SteveSchiets
11 months ago
https://capedge.com/transcript/1807794/2023Q4/CRDO
Bill Brennan
Sure, sure, absolutely.
So I think generally, I think that AI applications will create revenue opportunities for us across our portfolio.
I think the largest opportunity that we'll see is with AEC.
However, optical DSPs, there will definitely be a big opportunity there. Even linecard PHYs chiplets, even SerDes IP licensing will get an uplift as AI deployments increase.
So maybe I can start first with AEC.
Now it's important to kind of identify the differences between traditional compute server racks, which is kind of commonly referred to - use the front-to-end network, so basically a nick [ph] tour connection, the tour up to the leaf and spine [ph] network.
The typical compute rack would have 10 to 20 AECs in rack, meaning in rack connections from nick to tour. And you highlight the leading-edge lane rates today for these connections with compute servers is 50 gig per lane.
Within an AI cluster, in addition to the front-end network, which is similar, there's a back-end network referred to as the RDMA network. And that basically allows the AI appliances to be networked together within a cluster directly. And if we start going through the map, this back-end network has 5 to 10x the bandwidth as the front-end network.
And so the other important thing is to note within these RDMA networks, there are leaf spine racks as well.
And so if we look at the - if we look at one example of a customer that we're working with in deploying, the AI plant track itself will have a total of 56 ADCs between the front end and back-end networks. Each lease fine rack is a class track or at this aggregated chassis, which will have 256 ADCs.
And so when we look at it from an overall opportunity for AEC, this is a huge uplift in volume. The volume coincides with the bandwidth.
Now lane rates will quickly move certain applications will go forward at 50 gig per lane others will go straight to 100-gig per lane.
And so we see probably a 5x plus revenue opportunity difference between the typical - if you were to say apples-to-apples with the number of compute server racks versus an AI cluster.
So it's kind of extend of extend into optical. There is also a typically large - there's typically a large number of AOCs in the same cluster.
So you can imagine that the short in rack connections are going to be done with ADCs. These are three meters or less. But these appliances will connect to the to the back-end lease spine racks, these disaggregated racks, all of those connections will be AOCs. Those are connections that are greater than three meters.
And so if we look at this, this is all upside to, say, a traditional compute deployments where there's really no AOCs connecting rack to rack.
Okay? So when we look at the overall opportunity, we think that the additional AEC opportunity within an AI cluster is probably twice as large as twice as many connections as AOCs, but the AOC opportunity for us will be significant in a sense that AOCs represent the most cost-sensitive portion of the optical market.
And so it's also a lower technology hurdle since the optical connection is well defined, and it's within the cable.
So this is a really natural spot for us to be disruptive in this market. We see some of our planning on deploying with 400-gig AOCs. Others are planning to go straight to 800-gig AOCs.
So we view - AEC is the largest opportunity. Optical DSPs for sure will get an uplift in the overall opportunity set.
But also, I think that if we look at Tesla, as an example, that's an example of where, as they deploy, we're going to see a really nice opportunity for our chiplets that we did for them for that DOJoe [ph] supercomputer. And it's an example of how AI applications are doing things completely differently, and we view that long term, this will be kind of a natural thing for us to benefit from.
We can extend that to SerDes IP licensing. Many of the licenses that we're doing now are targeting different AI applications. And also, don't forget Linecards, the opportunity for the network OEMs and ODMs is also increasing. And of course, Linecards are something that go on those switch line cards that are developed.
So generally speaking, I think that AI will drive faster lane rates. And we've been very, very consistent with our message that as the market hits the knee in the curve on AI deployments, we're naturally going to see lane rates go more quickly to 100-gig per lane. And that's where we really see our business taking off.
So we're getting a really nice revenue increase from 50 gig per lane applications, but we really see acceleration this 100-gig per lane happens. And especially when you start thinking about the power advantages that all of our solutions offer compared to others that are doing similar things. Does that might have been more than you were looking for, but...
SteveSchiets
2 years ago
On AEC cables; These analysts will (BoA, GS) soon know there is also something like ACC cables. Credo expecting a product mix of 50% AEC revenues in the near future. That pretty big news for Spectra7 also, imo.
Vijay Rakesh
Yes. Hi, Bill and Dan. Great quarter and guide.
Just a quick question on the mix of revenues on the product side, when you look at the 4Q and 1Q, if you can use the split of AEC versus optical and Line Card?
Dan Fleming
Yes, thanks Vijay.
So, we don't -- unfortunately, we don't disclose specifics when it comes to our product line mix, but from our from a long-term perspective, the expectations that we've set out where we're trending toward those amounts, which longer term, we would expect AEC to be nearly half of our revenue. Again, IP, right now, we're not quite at the long-term model, of course, so we're slightly above the 10% to 15%. And then the Line Card business has been a very, very strong business for us, and it continues its strength and is growing. Longer term, it'll probably be in the 15% range. And then the balance, optical will be a significant contributor this year, especially towards the end of the year and then even more so next year.
So, hopefully, that gives you just kind of broad strokes, the direction that we're headed.
Vijay Rakesh
Got it. And on the AEC side, I know you guys have a pretty broad portfolio of MACsec and phase shift, and speed-shift and Nick-to-ToR, et cetera. I'm just wondering when you look at the company landscape and given your broad breadth of portfolio and you're qualified many of the enterprise hyperscale guys, what is the competitive moat? I mean, you guys think you have a 12 to 24-month or even longer lead-time when you compare -- when you look at the competition, who I would say, most of them are still trying to get their product in place, but if you can give some color on how you look at that space?
Bill Brennan
Yes, great question.
I think that -- first of all, I'll say that we're very happy that we see competitors validating the AEC product category. We feel like we do have a significant bleat [ph] in a sense that we've been working on this product category for -- going on four years. The approach that we've taken, I think, is unique in a sense that we have built an organization internally at Credo that's responsible for the design, the development, and ultimately, the delivery of the AEC System Solution. That means that at the end of the day, we're responsible for the entire AEC System Solution with a single throat to choke in a sense when we're dealing with customers.
Our manufacturing partners do a great job in what they do well. We had started by thinking that we could sell chips to copper cable companies, and it just simply was very clear very quickly that our approach is going to be more effective if we owned the entire system design from the firmware -- to the holes from the firmware, on the copper, to the test program development to the actual tester design and development. We felt that that was a much stronger approach than trying to rely on others to put all this together.
So, I do think that as we look at our business now, it's really taken shape in a sense that it's different than our original thinking, which was, hey, we'll just put together a 400-gig on each end of the cable -- 400-gig connector on each of the cable. The solutions we're delivering in volume right now are unique. These are truly system level solutions that are -- I think would be -- could be classified as the most advanced connectivity solution -- cable connectivity solutions ever delivered to the data center.
If you think about what we're shipping currently to our first hyperscalar, they were successful in deploying a dual tour architecture in a single rack, because of the intelligence in our cable, we've got the ability to sense when the ToR port is failing, our cable makes the decision to switch the data flow to the second port.
So, this is a level of intelligence in the cable that has never been delivered as a DAC never been delivered as an AEC or other optical connection.
And so this is the direction that our business has taken as a result of us owning and really building the capability internally. I can tell you that the second solution that we're developing is also very unique. It's not just a straight cable.
And so we expect that more and more innovation will be requested by our customer base, as they see this is now a category where they can think about systems solutions that they haven't even thought of before. There will be business, especially, in the switching and routing layer that is leading edge 800-gig ports, where you have 800-gig connectors on each end of the cable. That's going to be a robust business as well. But generally, we think that the approach we took is paying off right now, as you see us taking off as a clear leader in the space. But I will say that having competition is always a great thing.
So, we appreciate that multiple people are now investing and that it seems that the market is accepting AEC as really a de facto solution for short connections, meaning three meters and less.
Level up your trading with our powerful tools and real-time insights all in one place.
jablome
1 year ago
SmittyInvestor
2 years ago
Hot Features
Premium
