Credo Technology Group Holding Ltd (CRDO) Earnings Call Transcript & Summary

Good afternoon, everybody. Welcome to the Goldman Sachs Communacopia and Technology Conference. My name is Jim Schneider. I'm a semiconductor analyst here at Goldman Sachs, and it's my pleasure to welcome Credo. We have -- really happy to have CEO, Bill Brennan; and CFO, Dan Fleming with us today. Welcome, guys. Thanks for being here.

Thanks for having us.

You sit in a very hot part of the market right now tied to AI infrastructure within data centers specifically connectivity. So your words, what is the mission and vision of the company? How has the journey been so far? And what problem is Credo trying to solve more efficiently and effectively for everybody else?

Sure. So I've been with the company 12 years now. And when I joined the company, we had this vision that the data center was going to be driving an exponential need for more bandwidth. And it's taken many years, but it's on a super accelerated pace. So the mission has been pretty consistent and it's breaking bandwidth barriers by delivering highly reliable and energy-efficient connectivity solutions. And so that's what we've been at for the past decade for sure, and it's really in hyperdrive right now. So connectivity, like within the AI space, we all think about GPUs but it's connecting the GPUs, 10,000 or 100,000 or 1 million people are talking about at very high speed with great reliability and with great energy efficiency. That's what makes the magic happen with AI. And just to put reliability in perspective, the connection between the GPU and the first switching layer, there is no redundancy. And so when you get a single failure in that connection, a link flap is what the industry refers to these failures as, brings the entire cluster down. So the need for more reliability is absolutely a top priority, and that's been a very intense focus of our team, especially over the last 12 to 18 months as we've succeeded with the AEC product family. It's 1,000x more reliable than some of the optical connections that we're replacing.

Yes. One question we get all the time from investors is what is the total available market for AECs? And how do you kind of contextualize that in terms of the overall connectivity TAM for data centers?

I get that question a lot, too. I got that question 9 times today. What is the TAM? So the TAM is big. I think that the TAM is growing, and it's larger than we conceptualized back when we thought about this market. I can talk about the different parts of the TAM. But I think at a high level, it's -- I used to think about the market in terms of say, a $2 billion TAM. Now I think about the market in terms of like, say, a $5 billion to $10 billion TAM. And that's really driven by different parts within the network where AECs make a lot of sense. We started our business connecting mix TORs in general compute. Then we've talked a lot about switch racks that sit in the higher-level switching tiers within the front-end network. And then AI happened. And then we talk about scale-out network. We talk about the scale-up network. And so we see Ethernet, we're not fully penetrated by any means in these areas. And then on top of that scale-up is coming, and when we think in terms of volume, the front-end network, give that a 1x relative size. We view the scale-out opportunity as being a 10x size on actual connections, number of connections. And we think scale-up when it goes full rack scale to row scale, that represents another 10x in potential volume. So you can follow the analyst to get their view of the top down, but I can definitely make a case it's a large and growing market. And I don't think there's any kind of questions about is optical going to replace copper anytime soon. I've been fighting that whole conversation for the 12 years I've been here. I was advised don't do anything for copper, you're making a big mistake. And now we're actually replacing optical connections. And so I don't think there's any doubt over the next 5 to 10 years that it's a large market.

Yes. How do you think about what types of connections AECs ultimately displace? Is it optical, conventional copper? What is it?

Yes. So it's -- the product category, you can think about it, we came into a world that everything was connected by passive copper cables, certain lengths. And it got to a certain length where you couldn't make that connection anymore and it became optical. Those are the only 2 options. And so we walked into a marketplace that really fully believe that when passive copper cables kind of ran out of gas, every time you extend the length or you double the speed, you get more loss and less of an ability to do long connections. And so very much, we started by replacing DAC. Instead of it being optical, we replaced DAC. But now with AI and the priority for our reliability, we are replacing optical now as well. And so limited on length, 7 meters. So if a company were to figure out how to make optical connections 1,000x more reliable, that would be a very good market space to be in because you could offer that same thing at, I'd say, a row length, which is probably 15 meters or 20 meters.

Okay. On your AEC business, can you maybe talk to the primary use cases that are being adopted today? What have been the biggest kind of drivers of adoption? What other use cases do you think AECs can be -- can address in the future?

Yes. So there's kind of basic use cases, connecting NICs or whatever that server is, an appliance or general compute, computing that to the first switching layer. And then within the switching hierarchy, so that NIC to TOR is a huge market. You can see it's a huge market, especially because of AI, but it was a huge market even with general compute. Switch racks basically replace chassis. This is basically a chassis full of switches. Amazon for many years has been instead of deploying chassis, say, from the big networking companies, they're buying white box switches and stacking them vertically. And so replacing the backplane connection within a chassis with in-rack cables that are 3 meters or less. That's one big opportunity as well. And then when we think about things in terms of scale out and what liquid cooling is doing from the standpoint of how long the connections are, we had one customer that had an 18 rack deployment that was lightly populated. It wasn't very dense because of -- it was air-cooled and they didn't -- like the actual rack design couldn't source a lot of power. They redesigned it to compress 18 racks down to 6. And in that case, they had all optical connections. And so this is a different application. This is an expansion of the TAM from less than 3 meters to now row scale 7 meters. So many different spaces within the data center network that can take advantage of AECs.

Yes. And based on your conversations with customers, what kind of pushback are you getting, if any, on the adoption of AEC?

I'd say that once a customer really engages with us, they're pulling the product once the light goes on. There's a lot of inertia with the way things have been done over time. People have struggled with making DACs work for the next generation. They'll do incredibly difficult things to try to fight through the signal integrity and form factor problems. And then a lot of people just think the default is optical. So like we're still in a mode where establishing this as a de facto, but when it turns on when we engage with the customer the first time, the experience is incredibly easy. And we're hitting reliability. We're hitting energy efficiency, we're hitting costs. So these are all great things kind of in that order. I don't get a lot of pushback once the light goes on.

Yes. Yes. There are still customers where like it hasn't gone on?

Sure. That's called selling.

There you go. Our passive optical cables -- or copper cables excuse me, improving in performance at all or providing comparable ROI in any way?

Yes. So one of the things that we've seen is this kind of battle between copper and optical. The copper supply chain is not sitting still. When we had envisioned our 800 gig or 100 gig per lane family of AECs, we thought 3 meters is going to be the maximum length that we would achieve. And that was based on the copper that was available at the time, the characteristics of the copper. And it was also based on our DSP. Through improvements in the DSP and in the raw copper, we're now doing 7 meters. So that supply chain is not standing still. And so I expect that, that supply chain is going to continue to make progress. And we talk about 200 gig per lane, 1.6T ports, I think that ship has sailed. The ecosystem will look the same. It would be a tremendous TAM for copper and optical. And even talking with some of the copper companies, they believe 400 gig per lane is possible. And so it's -- I think that the future is definitely bright. That marketplace or that group of suppliers, they're doing amazing things at improving. And so hopefully, that gives you the color.

Yes. Yes. Maybe switching to the business trends a little bit. I think you ramped up your AEC revenue pretty rapidly. And I believe you're set to grow with 2 additional hyperscale customers in the back half of '26. Help us understand the visibility you have into the business the nature of the relationships in terms of forecasts, take-or-pay agreements, prepaid commitments and so on?

So coming out of COVID, one of the best things was that our -- the groups that we work with, the supply chain at our customers, they want us to have 12-month forecast. They want us to know what they're planning. And so we've got a good relationship with all our customers. Now within that 12 months, we don't have any take-or-pay or we don't have that kind of relationship. I don't want to overship. I want to basically match our shipments to consumption. That's just a philosophical way of doing business. I don't want to talk about a huge amount of inventory that's in place that I shoved in because I had a contract and I was able to do that. And so there's some fluidity within that 12-month forecast at a given customer. And especially when there's new deployments, those 12-month forecasts can be not hugely accurate. We've been on the right end of fluctuation with 2 of our customers recently. We had an enormous ramp, enormous pull-in, surprising amount of product that we were asked to ship in a very short period of time. So when they got their deployment plan together and they hit basically full green light, it was like a real like strong pull into the 12-month forecast. And then our fourth customer that we talked about, they're actually pulling in. And so on our call recently, Dan articulated that, although we didn't have what we would consider material revenue -- or we didn't have a 10% level revenue with this customer, we expect them to be a 10% customer for the year. So going from basically not shipping to 10% within 1 year. And if you think about our growth as a company, that's an amazing amount of product that they're looking for within a short period of time. So that's actually starting in the second quarter. And the fifth customer we've been engaged with for a long time, and it's a function of basically sorting out their deployment plans. I don't have a solid schedule yet, but that's -- I expect it to start towards the end of the year.

Great. Maybe help us think about how you conceptualize the opportunity for pricing uplift generation to generation on your AEC products?

You want me to talk about increasing pricing in a public forum?

Yes.

I'm kidding. But naturally, we have to say, go from a 400-gig solution to an 800-gig solution to 1.6T, fundamentally, it's going to cost more to build. And so first generation, when we're shipping 25 gig per lane, a lot of people were willing to do the work to use DACs. And there's a pretty big gap in pricing between passive copper cables and what we do. I mean it's fundamentally more expensive technology. And so we were kind of tethered on pricing to this very basic jumper cable basically, passive copper cable. There was only so high we could justify if they felt like they could get the job done. Now that we're really at a point where not a lot of people are struggling with DACs, so now we're being compared to optical. So we've kind of been untethered and almost lifted by optics pricing on -- like if we can save money compared to optics, I think that puts us in a good spot. But if we can be 1,000x more reliable, then we just -- we can start talking about the value of the product. I think Dan has talked about margins over the last couple of years. And I think we signaled that we'd see a national expansion as speeds would increase, and that's happened quite nicely.

Yes. Another question that I get a lot in terms of competition in the space is as more people enter the space because obviously, this kind of market is going to attract more competition, how do you balance competition for market share relative to gross margins? And then maybe just kind of give us the context about the value you provide to help keep commoditization in check?

Yes. So we're unique from the competitive landscape. We're the only true vertically integrated player in this market. We compete with groups of companies, chip companies and cable assembly companies, and they kind of team up and they go to market that way. I'm fully crystallized on our model delivering what I would think is a competitive moat. The way we try to compete is we try to, first of all, engage with customers at a level where we enable innovation. They want special things, we do it. We are highly focused on being first to deliver samples. We're highly focused on first to being qualified and first to ramping. If we can do a good job of that, I think the competitive dynamics will play in our favor long term. That's the competitive advantage we bring. In production, if we can be flawless with delivering and if there is an issue, if we can absolutely be that single throat to choke, that's going to pay off. So building relationships really takes many, many cycles, many years to do, and I feel like we're well on our way. So competitively, I'm more focused on making sure we satisfy the customers.

Yes. One of your innovations, you mentioned it before, a bit of a surprise to me initially in the market there's been your 7-meter AEC cable. Apart from your first customer, tell us about the interest in that specific product? And then maybe conceptually, how much longer you think you can make connection electrical without losing performance or signal integrity?

So I think 7 meters is going to be at max. I'm not going to -- maybe that's a challenge to the copper ecosystem to do better. But 7 meters at 100 gig per lane, when we talk about going to 200 gig per lane, probably goes from 7 to 5. And tell me the first part of your question again?

What's the customer interest been in the 7-meter?

Yes. So what drove by the interest was our customer was on this mission to build a ZeroFlap cluster. And if they looked at their linked flaps, these momentary intermittent disconnections in link, it actually shuts the entire cluster down. So they're on this mission to come up with a ZeroFlap cluster. And that they asked us to build 7-meter cables because the reliability -- we just don't have flaps with AECs. We don't have these interment failures that you get from laser-based optics. And so that was what drove the ask and what drove the possibility was liquid cooling and power sourcing. It's the example where I said 18 racks got compressed to 6. This trend is going to continue in the market. So this is the first of what I expect many customers to pursue. And it's such a fundamentally better product from the standpoint of reliability that I think the interest is going to be large, long term. And that applies to scale out and it applies to scale up as well.

Yes. I think the market has a view that your AEC revenue is somewhat levered to a couple or one custom ASIC program or custom ASIC programs in general. In other words, as your shipments -- their shipments go, yours will follow or maybe yours lead and theirs follow either way, especially your largest customer. Do you think that's a true statement?

No. I'll tell you why.

Yes. Yes. And then talk about how you may participate in merchant accelerator clusters as well?

Yes. So we're not a proxy for custom cloud silicon, just to be clear. Our third and fourth customer that we talk about, we're only connecting with NVIDIA gear. So that very clearly yesterday, at the AI Infra Show, Meta made a presentation. There's a nice photo of a rack with -- that's cabled out with purple cables, which are our cables. That's all NVIDIA gear. And so we're definitely not a proxy. We look at any GPU, any XPU as an opportunity. So we should be able to engage across the board.

Great. And you referenced or I referenced before the 2 additional hyperscale customers. Maybe talk about how you expect them to ramp relative to your lead customer?

Well, I think from the standpoint of the timing of the ramp or kind of the size of revenue that they can -- we can measure in different ways. And I do think that at a high level, you could tie the overall opportunity to CapEx. And if one guy is doubled the CapEx of another guy, it's hard to make the case unless we get fully penetrated. But if you say full penetration of both, I think the larger CapEx could be driving a larger revenue number for us. But as I think about any one of these hyperscalers, my feeling is that they could drive a significant revenue -- well over $100 million per year revenue with us. And even as we're approaching $1 billion run rate, kind of solidly be a 10% customer even as we keep growing in the future. So I think all of these relationships have the ability to ramp quickly and drive big numbers.

Maybe away from AECs for a moment. You've talked about the value proposition of Linear Receive Optics or LRO. From a power perspective, maybe help us understand how adoption has been the level of interest you've seen and how big this LRO offering can be in the context of your overall Optical DSP business?

Sure. So kind of shifting gears to optical DSPs. There's a really strong push in the market for better energy efficiency. And if you kind of break down an optical module and you say, how does the power look ratioed, the DSP is by far the highest power component, even half to even greater than half the total power consumption. And -- so if you make the case that for 1.6T that you hit a wall on power and that you need to do things differently, this is where the concept about LPO came from. It was a couple of years ago at the OFC conference. We kind of took away -- particularly with LPO, what you lose is you abandon interoperability. Signal integrity is not as good. Traditional test points for debug and so forth, you lose all of this. What you sacrifice to get lower power is questionable if it's even something you can really deploy in a world where you need interoperability. If you only end-to-end connection, it might be possible if you put the work in. What we came up with in response to that call to action for power was how do we keep everything you get with a full DSP, but get the power advantage of maybe reducing the amount of DSP that you're doing. And so Linear Receive Optics is basically doing instead of DSP on both ends of the fiber, you just do it on one end of the fiber, and we've proven it to be successful. But you're talking about a marketplace that is -- the way things are done are with full DSPs. And so we can think theoretically about moving, but there's a risk aversion to it. So we're doing both full DSP and LRO DSP solutions, and it's really allowing the customer to decide. If they can fit under the power ceiling and use a full DSP, they typically will. But we are having a lot of interest and how that plays out in that -- in the next 2- to 3-year time frame, we'll see. I'm agnostic, but what we're trying to do is solve problems for our customers.

Yes. And maybe just talk about the competitive landscape you see in that market, especially given Marvell's dominance in DSP and the fact that Broadcom obviously has a lot of scale to go after the opportunity if it becomes big?

Sure. Yes, a lot of respect for what Marvell has done, a lot of respect for them maintaining their incumbency, a lot of respect to Broadcom as well. When I think competitively about this space, we're really thinking along the lines of how do we move the market to more features. Instead of like saying we're going to just like stay in the cage and fight hand-to-hand combat on optical DSPs, we're thinking about how to go to higher-level ground, give the market something additional to what the market has today. And the hint that I'll drop is, it's all related to reliability. How do you improve through an optical DSP design, how do you improve reliability for laser-based optic system by 1,000x, right? And that's very much a system-level product. And so it's -- we'll talk more about that in the upcoming months. But our mindset is, yes, I can compete head-to-head. But if I can take the competition to a higher ground, I think we're going to fare even better.

Yes. Fair enough. I'm going to leave you out of the conversation. Let's talk about financials. Your gross margins, you've been trending above your long-term guidance of 63%, 65% despite lower IP revenue contribution. What's driving the strength in gross margins right now?

Yes. What we've laid out going back almost 4 years now as we went public is that we would experience an expanding margin due to increasing scale. We were maybe, you'd say, subscale 4 years ago. And that has played itself out quite well. In fact, even over just the last year, you see that pretty emphatically we went -- if you look Q1 of fiscal '26 versus our Q1 of fiscal '25, 274% year-over-year growth, and the product gross margin expanded over 500 basis points. Now we're at the point now, as you clearly point out, we are above our long-term expectation for gross margin and a scale story or the increasing scale story has played out as we expected it to. In fact, you might say we're kind of mission achieved there. If you look at sequentially from Q4 to Q1, 31% top line growth and an essentially flat product gross margin. It was up 12 basis points. But from here on out, what you might expect to see is more product mix variation. So there will be some variability quarter-over-quarter in gross margin. But we emphasize that we do not expect to change our long-term expectation. So 63% to 65% is our long-term expectation, even though there might be pockets of time like we're in right now where we might be at or above that long-term expectation.

Fair enough. And maybe talk about some of the key areas of R&D you're most focused on and whether you expect R&D intensity to go up at all?

Well, I'll speak to the dollars of it and then Bill can maybe speak to some of the activities. What we laid out for this fiscal year, for fiscal '26 is that OpEx is increasing about 50% year-over-year. We continue to add top flight engineers as fast as we can find them, and that trend will continue. The other thing I've mentioned is that over half of our R&D spend is really focusing on future programs, optical in particular. So those are the areas that we're focusing on that bears fruit 3 to 5 years from now. It's not a today thing.

Just sort of philosophically, you've -- well, let me just ask you broadly, long term, do you think Credo is going to remain a pure-play connectivity company? And do you have an ambition to diversify in other products in the future?

Yes. I envision us staying close to home. I mean I think if you were to talk about adjacencies that we could like make the case for adjacencies, you might say switching, you might say GPUs. And at this point, we feel like we can grow for a long time. And when we think about going from, say, $1 billion run rate to $5 billion and beyond, I think there's plenty of opportunities that we're going to pursue within the connectivity space. So we are looking at furthering our system level products. So when I think about that, I think of 3 tiers, and I've talked about this recently, but SerDes for us is core to everything we do. We think SerDes really matters. And we can talk about examples if we have time, which probably we don't. What we do is different and meaningful. The chips are going to be better, but the system level, like what we've shown with AECs, that system-level product has been really the key to our success as a company. And so when we think about optical investments in the future, I don't think in terms of the optical DSP. I think about building optical DSPs that enable better system-level products. And so I think you can see us going down that system-level path in multiple areas in optical. And I think you can think about us going down the system-level path, even for connectivity markets like GPU to memory. And so I think that's going to be the key to us moving the company from where we are to doing $5 billion and beyond in the future. So that's really what we're focused on. Now you might see us pursue inorganic growth in terms of the right resources, the right IP, even there may even be companies that have revenue that could be accretive. So generally speaking, we're heads up on really where we're going to take the company over the next 5 to 10 years. But we're not thinking -- at this point, we're not thinking switching and we're not thinking GPUs.

Yes. Very good. And then maybe lastly, you've met with a lot of investors today and presumably over the last several weeks since you reported. Maybe you can give us a sense about what is the one piece of the Credo story you think is underappreciated or missed or misunderstood by investors right now?

Although it's frustrating, I understand that when people think about the cable, they think of commodity. They think, well, obviously, that's going to be commoditized long term. So I don't think there's an appreciation for like the difficulty at a technology level to build these system-level solutions and really the differentiation we're bringing the advantages that we're bringing to our customers. And so I would say that over time, there's a stronger understanding. Dan is a great CFO. He should win awards for the things that he's done transforming our company. You get to do that when you're doing super valuable things with your customers. And so even when I told my mom that we were going to build cables, she said I'm very nervous. I'm worried. I have no idea what the semiconductor is, but I know what a cable is. And so getting over that perception is what I would love to do.

Excellent. Well, I think the stock price speaks volumes about what people think. So thanks very much for being here. We appreciate you coming.

Thanks. Really appreciate it.