Q3 2025 Earnings Call Transcript

Ladies and gentlemen, thank you for standing by. [Operator Instructions] And I would now like to turn the conference over to Dan O'Neil. Please go ahead, sir.

Good afternoon. Thank you for joining our earnings call for the third quarter of fiscal 2025. Today, I am joined by Bill Brennan, Credo's Chief Executive Officer; and Dan Fleming, our Chief Financial Officer. During this call, we will make certain forward-looking statements. These 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 is not possible for the company's management to predict all risks nor can the company assess the impact of all factors on its business of 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.

With that, I will now turn the call over to our CEO. Bill?

Thanks, Dan, and thank you for joining our earnings for the third quarter of fiscal '25. I'll start with an overview of our third quarter results and then discuss our outlook. Following my comments, our CFO, Dan Flemming, will provide Q3 financial details and our guidance for the fiscal fourth quarter. For the third quarter, Credo reported revenue of $135 million, up 87% sequentially and up 154% year-over-year. Credo's non-GAAP gross margin was 63.8%.

Credo achieved record revenue in Q3 as we saw the expected inflection point in our business. This ramp was led by our largest hyperscale customer as they scale production of AI platforms. Additionally, we received solidified forecasts and saw increased design activity for our products with additional hyperscalers and other customers, as the performance, reliability and power benefits of our connectivity solutions have become increasingly clear throughout the industry. In a world where data drives everything, the demand for faster, more reliable and energy-efficient connectivity continues to expand rapidly.

Key to Credo's competitive advantage is our multi-tiered innovation, which enables us to deliver a broad set of optimized solutions that are tuned at every level. The first tier of innovation is in our SerDes technology. Our SerDes technology is purpose-built to tackle the toughest bandwidth challenges, balancing speeds up to 200-gig per lane with exceptional performance and power efficiency. By leveraging advanced signal processing and a programmable design, we've created a flexible architecture that adapts to the unique needs of AI workloads, whether it's long reach data center links or ultrashort connections in dense compute environments.

The second tier of innovation is in our integrated circuit design, our IC designs, including retimers, DSPs and chiplets are engineered to deliver the best combination of performance, power and cost for a given application. Our LRO DSP is a great example of innovation on the customers' behalf to deliver a compelling IC solution optimized for power efficiency and optical links. The third tier of innovation is our system-level approach. We don't stop at chips.

Our best example of system-level innovation is Credo pioneering the active electrical cable market. By taking accountability for the system-level solution, we raised the bar to deliver end-to-end connectivity solutions that go beyond industry standards to deliver unique functionality and best-in-class reliability. We see this system-level approach, creating a larger opportunity for Credo with the emergence of AI clusters due to the intense demand for reliability and power efficiency. Finally, wrapped around each tier of innovation is our development and diagnostics software and firmware platform to deliver predictive signal integrity, link optimization and tuning.

From the SerDes to the system level, this software platform helps our customers navigate system development to achieve best performance, yields and reliability. As we look forward, we'll expand our solutions to the PCIe protocol with the same ingenuity that creates our differentiated Ethernet solutions. With the introduction of a full suite of PCIe products on the near-term horizon, Credo will soon be addressing a larger connectivity opportunity with AI scale out and scale up networks, substantially expanding our overall TAM. Now I'll discuss our business in more detail.

Regarding our AEC product line. As expected, our revenue served in the third quarter, driven by our largest hyperscale customer. Compared to alternatives, the benefits of AECs have become clearer. More than ever, data centers are highly focused on back-end network reliability.

With billions of hours operating in the field, AECs have become the de facto standard for intra-rack connections for NIC to TOR and switch to switch applications. However, we are now seeing a new expansion of AEC usage. Our ZeroFlap AECs deliver more than 100x better reliability than laser-based optical solutions. And as a result, we're seeing AECs replacing optics for rack to rack solutions for lengths up to 7 years.

We continue to make significant progress with additional hyperscalers for our Ethernet AEC solutions. We've achieved volume production with 3 hyperscalers, and we're in qualification with 2 additional hyperscalers, affecting production in fiscal '26. With broad traction, we feel confident we'll continue to see increasing diversification of our revenue base across more customers in the coming quarters and years. Additionally, Credo continues to make progress with our PCIe AEC solutions, our Gen6 64 gig PAM-4 AECs will deliver the same compelling benefits for AI scale-up networks as deployments move to rack scale architectures.

Credo will demonstrate our PCIe AECs at NVIDIA's GTC show later this month. We expect customer design engagements and qualifications for our PCIe AECs in the upcoming quarters with a significant revenue opportunity in the upcoming years. For our AEC business in total, we expect continued revenue growth based on customer forecasts, new qualifications, new design engagements and TAM expansion. Now I'll turn to our optical business.

Our optical DSP business is on track to achieve the growth objectives we set out at the beginning of fiscal '25. We have opportunities across the global customer base with revenue currently driven by 50-gig and 100-gig per lane designs for AOC and transceiver applications at port speeds up to 800 gig. Credo is actively engaged in opportunities with more than 10 transceiver vendors for multiple hyperscale end users. We work with our optical transceiver partners to provide full DSP and LRO options to meet a wide range of networking architectures. We see a large and growing market for these offerings as well as for 1.6T port deployments in the future.

With our recent 3-nanometer tape-out, Credo is well positioned for these leading-edge opportunities with our 200-gig per lane DSPs, where we expect to again have a compelling combination of performance, power and features. We see the market opportunity for optical connectivity continuing to be very dynamic as reliability and energy efficiency become more important. As a result, we've seen an increasing opportunity for Credo to deliver system-level advantages to our partners, activating Credo's third tier of innovation I outlined earlier. Next month, at the OFC conference in San Francisco, we'll demonstrate a full suite of optical solutions, including 200-gig per lane in conjunction with our optical module partners.

Based on all of our progress, the breadth of customer engagements and the expanding market opportunity, we remain excited about the increasing revenue prospects given our role as an innovator in the optical connectivity market. Now regarding our retimer business. Credo continued to gain momentum with our retimer business in the third quarter. Over the past several years, Credo has established leadership in the Ethernet retailer market, delivering advanced capabilities such as MAX second encryption, gearboxing and other software-enabled functionality.

Existing customer wins and future opportunities here include 100-gig and 200-gig per lane applications for both traditional switching and increasingly for AI servers requiring retailers for scale-out networks. This year, Credo has entered the market for PCIe retimers used in scale up networks. Our strategy is in alignment with our 3-tier innovation approach. We believe credos PCIe SerDes IP will establish new benchmarks for the combination of latency, reach performance and power and that our implementation of the 2 CAN PCIe retimer will deliver compelling advantages to our customers.

In February, Credo participated in the PCI SIG compliance workshop in Taipei, and we are pleased that our 2 CAN retimer achieved full PCIe compliance. It is notable that Credo is only the second vendor to achieve this level of compliance certification for PCIe Gen5. This significant milestone demonstrates our capability to bring best-in-class PCIe products to market. Credo will be added to the PCI SIG integrators list in the coming weeks. During Q3, we engaged with key customers who evaluated our PCIe silicon.

I'm pleased to say that the feedback was very encouraging, and we received our first platform commitment from a large AI server ODM. We are on track for production revenue in calendar year 2026. Market forecasters believe the TAM for PCIe retimers will exceed $1 billion by 2027, and Credo is very well positioned to compete for material market share. In summary, I'd like to first comment about our team's incredible execution over the past quarter. To successfully navigate a ramp of this magnitude requires extremely tight operational control, supply chain coordination and customer communication.

Just as Credo works tirelessly with customers to innovate on solving their pressing connectivity needs, the Credo team is clearly rising to the occasion to deliver on the significant demand ramp we're experiencing. As we more broadly ramp customers across our products, we will continue to closely manage our execution. I remain enthusiastic about the expanding market opportunity for high-speed connectivity, driven by the promise of AI and the investment it's spurring. Credo's tiered approach to innovation has and will continue to be an advantage as we serve our customers.

Based on our progress with customers and the increasing demand for leading-edge connectivity solutions, Credo remains on track for continued scaling of revenue and profit. I'll now turn the call over to our CFO, Dan Fleming, who will provide more detail.

Thank you, Bill, and good afternoon. I'll first review our Q3 results and then discuss our outlook for Q4 of fiscal year '25. In Q3, we reported revenue of $135 million, up 87% sequentially and up 154% year-over-year and well above the high end of our guidance range. Our product business generated $132 million of revenue in Q3. up 91% sequentially and up 155% year-over-year. Notably, our AEC product line grew strong triple digits sequentially to achieve new record revenue levels.

Our product business, excluding product engineering services, generated another record at $129.4 million of revenue in Q3. 101% higher than our previous product record in the prior quarter. Our IP business generated $3 million of revenue in Q3. As demonstrated by our product revenue ramp, we are seeing substantial opportunities with customer programs on the product side which we are prioritizing. This prioritization does not impact our long-term model for company-wide non-GAAP gross margin of 63% to 65%. Our largest end customer was 86% of revenue in Q3. As a reminder, customer mix will vary from quarter-to-quarter, and we continue to make progress in diversifying our customer base.

As we shared last quarter, we had 7 customers that contributed more than 5% of revenue. And going forward, we expect that 3 to 4 customers will be greater than 10% of revenue in the coming quarters and fiscal year as additional hyperscalers ramp to more significant volumes, as Bill described. [Audio Gap]

[Operator Instructions] And your first question comes from the line of Vivek Arya with Bank of America Securities.

For the first one, if you could give us a sense of how large was the largest customer. I think Dan mentioned the number I didn't catch it. But I guess I have 2 parts to the question. One is, where are you in the adoption of AEC at that customer?

And if you exclude that customer, how are you looking at the growth of your business among other customers? Because depending on that customer concentration, rightly, we get a slightly different trend outside of that large customer.

Yes. So Vivek, so what I mentioned in the prepared remarks is our largest customer was 86% of revenue. So let me just give you some historical context there. As we entered our fiscal '25, we described a second half inflection point, really to be driven by our largest hyperscaler, which is exactly what we saw play out with that 87% sequential growth into Q3. So as far as customer concentration goes, Q3 was a bit of an outlier for us. especially considering the customer diversity that we had in Q2, and what we expect in the coming quarters.

And again, as I mentioned in the prepared remarks, we expect 3% to 4%, 10% plus end customers in the coming quarters and fiscal year, which is simply based on the forecast that we received from our customers. So just broadly speaking, there's clearly a broad-based need for our innovative solutions as lane speeds are increasing at these hyperscale deployments. And our AEC opportunities, in particular, are expanding in the back-end network from originally, the scale-out network to now the scale-up network as well, which represents a significant expansion of our TAM. So I'll let Bill add some additional color to those comments on customer diversification.

Yes. First thing I would say is that my takeaways are that this is really a great confirmation of the AEC product category and market and the overall TAM. I think that it's also a great confirmation of Credo's leadership in the space. And so as we've talked about on past calls, we see each one of hyperscalers capable of driving a very sizable TAM as we take our engagements from first deployment to a more populated accretive AMC deployment longer term.

So I think if we look at our customer base, it's broadening. I mentioned that we've taken now 3 customers, 3 hyperscalers to volume production. And we've got 2 additional hyperscalers in qualification and expecting ramp in our fiscal '26. So we expect to see really solid customer diversity long term within our AEC business. But also I think that that I'll add to that, that the expanding opportunity that we see for optical as well as PCIe drive really further diversity across products and end customers as we look towards the future.

Got it. So my follow-up, Bill, if let's say, more of the market turns towards inference rather than training, right, whether more [Audio Gap], but [Audio Gap] of the compute build-out. What does that do to these air clusters? And what does that do to connectivity requirements and the AEC potential? Is that good, bad, neutral for AEC?

Yes. I think as the inference market really takes off, I think if anything, we would see a larger opportunity from an AEC standpoint. And that's just based on sheer number of deployments related to inference. So I think that as we look at the customer activity going forward. There's a lot more focus now on inference, and we're seeing really an uptick in the amount of activity.

And your next question comes from the line of Tom O'Malley with Barclays.

So I think a lot of the tone of the call here has been about like a broadening of the portfolio. You're talking about PCIe retimers. You're talking about PCIe cables. To push you maybe 1 step further, if you're looking in the PCIe realm, an area of a lot of value is in the switching ecosystem as well.

Do you guys have plans to move into the switching ecosystem? Is that like a natural progression from kind of the road traveling down already? And maybe talk about the challenges of doing retailers versus moving to switching and the time line that it would take to transition from product in the market today to maybe some new products on the switching side.

Sure. I appreciate the question. So we've made a lot of progress with PCIe over the past quarters. I think the comment that I made earlier was regarding credo passing special interest group, the PCI SIG most recently in February.

That's a great confirmation of the technology at confirmation of interoperability for Gen5. So as we move towards Gen6, I think we're increasingly bullish on our ability to compete market. We see that for us, first step is really going after the retimer market as well as the PCIe AEC market a scale-up architecture has been to rack scale. I think it's a really natural progression to think about moving to PCIe switching. There's different challenges associated with that, but it is really not such a huge step to transition from retimers to the type of switches that are being built and talked about being deployed in the future.

So I look at that as a possible spot for us to grow into over time. Right now, we're very much focused on getting things right related to the retimer as well as the ADCs that we're building.

I appreciate you answering despite you just announcing the prior product before, so I appreciate it. But just in terms of the market outlook, so you're talking about the scale up networks and also the scale-out network you have PCIe cables, potentially more aligned with the scale up network. You've got some of the AEC product, particularly looking more at the scale-out network. Can you talk about the opportunity size?

Because I think we're going to see in a couple of weeks here, like maybe a redefinition of what scale up and scale out is? Is it 1 rack, is it multiple racks, maybe your thoughts on scale up versus scale out, where does the opportunity set life for you? Are you indifferent? Just want to kind of to get your temperature on that transition.

Sure. So we've talked about the opportunity in the past to scale out, we really see being an Ethernet protocol. And so we see that continuing on the path that we're -- that we've been on as the market moves to 100-gig per lane and this market will subsequently move to 200-gig per lane solutions. So scale up is really probably more interesting in the sense of talking about where we are today and where we're going.

And we do see by the innovations previously shown by NVIDIA and what we expect coming up that there's a lot of activity. It's really a dynamic space from their perspective. Now they're a little bit different from the market broadly. But what we see within the customer base that we're talking to is that this scale-up network that has really been a network that exists within an AI appliance really going rack scale.

And then, of course, there's an opportunity for it to go rose-scale long term. We've talked about the volumes being larger than larger than the scale-out network opportunities. So we really see this as a big new TAM. As the market moves from Gen5 to Gen6, we're talking about moving from 32 gig NRZ, which is really very old technology, and it is really not competitive if you compare it to the market leader from a bandwidth standpoint per lane.

Gen6 represents the move to 64-gig PAM-4 modulation. So that's a step-up in complexity and difficulty from the existing suppliers in the market moving towards that direction for Credo. We've been there for many years. So we feel like we're going to bring the same compelling advantages that we've brought to Ethernet.

The interesting part of the future is that, when you look at that scale of network and the opportunity for improving performance with an increase in bandwidth, we've talked about the ultra accelerator linked to UAL conversation in the market, taking the scale-up network from 64 gig all the way to 224 gig. And so we're going to be in a unique position to take advantage of that transition when that happens with the other market leaders that are outside of the NVIDIA ecosystem. So we're pretty excited, absolutely about both opportunities, but I would say the scale-up opportunity is probably larger if we look out a 2- to 3-year time frame.

And your next question comes from the line of Karl Ackerman with BNP Paribas.

Two questions, if I may. The first one, I guess, what's driving the uptick in gross margins in the April quarter? Is it IP licensing revenue related? Or is there some other thing that we should be thinking about?

Yes. So to sum it up, we're really seeing a huge benefit from scale, which we've talked about or we've foreseen it for quite some time. And as you know, our overall gross margin in Q3 was almost 64%, up a little bit from last quarter. But if you dig into the numbers, the thing that's more indicative of what's happening underneath is if you look at our product gross margin, excluding product engineering services, our gross margin was 62.4%.

So that was up over 200 basis points sequentially, up over 900 basis points year-over-year, principally driven by scale. It's really as simple as that. The other lesser factor that I'll mention we have a warrant with Amazon, the contra revenue associated with that kind of rolled off during the quarter. as we attain that as we fulfill the $200 million of total gross revenue shipments to them in the quarter. So that was also accretive to margin, which will continue to be accretive next quarter as well.

Got it. Just a follow-up on your prepared comments, you spoke about how you may have 3 or 4 customers that would be 10% plus in the coming quarters or in calendar '26. Obviously, the first 3 today are AEC related. Is the fourth one also AEC-related, or would that just perhaps broadening into other aspects in product portfolios of your business?

Yes. Of the ones we've referred to in those comments, they are AEC related. We've had, by far, is the largest driver in absolute dollars of our revenue growth. Now having said that, if I just look simply year-over-year from fiscal '24 to '25, for instance, we've had our main 3 product categories between AEC our retimer business or Ethernet retimer business and Optical, they're all experiencing sequential growth, but AEC just has -- it's growing from a bigger base into an even bigger base as we go into the future.

And your next question comes from the line of Quinn Bolton with Needham & Company.

If I've got my numbers right, revenue outside of your largest customer went from about $48 million in October to about $19 million in January. You talked about kind of revenue diversifying again over the next few quarters and into fiscal '26 with 3 to 4 customers that could be over 10%. Can you just give us some sense. What do you expect your largest customer to do?

Does it stay 80-plus percent of revenue? Are you anticipating that, that pulls back and that you see ramps at some of the other customers in the April quarter. Just any sense of revenue rediversifying would be helpful.

Yes. We've talked in the past about Actually, Amazon is a great example. So our largest hyperscaler. If you look at their Q1 revenue, was $30 million, then it went down a bit in Q2. Now it obviously surged in our Q3. Our internal expectation is they'll probably be in the same ZIP code as to where they were in absolute dollar terms this -- in Q3, or where they were in Q3. So if you look at that being what it is and knowing that we guided 19% sequentially up quarter-over-quarter into Q4 at the midpoint.

That would imply that maybe they're 2/3rd of our revenue in Q4 would be what that math would imply to apply.

Got it. And then I guess just looking at that customer, obviously, it's a significant percentage of revenue in the January and April quarters. I think in the past, you've talked about having kind of 12-month rolling forecast from some of your larger hyperscale customers. Just generally speaking, can you give us any sense -- what gives you the confidence that the surge you've seen here in January and April may not just be somewhat related to an inventory build ahead of server deployments that at some point in fiscal '26, you go into kind of a depletion mode or just a deployment mode where you could see I guess, for a lack of a better term, an air pocket at that customer at some point next year?

Just any help to hold hands would be appreciated.

Yes, sure. So I guess -- when we talk about visibility, I think it's very normal at this point to get 12-month forecast from customers. I think this is one of the positives that came out of COVID is that really a top priority for our customers is to secure their supply across all materials, including the things that Credo supplies. And so when there is a sizable ramp, it's really imperative to work together with our customers as well as our supply team to ensure that we can deliver flawlessly.

And so I would say across the board from a customer perspective, we're getting good visibility. In the case where there is a new ramp, this is where things become more high definition as you get closer to the actual ramp. There are some things that can happen we've seen pushouts to schedule. We've seen pull-ins.

We've seen increases as we saw this quarter in the level of deployments as customers around. But generally, as we look towards the future, we see several customers that are at different stages of deployments. And so if we look at the numbers and compare, say, Q2 to Q3, you saw a pretty big shift, pretty good move. And we indicated that, say, for the first customer that we ramped with over the last several years, we talked about a period of time where supply and consumption and those curves will meet.

Well, that happened in Q3. We also talked about that customer returning to historic levels, which we expect to be over $100 million run rate, say, as we look at fiscal '26. We've got a new customer that we're ramping the size of allocations that they're getting from a GPU standpoint suggests that they're going to be a very large customer in our fiscal '26 as well. I mentioned the 2 new customers, these would be customers that are a little less a little less definition on the exact timing and size of the ramp, but both are promising in the sense that we're seeing the level of deployments they're talking about being sizable.

So as you look at the next couple of quarters, I'm sure. We're going to have a decreasing level of concentration, but it's still going to be something that you would say that's a large amount of concentration, but it's going to be balanced out over the second half of fiscal '26.

Understood. I'd rather see you guys have the Amazon revenue than not. So congratulations on the nice results.

These are good issues.

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

And congrats on the record results. Bill, in your prepared remarks, you talked about several layers or multitiered innovation. And the one thing that obviously is that you talk about is the system-level approach. That's obviously very obvious on the AEC part of your business.

But as you venture into some of these new segments and especially on PCIe. Will you take more of a system business model approach there as well, or should we assume that's going to be primarily selling chips?

Well, I think the fact that at GTC in a few weeks, we're going to be demonstrating a full rack multiple AECs that are active, that are live. We're going to be demonstrating our AEC capability. And it may actually be it might even be the first demo of its kind in the AEC space. So yes, absolutely.

From day 1, we've been planning on pursuing the same path with PCIe AECs as we have with Ethernet, and no change there. We're even planning as the scale of architecture could potentially grow scale. We're also investing in the optical space as well.

That's very interesting. And then my follow-up, LRO, I haven't heard a whole lot about it since you obviously introduced it. So any update you can give us there? If you do have some stuff that you're going to talk about at OFC, I get it. But any update we can get on the row would be really helpful.

Right. We continue to offer customers both full DSP solutions and LRO solutions. We're agnostic to their choice. We see customers continuing to purchase traditional full DSP solutions where they can fit within the power ceiling.

In the case, we do have cases where power is a deciding factor. And I think with all of the customers we're talking to, LRO really becomes that de facto solution compared to, say, LPO. I think we've all seen that the discussion around LPO has decreased pretty substantially. The conversation with LRO, I think is -- it's a real opportunity to get the kind of signal integrity, [indiscernible] and interoperability and being able to save that kind of power, those sub 10 watts for an 800 gig port [indiscernible] to the side.

And your next question comes from the line of Vijay Rakesh with Mizuho.

Bill and Dan, great quarter, congratulations. Just on the AEC side, I know you mentioned 2 new customers to new hyperscalers that you'll be working with. On the visibility, do you see them happening to -- as you look out 12 to 18 months, you see that ramping pretty nicely to significant volumes. How do you see that ramp. And I have a follow-up.

Yes, sure. The customer-facing team at Credo is hard at work, and that that includes our applications engineering group. And I think it's a function of exactly when we intercept on the ramp versus if we do. Both of these customers, there are there are design engagements that are absolutely identified and committed to.

So I think it's a function of exactly when the ramp starts. And that's when I make the comment that fiscal '26 is what we expect I think if we wanted to be somewhat, I guess, on the conservative side, we'd say second half of fiscal '26. But we're getting to the point where our confidence is quite high with the additional 2 customers that are in qualification.

Got it. And then on the PCIe scale up side, really great to see you guys ramp so fast since you so quickly after [indiscernible]. You mentioned 1 server ODM. How does that pipeline look? And do you have to qualify the server ODM or the or the GPU [indiscernible] guide, just [indiscernible].

Yes. So I understand the timing of our ramp for PCIe, it will still take some time. We're really targeting the Gen6 market, which is that's really a calendar year '25 design cycle and a calendar '26 production ramp cycle. And so it's really somewhat out in time.

Even though we're demonstrating we're passing interoperability test, we're absolutely ready to compete. It's going to be some time. So another question would be, are there going to be Gen5 opportunities that we're able to pursue. And of course, as our solution is brought to market, we're open for business, and we'll bring advantages to Gen5, an easy upgrade to Gen6. And so -- what we're seeing from a customer perspective is that there is independence of making decisions.

And there is a need for a more broad supply chain than exists for Gen5.

And your next question comes from the line of Sean O'Loughlin with TD Cowen.

Congrats on a nice set of results and guide. I'd like to trend a little bit and ask about front-end networking. The original -- obviously, the original AEC deployment was on a front-end solution. And I wonder whether or how you and your customers are thinking about that front-end opportunity that still potentially exists?

Is it just that in end networking is not where the innovation is happening in the space, and so customers are focused on the back end? Or is it a matter of time or some hybrid of the both? And then to squeeze an extra question on top of that, is there really a difference in the product that you ship to front end or go-to-market for the front end versus back end?

So from a front-end network perspective, it applies to general compute and it applies to AI clusters. So the front-end connections are pretty similar between the 2. Now with general compute, that's the only connection to the network, the front-end network. And if we think about the traditional general compute space. It's really a question about really the x86 road map.

And that has typically moved at a slower pace than what's happening within AI clusters. And so it's very common still today to see 25-gig per lane front-end network connections. In fact, the one that we've shipped in high volume is of 4 lanes of 25 gig. So it's a 100-gig port connecting to that front-end network.

The unique feature that we brought was that our lead customer implemented a dual tour design to achieve levels of availability, the SLAs of Five9. So it was a way for them to to really achieve a pool that had been in place for 5 to 10 years even. So that unique functionality is what turned on that business for the front-end network connection at 100-gig ports. As we look at that general compute, there's a natural move to 50-gig per lane and 100-gig per lane. But in contrast, if we look at AI, just the nature of the application is causing customers to move to the fastest connection so 100-gig per lane today and really putting pressure on delivering 200-gig per lane solutions in the future.

And within AI clusters, those front-end connections exist, and at the best case, albeit the same line rates as we see in the back end, but probably most likely, there'll be a lower line rates. And so our largest customer is using our Credo AEC solutions for both back-end and front-end connections.

Great. Super helpful. And then just as a quick follow-up, you mentioned the 3 hyperscalers that are achieving volume shipments today. I think last quarter, you talked about the 2 hyperscale, obviously, that we all know about. And then in the third emerging hyperscale, I just want to clarify that, that is not the third hyperscaler that you talked about today, but is, in fact, a different customer.

Yes. So as it relates to this concept of emerging hyperscalers, it really boils down to the amount of spend. And if you look at the allocation that is being given to this kind of the lead "emerging" hyperscaler. We really are classifying them now as a hyperscaler. So we're not drawing the line between the two.

And your next question comes from the line of Christopher Rolland with Susquehanna.

I guess -- yes, as it concerns your largest customer here and maybe any of these large ramps going forward, I think a lot of these things are project-based. We have very large deployments, sometimes 200,000 GPUs in just a couple of months. So as we think about these kind of large, lumpy deployments, should we be expecting these customers, for example, in this case, a pretty strong hand off in the next few quarters to other customers? Like Bill, maybe if you could talk to how project-based these revenue ramps are, that would be great.

Yes. I think that when I -- I guess when I think about your question, we talked about project based, of course, all of these programs that we're involved with are projects that have been planned for quite some time. And going through the development internally within hyperscaler customers, it's a long process to architect and ultimately bring all of the gear up to a point where they're ready for qualification. Qualification takes a significant amount of time and then planning the ramp takes time as well.

And so we're at the point where we've kind of gotten through that process with or nearly through that process with many of the customers. I don't really see a handoff so much as I see a diversification in the shipments that we'll be seeing throughout fiscal '26. This is just a function of where we are with each customer. But I don't think we'll see a customer get to a point where they are done with a project. There will be a transition from say, 50-gig per lane technology to 100-gig per lane technology.

And that, as we've talked about in the past is we are absolutely queued up and ready to ramp. We're the shortfall in the tent, so to speak. That all of that technology that need to be developed internally accretively is done, and we're ready to ramp as soon as our customers are. And so I don't see us getting to the point where we say that, "Hey, that project is over and now there's some sort of an air pocket." The key is making sure that we're first, making sure that we're first to deliver next-generation samples, we're first to get qualified, and we're first around production, which really is our strength at this point given the fact that we're taking accountability for the entire system solution.

Great. And maybe as a second question, analog, copper cable solutions seem to have missed perhaps the B-300 cycle. Is that an opportunity for you guys? Do you think you're gaining share and design wins just as that technology did not ramp, does that increase your TAM and your ramp expectation?

Yes, I think that TAM really existed within the NVIDIA ecosystem. The TAM that we're looking at, we've not seen really any customer think seriously about about using amplified solutions. So this referring to the analog amplified solution. We haven't really been competing head-to-head with any of those solutions really at any other customer.

We really see it within the NVIDIA ecosystem. And what role do we play within that ecosystem, that is one that we're pretty conservative in talking about. When there is an opportunity for us to add value, we're absolutely ready, but it's not something that's built into any kind of forecast that we've got.

And your next question comes from the line of Richard Shannon with Craig-Hallum.

Bill and Dan, maybe I'll just ask a simple question here that I think you get most quarters here, Bill, but I just want to get your latest thoughts on competition in the AEC space. Obviously, you've got a very high share, you're doing exceptionally well here. It certainly makes sense for some of your large customers to attempt to do dual sourcing. Have you seen any evidence of that, either attempts or see that coming here anytime soon just kind of your latest thoughts on that, please?

Yes, I think the fact that the AEC product category is now really de facto for in-rack connectivity. Naturally, there is a desire for multiple sources in the market. But the way that we think about competition, it's really -- our objective is always to be the best possible partner to our customers. That means delivering on the innovations first, delivering on the most reliable solutions, first passing qualification first and getting our customers into production predictably.

And so that's -- our focus with our customers is really on creating relationship with them where they're counting on us to help them ramp. And having the second source thing, something that's kind of a trailing edge effort. So that's really the way that we think about competition. Now I will mention again that being accountable for the system solution allows us to deliver faster, more predictable way.

Owning every level from the SerDes to the ICs and ultimately, to the cable system. It gives us a big advantage from our perspective. That's what we're seeing across the board with customers. With that said, with a lot of respect for competition, but we don't see any significant changes on the competitive front.

Okay. My second question, really just looking at the large AEC opportunities here and clearly the largest customer is really drive a lot of content here, both as you said, both in the back end and the front end. I think maybe just thinking about from a back-end perspective, are you seeing any other hyperscalers getting close to adopting and even qualifying your AECs in the back end and kind of thought process on timing here? I'm assuming it's somewhere in fiscal '26 with that after, but maybe just kind of your latest thoughts there as well.

With our new customers, we see it playing out the same way that it's played out with our existing customers. And it's always starting with the first -- starting with first project or a first SKU that we're working on, starting in a given rack and then really expanding from there. And so one of the customers that's new. We started with a switch rack, so a disaggregated chassis, a relatively low hurdle 50-gig per lane type of project.

But it quickly turned into 2 additional SKUs that are targeted for AI clusters that they'll be building in the future. And those are going to 100-gig per line speeds. And so we see there's always got to be a first program, but the execution that we have on that first program will always lead to a deeper relationship. And we see that playing out the same way with our fourth and fifth [indiscernible] customers as we did with the first 3.

And your next question comes from the line of Suji Desilva with ROTH Capital.

Bill and Dan, congrats on the progress here. I just want to understand from the lead customer and the future 10% customers. Just to clarify whether the ramp is today a single program and expected to be a single program across those 10% customers or whether you get into the point where you have multiple projects across these customers in the pipeline that might diversify you for each customer?

Yes. I would say that if you look at any of our customers, there's multiple projects for the first 3 that we're ramping. There's multiple projects. There's multiple SKUs. And I expect that to be the same with the additional customers that we add, although we always are going to start with the first project.

Okay, that's helpful, Bill. And then the opportunity and specifically in back end scale up, I just want to be clear here, is that an opportunity only on in-house ASICs, or is that opportunity as well on merchant GPUs in the marketplace?

It's an opportunity across the board. And so I mean, we can talk about the NVIDIA ecosystem and that's probably part of the TAM that we're not really thinking about too aggressively as we start. But even for the deployments that use NVIDIA GPUs and use using the kind of the open protocols in the market, PCIe will be an opportunity there as well. So hopefully, that gives you an answer to your question.

There are no further questions at this time. And Mr. Brennan, I will turn the call back over to you.

Well, thanks for joining us today. Really appreciate the interest and support. As we look ahead, we're absolutely laser-focused on executing on our strategy, delivering value to our customers and driving long-term growth and profitability. We look forward to the call backs. Thanks.

And ladies and gentlemen, this concludes today's call, and we thank you for your participation. You may now disconnect.