I know there was prior discussion on the board around Infiniband vs. Ethernet most of which seemed to lean very positively toward Infiniband. I’m no expert. For anyone that may be interested, I did want to post some of Jayshree’s comments during the Q&A portion of the call.
Before I do that, ANET has founds itself into the top spot in my portfolio after cutting my CRWD position in half. I do have a lot of faith in management having followed them for many years.
In terms of the quarter, revenue grew about 16% and EPS grew double that at just over 32%. Arista always has extremely high operating margins and cash yields and this quarter was no different.
Having listened to management for many years, I tend to believe growth will accelerate over the next 24 months. I realize that review is just woeful when it comes to fine details.
With that said, I was hoping to add some comments from Jayshree that are far beyond my ken. I know there are folks on this board who are quite knowledgeable on the technical aspects. For those I might suggest reading the transcripts as there was quite a bit of technical discussion.
Response from the very first question asked on the call:
Sure. Michael, I think as the GPUs get faster and faster, obviously, the dependency on the network for higher throughput is clearly related. And therefore, our timely introduction of these 800-gig products will be required, especially more for Blackwell. In terms of its connection and modularity with NVLink and 72 port, there’s always been a market for what I call scale up, where you’re connecting the GPUs internally in a server and the density of those GPUs connecting in the past has been more PCIe and XL and now NVLink, and there’s a new consortium now for called UAL that’s going to specify that, I believe, eventually, by the way, even there, Ethernet will win.
And so, that density depends more on the AI accelerator and how they choose to connect. As I’ve often said, it’s more a bus technology. So eventually, where Arista plays strongly both on the front end and back end is on the scale out, not on the scale up. So, independent of the modularity, whether it’s a wrap-based design, a chassis, or multiple RU, the ports have to come out Ethernet and those Ethernet ports will connect into scale-out switches from Arista.
More later about Infiniband:
Well, I think there are a lot of market studies that point to today is still largely InfiniBand. You remember me, Ben, saying we were outside looking in just a year ago. So, step one is we’re feeling very gratified that the whole world, even InfiniBand players have acknowledged that we are making Ethernet great again. And so, I expect more and more of that back end to be Ethernet.
One thing I do expect, even though we’re very signed up to the $750 million number – at least $750 million, I should say, next year, is it’s going to become difficult to distinguish the back end from the front end when they all move to Ethernet. For this AI center, as we call it, is going to be a conglomeration of both the front and the back. So, if I were to fast forward three, four years from now, I think the AI center is a supercenter of both the front end and the back end. So, we’ll be able to track it as long as there’s GPUs and strictly training use cases.
But if I were to fast forward, I think there may be many more edge use cases, many more inference use cases and many more small-scale training use cases which will make that distinction difficult to make. Thank you, Ben.
Answer to a question asking if ANET is seeing Spectrum X by NVDA and if so, how ANET is faring.
Yeah. Well, first, I just want to say, when you say competitive environment, it’s complicated with NVIDIA because we really consider them a friend on the GPUs, as well as the mix, so not quite a competitor. But absolutely, we will compete with them on the Spectrum switch. We have not seen the Spectrum except in one customer where it was bundled.
But otherwise, we feel pretty good about our win rate and our success. For a number of reasons, great software, portfolio of products, and architecture has proven performance, visibility features, management capabilities, high availability. And so, I think it’s fair to say that if a customer were bundling with their GPUs, then we wouldn’t see it. If a customer were looking for best of breed, we absolutely see it and win it.
I’d be interested to hear if anyone has thoughts on Jayshree’s claims.
Best reagrds,
A.J.