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

Well, welcome back to the Barclays Global TMT Conference. I'm Tom O'Malley, its mid-cap U.S. semiconductors and semiconductor capital equipment analyst. Pleased to have Bill Brennan and Dan Fleming here, CEO and CFO of Credo. Thank you for being here, guys.

Absolutely.

Thank you.

Great. So why don't we start off, I think this is always very useful, particularly for companies that we cover that are bit younger. You went public around a year ago and it's been quite an adventure in the semiconductor space. But why don't you kick off and just remind us what were the goals when you were going public as a company? How that direction has changed over the first year, and kind of where your focus is today as a business going forward?

Sure. Yes. Thanks for having me. So I guess if we go back to the IPO, the real objective behind the IPO was, number one, raise the money. We've got several expensive developments in flight. And so it's important that we kind of take the money piece of the equation and take it off the table. Also, look, this now gives us the ability to grow in many different directions. Being a publicly traded company is good from the capital markets perspective, but also it's good from the standpoint that -- and this is one that surprised me a bit that I think the hurdle for customers accepting us and giving us more business. Now that we're public, they can kind of see into everything. They know where we're going to end up. We've ended up as a public company. We're going to continue to grow as a public company. Also for employees, one of the big shifts over -- going from private to public, morale is extremely high on our team right now. If you can imagine, we've been at this for a while. And if you've got an employee that's 5 years, 7 years, 9 years, and you've been working towards this crescendo of becoming an IPO, publicly traded company, we crossed that threshold. And so if we didn't go through at the IPO successfully, I'd find myself not trying to raise money in a pretty bad market. My employee morale would be tough. And my customers would probably think there's more risk in engaging with us because they're uncertain about our future. So the transformation has been extremely big in the sense that my world has changed a bit because now I have a very firm 90-day calendar. And so I've gone from maybe 1080p to 8K as far as resolution and focus. And I have to say that our team has really stepped up in a sense that execution has been really the key to being a predictable company now that we're publicly traded. So a big shot up to team Credo in a sense that our execution this year has been flawless. And that's the result of the entire team really getting more focused. So a little bit of background on Credo if I can. Credo is a high-speed connectivity company. All of our solutions are driven by the need for speed, different markets across the industry are going much faster in connectivity speed. Ethernet was really the first market where we saw an exponential speed increase, and we measure things in terms of how much data is being communicated over a single lane. And so the world has gone from 10 gigabits per second per lane to 25 to 50 to 100, and now we're even working on 200. That exponential curve becomes an opportunity, and we exist because of the need for this speed, extremely difficult technology and a big opportunity to differentiate. And so we've really emerged as a player as the lane rates have gone to 50-gig and 100-gig. Ethernet is the first market standard that we built solutions for because they needed the fastest speeds. Ethernet is used in the data center and the switching and routing layer, it's ubiquitous, and we make a solution for every Ethernet connection in the data center, copper connections, optical connections as well. You'll see us enter new markets over time as those standards like PCIe and USB move up the exponential speed curve to the point where our differentiation will deliver real market value to our customers. That's a little background on the company.

No, that's great. So I guess just looking at the need for speed in the data center, right? So could you just describe, I think most people that have looked at the optical interconnect space over a long period of time as I've really focused on the DSP market and the insides of the world, enabling the 200G ports with the TPU and then now the 400G. So you have a part of your business that addressed that. Why don't you first talk about your mission to identify that market there and kind of be the other player in that market? And then two, there's been this development of the AEC market, which is something that really didn't come into many investors' minds until really the last couple of years. So why don't you talk about that first opportunity, which may be a little more well known, and then the AEC side, which I think is really exciting today?

Sure. So Inphi and now Marvell has done a great job of being the pioneer in the DSP market for the optical modules. And so they were the first mover with Google on the 200-gig, so 50-gig PAM4, and they were the first mover with Amazon. Being the new entrant, we're the disruptor, we're the challenger. And so how do we be successful? We've got to deliver solutions that are disruptive. We've got to have great performance, great power, great price, and we have to have time to market improvement. We were definitely later than. So the time to market was there for the first products. But I think we've established the fact that from a performance perspective, we're equal or better. From a power perspective, equal or better. We're -- we have a much lower cost profile because we use more mature process technology, and we've got small die sizes. So my ability on the pricing side of the market, that's in my favor. For next generation of solutions like the 800-gig add time to market, and you'll see us, I think, become a first mover at some data center players, meaning first 800-gig solutions deployed. So I feel really good about where we are. It's taken longer naturally than expected. And that's just a result of Marvell doing a great job. And us having to learn the market and ultimately getting to the point where the compelling value that we can add or bring to the market is -- it kind of matches with where the end customers are, which are the hyperscalers. Now AEC has kind of flipped that whole conversation because we've always looked at every connection in the data center, optical as well as copper. And so for short cabled connections, for the past 20 years or so, the default solution has been passive copper, the term used as DAC, D-A-C. As you go faster on copper because of such a lossy material, there is an effect that if you use the same wire gauge, if you use the same copper wire, the distance you can travel is shortened. So if you move from 25-gig to 50-gig, the distance you can travel might be 5 meters at 25-gig; it gets cut to 2.5 meters with 50-gig; at 100-gig, it will get cut to a meter. And so we saw the market was assuming that when DACs were no longer usable when they became [ obviously ] it would be natural that the world would shift to optical connections, short AOCs. And so we started investing a long time ago, more than 5 years ago, on building an active electrical cable. And at a product spec perspective, it looks very similar to an AOC in a sense that we're doing [indiscernible] data recover on each 1 of the cable. The difference is we're not making the electrical to optical conversion and then the optical to electrical conversion on the other end of the cable. We're staying in the electrical domain. So where you find a DSP chip in an optical module, you also find a DSP chip in an AEC, so that's the only major component. If you look -- if you open these modules and looked at them, you'd see it'd be really clear this 1 is lower cost, the 1 that has 1 chip versus many chips and tosses and roses and laser drivers and TIAs in addition of the DSP. So it's clearly going to be lower cost and clearly going to be lower power because we're not making that optical conversion. And so we've successfully pioneered the market. The market is equal or larger than the optical market that we're pursuing. We are the pioneer. We're the first mover. Now we've actually invested vertically. I've got a whole team, 100 people or so that are focused every day on ADC solutions. We're a consumer of our DSP in a sense. And I think that the traction that we've seen in the market, if I compare it to a year ago when we achieved our IPO, probably one of the biggest differences in the market is that this product category is very, very well accepted. There's even market forecasters that are pegging this market as a multibillion-dollar market in the 26th kind of time frame. So we've -- in the last year, we've successfully ramped our first customer. We've successfully passed qualification with our next large hyperscaler customer. And we've got good traction with the third and fourth player that should layer in over the next 1 to 2 years.

Helpful. Yes. And it's a perfect segue into my next question where, if you look at the the semiconductor market today in many industries in tech today, it's a trying time, but there's not many that you can point to when you look at forward estimates that have 50% growth for '23. So we've talked on the optical business a bit. We've talked on the AEC business that you obviously have some IP and you have some traditional Line Card business. Why don't you talk about what's driving that 50% growth and then maybe spend some time on those other 2 businesses in case people haven't heard about them in the past. And what that brings obviously, with the IP technology across all those other product lines as well?

Sure. So the Credo's growth is definitely being driven by the AEC products. These are large-scale deployments that we're involved with, and it's quickly become more than half of our revenue. And so when I think about the next couple of years, I believe we're going to continue to grow faster than market. I believe we're going to grow faster than other semiconductor companies in the market as well because we've hit this inflection point. And we're deploying the next-generation technology for the data center partners that we're engaged with. And so I would also comment that we've got lots of other irons in the fire. We've got optical business that will be increasing our Line Card PHY business that will be increasing. But related to the IP business, it is strategically important to us. Actually, we've always been a product company. There's nobody in the company that's ever been part of an IP licensing company in the past. But we learned how to license IP when it became obvious that what we were doing is enabling and valuable in the market. And so our first contract was an IP contract as a company. Our second contract was a product contract. So we really haven't made the transition to products, but there's a perception that we are an IP company first. So the way that I view it today is it continues to be strategically important because it puts us right in the leading edge of next-generation development for the kind of chips that we connect to. So it's not like we would build the chips where we're licensing our IP. Tesla is a good example. If I can pull one example from the 40 different licenses that we've engaged in -- or 40 different customers we've engaged with over time. So Tesla talked about their Dojo supercomputer. At the core of their Dojo supercomputer is the D1 chip, the D1 SoC. They licensed our IP, they licensed our 112-gig XSR IP, and they integrated 576 lanes on the D1 SoC. When we talk about what kind of integration is that, we talk about the 51.2T switch, that's coming. That's 512 lanes of 100-gig series. So the D1 ship has 576 lanes of 100-gig series. So this is by far the highest bandwidth device ever delivered in the market. And so they also engaged us in a chiplet effort. So if you look at their tile, the D1 SoC, I think there's 16 per tile. And then if you look at the outside of the tile, how they communicate off the tile, they use our chiplets. So we built a 3.2 terabit per second chiplet that connects with the D1 and then goes off tile. So it's really an impressive technology demonstration, and we -- it put us right on the leading edge of what's happening in the future. So it's strategically important for us to stay involved and be on these leading-edge developments because ultimately, at the end of the day, we're going to sell them AECs. We're going to sell them other products that we're building as they're deploying. So it's kind of a 2-year look ahead into what we'll be connecting with eventually. As we've guided the business, we see it trending down as we grow our revenue. Of course, we see the IP part of our business trending down to the 10% to 15% of our revenue type of level.

Helpful. Yes. And I just wanted to dive a bit deeper. Obviously, that was a good deep dive into those and a good example of where you're seeing that deployed. On the AEC side, I think you've -- there's been almost a popularity of the Microsoft when you guys have announced that, it's out in the market. There's a novel deployment of the AED that you guys are providing them with something that they couldn't do else with another vendor. Can you just -- for people that don't know the story, can you explain what you're doing for them and why that's such a useful cost-saving feature and how you enable it?

Sure, sure. So Microsoft, by the way, is a great customer because they like to talk publicly about the technology, they like to contribute their innovations to the market in general. So we can speak freely about this. So we engaged Microsoft very early in looking at the AEC product category. And we kind of proved that we could build a switch rack for them, that we have the necessary ingredients to help them deploy that. And this is switch rack, they've stayed with chassis, which is fine. But what we did was we convinced them to -- we convince them that the product category was real and that there was a possibility if they invited us to collaborate with them that we could solve a problem they have been working on for literally more than 5 years if you follow their patent history. So they've published papers on the #1 failure, the Pareto chart of failures in the data center. And the #1 failure on the Pareto chart was toward court failures. And so to get the highest levels of SLA or availability, they found themselves having to deploy a dual rack, a redundant rack. So if you build a server rack with a top of rack switch, you can achieve a certain level of availability. To achieve the highest level, which customers will pay for, they had to deploy a second rack side by side [ duty ] cycle far less than 1%. It only would be active if they had a tour port failure. So it's a huge expense from a square footage standpoint in the data center, but it's also the expense of the equipment, the power it takes to run the equipment, very, very inefficient. So what they were trying to do is come up with a way to have a hitless redundancy within a given rack by going to 2 tours. To be hitless -- truly hitless like there's really complete transparency, meaning that if you have a tour port failure that the data flow is switched to the redundant tour in less than a microsecond, you really had to have a unique solution. And so they asked us, could we build this switch cable to be able to sense when a tour port was failing and switch the data less than a microsecond, completely transparent to the user? And so we kicked off this development, and we achieved the objective. And now so as they deploy server racks, each 1 has highest levels of availability and they freed up a bunch of floor space to be revenue-generating floor space. So it's -- the approach was -- surprised me a little bit because I never thought that we'd have a customer ask us to truly innovate on a connectivity solution like this. But more and more, as we talked about maybe our second customer, I think they saw what Microsoft did and said, hey, you can actually innovate with a cable. You can innovate with a connectivity solution. You can make a smart fabric in that sense. And more and more, we see folks in the data center engineering community thinking about innovation as it relates to a cabled solution. So really proud of that engagement.

So there's clearly wins you guys are getting and you're interacting and solving problems for these hyperscalers in tandem, and that's very unique. In terms of competition, there's obviously the large guy, Marvell that you've talked about. Is there anyone else that you feel like could come into this market and disrupt? And then can you just talk about the decision to do the cabling yourself as a whole versus others who would just provide a chip. That's always a debate. Like obviously, there's a cost of goods that comes into the equation when you're doing the cable yourself.

Sure. So maybe I'll go back to our first engagement. Our first engagement was with an OEM that was looking to implement this AEC solution. And at the time, we were promoting our chip to copper cable companies. And so we were going beyond the copper cable company and trying to develop the market by talking to end users, but we agreed on this first relationship. We were going to sell chips, they were going to sell cables. But what we quickly found within 10 or 12 weeks was that we were designing the entire cable. And if this were really to take off, I needed to make an investment. For this market to take off, I needed to own the solution from start to finish. And so we made the decision early on to flip the model. And as it relates to us selling a cable, it is a very deep system solution. We've got, as I mentioned, more than 100 people working in a dedicated fashion on AECs. And it just makes sense. So we went out and found our own cable manufacturing partners. And the way that I look at the decision going back 5 years ago, I think it was the absolute right decision. If we think about the relationship -- our first relationship designing the cable, passing qualification, ramping production, a service agreement that involves 24/7 engineering support involves 1-day FA, 24-hour FA on any failure, a 3-day software bug fix. These types of things, only my team is able to sign up to those things. If I were to sell my chip to a copper cable company that has -- this isn't built with the expertise needed to manage that relationship. And that's why we've seen, I think, the competition although I'm very encouraged that there's a lot of competitors talking about energy market. I feel like one of the big competitive moats that we've got is the fact that we own the system solution from cradle to grave, so to speak. And it's proven in the first 4 engagements that we've had, it's really proven to be the key as it relates to actually being able to deliver exactly what the customers are asking for. With that said, I have a lot of respect for Marvell. And I would say the only other competitor that would be capable, really shaking things up is Broadcom. And I think there's always a question mark about Broadcom because I think they see a world that's going to co-package optics sometime in the future. And so this would -- this might cause kind of the market, the markets kind of question that -- their commitment on co-package optics. So in any case, either Marvell or Broadcom, I think will have to make an investment in the system capability for them to really see success in the market.

So we've done a little bit of a deep dive into the product side. You guys are uniquely positioned to see what the market looks like from a hyperscaler perspective and then also somewhat on the telecom side, but just starting with the hyperscale world. What's your view of the health of the market right now? What's your assumption for '23? Obviously, with the secular driver, you're able to see growth in some instances, despite what the broad market does. But just your view on what you think the large guys are thinking right now? And have you seen any change in those plans over the past quarter or so?

Yes, that's a popular question right now. It seems like we're almost talking ourselves into a decrease in spending. With that said, I'm definitely not the expert on the industry spend overall. I can tell you that we're in a good spot right now because if you think about how things will play out for us at each 1 of our customers, once we achieve qualification on a new SKU that they'll deploy, we're going to go through a 12- to 24-month period of ramping before we reach steady state. And steady state will run like that on a given program, we'll run for 1, 2, 3 years and then ramp down and ramp up on a new program. We're nowhere near steady state right now. And so first customer. Second customer, we're just entering that 12- to 24-month period. So we're in a bit of a unique position because we're growing into next-generation deployments. Those next-generation deployments are not going to decrease in their strategic importance. There's a competition at the data center level to offer the most advanced technology and the most advanced services, that's not going to slow down even if spending slows down. So I'm somewhat bullish about our position as it relates to this cycle that it seems like we're going into.

Got it. Yes. I wanted to pivot to the financial side to get everyone involved in here as well. But so let's just talk, obviously, on company, but in the most recent quarter, there was some movement in the IP side, which impacted gross margins. There's obviously a transition as product side grows where IP is going to become a smaller portion of the mix. And that naturally is going to be a headwind on the gross margin line given just what IP does for your business. Can you just talk about one, why is that business lumpy on the IP side? Is that just revenue recognition? Or is there other factors involved there? And two, what do you think the run rate gross margin should look like for the business over the long term, just given the change of mix?

Sure. So from a gross -- overall gross margin perspective, our long-term model has always been 63% to 65%. And as Bill mentioned, our revenue mix in that model of IP is 10% to 15%. So if you kind of look back into the math, knowing that IP is nearly 100% gross margin, that gives you product gross margin of around 60%. The underlying trend that we've seen since FY '22 over the last couple of years, over the last 18 months, let's say, is we have been -- if you look at specifically our product gross margins, there has been an expansion story purely due to us gaining scale, and that continues through today. Longer term, there are additional levers that we see that will continue to increase that scale. We ended up Q2 of our fiscal '23 at a gross margin of about 53% product-wise. And as I mentioned, long-term model, that will trend up to about 60% due to, again, increasing scale. There's mix changes that will occur over time as well as optical becomes a larger contributor to our overall revenue mix. Gross margins on optical are pretty well known since Inphi was a very established company there. So there's a lot of different movement that will happen over the next couple of years as we're continuing to expand those margins.

And another point that's often asked of larger companies, but wanted to get your take as well. Headed into a year that you could see some weakness from the broad market. Is there areas of spend that you can be more flexible on? Obviously, from a capital perspective, from a human capital perspective, you've talked about focusing people on the AEC market. In terms of cost, how are you thinking about your spend into next year? And where is most of that being directed today?

Well, specifically with OpEx, it's a pretty straightforward equation. More than half of our spend is human capital. And that's obviously a very critical resource that we need in order to innovate on these solutions that we're delivering to the market. So -- and everything else, facilities, et cetera, are kind of a derivative of employing people, R&D lab space. So it's a relatively simple equation. One other thing I'd be able to answer on your previous question was, from an IP revenue recognition standpoint, there are some things to be aware of there because we've always said that there is a high degree of variability quarter-over-quarter in IP revenue. And that's really due to the way that we recognize revenue under ASC 606, where we have -- we might have -- any IP contract might be a 3-year commitment in terms of our resources and how we're delivering that IP to a customer. But by and large, the majority of the value of that contract is recognized at a single point in time when we deliver the final database to that customer. So really, in the old days, prior to ASC 606, which wasn't that long ago, we were able to recognize revenue more on a milestone base, which has the effect of smoothing out that revenue, but that's no longer the case.

Got it. Pivot anyway, just a broad overview to kind of end here. So you've obviously found yourself well positioned in the Ethernet market. There's a PCIe transition going on, and you also hear other players in the semi market talk about CXL. So as you're making these transitions, do you think that's cutting into the Ethernet TAM? Or maybe on a more positive note, how are you addressing those 2 changes in the market? And what can you provide from a value perspective?

Right, right. So when we talk about the network in the data center, we typically talk about the network above the [ neck ]. So the server, if we talk about that as the box, the network is really outside the box. When I look at PCIe and CXL and UCIe, which is another standard driven by Intel for chiplets, I think about the network inside the box, okay? And so if we look at where the world is going inside the box, you might have 80 lanes of PCIe running at 8 gigabits per second per lane Gen 3. Where the world is going over the next 3 to 5 years is an increase in speeds, the 64-gig Gen 6, to [ 8x ]. And because of disaggregation at a chip level within the server, the number of lanes is going to go from 80 to 400. So where we see less than a terabit per second of overall network bandwidth inside the box is growing to 25 terabits per second over the next 3 to 5 years. Tremendous opportunity for PCIe retimers as well as UCIe chiplets, which are really PCIe chiplets that connect inside of MCM package with a specific interface to the main die. So I look at this as being highly complementary to our Ethernet business. And so I see us moving from above the server to inside the server over the next 3 to 5 years. What will make us unique is our application-specific SerDes mentality. We will be lower power. We will be lower cost. We will be an enabler what I think other companies that don't have such an acute focus on the SerDes technology won't be able to offer as compelling solutions as we will. And that's the reason we're winning in Ethernet. We're going to take the same core capabilities and use that in the PCIe market.

Great. Well, with that, we're out of time here. I really appreciate you both being here and good luck for the rest of the year and into next.

Thank you very much.

Yes, thank you.