number has changed pretty drastically from a couple of years ago to today. Things take time to ramp. And so when we think about customer # 5, we think about initial revenue this year. There's not a chance that, that will be a 10% customer this fiscal year. And I do think that the customer has the potential of being a 10% customer in -- especially if we think about it at a quarterly level first and then an annual level secondarily. No question that they've got the type of size to be able to drive that type of number, but it's going to take time.
Operator: And your last question comes from Sebastien Naji with William Blair.
Sebastien Cyrus Naji: Congrats on a nice set of results. I'll ask both my questions together in the interest of time. The first one is Credo has an advantage in that it's agnostic to the underlying compute, could be merchant GPU racks, ASIC racks, even CPU servers. So could you maybe talk a little bit about how much of your business is tied to ASIC versus merchant silicon deployments today and where that share might go over the next year? And then my second question, I just wanted to ask about the timing of purchases from your customers. How aligned are AEC purchases with the GPU or ASIC purchases? And do customers typically purchase ahead or one after the other? And is there any risk of inventory build if they're purchasing ahead of GPU orders?
William Brennan: We're not really able to break out the percentage of GPUs that are, say, internally developed or commercially available in the market. We don't actually track it that way. I can say we're -- the first comment that you made is absolutely accurate that we're agnostic to the type of GPU that's being deployed. I will say that from a deployment standpoint, the second question you asked, the -- my belief is that everything is ordered in a way that would have all of the necessary components being delivered at the same time. It's a very, very complex supply chain that our customers are dealing with. But I think they're ordering the connectivity solutions right along with the GPUs. So we have pretty good visibility within the supply chain from delivering our products all the way to having those products deployed. And so we get a good sense of the type of inventory that exists within the partners that each one of our customers use to stage inventory. And we feel pretty good about -- we feel pretty good that there's not really a whole lot of product that's sitting in inventory right now.
Operator: And with no further questions, Mr. Brennan, I turn the call back over to you for closing remarks.
William Brennan: Yes. Thank you. I really appreciate the strong interest in Credo. We'll talk to you all soon. We're a little bit off schedule, so the call might be a bit rush and a bit delayed. But again, really appreciate it. Thank you very much.
Operator: This concludes today's conference call. You may now disconnect.