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

Okay. Good morning, everybody. Welcome to the Goldman Sachs Communacopia Technology Conference. My name is Jim Schneider. I'm the semiconductor analyst here at Goldman Sachs. It's my pleasure to welcome Credo Technologies' CEO, Bill Brennan; and CFO, Dan Fleming. Welcome, guys to the stage.

Thank you.

For investors who may be new to the Credo story, how would you describe your positioning in the AI infrastructure layer? And what are your biggest and most defendable competitive advantages?

Sure. So Credo is a pure-play high-speed connectivity solutions company. The role we play in AI infrastructure is moving data reliably and efficiently across the network. And specifically, our business has really taken off as AI clusters have taken off. Given the fact that we connect GPU to GPU, GPU to switches, switches to switches. We play that role. We help customers optimize the valuable compute resources. So our North Star is really reliability and power efficiency. And if we can deliver solutions that can help customers get clusters up more quickly in days versus weeks. There's a compelling financial impact to that. And if we can keep clusters up at 99.9% availability. There's also a huge return on that. We focus on a broad range of solutions from millimeter length all the way to kilometer length. Our flagship product is really a system-level solution where -- if you think about us as a semiconductor company, we have very unique and differentiated leading-edge SerDes IP. With that platform, we build semiconductor solutions. But beyond that, we have gone to the system level. designing complete connectivity solutions. And you'll see that more and more as we develop as a company. One of the things that for the new people to the story, incredible story during the last 3 to 4 years. If we go back to our we were sub-$200 million in revenue. And over the past 2 years, we've more than doubled and then more than tripled. We actually had 6 quarters in a row where we achieved more than 20% sequential growth over that period of time. And that puts us at a base going into fiscal '27 at more than $1.3 billion. We've indicated that we're going to grow at a 10% or greater sequential level quarterly throughout the year. And I think as we look at the markets that we address now, you can measure those markets, that market opportunity and tens of billions of dollars. And so I think the growth story, although we've had an incredible run so far, we're still at the early stages -- the growth story.

Yes. Fair enough. Now there's been a lot of investor at round this into kind of like the idea of copper versus optical. I realize that's oversimplifying things here a lot, but I mean how would you frame it for people in terms of the time line for the transition between the two? And at what point do you think copper starts to hit these physical limitations?

Yes, I think -- I've been in this role for 13 years. And the first time I got that question was about 12 years ago. As I talked about Credo being a company that was going to develop connectivity solutions for both copper and optical. At that point, the number of copper connections in the data center was greater than 95%. And today, the same ratio exists. With that said, we've been very successful extending the life of copper over the last 5 or 6 years. we've grown into what is going to be a multibillion-dollar revenue company. But our investments over the last couple of years have been almost exclusively from a real innovation and disruption standpoint in the optical area. Customers will decide on what type of connectivity solutions they use based on the requirements of that connection. For short connections where copper can be used, the debate is over. copper will be used because it's fundamentally more reliable, more power efficient and a lot of times lower cost. When you've got connections that are longer than you can achieve with copper, of course, optical is the is the default choice. With that comes opportunity because with laser-based solutions, there's a fundamental reliability challenge as compared to copper, which is probably 100 or 1,000x more reliable. And with reliability becoming much more important as AI clusters become a much bigger part of the market and considering that the connection between the GPU or Nick and that first switching layer that there's no redundancy. And now you've got tens of thousands of GPUs all interconnected together and dependent on each other. There's an increasing requirement for better reliability. And so and that's our north star, right? That's our differentiation and get into that more. But as it relates to the debate, I think -- we believe it's going to be heterogeneous. We've got a complete portfolio of solutions from millimeter to kilometer from copper to optical. And even with optical, we've got different flavors of light sources, micro emitters as well as laser based. So we're agnostic to the conversation, although some people identify us as a copper company. We self-identify as a connectivity company that is agnostic to the medium.

Yes. And maybe to that point, if we are back on stage and again in 5 years, what is the thing that you think investors are going to be surprised at looking back?

Well, I think there will be a very strong understanding at that point that we're a broad-based connectivity company. that just as I emphasized, you'll see us doing very well with products that are laser-based, reliability differentiated. You'll see us with different light sources. You'll see us with copper solutions as well. You'll see us as a company that is not only addressing the pluggable market broadly. You'll see us playing a role in a big way and scale up networks. And you'll see us being a company that's helping to break the memory wall, the fan-out, the physical fan-out issues that exist and kind of a more popular topic in the upcoming years. I think you'll -- 5 years from now, I think people will be surprised that although we scaled tremendously over the last 3 or 4 years, that's really the beginning of the story. And that 5 years from now, I expect to be a much larger company.

More shorter term, if you look at investor expectations, for your business heading into 2027 or calendar '27, whichever fiscal calendar. If you were to outperform, what would be the areas of upside for the business as you see it? And maybe talk about the one or two operational kind of key drivers for the company that you're using to kind of execute or you're driving the company towards?

Yes, I would say, number 1 is execution. And if you look at the path that we've been on the last few years, execution is, by far, the most important thing. Scaling like we have as a company and broadening our portfolio and making sure that we deliver the same level of quality, reliability, responsiveness to customers, it's really an execution game. And in the upcoming years, that's going to continue as the #1 key priority. The -- I would say the ramp of our optical portfolio is probably next on that list. We're in the very early stages. Of course, we've been an optical DSP component supplier for quite some time, and that business is going very well. We acquired a team dust with Onex that focuses on leading in silicon photonics picks that will broaden our components portfolio, but also, we're going to be in the module is we're going to build 0-flat optics, which are a very differentiated optical transceiver solution that addresses the -- that reliability challenge. So I think that -- I guess the question was about outperforming?

Yes.

Yes. So I think I wouldn't be surprised if we were to outperform, I think you could see contributions from expanding our footprint with our core business today, which is AECs, expanding in the application area on the customer front as well. deepening existing partnerships, expanding in the neocloud area. I think a faster ramp, and we got to be careful about the time frame we talk about, but I think there's a great opportunity to look at what we're doing in the optical front and see outperformance over a given period of time. And really with the new efforts that we've got with micro emitter technology with AC as well as the memory well, we'll -- in the fiscal '28 time frame, we're going to see those products take off, and there's a great opportunity to outperform there as well.

Yes. Okay. maybe let's talk about the optical business for a second then which you just raised. What is driving the customer adoption of your zero flat optical products today how is that customer aperture broadened over like last few quarters? And how do you expect the customer base to sort of broaden further across hyperscalers, neoclouds and enterprises over the next year or so?

Sure. So maybe a little bit of background on how we landed on this idea. And this is all public information that we've talked about. So XAI was a customer that we paid a lot of attention to before they actually started changing the competitive dynamics in the industry as they are today. And they were building in another data center at the time, and they were struggling mightily with Link Lapse. And Link Lapse is a case where you've got a connection that doesn't hard fail, but it disconnects and reconnects in effect, flaps. And it really screw things up from a data standpoint because you get that data that's propagated across the cluster, needing to bring the cluster down and reset it. So they were complaining about their ability to bring a cluster up quickly. they were complaining about keeping that utilization or that availability above even 90%. They weren't able to achieve that. And so they came to us and said, "Look, we're moving into our own data center. It's going to be liquid cooled. We're going to source tons of power to the racks. It's a completely different scenario than the air cooled kind of small power sourcing that they were dealing with." And what we want you to do is extend your AEC family from 3 meters to 7 meters because if you can do 7 meters, we can cover every connection in the row and we can build a 0-flat cluster. And so in this case, back to the litigation of our copper and optical. The only transition between the 2 has been copper replacing optical or length of the 7 meters. And so the bottom line is I visited Memphis over the summer. And even the folks that I was meeting with, they were talking about they could bring a cluster up in 5 days, if you can do that versus 6 to 8 weeks, we all know what the value of a month of this compute is worth. Again, you can measure a month in $1 billion increment from a revenue standpoint. So it's compelling, but they were saying they've achieved that. They've achieved 99.9% uptime. After we made this transition, Oracle came to us and said, "Look, I mean we'd love to address these problems as well." But our connections between the NIC and the first switch are longer than 7 meters, and we can't redesign. I mean we've got what we've got. And so we went to work with them thinking about how could we identify Lincoln stabilities early and then mitigate in an orderly way by taking likely link flaps out of the cluster, take that link down in an orderly fashion, so it didn't disrupt the whole cluster. And so some of the work that we've done with Microsoft maybe going back 5 years ago, where we designed a cable that was smart enough to be able to identify when a tour pork was likely to fail and when it was likely to fail, transitioning basically, changing the data flow to a redundant tour and help them build a single rack architecture that achieved Five9's reliability. What was involved with that was a huge amount of effort on designing a device that was capable of doing the telemetry required to sense and make decisions and act in a cable solution, really, really smart solution. So we kind of leverage that background. And we started thinking about, well, if we did design a custom DSP that could light up telemetry on a continuous real-time basis for each link between a GPU and a switch, which there's actually 6. There's 3-in-1 direction and 3 on the other. To do that, it was a very special DSP design. And along with that, we had to develop a software platform. And the types of things that we're measuring on a continuous basis are very deep technical measurements related to the surging season. So going back to that being an expertise of ours. So things like i height, S&R, post-tech bid air rates, histograms, all of these things are what we're able to do and ultimately creating a yellow almost check engine light for the link where previously we've been green or red. Basically, it's either good or it's failing. In addition to that, they said, "Hey, when we're bringing a cluster up, it's tough for us to sense dust on the fiber plant. Is there something we can do to identify multipath interference?" Because you've got a purple of dust, like will actually bounce back the other direction and create this signal integrity degradation. And so being able to identify which links need to be addressed, even ESD demos, not the type that would actually fail a transceiver but would become a latent defect that would become a problem later. So we actually went way beyond way beyond what the original ideas were. And so now they -- we've brought a product to market that not only can bring clusters up in the same kind of time period days versus weeks is not playing the cat and mouse game of trying to identify which connections are failing. But really, we've accomplished the goal of being able to identify instabilities and mitigate proactively. That's the reason that the conversation is a rich conversation with every customer that we talk to, and I think that if anything, the number of conversations are expanding, and we've given an indication about the very fast ramp that we expect this fiscal year. And I think that's just the beginning of the ramp.

Yes. Okay. Then maybe ALCs talk about that as a driver for you. What feedback have you been hearing from customers on that front? And sort of are deployments kind of moving from evaluation phase to production faster than expected? And maybe talk about the kinds of customers you're seeing the strongest demand signals from?

Yes. So why are we doing this? It's interesting because if you just consider the reason that people love copper, it's just bulletproof reliability at a core technology level and very power efficient. A micrometer technology, we've been looking at working in the space for 3 years now because at a core level technology, wide and slow can get that base level bulletproof reliability and have much better power efficiency than laser-based optics. And so that was really the reason that we started investing 3 years ago. And I will say that what's compelling about the technology is also that it will reach 30 meters. So we're going from, say, a 7-meter reach with AECs extending from 7 to 30 with an equal level of reliability and power efficiency. That's the promise of the technology. And again, I think it speaks to the fact that we're agnostic on medium, we're agnostic on light source. What we're trying to do is give our customers optionality and let them choose. So the conversation is quite easy with customers, especially another benefit you get from the technology is that actual cable that's 75% smaller from a total volume if you look volumetrically. So it's a much more elegant solution. If you look at these 7-meter AECs, they're pretty bulky at the speeds at 200 gig per lane. And so the conversation is quite easy. You've talked about the promise of the technology and the customers know that we're going to put these products, these system-level products, these cabled solutions through the same rigor that we've put our AEC through and that we've put our optics though. We're talking about a rigor that we've built, the qualification capability, where we take our customer switches, we take our customers' mix. And we run live traffic, and we very over every corner imaginable. And the whole goal of that process is to break the link and then strengthen the link through firmware. When we deliver the solution to customers, they will simply plug it in and qualify it. And so the interest in the product is very high. And we'll -- not to get too ahead of the announcements, but we'll be demonstrating an OCP next month. And from there, we expect to go through qualification and then be in production in FY '28.

Okay. Excellent. Then if you sort of roll up all the things we talked about with optical together, based on customer engagements, your visibility, how do you encourage investors to think about the growth rate specifically for optics as we progress through fiscal '27 and into '28.

It's my expectation that I would like investors to have is that it's going to be a very fast growth rate. I've indicated previously that by the end of this fiscal year, we'll be producing hundreds of thousands of units per month. And then we will increase that production rate to double and then even triple by the end of fiscal year '28. So the ramp is going to be quite fast. It's going to be quite significant. And we expect to even go beyond that in fiscal '29.

Yes. Okay. SP1 Maybe talk about AECs for a second, sort of the core bread and butter business. What are you seeing in terms of the near-term AEC demand across customers or kind of the deployment time lines, purchasing patterns and visibility changing in any way as sort of the customers move to larger and larger scale infrastructure builds?