Welcome to the Credo Technology session here at the 2024 Stifel Cross Sector Insight Conference. My name is Tore Svanberg. I cover analog, connectivity, and processors, some of the other companies. For those of you who were in the previous session, it was all about AI at the edge. We're gonna switch to AI more at the core and definitely more from processors to connectivity. It's my great pleasure to introduce Credo. So from the company, we have Bill Brennan, who's the company's Chief Executive Officer, and walking up on stage right on time is Dan Fleming, the Chief Financial Officer. So the format for this particular session is a fireside chat. We will allow for some time at the end for the audience to ask questions as well.
So with that, we'll get started. Thank you. So Bill and Dan, thank you so much for coming. Great to see you!
Thanks for having us.
We usually just start off by giving a very general introduction.
Mm-hmm.
to investors, you know, what Credo's all about, especially to those people that may not know the company that well.
Sure. So, Credo focuses on high-speed connectivity. We're really a pure-play, high-speed connectivity company. The first market that we've pursued is within the data center, and it's with the Ethernet protocol. So, somewhat unique in a sense that we focus on all connections, all Ethernet connections in the data center, both copper as well as optical. We build, as you know, a full family of products that include active electrical cables for connections that would be in-rack, 3 meters or less. We build optical DSPs for our partners that are building AOCs or transceivers. We build devices that sit on line card switch line cards, and we also even sell IP. We're kind of a leader in the chiplet market for SerDes.
The core to the company is really our SerDes technology. The reason that I joined the company 11 years ago was because of the differentiation that our our team was doing even back then, related to the SerDes IP. When you think about connectivity, it all starts with the SerDes. And so if you think about any kind of connectivity solution, there's a transmitter and a receiver, a serializer and a deserializer. We're unique in a sense that the platform that we've built on SerDes IP is really probably the most application-specific portfolio in the industry. So we view every application differently, and we try to optimize around performance, around power, and, you know, of course, around speed.
As we look into the future, we're also gonna be targeting other protocols like PCIe and even USB as the edge. Typically, people have been talking about consumer, but there's gonna be great, you know, great developments that happen on the edge, and we're really a leader in that space, you know, our first contract on USB IP was for 80 Gbps PAM3, for those that want to get technical.
Great.
So I will say that for us, wind in our sails is speed and width. So when we talk about the need for bandwidth, really, AI has put that front and center. To make an AI cluster higher performance, it's all about: How do you increase the bandwidth in the connectivity between the GPUs? How do you, you know, how do you take the thousands of GPUs that are being networked together to operate as one large supercomputer? If you can feed more bandwidth, more data, you're gonna get a higher performance cluster. And so what that means is faster and wider. So we're talking about going to the fastest lane rates, which is high hurdle from a technology standpoint, and we're talking about an increased density or number of connections.
And so that's really been, you know, again, you know, we're a beneficiary. It's wind in our sails from that standpoint, and that's why we're, you know, pretty strongly associated with the opportunity in AI.
Right. No, thank, thank you for that. And to, to follow up on that, and you talked about some of the products, so, given the company's size, you're pretty unique in the sense that you do chips, systems, chiplets, licensing. Help us understand why the company decided to go that route, because, you know, a lot of other, you know, SerDes companies obviously decided to do chip only. So if you could elaborate on that, that'd be great.
You know, I think our mindset is, uh, not to limit ourselves. You know, I think we look at, you know, connectivity solutions in general, and we say: How can we, you know, deliver the best overall solution to our customers? And so, delivering half of the solution, only going halfway, I don't think the AEC market would exist if we didn't pioneer the full system-level solution. And when I think about, you know, new opportunities in the future, I don't have a limited mindset as to how deep or how far we're gonna go, in the system solution. We can talk about other things, that we've done, that you wouldn't necessarily think, you know, why is a chip company doing that? But...
That's why I think of ourselves as a solutions provider versus limiting us to just the chip.
Right. Great. So maybe zooming into some of the product lines specifically, and let's start with AEC, which is the largest segment. It's pretty interesting, you know, when the company went public, I don't think a lot of investors had heard about AECs. Even your competitors certainly weren't talking a whole lot about it. You know, subsequently, a few of your competitors have now entered into the market, right? So, can you talk a little bit about some of the inflection points that you're seeing with the AEC adoption? Obviously, we understand AI. You know, to your point, we understand, you know, speeds. But talk a little bit about some of those inflection points that are coming for the AEC market to adopt more meaningfully.
Sure.... Yeah, so I think, by the way, it's, it's, it's comfortable for me to have competition. If you're running a race and you're by yourself-
Right.
You know, I, I think that that doesn't ultimately lead to a large market.
Right.
So I'm very comfortable that others are investing now. You know, I think from, you know, the catalysts in, in this space that caused the inflection point are important to understand, and there's multiple catalysts at a technology level. You know, first one being speed. If you're talking about copper being the medium, as you go faster, the signal degrades faster over a shorter distance. And so we've talked about at 25 gigabits per second per lane, or a 100 gig port, you can really manage that link without any, without any DSPs in, in, in the, in the cable. As the market moved to 50 gig per lane, speed was a catalyst, and we see most of the market converting at 50 gig per lane to an active electrical cable solution.
At 100 Gbps per lane, for sure, that is absolutely, for short, in-rack connectivity, that's a de facto solution.
Mm.
It's really driven by speed. But another catalyst is functionality. And so what we've done with our customers is we've opened the door to innovation. And so a lot of the solutions that we're selling and developing right now are customer-specific in a sense that we're developing hardware features and firmware features that make their system solution more efficient, whether that be, you know, from a manageability, from a telemetry, from a diagnostic perspective, or from a firmware upgrade perspective, or just in mixing and matching hardware with... You know, we've got cables with two connectors, three connectors, four connectors, even five connectors. And so functionality has been an absolute innovation at a spec level, has been probably the largest catalyst, you know, beyond the fact that speed is going to push us there anyway.
Right. And, you know, unfortunately, you know, we investors, we tend to look at the world a little bit black or white.
Mm-hmm.
And it's pretty interesting how, if you look at this year, whether it's GTC or OFC, all of a sudden, active copper cables consider like the new best technology, technology that's been around for a very, very long, very long time. And, you know, as, as we know, it's not always about ACCs versus AOCs versus AC. Sometimes you need all of them. But, you know, maybe you could talk a little bit, because obviously, you talked a lot of investors. Why, why do you think there is sort of this obsession with the ACC market right now, and how, how do you see that evolving, especially when we start talking about bigger deployments at 100 gig per lane and especially eventually, 200 gig deployment?
Yeah, so I would say that ACC and LPO are equivalent.
Right.
Both depend on a world with no DSP. And I think that in the optical space, you know, coming out of OFC this year, I think it's pretty safe to say that the momentum around optical solutions with no DSP is gone. I don't think anybody. I think Andy Bechtolsheim himself, you know, basically, he was the guy that started the fire two OFCs ago-
Right.
and he kind of put the fire out this year and basically said that the idea of an optical solution with no DSP is really a very small percentage of the market. ACC is very, you know, very analogous to that. It's building, you know, building a copper cable solution with no DSP. And I don't see it as a big market. I think, you know, if it exists, it's gonna exist in a very limited way, and it's got to be with a company that is gonna take full ownership of the end-to-end connection. And so NVIDIA is the place where, you know, people talk about ACCs, and that's because they literally own the link, both sides.
Right.
And so if you want to do the work, you can make it work at 100 gig per lane. But I think even if you talk about 200 gig per lane, I don't think there's a future.
Right.
I don't see ACCs anywhere else in the market.
Right. Just one last question on ACC before I move on to PAM4 DSP. So, and this goes back a little bit to the system-level product. Some of your competitors are selling only DSPs, others are selling, you know, these sort of paddle board-type solutions. Based on your conversations with customers, you know, even longer term, where, you know, what's, what's the way that we should think about being the most ideal solution? And, and where are some of the bottlenecks, you know, when working with some of the cable manufacturers if you're only selling chips?
Yeah, I would say that we're really driven by delivering a good experience with our customers. And, you know, having the conversation with one company feels right. Like, if you want to buy a solution, you buy it from one party. You don't call one party and say, "Hey, would you work with another party and deliver something that I would like, but I don't want to buy this from you. I want to buy that from you, but I want..." Like, who's going to be responsible if there's a problem? You know, who's going to be responsible, you know, for a schedule delay or whatever the case comes up. So it feels more comfortable at a customer level to have a single, you know, company that you're dealing with.
And so I can say that I'm very comfortable that, that, you know, the relationships that we have, and I would say, you know, we probably have 10 different customer relationships, it works. We're delivering a good experience, and, you know, I think that, I'd be really scratching my head if I were, you know, going to an end customer saying, "I've got a chip that I'll sell to another company, and promise we'll put a system together quickly, and, and, and it'll be, it'll, it'll be, I'm not responsible for the end system."... So if you have a problem, don't call me.
Right.
You know, anyway, I just think the idea, I don't know why there's so much focus on this particular product. Where, say, in the optical module world, this conversation doesn't exist.
Right.
Right?
Yeah.
Pick another, pick another product, pick another solution. This whole disaggregation of the supply chain for a single solution just doesn't happen anywhere else.
Right.
So I'm really happy with where we are. From a competitive level, we're trying to raise the competitive bar every day. How can we be faster? How can we be more dynamic in our innovation? How can we be qualified first, and how can we deliver flawlessly? If we do those things, this conversation about model is going to disappear.
Well, that makes sense. So moving on to the PAM4 DSP market. So here's an area where, you know, you're not the pioneer, you're actually coming from behind. There's one company that has very dominant share, has had a dominant share for a long, long time. What's going on here, and why do you see new opportunities for Credo? I mean, I understand the 800G market, of course, is big and, you know, that probably requires more suppliers. But, other than, you know, multi-sourcing, you know, what are some of the reasons why Credo is starting to see success in the PAM4 DSP market?
Yeah, so I think, first of all, in any market that we talk about, the end customers that we're serving, they want multiple sources, right? You know, the conversation about AECs, right, there's competition. They want competition as well on optical DSPs. Now, we can deliver excellent solutions, and excellent can be measured by the performance, the signal integrity, the power, the pricing that we're able to achieve. And so naturally, you know, there's been momentum that we've developed over time as our solutions have become, you know, better and more perfected.
But I will mention, you know, a couple of OFCs ago, when the world was talking about a no-DSP world, the message that we heard was, "Power is more and more important now." And so we looked at the solutions that we developed and said, "How could we significantly reduce the power?" And so we immediately started working on what we call the linear receive optics, basically, doing a DSP just on the transmit side of the link. And we thought if we could show that doing a single DSP instead of two within the optical module, if we could deliver the signal integrity, we could maintain industry standards, we could maintain IEEE compliance, we could maintain interoperability. It would be transparent to the user.
It would just be that you could get an optical module at an 800 gig level at 10 watts versus 15 watts. That would be directly addressing the call to action for lower power. And so we really set off on this innovation to redesign our DSP. Really, we moved into a space that was unoccupied. And it, you know, one year later at OFC, we're showing module designs with 3 different partners, showing, you know, actual evidence of the promise. We had an 800 gig module there that was less than 10 watts, and we had two others that were just a little over 10 watts, AOCs versus transceivers. And so I think we're out first, right? And you know that it's a good idea when Marvell announces a solution at OFC.
They call it TRO. They I guess that's unique, but it's the same functionality. And so the idea about the market wanting lower-power solutions, it's real, it's front and center. And when we think about our next-generation design, you know, previously, we've been talking about doing more with less, using older process geometry to be more competitive in the market. So if you can deliver better performance, equal power, and do it with older technology, it puts you in a much more competitive space. We've shifted now. We're making decisions solely based on power. So we're moving directly to 3-nanometer for 200 gig per lane solutions because of power. If we look at the 5-nanometer solutions in the market, these are 16-18-watt solutions.
You know, I just said 800 gig module at 10 watts, and the DSP for a 1.6 T module is 16-18 watts. You know, that is not going to fly.
Right.
Right? And so that's the reason we decided to go to a more advanced process and deliver a 10-watt DSP and a five-watt LRO DSP. So power is really, is really driving our decision-making now. And that's, again, this is along the lines of innovation.
Right
... and differentiation. From a competitive space, it's how can we make our customers more successful using our solutions.
Right
... than they would be with our competitors?
Right. So I'm going to leave the Line Card PHY discussion to another time, but I wanted to wrap up on the product or the segment with chiplet and licensing. So you have two really big customers today. One's a consumer one, one's a sort of AI auto one, AI/auto. But what's very interesting is that you're actually assuming that this segment will remain 10%-50% of revenue longer term. That would imply that that's actually a really strong growth business, because you are a strong growth company. So help us understand a little bit what's going to drive that growth. Is it going to be those two customers and their continuous innovation with speeds and interfaces, or are you also working with potential other customers?
Yeah, so we're, if you look at our funnel, so to speak, of opportunities, it's growing.
Right.
The number of designs that are being done at very high speed, just from an IP licensing standpoint-
Yep
... and the number of companies that are considering chiplets as a way of, you know, piecing together the overall solution that they're working on, that number is growing. AI is driving a lot of it.
Right.
Right? So if we look at our funnel, it's you know, a wide range of companies. Some of these companies you've never heard of-
Okay
... you know, but they're well-funded. And you know, Arista is absolutely critical, fundamental to the success of the company.
Right.
We feel like we can maintain, you know, this percentage for this upcoming year. But over time, look, when we're $1 billion in revenue, I'm not going to be, you know, sitting here talking about 10%-15% being IP licensing. That market is just not that big.
Right.
So it will transition over time as we continue to grow. But I think, you know, as we talk about our fiscal 25, I, I think we're, yeah, we're absolutely bullish on the space.
Great. So maybe shifting gears and talk a little bit more about the near term. You reported last week, notably, you talked about 70% of your revenue is now being tied to AI deployments. You're expecting that particular business to, you know, grow 100% year-over-year. Can you talk a little bit about some of the drivers there? You did mention a few things on the call, but I still think there's some confusion out there with investors because they're trying to do the math, right? And adding it all up. So, I don't know, Dan or you, Bill, if you can clarify, you know, what's really behind the 70% and how does it double?
Yeah, what we said was, three-quarters of our revenue in Q4, and year-over-year, Q4 to Q4 of our fiscal 2025, you know, April-ended next year, the AI piece will grow 100%. And that's really built upon, you know, our, our bottoms-up forecasting. We have one of our, emerging 10% customers, which you've seen in the last couple of quarters, grow as a percentage of total revenue. They—we expect them to continue to grow. That's a—and that's largely driven by specific AI programs at that company, that hyperscaler. So that's, that's a, that's a big piece of, of how we communicated.
Mm-hmm.
that 100% year-over-year growth. But there's other things as well. Talk about the IP pipeline, as Bill was mentioning. Largely, the growth in IP, you know, exclusive of the, the USB contract that we have, that's driven by AI, AI opportunities. 100 gig lanes at Line Card PHYs, that's driven by AI opportunities, ultimately, whether we're selling to Arista, Cisco, or to whomever.
Mm-hmm.
So, boiling it down more simply, if you think of the hyperscale universe right now and what they're spending their, their CapEx on, it's largely AI deployments.
Right.
So it kind of makes perfect sense that, that would be driving our growth at this point in time.
Right. So, to make it simpler for us, then let's just look at the 25% versus the 75. So the 25% would be your consumer licensing, licensing customer-
Correct.
It would be the compute side of your AEC business, and then maybe a little bit of Line Card PHY. Is that how we should think about it?
Yeah, that's, that's fair. The two largest pieces that are not AI is the USB contract-
Right
... and then a portion of our front-end network deployments that we're going into.
Right. Great. So Bill, back, back to you, and you know, this is related to some of this growth that's coming. I think the one thing that investors are always nervous about is how long this CapEx cycle from the hyperscalers are going to last, right? So, you know, I think some people said, "Hey, it's going to last at least three years." Some people are saying, "Hey, you know, maybe by middle of next year, there's going to be an issue." What's your view based on conversations you've had with your customers? Because you're now talking to all of them, right?
Sure. We're talking about, and when you talk, when you're saying all of them, you mean the five US hyperscalers?
Yes, correct.
Yeah. You know, I think that I don't have any wisdom to add on CapEx cycles. I know that the energy that we're all feeling. We're at the early innings of this playing out, and there might be some, you know, variability in the CapEx number. But I think, you know, generally, I would say that I think the customer space is going to become even more interesting than the five. Just this week, there's been talk about X and xAI-
Mm-hmm
... and what are the plans on CapEx there?
Mm-hmm.
We're very well positioned. We've talked about Twitter for a while being one of these, the examples of customers that naturally made the decision to use AECs. So I think, you know, you can talk about that. There's other companies that are being funded by different, you know, entities globally that are talking about large CapEx numbers. So I think this space, I think, doesn't have bounds.
Mm-hmm.
I don't think it's limited to what's the CapEx cycle for Microsoft and Amazon. I think we're kind of moving past that as an industry.
Mm-hmm.
You know, and so I think the bullishness that we all feel, I think it's, you know, I think we should be thinking that way.
Okay.
I think that we'll see that the industry even grows in areas that we're not really including in our thinking right now.
Mm-hmm. Mm-hmm.
When people talk about surging and spending and building capabilities, I don't think it's limited to the top five anymore.
Okay. No, that's fair. Maybe drilling down to some very specific questions, and I mean, I don't know, Bill, if you can even comment on this, right? But yesterday, there was a lot of back and forth with your largest chiplet customer at their chip strategy or GPU strategy. I assume that's not anything that impacts you from a Credo's perspective, right? I mean, it's kind of irrelevant. I just want to hear from you if that's the fact.
Yeah, so I can't say that we know exactly what their plans are. You know, I can tell you that I do understand the sense of urgency.
Right
... on the next generation solution that we were chosen to provide. And so I think we're very well positioned with the first generation. We're very well positioned for the second. I don't think there's going to be a slowdown in the investment... internally-
Right
You know, within this, AI is critical for full, full self-driving.
Right.
But I think what you might be seeing is that if you've got capacity with both available and you shift certain capacity over to, you know, another one of your companies to go build out a vision that you've got for that company, you know, I think that would be a kind of natural conclusion.
Right.
And so, you know, for us, we're well positioned at both.
Right.
So I think it's a benefit for us.
Got it. The other more specific question that I had is on LRO. I think we're now maybe 8 months since you announced that product. Obviously, there was a lot of discussions at OFC. I think the one concern investors have is that, you know, this could potentially be LPO. And I know they're two completely different architectures, right? But, you know, could this be LPO all over again, where, yeah, you know, great, one retimer or one DSP, but you know what? It's still not good enough. We're still gonna need two. So, based on your discussions with customers, what's the viability of LRO, and when could we start to see some early revenues from that product?
Yeah, so I think, you know, ultimately the market will decide.
Right.
Right? The market has decided on LPO. The reason that there's a lot of momentum right now is, you know, from announcing the product in November to demonstrating with 3 partners at OFC with really solid solutions, and solid meaning, you know, right error rates, the interoperability, IEEE compliance, all of the reasons that LPO is not gonna take off. There is no pitfall, there's no shortfall with an LRO solution. It's really about power.
Mm-hmm.
You know, it's about power, and the bottom line is we're gonna deliver full DSP and LRO-
Mm-hmm
... for 800 gig we've delivered already, and for 1.6 T. Ultimately, we're trying to enable the market, and the customers will decide. What we're trying to do is give them optionality. What, what I've seen is that for the 800 gig market, just since OFC, we've expanded the number of, of design engagements that we're involved in, so it's more than a handful now that are, you know, full steam ahead. End customers are now open-minded, but they wouldn't even think about LPO. I mean, they, you know, it, it's somebody else's idea, not theirs. They didn't want anything to do with it, but they're very open-minded to LRO. It doesn't look any differently. You plug it in, you use it the same exact way. There's nothing special that you need to do.
Right.
It's just that you save 5 W if you use this versus something else.
Right. That makes sense. So I know we're running out of time. I just want to make sure I address the audience to see if you have any questions. Robert?
The view on, sort of the sustainability of kind of more DSPs inside the data center, you know, we're seeing interface speeds with very fast data. So you kind of they might see coherent, coherent like DSPs.
You know, there may be a time in the future, but I don't think it's soon. So I think, I think the existing ecosystem will definitely be extended.
Yeah, so just to repeat the question, it was a question about coherent technology penetrating inside the data center. And yeah, I would agree with Bill that based on what we heard at OFC, we're still several years away from that potentially happening.
Right.
Right. Any other questions? A question right here.
Yeah, the LRO, does that shrink the DSP TAM or does it grow?
I'll, I'll repeat the question for the webcast. The question is, does LRO shrink or expand the PAM4 DSP market?
You know, I would say that, there's not any, you know, clear reason why it would shrink the market. You know, what the solution would deliver is lower power. Now, if you design it and you optimize it for size, you optimize it for cost, there's an opportunity to improve compared to a full DSP. But I don't think, naturally tying it to TAM size is, a good conclusion.
So, does that mean it actually eats into ACC or AOC? Because, I mean, you know, if you're going from two DSPs to one, I mean, theoretically, it's a lower-cost solution.
Well, it's still one chip per module.
Okay.
You know, it's not like, in a given module, a DSP is being done on both directions of traffic. So you know, there's bidirectional traffic in each module. And so what we're talking about doing is still, there's a single chip, but we're doing a DSP just on the transmit side in the module. So it's not gonna decrease the units for sure.
Yeah.
And-
And I guess from Credo's perspective, I mean, this is, this could be a $2 billion market, and you know, you haven't gotten that much share yet, right? So I mean, it's, it's a great opportunity for the company.
Absolutely.
Right. Right. Great. All right, thank everyone for coming to the Credo session. Enjoy lunch and, Bill, Dan, and Dan, thank you so much for coming to our conference.
Thank you.
Thank you.
Appreciate it.